WO2017107494A1 - Method and device for recognizing badminton racket swinging motion - Google Patents

Method and device for recognizing badminton racket swinging motion Download PDF

Info

Publication number
WO2017107494A1
WO2017107494A1 PCT/CN2016/093071 CN2016093071W WO2017107494A1 WO 2017107494 A1 WO2017107494 A1 WO 2017107494A1 CN 2016093071 W CN2016093071 W CN 2016093071W WO 2017107494 A1 WO2017107494 A1 WO 2017107494A1
Authority
WO
WIPO (PCT)
Prior art keywords
matching
template
preset
badminton
valid data
Prior art date
Application number
PCT/CN2016/093071
Other languages
French (fr)
Chinese (zh)
Inventor
宋志聪
叶景泰
Original Assignee
深圳市酷浪云计算有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市酷浪云计算有限公司 filed Critical 深圳市酷浪云计算有限公司
Publication of WO2017107494A1 publication Critical patent/WO2017107494A1/en

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
    • A63B71/06—Indicating or scoring devices for games or players, or for other sports activities
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B67/00—Sporting games or accessories therefor, not provided for in groups A63B1/00 - A63B65/00
    • A63B67/18—Badminton or similar games with feathered missiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing

Definitions

  • the present invention relates to the field of interactive application technologies, and in particular, to a badminton swing motion recognition method and apparatus.
  • the existing badminton swing motion tracking and recognition is mainly divided into image video based recognition technology and sensor based recognition technology.
  • the former uses a high-speed camera to capture a marker point set on a badminton racket to obtain racket motion information, which is dependent on image and video processing.
  • the sensor-based recognition technology uses sensors to measure the original data such as the motion trajectory space, velocity and rotation angle during the swing process, and identifies the badminton swing motion based on the feature values.
  • image recognition-based recognition technology requires the use of high-speed cameras, which are expensive and inconvenient to carry. They are only suitable for professional training scenarios, and the requirements for professional requirements and processing capabilities of hardware platforms are very high, and the scope of application is limited. .
  • sensor-based recognition technology is not limited in cost and scope of application.
  • information such as acceleration and angular velocity of various badminton swing movements are similar, it is difficult to extract each action type.
  • the eigenvalues result in higher complexity and lower separability, and the resulting misjudgement of the badminton swing motion recognition result is too much.
  • a method for identifying a badminton swing motion comprising:
  • the endpoint detects the original data to obtain a valid data segment
  • the method before the matching the valid data segment and the preset template by the dynamic time warping algorithm, the method further includes:
  • the step of distinguishing the valid data segments by preset feature values includes:
  • Reading a preset feature value the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
  • the action type corresponding to the feature value is a differentiated action type.
  • the step of matching the valid data segment and the preset template by a dynamic time warping algorithm includes:
  • the template preset by the action type template library is matched with the time series by a dynamic time warping algorithm to obtain a matching distance between the time series and any template.
  • a matching result is generated according to the minimum matching distance.
  • the method further includes:
  • test data Collecting a preset number of test data for the action type in the badminton swing, the test data being collected by the sensor on the badminton racket during the swing of the same swing action;
  • a test data is used as a temporary template matching to obtain a matching distance with other test data, and the operation obtains a sum of matching distances between the temporary template and other test data;
  • test data with the smallest sum of matching distances is selected as a preset template of the action type, and saved.
  • a badminton swing action recognition device includes:
  • a raw data obtaining module configured to acquire raw data generated by tracking a badminton racket, wherein the raw data is continuously collected by a sensor on the badminton racket during a swinging process;
  • An endpoint detection module configured to detect, by the endpoint, the valid data segment by using the original data
  • a template identification module configured to match the valid data segment and the preset template by a dynamic time warping algorithm, where the preset template corresponds to an action type of a badminton swing action;
  • the result output module is configured to obtain a corresponding badminton swing motion recognition result according to the matching result.
  • the apparatus further includes:
  • An eigenvalue distinguishing module configured to pre-determine the valid data segment by using preset feature values
  • the determining module is configured to determine whether the valid data segment can be distinguished by the feature value, and if yes, outputting the differentiated action type to the badminton swing action recognition result, and if not, notifying the template identification Module.
  • the feature value distinguishing module includes:
  • a reading unit configured to read a preset feature value, wherein the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
  • Aligning unit configured to compare the feature value and the valid data segment, in the valid data segment and When the feature values match, the action type corresponding to the feature value is a type of action obtained by distinguishing.
  • the template identification module includes:
  • a posture fusion unit configured to perform posture fusion in the valid data segment to obtain a time series of three-axis gravity components
  • a matching unit configured to be matched with the time series by the dynamic time warping algorithm by the template preset in the action type template library, to obtain a matching distance between the time series and any template
  • the distance determining unit is configured to extract a minimum matching distance, determine whether the minimum matching distance is smaller than a threshold, and if yes, generate a matching result according to the minimum matching distance.
  • the apparatus further includes:
  • test data acquisition module configured to collect a preset number of test data for the action type in the badminton swing, wherein the test data is collected by the sensor on the badminton racket during the swing of the same swing action;
  • the test data matching module is configured to use a test data as a temporary template matching to obtain a matching distance between the test data and the other test data, and obtain a matching distance between the temporary template and the other test data. The sum of the matching distances;
  • the distance selection module is configured to select test data with the smallest sum of matching distances as a preset template of the action type, and save.
  • the original data generated by the tracking badminton racket is acquired, and the original data is continuously collected by the sensor on the badminton racket during the swing process, and the endpoint detection is performed in the original data.
  • the effective data segment is extracted, and the effective data segment and each preset template are matched by the dynamic time warping algorithm to obtain a corresponding matching result.
  • the action type of the badminton racket corresponding to the valid data segment can be obtained, and the corresponding type is obtained.
  • the badminton swing motion recognition result depends on the matching of the entire valid data segment with the template in this process.
  • the matching process is through dynamic time.
  • the realization of the regularization algorithm will ensure the accuracy in the matching process.
  • the badminton swing motion recognition will improve the recognition rate and effectively avoid the misjudgment in the badminton swing motion recognition.
  • FIG. 1 is a flow chart of a method for recognizing a badminton swing motion in an embodiment
  • FIG. 2 is a flow chart showing a method for recognizing a badminton swing motion in another embodiment
  • FIG. 3 is a pre-determination of the valid data segments by the preset feature values in FIG. 2;
  • FIG. 4 is a flow chart of a method for matching a valid data segment and a preset template by a dynamic time warping algorithm in FIG. 1;
  • Figure 5 is a flow chart showing a method for recognizing a badminton swing motion in another embodiment
  • FIG. 6 is a schematic structural view of a badminton swing motion recognition device in an embodiment
  • FIG. 7 is a schematic structural view of a badminton swing motion recognition device in another embodiment
  • FIG. 8 is a schematic structural diagram of a feature value distinguishing module in FIG. 7;
  • FIG. 9 is a schematic structural view of a template identification module of FIG. 6;
  • FIG. 10 is a schematic structural view of a method for recognizing a badminton swing motion in another embodiment.
  • the badminton swing motion recognition method is as shown in FIG. 1 and includes:
  • step 110 the original data generated by the tracking badminton racket is acquired, and the original data is continuously collected by the sensor on the badminton racket during the swinging process.
  • the raw data is continuously collected and output by the sensor during the swing process, which embodies the various parameters related to the badminton swing posture, such as the motion trajectory, speed, and angle during the swing.
  • the acquisition of this raw data continues during the swing until the end of the swing.
  • step 130 the endpoint detects the original data to obtain a valid data segment.
  • the original data After obtaining the original data in the swing process, since the original data corresponds to the entire swing process, and the badminton swing action is only a part of the entire swing process, the original data needs to be segmented. , extracting a piece of data corresponding to the effective swing action, that is, extracting the valid data segment.
  • endpoint detection is performed by a double threshold comparison method, and the endpoint of the effective swing is determined by the angular extent of the X-axis on the sensor.
  • a large angular velocity amplitude T n is set to confirm the effective swing motion, which is compared as a threshold, and then set.
  • a slightly smaller angular velocity amplitude T 1 is used as the end point of the cut shot, thereby extracting the valid data segment corresponding to the valid swing.
  • Step 150 Match the valid data segment and the preset template by a dynamic time warping algorithm, and the preset template corresponds to the action type of the badminton swing action.
  • a set of action type template library is preset, and the action type template library stores a plurality of preset templates, each template has a corresponding action type, and the action types corresponding to each template are different.
  • the matching process will be performed with the valid data segment as input.
  • the dynamic data segment is matched with each preset template by a dynamic time warping algorithm to learn, according to the matching result between the valid data segment and each preset template, which template the valid data segment is associated with. Approximation, and the action type corresponding to the template is the corresponding badminton swing action recognition result.
  • the dynamic time warping algorithm is used to match the valid data segments and templates.
  • the dynamic time warping algorithm is used to solve the problem of the length between the valid data segment and the template. If the match between the Euclidean distances is matched by the Euclidean distance, even if the similarity between the valid data segment and a template is high, the Euclidean distance obtained is large, and the dissimilar matching result is obtained. This is because the Euclidean distance is sensitive to changes in the length of time and cannot be accurately matched by the Euclidean distance.
  • the problem of the length and length of the valid data segment and the template can be solved, and then the matching between the valid data segment and the template is performed, and the matching distance is accurately obtained.
  • Step 170 Obtain a corresponding badminton swing motion recognition result according to the matching result.
  • the original data and the valid data segments obtained by the truncation are all in the form of three-axis gravity components.
  • the form of the triaxial gravity component is more obvious for the forehand smash and the forehand picking, and the separability Larger,
  • the method as described above further includes:
  • Step 210 Pre-determine the valid data segments by preset feature values.
  • the feature value is used to distinguish a specific badminton swing action, that is, if the valid data segment matches the feature value, it indicates that the feature value can distinguish the valid data segment, and the corresponding badminton swing action is the feature value.
  • step 230 it is determined whether the valid data can be distinguished by the feature value to obtain the action type. If yes, the process proceeds to step 250. If not, the process returns to step 150.
  • the valid data segment can be distinguished by the preset feature value, it is no longer necessary to match the preset template through the dynamic time warping algorithm, and directly output the differentiated action type.
  • valid data segments cannot be distinguished by the preset feature values, they can only be processed by the matching as described above.
  • step 250 the output type obtained by the output distinction is a badminton swing motion recognition result.
  • the template matching is performed after the differentiation, and the optimization process effectively reduces the number of templates required for the preset, thereby correspondingly reducing the number of matching between the valid data segment and the template, so as to improve the recognition speed.
  • the feature value is pre- Differentiating will eliminate the need to preset the corresponding template for action types such as smashing and picking, and no matching processing is required.
  • Badminton swing action corresponds to a variety of action types, single ball action, such as smashing, picking, flat pumping and other types of action, plus positive and negative hand movements and other non-standard movements, therefore, through The introduction of the eigenvalues will not require the creation of a template corresponding to all the action types, nor the matching of all the action types, thereby greatly reducing the time spent and improving the speed efficiency of the matching.
  • the characteristic value will be in the form of a gravity component, for example, it may be a gravity component corresponding to a maximum speed point when the badminton swings, and the gravity component corresponding to the maximum speed point can distinguish the handball from the hand.
  • the ball in turn, separates the action of the buckle and the pick.
  • step 210 is as shown in FIG. 3, and includes:
  • Step 211 reading a preset feature value, the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type.
  • the feature value storage is performed in advance, that is, any action type that can be distinguished by the feature value has a corresponding feature value.
  • Step 213 Compare the feature value and the valid data segment, and the action type corresponding to the feature value when the valid data segment matches the feature value is the action type that is distinguished.
  • the effective data segment and the feature value are first compared one by one. If the valid data segment matches the feature value corresponding to an action type, the valid data segment is located.
  • the badminton swing action belongs to this type of action, and here, the badminton swing action recognition result can be directly obtained.
  • the step 150 includes:
  • step 151 the pose fusion is performed in the valid data segment to obtain a time series of the three-axis gravity component.
  • the raw data is output by three-axis acceleration and a three-axis gyroscope.
  • Step 153 when the template preset in the action type template library passes the dynamic time and the time series respectively
  • the regularization algorithm performs matching to obtain the matching distance between the time series and any template.
  • the dynamic time warping algorithm is matched with the template preset in the action type template library one by one, and correspondingly, the preset template is also a three-axis gravity component.
  • the preset template is also a three-axis gravity component. The corresponding form.
  • the matching distance between each gravity component is separately calculated, and the sum of the three matching distances obtained is set between the time series and the template. Matching distance.
  • the three-axis gravity components corresponding to the template are: Templet_x, Templet_y, and Templet_z
  • the three-axis gravity components corresponding to the time series are test_x, test_y, and test_z, which are between Templet_x and test_x, between Templet_y and test_y, and The match between Templet_z and test_z is obtained, and the matching distances DTW (Templet_x, test_x), DTW (Templet_y, test_y) and DTW (Templet_z, test_z) are obtained respectively.
  • step 155 the minimum matching distance is extracted, and it is determined whether the minimum matching distance is smaller than the threshold. If yes, the process proceeds to step 157, and if not, the process ends.
  • the matching distance is obtained by matching between the valid data segment and each template. Therefore, among the multiple matching distances obtained by the matching, the matching distance with the smallest value is extracted to obtain the minimum matching distance.
  • a threshold for performing matching distance determination is set in advance, and the threshold value is used to measure whether the degree of similarity between the valid data segment and the closest template is acceptable for the recognition of the badminton swing action, if the judgment is obtained If the minimum matching distance is less than the threshold, it indicates that the matching operation obtains the matching result of the template matching the valid data segment with the minimum matching distance.
  • Step 157 Generate a matching result according to the minimum matching distance.
  • the above method also includes a process of template construction. Specifically, as shown in FIG. 5, the method as described above further includes:
  • Step 310 Collect a preset number of test data for the action type in the badminton swing, the test The test data was collected by the sensor on the badminton racket during the swing of the same swing action.
  • a preset number of test data is respectively collected for the action type of the badminton swing, and the test data is the tester's hand holding the badminton racket, and the sensor outputs according to the currently specified action type.
  • the preset number of test data obtained after completing a preset number of badminton swing actions according to the specified action type will be used to perform template construction for this specified action type.
  • test data is also in the form of a three-axis gravity component.
  • Step 330 For a preset number of test data of the same action type, a test data is used as a temporary template matching to obtain a matching distance with other test data, and the operation obtains a sum of matching distances between the temporary template and other test data.
  • test data There are multiple types of badminton swing action, and for a type of action, a preset number of test data is collected and matched in these test data to obtain the best test data as a template.
  • test data is in the form of three-axis gravity components, namely, V x , V y , and V z .
  • Each test data is used as a temporary module, and is matched with other test data by a dynamic time warping algorithm to obtain corresponding Match the distance and calculate the sum of all matching distances, that is, the total matching distance corresponding to the temporary template.
  • Step 350 Select test data with the smallest matching distance as the preset template of the action type, and save it.
  • test data with the smallest total matching distance will be the most similar to the other test data corresponding to the action type, which is the optimal data used as a template.
  • the matrix mesh is constructed based on the valid data segment and the template, so as to perform the correspondence between the valid data segment and the template, and the obtained valid data segment and The mapping between templates is matched by the corresponding operation.
  • this will also be optimized to limit the internal search matching path of the algorithm to improve the matching speed and match power.
  • a lookup table of the length is established to save the length in the table and the lower boundary of the corresponding search range. For each match, you only need to derive the upper bound of the limit range based on the slope.
  • the second data extraction may be performed therein to reduce the length of the matching sequence, for example, the odd number is discarded and the even number is discarded to reduce the length of the matching sequence.
  • greatly reducing the matching time in this case, almost half of the matching time will be reduced compared to before, further improving the recognition efficiency.
  • a badminton swing motion recognition device is further provided, as shown in FIG. 6, including a raw data acquisition module 410, an endpoint detection module 430, a template recognition module 450, and a result output module 470, wherein:
  • the original data obtaining module 410 is configured to acquire raw data generated by the tracking badminton racket, and the original data is continuously collected by the sensor on the badminton racket during the swinging process.
  • the endpoint detection module 430 is configured to detect the original data by the endpoint to obtain a valid data segment.
  • the template identification module 450 is configured to validate the data segment and the preset template by using a dynamic time warping algorithm, and the preset template corresponds to the action type of the badminton swing action.
  • the result output module 470 is configured to obtain a corresponding badminton swing motion recognition result according to the matching result.
  • the apparatus as described above further includes a feature value distinguishing module 510 and a determining module 530, wherein:
  • the feature value distinguishing module 510 is configured to pre-determine the valid data segments by preset feature values.
  • the determining module 530 is configured to determine whether the valid data segment can be distinguished by the feature value, and if yes, output the differentiated action type to the badminton swing motion recognition result, and if not, notify the template recognition module 450.
  • the feature value distinguishing module 510 includes a reading unit 511 and a comparing unit 513, where:
  • the reading unit 511 is configured to read a preset feature value, and the preset feature value is used to identify a badminton swing action, which corresponds to a partial action type.
  • the comparison unit 513 is configured to compare the feature value and the valid data segment in the valid data segment and the feature value
  • the action type corresponding to the feature value when matching is the action type that distinguishes the obtained action.
  • the template identification module 450 includes a gesture fusion unit 451, a matching unit 453, and a distance determination unit 455, wherein:
  • the attitude fusion unit 451 is configured to perform posture fusion in the valid data segment to obtain a time series of the three-axis gravity component.
  • the matching unit 453 is configured to match the time series of the template preset in the action type template library with the time series by a dynamic time warping algorithm to obtain a matching distance between the time series and any template.
  • the distance determining unit 455 is configured to extract a minimum matching distance, determine whether the minimum matching distance is smaller than a threshold, and if yes, generate a matching result according to the minimum matching distance.
  • the apparatus as described above, as shown in FIG. 10, further includes a test data acquisition module 510, a test data matching module 530, and a distance selection module 550, wherein:
  • the test data collection module 510 is configured to collect a preset number of test data for the action type in the badminton swing, and the test data is collected by the sensor on the badminton racket during the swing of the same swing action.
  • the test data matching module 530 is configured to use a test data as a temporary template matching to obtain a matching distance between the test data and the other test data, and obtain a matching distance between the temporary template and the other test data. Match the sum of the distances.
  • the distance selection module 550 is configured to select the test data with the smallest matching distance as the preset template of the action type, and save it.
  • a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
  • the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.

Landscapes

  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Image Analysis (AREA)

Abstract

A method for recognizing a badminton racket swinging motion, comprising: obtaining original data generated by tracking a badminton racket, wherein the original data is continuously collected by a sensor on the badminton racket in a swinging process; performing endpoint detection on the original data to obtain a valid data segment; matching the valid data segment with a preset template by using a dynamic time warping algorithm, wherein the preset template corresponds to a motion type of the badminton racket swinging motion; and obtaining a recognition result of the badminton racket swinging motion according to a matching result. Also provided is a device for recognizing a badminton racket swinging motion.

Description

羽毛球挥拍动作识别方法和装置Badminton swing action recognition method and device 技术领域Technical field
本发明涉及交互应用技术领域，特别涉及一种羽毛球挥拍动作识别方法和装置。The present invention relates to the field of interactive application technologies, and in particular, to a badminton swing motion recognition method and apparatus.
背景技术Background technique
羽毛球运动中，随着其周边设备的不断扩充，跟踪并识别羽毛球挥拍过程已经成为羽毛球运动中的重要交互式应用。In badminton, with the continuous expansion of its peripheral equipment, tracking and identifying the badminton swing process has become an important interactive application in badminton.
现有的羽毛球挥拍动作跟踪和识别主要分为基于图像视频的识别技术和基于传感器的识别技术。前者是利用高速摄像机拍摄设置在羽毛球拍上的标志点，从而获取球拍运动信息，此过程依赖于图像和视频处理。The existing badminton swing motion tracking and recognition is mainly divided into image video based recognition technology and sensor based recognition technology. The former uses a high-speed camera to capture a marker point set on a badminton racket to obtain racket motion information, which is dependent on image and video processing.
而基于传感器的识别技术则是要利用传感器来测量出挥拍过程中的运动轨迹空间、速度和旋转角度等原始数据，对原始数据根据特征值识别羽毛球挥拍动作。The sensor-based recognition technology uses sensors to measure the original data such as the motion trajectory space, velocity and rotation angle during the swing process, and identifies the badminton swing motion based on the feature values.
然而，基于图像识别的识别技术需要使用高速摄像机，而高速摄像机价格昂贵，携带不便，只适用于专业训练场景下，并且对硬件平台的专业性要求和处理能力均要求非常高，运用范围受局限。However, image recognition-based recognition technology requires the use of high-speed cameras, which are expensive and inconvenient to carry. They are only suitable for professional training scenarios, and the requirements for professional requirements and processing capabilities of hardware platforms are very high, and the scope of application is limited. .
另一方面的，基于传感器的识别技术虽然成本和运用范围均不会受到局限，但是由于各种羽毛球挥拍动作之间作为特征值的加速度和角速度等信息都比较相似，比较难提取各动作类型的特征值，因此造成较高的复杂性和较低的可分性，进而所得到的羽毛球挥拍动作识别结果的误判过多。On the other hand, sensor-based recognition technology is not limited in cost and scope of application. However, because the information such as acceleration and angular velocity of various badminton swing movements are similar, it is difficult to extract each action type. The eigenvalues result in higher complexity and lower separability, and the resulting misjudgement of the badminton swing motion recognition result is too much.
发明内容Summary of the invention
基于此，有必要提供一种羽毛球挥拍动作识别方法，所述方法能够提高识别率，避免出现羽毛球挥拍动作识别中的误判。Based on this, it is necessary to provide a badminton swing motion recognition method, which can improve the recognition rate and avoid misjudgment in the badminton swing motion recognition.
另外，还有必要提供一种羽毛球挥拍动作识别装置，所述装置能够提高识别率，避免出现羽毛球挥拍动作识别中的误判。In addition, it is also necessary to provide a badminton swing motion recognition device which can improve the recognition rate and avoid misjudgment in the badminton swing motion recognition.
为解决上述技术问题，将采用如下技术方案： In order to solve the above technical problems, the following technical solutions will be adopted:
一种羽毛球挥拍动作识别方法，包括：A method for identifying a badminton swing motion, comprising:
获取跟踪羽毛球拍产生的原始数据，所述原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到；Obtaining raw data generated by the tracking badminton racket, the raw data being continuously collected by the sensor on the badminton racket during the swinging process;
端点检测所述原始数据得到有效数据段；The endpoint detects the original data to obtain a valid data segment;
通过动态时间规整算法匹配所述有效数据段和预置的模板，所述预置的模板对应于羽毛球挥拍动作的动作类型；Matching the valid data segment and the preset template by a dynamic time warping algorithm, where the preset template corresponds to an action type of a badminton swing action;
根据匹配结果得到对应的羽毛球挥拍动作识别结果。According to the matching result, the corresponding badminton swing motion recognition result is obtained.
在其中一个实施例中，所述通过动态时间规整算法匹配所述有效数据段和预置的模板之前，所述方法还包括：In one embodiment, before the matching the valid data segment and the preset template by the dynamic time warping algorithm, the method further includes:
通过预置的特征值对所述有效数据段预先进行区分；Predetermining the valid data segments by preset feature values;
判断所述有效数据段是否能够通过特征值区分得到动作类型，若为是，则输出区分得到的动作类型为羽毛球挥拍动作识别结果，若为否，则Determining whether the valid data segment can be distinguished by the feature value, and if yes, outputting the differentiated action type is a badminton swing action recognition result, if not, then
进入所述通过动态时间规整算法匹配所述有效数据段和预置的模板的步骤。Entering the step of matching the valid data segment and the preset template by a dynamic time warping algorithm.
在其中一个实施例中，所述通过预置的特征值对所述有效数据段预先进行区分的步骤包括：In one embodiment, the step of distinguishing the valid data segments by preset feature values includes:
读取预置的特征值，所述预置的特征值用于识别部分羽毛球挥拍动作，对应于部分动作类型；Reading a preset feature value, the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
比对所述特征值和有效数据段，在所述有效数据段与所述特征值相符时所述特征值对应的动作类型为区分得到的动作类型。Comparing the feature value and the valid data segment, when the valid data segment matches the feature value, the action type corresponding to the feature value is a differentiated action type.
在其中一个实施例中，所述通过动态时间规整算法匹配所述有效数据段和预置的模板的步骤包括：In one embodiment, the step of matching the valid data segment and the preset template by a dynamic time warping algorithm includes:
在所述有效数据段中进行姿态融合得到三轴重力分量的时间序列；Performing a pose fusion in the valid data segment to obtain a time series of three-axis gravity components;
由动作类型模板库中预置的模板分别与所述时间序列通过动态时间规整算法进行匹配，得到所述时间序列与任一模板之间的匹配距离；The template preset by the action type template library is matched with the time series by a dynamic time warping algorithm to obtain a matching distance between the time series and any template.
提取最小匹配距离，判断所述最小匹配距离是否小于阈值，若为是，则 Extracting a minimum matching distance, determining whether the minimum matching distance is less than a threshold, and if yes,
根据所述最小匹配距离生成匹配结果。A matching result is generated according to the minimum matching distance.
在其中一个实施例中，所述方法还包括：In one embodiment, the method further includes:
为羽毛球挥拍中的动作类型采集预设数量的测试数据，所述测试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到；Collecting a preset number of test data for the action type in the badminton swing, the test data being collected by the sensor on the badminton racket during the swing of the same swing action;
对同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到所述临时模板与其它测试数据之间的匹配距离之和；For a preset number of test data of the same action type, a test data is used as a temporary template matching to obtain a matching distance with other test data, and the operation obtains a sum of matching distances between the temporary template and other test data;
选取匹配距离之和最小的测试数据作为所述动作类型的预置模板，并保存。The test data with the smallest sum of matching distances is selected as a preset template of the action type, and saved.
一种羽毛球挥拍动作识别装置，包括：A badminton swing action recognition device includes:
原始数据获取模块，用于获取跟踪羽毛球拍产生的原始数据，所述原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到；a raw data obtaining module, configured to acquire raw data generated by tracking a badminton racket, wherein the raw data is continuously collected by a sensor on the badminton racket during a swinging process;
端点检测模块，用于端点检测所述原始数据得到有效数据段；An endpoint detection module, configured to detect, by the endpoint, the valid data segment by using the original data;
模板识别模块，用于通过动态时间规整算法匹配所述有效数据段和预置的模板，所述预置的模板对应于羽毛球挥拍动作的动作类型；a template identification module, configured to match the valid data segment and the preset template by a dynamic time warping algorithm, where the preset template corresponds to an action type of a badminton swing action;
结果输出模块，用于根据匹配结果得到对应的羽毛球挥拍动作识别结果。The result output module is configured to obtain a corresponding badminton swing motion recognition result according to the matching result.
在其中一个实施例中，所述装置还包括：In one embodiment, the apparatus further includes:
特征值区分模块，用于通过预置的特征值对所述有效数据段预先进行区分；An eigenvalue distinguishing module, configured to pre-determine the valid data segment by using preset feature values;
判断模块，用于判断所述有效数据段是否能够通过特征值区分得到动作类型，若为是，则输出区分得到的动作类型为羽毛球挥拍动作识别结果，若为否，则通知所述模板识别模块。The determining module is configured to determine whether the valid data segment can be distinguished by the feature value, and if yes, outputting the differentiated action type to the badminton swing action recognition result, and if not, notifying the template identification Module.
在其中一个实施例中，所述特征值区分模块包括：In one embodiment, the feature value distinguishing module includes:
读取单元，用于读取预置的特征值，该预置的特征值用于识别部分羽毛球挥拍动作，对应于部分动作类型；a reading unit, configured to read a preset feature value, wherein the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
比对单元，用于比对所述特征值和有效数据段，在所述有效数据段与所 述特征值相符时所述特征值对应的动作类型为区分得到的动作类型。Aligning unit, configured to compare the feature value and the valid data segment, in the valid data segment and When the feature values match, the action type corresponding to the feature value is a type of action obtained by distinguishing.
在其中一个实施例中，所述模板识别模块包括：In one embodiment, the template identification module includes:
姿态融合单元，用于在所述有效数据段中进行姿态融合得到三轴重力分量的时间序列；a posture fusion unit, configured to perform posture fusion in the valid data segment to obtain a time series of three-axis gravity components;
匹配单元，用于由动作类型模板库中预置的模板分别与所述时间序列通过动态时间规整算法进行匹配，得到所述时间序列与任一模板之间的匹配距离；a matching unit, configured to be matched with the time series by the dynamic time warping algorithm by the template preset in the action type template library, to obtain a matching distance between the time series and any template;
距离判断单元，用于提取最小匹配距离，判断所述最小匹配距离是否小于阈值，若为是，则根据所述最小匹配距离生成匹配结果。The distance determining unit is configured to extract a minimum matching distance, determine whether the minimum matching distance is smaller than a threshold, and if yes, generate a matching result according to the minimum matching distance.
在其中一个实施例中，所述装置还包括：In one embodiment, the apparatus further includes:
测试数据采集模块，用于为羽毛球挥拍中的动作类型采集预设数量的测试数据，所述测试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到；a test data acquisition module, configured to collect a preset number of test data for the action type in the badminton swing, wherein the test data is collected by the sensor on the badminton racket during the swing of the same swing action;
测试数据匹配模块，用于对同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到所述临时模板与其它测试数据之间的匹配距离之和；The test data matching module is configured to use a test data as a temporary template matching to obtain a matching distance between the test data and the other test data, and obtain a matching distance between the temporary template and the other test data. The sum of the matching distances;
距离选取模块，用于选取匹配距离之和最小的测试数据作为所述动作类型的预置模板，并保存。The distance selection module is configured to select test data with the smallest sum of matching distances as a preset template of the action type, and save.
由上述技术方案可知，羽毛球挥拍过程中，将获取跟踪羽毛球拍产生的原始数据，该原始数据是由羽毛球拍上的传感器在挥拍过程持续采集得到的，在原始数据中进行端点检测，以提取有效数据段，通过动态时间规整算法对有效数据段和每一预置的模板进行匹配得到相应的匹配结果，根据匹配结果即可获知有效数据段所对应的羽毛球拍的动作类型，进而得到对应的羽毛球挥拍动作识别结果，在此过程中依赖于整个有效数据段与模板的匹配，因此，并不需要通过单一的特征值来实现羽毛球挥拍动作的识别，并且此匹配过程是通过动态时间规整算法实现的，将得以保证了匹配过程中的精准性，在此基础上所得到的羽毛球挥拍动作识别将提高了识别率，有效避免了羽毛球挥拍动作识别中的误判。 It can be known from the above technical solution that during the badminton swing process, the original data generated by the tracking badminton racket is acquired, and the original data is continuously collected by the sensor on the badminton racket during the swing process, and the endpoint detection is performed in the original data. The effective data segment is extracted, and the effective data segment and each preset template are matched by the dynamic time warping algorithm to obtain a corresponding matching result. According to the matching result, the action type of the badminton racket corresponding to the valid data segment can be obtained, and the corresponding type is obtained. The badminton swing motion recognition result depends on the matching of the entire valid data segment with the template in this process. Therefore, it is not necessary to realize the recognition of the badminton swing motion by a single feature value, and the matching process is through dynamic time. The realization of the regularization algorithm will ensure the accuracy in the matching process. On the basis of this, the badminton swing motion recognition will improve the recognition rate and effectively avoid the misjudgment in the badminton swing motion recognition.
附图说明DRAWINGS
图1是一个实施例中羽毛球挥拍动作识别方法的流程图；1 is a flow chart of a method for recognizing a badminton swing motion in an embodiment;
图2是另一个实施例中羽毛球挥拍动作识别方法的流程图；2 is a flow chart showing a method for recognizing a badminton swing motion in another embodiment;
图3是图2中通过预置的特征值对有效数据段预先进行区分；3 is a pre-determination of the valid data segments by the preset feature values in FIG. 2;
图4是图1中通过动态时间规整算法匹配有效数据段和预置的模板的方法流程图；4 is a flow chart of a method for matching a valid data segment and a preset template by a dynamic time warping algorithm in FIG. 1;
图5是另一个实施例中羽毛球挥拍动作识别方法的流程图；Figure 5 is a flow chart showing a method for recognizing a badminton swing motion in another embodiment;
图6是一个实施例中羽毛球挥拍动作识别装置的结构示意图；6 is a schematic structural view of a badminton swing motion recognition device in an embodiment;
图7是另一个实施例中羽毛球挥拍动作识别装置的结构示意图；7 is a schematic structural view of a badminton swing motion recognition device in another embodiment;
图8是图7中特征值区分模块的结构示意图；8 is a schematic structural diagram of a feature value distinguishing module in FIG. 7;
图9是图6中模板识别模块的结构示意图；9 is a schematic structural view of a template identification module of FIG. 6;
图10是另一个实施例中羽毛球挥拍动作识别方法的结构示意图。FIG. 10 is a schematic structural view of a method for recognizing a badminton swing motion in another embodiment.
具体实施方式detailed description
体现本发明特征与优点的典型实施方式将在以下的说明中详细叙述。应理解的是本发明能够在不同的实施方式上具有各种的变化，其皆不脱离本发明的范围，且其中的说明及图示在本质上是当作说明之用，而非用以限制本发明。Exemplary embodiments embodying the features and advantages of the present invention will be described in detail in the following description. It is to be understood that the invention is capable of various modifications in the various embodiments and this invention.
如前所述，为实现各种场景下羽毛球挥拍动作识别的应用，大都采用基于传感器的识别技术来实现，也将是说，将特定的传感器置于羽毛球拍上，以对羽毛球拍在空间中的运动进行跟踪，并将由此所得到的原始数据进行特征提取，分类规划等处理方可得到相应的羽毛球挥拍动作识别结果。As mentioned above, in order to realize the application of badminton swing motion recognition in various scenarios, most of them are implemented by sensor-based recognition technology. It will also say that a specific sensor is placed on the badminton racket to take the badminton racket in space. The motion in the middle is tracked, and the original data obtained by the method is extracted, and the classification and planning processing can obtain the corresponding badminton swing motion recognition result.
但是，此识别处理过程由于是基于特征值所得到的，因此误判率较高，特别是对于不同的挥拍者而言，由于不同挥拍者挥拍动作习惯的差异性，导致了即便同一动作类型，所对应的特征值也是有非常大的差异，因此，在现有的识别处理基础上亟待解决识别率低下的缺陷，以保证羽毛球挥拍动作识别在各种场景下的应用。However, since this recognition process is based on the feature value, the false positive rate is high, especially for different swingers, because of the difference in swinging action habits of different swingers, even if the same The type of action, the corresponding feature value is also very different. Therefore, on the basis of the existing recognition processing, it is urgent to solve the defect of low recognition rate to ensure the application of badminton swing motion recognition in various scenarios.
由此，特提出了一种羽毛球挥拍动作识别方法。在一个实施例中，具体的，该羽毛球挥拍动作识别方法如图1所示，包括：Therefore, a badminton swing motion recognition method is proposed. In one embodiment, specifically, the badminton swing motion recognition method is as shown in FIG. 1 and includes:
步骤110，获取跟踪羽毛球拍产生的原始数据，该原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到。 In step 110, the original data generated by the tracking badminton racket is acquired, and the original data is continuously collected by the sensor on the badminton racket during the swinging process.
原始数据是传感器在挥拍过程中持续采集并输出的，其体现了挥拍过程中的运动轨迹、速度、角度等各种与羽毛球挥拍姿态相关的参数。该原始数据的采集在挥拍过程中持续进行，直至挥拍结束。The raw data is continuously collected and output by the sensor during the swing process, which embodies the various parameters related to the badminton swing posture, such as the motion trajectory, speed, and angle during the swing. The acquisition of this raw data continues during the swing until the end of the swing.
步骤130，端点检测原始数据得到有效数据段。In step 130, the endpoint detects the original data to obtain a valid data segment.
在获取得到挥拍过程中的原始数据之后，由于此原始数据是与整个挥拍过程相对应的，而羽毛球挥拍动作仅仅是整个挥拍过程中的一部分，因此，需要对原始数据进行分割处理，提取出有效挥拍动作所对应的一段数据，即进行有效数据段的提取。After obtaining the original data in the swing process, since the original data corresponds to the entire swing process, and the badminton swing action is only a part of the entire swing process, the original data needs to be segmented. , extracting a piece of data corresponding to the effective swing action, that is, extracting the valid data segment.
此提取过程将通过端点检测来实现。具体的，将通过基于双门限比较法来进行端点检测，通过传感器上X轴的角度度幅值来确定有效挥拍的端点。在优选的实施例中，在此通过基于双门限比较法来进行的端点检测中，设置一较大角速度幅值Tn来确认有效的挥拍动作，以此作为阈值来比较，然后又设定一个稍微较小的角速度幅值T1作为截拍的终止点，由此便提取得到有效挥拍所对应的有效数据段。This extraction process will be implemented by endpoint detection. Specifically, endpoint detection is performed by a double threshold comparison method, and the endpoint of the effective swing is determined by the angular extent of the X-axis on the sensor. In a preferred embodiment, in the endpoint detection by the double threshold comparison method, a large angular velocity amplitude T n is set to confirm the effective swing motion, which is compared as a threshold, and then set. A slightly smaller angular velocity amplitude T 1 is used as the end point of the cut shot, thereby extracting the valid data segment corresponding to the valid swing.
原始数据中有效数据段的截取将是每一有效挥拍动作识别的关键，因此通过对原始数据进行端点检测将极大地保障了后续最终得到的羽毛球挥拍动作识别结果的精准性。The interception of the valid data segments in the original data will be the key to the identification of each valid swing motion. Therefore, the endpoint detection of the original data will greatly ensure the accuracy of the subsequent badminton swing recognition results.
步骤150，通过动态时间规整算法匹配有效数据段和预置的模板，预置的模板对应于羽毛球挥拍动作的动作类型。Step 150: Match the valid data segment and the preset template by a dynamic time warping algorithm, and the preset template corresponds to the action type of the badminton swing action.
预置了一套动作类型模板库，该动作类型模板库中存储了若干个预置的模板，每一模板均有对应的动作类型，并且各模板之间所对应的动作类型各不相同。A set of action type template library is preset, and the action type template library stores a plurality of preset templates, each template has a corresponding action type, and the action types corresponding to each template are different.
在此，将以有效数据段为输入进行匹配处理。具体的，通过动态时间规整算法将有效数据段与预置的每一模板进行匹配，以根据有效数据段与预置的每一模板之间的匹配结果来获知该有效数据段与哪一模板相近似，而该模板所对应的动作类型即为对应的羽毛球挥拍动作识别结果。Here, the matching process will be performed with the valid data segment as input. Specifically, the dynamic data segment is matched with each preset template by a dynamic time warping algorithm to learn, according to the matching result between the valid data segment and each preset template, which template the valid data segment is associated with. Approximation, and the action type corresponding to the template is the corresponding badminton swing action recognition result.
进一步的，由于每一次所得到的有效数据段的长度是不一样的，并且有效数据段与模板之间的长度也是各不相同的，换而言之，即时间长度各不相同，因此，需要采用动态时间规整算法进行有效数据段和模板之间的匹配。动态时间规整算法用以解决有效数据段和模板之间长度不等的问题，相比通 过欧式距离所进行的匹配，如果通过欧式距离对此进行匹配，那么即便有效数据段与一模板之间相似度很高，但所求出的欧式距离却很大，进而得到不相似的匹配结果，这是由于欧式距离对时间长度的变化很敏感，无法通过欧式距离来实现准确匹配。Further, since the length of the valid data segment obtained each time is different, and the length between the valid data segment and the template is also different, in other words, the lengths of time are different, therefore, it is required The dynamic time warping algorithm is used to match the valid data segments and templates. The dynamic time warping algorithm is used to solve the problem of the length between the valid data segment and the template. If the match between the Euclidean distances is matched by the Euclidean distance, even if the similarity between the valid data segment and a template is high, the Euclidean distance obtained is large, and the dissimilar matching result is obtained. This is because the Euclidean distance is sensitive to changes in the length of time and cannot be accurately matched by the Euclidean distance.
而对于动态时间规整算法则能够解决有效数据段和模板之间长短不一的问题，进而适用于进行有效数据段和模板之间的匹配，并准确得到相应的匹配距离，匹配距离越小，则有效数据段与模板越相似。For the dynamic time warping algorithm, the problem of the length and length of the valid data segment and the template can be solved, and then the matching between the valid data segment and the template is performed, and the matching distance is accurately obtained. The smaller the matching distance is, the smaller the matching distance is. The more similar a valid data segment is to the template.
步骤170，根据匹配结果得到对应的羽毛球挥拍动作识别结果。Step 170: Obtain a corresponding badminton swing motion recognition result according to the matching result.
通过如上所述的过程，在端点检测和动态时间规整算法的配合下，实现了任意挥拍过程中动作的精准识别，在此基础上所进行的羽毛球挥拍动作识别能够应用于各种场景，并且由于复杂性不高，且成本较低，因此，易于进行推广。Through the process as described above, with the cooperation of the endpoint detection and the dynamic time warping algorithm, the accurate recognition of the motion during the arbitrary swing process is realized, and the badminton swing motion recognition performed on this basis can be applied to various scenes. And because of the low complexity and low cost, it is easy to promote.
在如上所述的过程中，原始数据和截取得到的有效数据段均是三轴重力分量的形式，三轴重力分量的形式对于正手扣杀和正手挑球而言存在较为明显，可分性较大，In the process as described above, the original data and the valid data segments obtained by the truncation are all in the form of three-axis gravity components. The form of the triaxial gravity component is more obvious for the forehand smash and the forehand picking, and the separability Larger,
在一个实施例中，如图2所示，该步骤150之前，如上所述的方法还包括：In an embodiment, as shown in FIG. 2, before the step 150, the method as described above further includes:
步骤210，通过预置的特征值对有效数据段预先进行区分。Step 210: Pre-determine the valid data segments by preset feature values.
特征值用于区分特定的羽毛球挥拍动作，也就是说，如若有效数据段与特征值相符，则说明此特征值能够区分这一有效数据段，所对应的羽毛球挥拍动作就是特征值所能够识别的动作类型。The feature value is used to distinguish a specific badminton swing action, that is, if the valid data segment matches the feature value, it indicates that the feature value can distinguish the valid data segment, and the corresponding badminton swing action is the feature value. The type of action identified.
步骤230，判断有效数据是否能够通过特征值区分得到动作类型，若为是，则进入步骤250，若为否，则返回步骤150。In step 230, it is determined whether the valid data can be distinguished by the feature value to obtain the action type. If yes, the process proceeds to step 250. If not, the process returns to step 150.
如果通过预置的特征值能够对有效数据段进行区分，则不再需要与预置的模板通过动态时间规整算法进行匹配，而直接输出所区分得到的动作类型即可。If the valid data segment can be distinguished by the preset feature value, it is no longer necessary to match the preset template through the dynamic time warping algorithm, and directly output the differentiated action type.
如果通过预置的特征值不能够对有效数据段进行区分，则只能通过如前所述的匹配进行处理。If the valid data segments cannot be distinguished by the preset feature values, they can only be processed by the matching as described above.
步骤250，输出区分得到的动作类型为羽毛球挥拍动作识别结果。In step 250, the output type obtained by the output distinction is a badminton swing motion recognition result.
在此过程中，通过引入特征值对有效数据段预先进行区分，在确定无法 区分后才进行模板匹配，此优化过程将有效地减少了所需要预置的模板数量，进而相应减少了有效数据段与模板之间进行匹配的次数，以提高识别速度。In this process, by distinguishing the valid data segments by introducing feature values, it is determined that The template matching is performed after the differentiation, and the optimization process effectively reduces the number of templates required for the preset, thereby correspondingly reducing the number of matching between the valid data segment and the template, so as to improve the recognition speed.
实际中，尽管寻找将羽毛球挥拍动作区分开来的特征值较难，但是对于一些动作，如扣杀和挑球等动作类型，还是能够通过特征值来区分识别的，因此，通过特征值预先进行区分，将使得不需要为扣杀和挑球等动作类型预置相应的模板，也不需要进行匹配处理。In practice, although it is difficult to find the feature value that distinguishes the badminton swing action, for some actions, such as smashing and picking the ball, the type of action can be distinguished by the feature value. Therefore, the feature value is pre- Differentiating will eliminate the need to preset the corresponding template for action types such as smashing and picking, and no matching processing is required.
羽毛球挥拍动作所对应的动作类型有多种，单是大球动作，如扣杀、挑球、平抽等动作类型，并且加上正反手动作以及其它的一些不规范动作，因此，通过特征值的引入，将不需要对于所有的动作类型都建立与之对应的模板，也不需要对所有的动作类型都进行匹配，进而极大地减少了耗费的时间，提高了匹配的速度效率。Badminton swing action corresponds to a variety of action types, single ball action, such as smashing, picking, flat pumping and other types of action, plus positive and negative hand movements and other non-standard movements, therefore, through The introduction of the eigenvalues will not require the creation of a template corresponding to all the action types, nor the matching of all the action types, thereby greatly reducing the time spent and improving the speed efficiency of the matching.
在优选的实施例中，该特征值将是重力分量的形式，例如，其可为羽毛球挥拍时最大速度点所对应的重力分量，此最大速度点所对应的重力分量能够区分上手球和下手球，进而将扣和挑的动作区分开。In a preferred embodiment, the characteristic value will be in the form of a gravity component, for example, it may be a gravity component corresponding to a maximum speed point when the badminton swings, and the gravity component corresponding to the maximum speed point can distinguish the handball from the hand. The ball, in turn, separates the action of the buckle and the pick.
进一步的，在本实施例中，该步骤210如图3所示，包括：Further, in this embodiment, the step 210 is as shown in FIG. 3, and includes:
步骤211，读取预置的特征值，该预置的特征值用于识别部分羽毛球挥拍动作，对应于部分动作类型。Step 211, reading a preset feature value, the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type.
预先进行了特征值存储，也就是说，可通过特征值区分的部分动作类型中，任一动作类型均有对应的特征值。The feature value storage is performed in advance, that is, any action type that can be distinguished by the feature value has a corresponding feature value.
步骤213，比对特征值和有效数据段，在有效数据段与特征值相符时特征值对应的动作类型为区分得到的动作类型。Step 213: Compare the feature value and the valid data segment, and the action type corresponding to the feature value when the valid data segment matches the feature value is the action type that is distinguished.
在进行端点检测以提取得到有效数据段之后，将首先进行有效数据段与特征值的逐一比对，如果该有效数据段与某一动作类型所对应的特征值相符，则说明此有效数据段所在的羽毛球挥拍动作是归属于这一动作类型的，在此，可直接得到羽毛球挥拍动作识别结果。After the endpoint detection is performed to extract the valid data segment, the effective data segment and the feature value are first compared one by one. If the valid data segment matches the feature value corresponding to an action type, the valid data segment is located. The badminton swing action belongs to this type of action, and here, the badminton swing action recognition result can be directly obtained.
在一个实施例中，如图4所示，该步骤150包括：In one embodiment, as shown in FIG. 4, the step 150 includes:
步骤151，在有效数据段中进行姿态融合得到三轴重力分量的时间序列。In step 151, the pose fusion is performed in the valid data segment to obtain a time series of the three-axis gravity component.
对有效数据进行姿态融合处理后得到三轴重力分量所分别对应的时间序列。其中，该原始数据是三轴加速度和三轴陀螺仪所输出的。After performing the pose fusion processing on the valid data, a time series corresponding to the three-axis gravity components is obtained. The raw data is output by three-axis acceleration and a three-axis gyroscope.
步骤153，由动作类型模板库中预置的模板分别与时间序列通过动态时 间规整算法进行匹配，得到时间序列与任一模板之间的匹配距离。 Step 153, when the template preset in the action type template library passes the dynamic time and the time series respectively The regularization algorithm performs matching to obtain the matching distance between the time series and any template.
对于任一需要进行识别的有效数据段，将通过动态时间规整算法与动作类型模板库中预置的模板逐一进行匹配，其中，与之相对应的，该预置的模板也为三轴重力分量所对应的形式。For any valid data segment that needs to be identified, the dynamic time warping algorithm is matched with the template preset in the action type template library one by one, and correspondingly, the preset template is also a three-axis gravity component. The corresponding form.
具体的，按照三轴重力分量，在时间序列和模板之间，将分别运算每一重力分量之间的匹配距离，并将所运算得到的三个匹配距离之和置为时间序列与模板之间的匹配距离。Specifically, according to the three-axis gravity component, between the time series and the template, the matching distance between each gravity component is separately calculated, and the sum of the three matching distances obtained is set between the time series and the template. Matching distance.
详细运算过程如下所述：The detailed operation process is as follows:
模板所对应的三轴重力分量分别为：Templet_x、Templet_y和Templet_z，时间序列所对应的三轴重力分量为test_x、test_y和test_z，对此分别在Templet_x和test_x之间、Templet_y和test_y之间，以及Templet_z和test_z之间进行匹配，分别得到匹配距离DTW(Templet_x，test_x)、DTW(Templet_y，test_y)和DTW(Templet_z，test_z)。The three-axis gravity components corresponding to the template are: Templet_x, Templet_y, and Templet_z, and the three-axis gravity components corresponding to the time series are test_x, test_y, and test_z, which are between Templet_x and test_x, between Templet_y and test_y, and The match between Templet_z and test_z is obtained, and the matching distances DTW (Templet_x, test_x), DTW (Templet_y, test_y) and DTW (Templet_z, test_z) are obtained respectively.
此时，有效数据段和模板之间的匹配距离Distance＝DTW(Templet_x，test_x)+DTW(Templet_y，test_y)+DTW(Templet_z，test_z)。At this time, the matching distance between the valid data segment and the template is Distance=DTW(Templet_x, test_x)+DTW(Templet_y, test_y)+DTW(Templet_z, test_z).
步骤155，提取最小匹配距离，判断最小匹配距离是否小于阈值，若为是，则进入步骤157，若为否，则结束。In step 155, the minimum matching distance is extracted, and it is determined whether the minimum matching distance is smaller than the threshold. If yes, the process proceeds to step 157, and if not, the process ends.
通过前述过程，有效数据段与每一模板之间均通过匹配得到相应的匹配距离，因此，匹配得到的多个匹配距离中，将提取其中数值最小的匹配距离，以得到最小匹配距离。Through the foregoing process, the matching distance is obtained by matching between the valid data segment and each template. Therefore, among the multiple matching distances obtained by the matching, the matching distance with the smallest value is extracted to obtain the minimum matching distance.
预先设置了用于进行匹配距离判断的阈值，该阈值将用于衡量有效数据段与最相近的模板之间的相似程度对于羽毛球挥拍动作的识别而言，是否是可接受的，如果判断得到最小匹配距离小于阈值，则说明匹配运行得到最小匹配距离的模板与有效数据段匹配的匹配结果。A threshold for performing matching distance determination is set in advance, and the threshold value is used to measure whether the degree of similarity between the valid data segment and the closest template is acceptable for the recognition of the badminton swing action, if the judgment is obtained If the minimum matching distance is less than the threshold, it indicates that the matching operation obtains the matching result of the template matching the valid data segment with the minimum matching distance.
如果判断得到最小匹配距离仍然大于阈值，则说明有效数据段所在的羽毛球挥拍动作不属于动作类型模板库中任何一种动作类型。If it is determined that the minimum matching distance is still greater than the threshold, it indicates that the badminton swing action in which the valid data segment is located does not belong to any action type in the action type template library.
步骤157，根据最小匹配距离生成匹配结果。Step 157: Generate a matching result according to the minimum matching distance.
在另一个实施例中，如上方法还包括了模板构建的过程。具体的，如图5所示，如上所述的方法还包括：In another embodiment, the above method also includes a process of template construction. Specifically, as shown in FIG. 5, the method as described above further includes:
步骤310，为羽毛球挥拍中的动作类型采集预设数量的测试数据，该测 试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到。Step 310: Collect a preset number of test data for the action type in the badminton swing, the test The test data was collected by the sensor on the badminton racket during the swing of the same swing action.
在进行羽毛球挥拍动作识别之前，针对羽毛球挥拍的动作类型，分别采集预设数量的测试数据，测试数据是测试者手握羽毛球拍，按照当前所指定的动作类型挥拍后传感器所输出的数据，按照指定的动作类型完成了预设数量次数的羽毛球挥拍动作之后所得到的预设数量的测试数据将用于进行这一指定的动作类型的模板构建。Before performing the badminton swing motion recognition, a preset number of test data is respectively collected for the action type of the badminton swing, and the test data is the tester's hand holding the badminton racket, and the sensor outputs according to the currently specified action type. Data, the preset number of test data obtained after completing a preset number of badminton swing actions according to the specified action type will be used to perform template construction for this specified action type.
需要说明的是，与前述原始数据、有效数据段所对应的，测试数据也为三轴重力分量的形式。It should be noted that, corresponding to the foregoing original data and valid data segments, the test data is also in the form of a three-axis gravity component.
步骤330，对于同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到临时模板与其它测试数据之间的匹配距离之和。Step 330: For a preset number of test data of the same action type, a test data is used as a temporary template matching to obtain a matching distance with other test data, and the operation obtains a sum of matching distances between the temporary template and other test data.
羽毛球挥拍动作的动作类型有多个，而对于一动作类型，都将进行预设数量的测试数据的采集，并在这些测试数据中进行匹配，以得到最佳的测试数据作为模板。There are multiple types of badminton swing action, and for a type of action, a preset number of test data is collected and matched in these test data to obtain the best test data as a template.
具体的，测试数据为三轴重力分量的形式，即Vx、Vy和Vz，每一测试数据均作为临时模块，与其它测试数据进通过动态时间规整算法分别进行匹配，以得到对应的匹配距离，并由此运算所有匹配距离之和，即该临时模板所对应的总匹配距离。Specifically, the test data is in the form of three-axis gravity components, namely, V x , V y , and V z . Each test data is used as a temporary module, and is matched with other test data by a dynamic time warping algorithm to obtain corresponding Match the distance and calculate the sum of all matching distances, that is, the total matching distance corresponding to the temporary template.
步骤350，选取匹配距离之和最小的测试数据作为动作类型的预置模板，并保存。Step 350: Select test data with the smallest matching distance as the preset template of the action type, and save it.
该总匹配距离最小的测试数据将是与动作类型所对应的其它测试数据中最为相似的，其是用来作为模板的最优数据。The test data with the smallest total matching distance will be the most similar to the other test data corresponding to the action type, which is the optimal data used as a template.
需要说明的是，如上所述的模板构建以及原始数据的处理中涉及的匹配过程中，都是通过动态时间规整算法来实现的，以保证其精准性。It should be noted that the matching process involved in the template construction and the processing of the original data as described above is implemented by a dynamic time warping algorithm to ensure the accuracy.
具体的，在此动态时间规整算法所实现的匹配中，将基于有效数据段和模板构建矩阵网格，以此来进行有效数据段和模板之间对应关系的，进而所得到的有效数据段和模板之间的映射，以相应运算得到匹配距离，匹配距离越大则相似度越低。Specifically, in the matching implemented by the dynamic time warping algorithm, the matrix mesh is constructed based on the valid data segment and the template, so as to perform the correspondence between the valid data segment and the template, and the obtained valid data segment and The mapping between templates is matched by the corresponding operation. The larger the matching distance, the lower the similarity.
在此动态时间规整算法所实现的匹配中，在优选的实施例中，还将对此进行优化，以对算法内部搜索匹配路径进行限制，以提高匹配速度和匹配成 功率。In the matching implemented by the dynamic time warping algorithm, in a preferred embodiment, this will also be optimized to limit the internal search matching path of the algorithm to improve the matching speed and match power.
具体的，在有效数据段和模板的长度，或者测试数据之间的长度较为接近，且长度已知时，建立长度的查找表，以在表中保存长度与对应搜索范围的下边界。每一次匹配，只需要根据斜率推导出限制范围的上边界即可。Specifically, when the length of the valid data segment and the template, or the length between the test data is relatively close, and the length is known, a lookup table of the length is established to save the length in the table and the lower boundary of the corresponding search range. For each match, you only need to derive the upper bound of the limit range based on the slope.
如上所述的过程中，对于进行匹配的有效数据段和测试数据，其都可在其中进行二次提取，以减少匹配序列长度，例如，采用取奇数点舍弃偶数点的方式来减少匹配序列长度，进而极大减少匹配的时间，在此情况下，将相比之前减少了几乎一半的匹配时间，进一步提高了识别效率。In the process as described above, for the valid data segment and the test data to be matched, the second data extraction may be performed therein to reduce the length of the matching sequence, for example, the odd number is discarded and the even number is discarded to reduce the length of the matching sequence. , in turn, greatly reducing the matching time, in this case, almost half of the matching time will be reduced compared to before, further improving the recognition efficiency.
在一个实施例中，还相应地提供了一种羽毛球挥拍动作识别装置，如图6所示，包括原始数据获取模块410、端点检测模块430、模板识别模块450和结果输出模块470，其中：In an embodiment, a badminton swing motion recognition device is further provided, as shown in FIG. 6, including a raw data acquisition module 410, an endpoint detection module 430, a template recognition module 450, and a result output module 470, wherein:
原始数据获取模块410，用于获取跟踪羽毛球拍产生的原始数据，该原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到。The original data obtaining module 410 is configured to acquire raw data generated by the tracking badminton racket, and the original data is continuously collected by the sensor on the badminton racket during the swinging process.
端点检测模块430，用于端点检测原始数据得到有效数据段。The endpoint detection module 430 is configured to detect the original data by the endpoint to obtain a valid data segment.
模板识别模块450，用于通过动态时间规整算法有效数据段和预置的模板，预置的模板对应于羽毛球挥拍动作的动作类型。The template identification module 450 is configured to validate the data segment and the preset template by using a dynamic time warping algorithm, and the preset template corresponds to the action type of the badminton swing action.
结果输出模块470，用于根据匹配结果得到对应的羽毛球挥拍动作识别结果。The result output module 470 is configured to obtain a corresponding badminton swing motion recognition result according to the matching result.
在一个实施例中，如图7所示，如上所述的装置还包括特征值区分模块510和判断模块530，其中：In one embodiment, as shown in FIG. 7, the apparatus as described above further includes a feature value distinguishing module 510 and a determining module 530, wherein:
特征值区分模块510，用于通过预置的特征值对有效数据段预先进行区分。The feature value distinguishing module 510 is configured to pre-determine the valid data segments by preset feature values.
判断模块530，用于判断有效数据段是否能够通过特征值区分得到动作类型，若为是，则输出区分得到的动作类型为羽毛球挥拍动作识别结果，若为否，则通知模板识别模块450。The determining module 530 is configured to determine whether the valid data segment can be distinguished by the feature value, and if yes, output the differentiated action type to the badminton swing motion recognition result, and if not, notify the template recognition module 450.
进一步的，在本实施例中，如图8所示，特征值区分模块510包括读取单元511和比对单元513，其中：Further, in this embodiment, as shown in FIG. 8, the feature value distinguishing module 510 includes a reading unit 511 and a comparing unit 513, where:
读取单元511，用于读取预置的特征值，该预置的特征值用于识别产吩羽毛球挥拍动作，对应于部分动作类型。The reading unit 511 is configured to read a preset feature value, and the preset feature value is used to identify a badminton swing action, which corresponds to a partial action type.
比对单元513，用于比对特征值和有效数据段，在有效数据段与特征值 相符时特征值对应的动作类型为区分得到的动作类型。The comparison unit 513 is configured to compare the feature value and the valid data segment in the valid data segment and the feature value The action type corresponding to the feature value when matching is the action type that distinguishes the obtained action.
在一个实施例中，模板识别模块450如图9所示，包括姿态融合单元451、匹配单元453和距离判断单元455，其中：In one embodiment, the template identification module 450, as shown in FIG. 9, includes a gesture fusion unit 451, a matching unit 453, and a distance determination unit 455, wherein:
姿态融合单元451，用于在有效数据段中进行姿态融合得到三轴重力分量的时间序列。The attitude fusion unit 451 is configured to perform posture fusion in the valid data segment to obtain a time series of the three-axis gravity component.
匹配单元453，用于由动作类型模板库中预置的模板分别与时间序列通过动态时间规整算法进行匹配，得到时间序列与任一模板之间的匹配距离。The matching unit 453 is configured to match the time series of the template preset in the action type template library with the time series by a dynamic time warping algorithm to obtain a matching distance between the time series and any template.
距离判断单元455，用于提取最小匹配距离，判断最小匹配距离是否小于阈值，若为是，则根据最小匹配距离生成匹配结果。The distance determining unit 455 is configured to extract a minimum matching distance, determine whether the minimum matching distance is smaller than a threshold, and if yes, generate a matching result according to the minimum matching distance.
在一个实施例中，如上所述的装置如图10所示，还包括测试数据采集模块510、测试数据匹配模块530和距离选取模块550，其中：In one embodiment, the apparatus as described above, as shown in FIG. 10, further includes a test data acquisition module 510, a test data matching module 530, and a distance selection module 550, wherein:
测试数据采集模块510，用于为羽毛球挥拍中的动作类型采集预设数量的测试数据，测试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到。The test data collection module 510 is configured to collect a preset number of test data for the action type in the badminton swing, and the test data is collected by the sensor on the badminton racket during the swing of the same swing action.
测试数据匹配模块530，用于对同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到临时模板与其它测试数据之间的匹配距离之和。The test data matching module 530 is configured to use a test data as a temporary template matching to obtain a matching distance between the test data and the other test data, and obtain a matching distance between the temporary template and the other test data. Match the sum of the distances.
距离选取模块550，用于选取匹配距离之和最小的测试数据作为动作类型的预置模板，并保存。The distance selection module 550 is configured to select the test data with the smallest matching distance as the preset template of the action type, and save it.
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。A person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium. The storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.
虽然已参照几个典型实施方式描述了本发明，但应当理解，所用的术语是说明和示例性、而非限制性的术语。由于本发明能够以多种形式具体实施而不脱离发明的精神或实质，所以应当理解，上述实施方式不限于任何前述的细节，而应在随附权利要求所限定的精神和范围内广泛地解释，因此落入权利要求或其等效范围内的全部变化和改型都应为随附权利要求所涵盖。 While the invention has been described with respect to the preferred embodiments the embodiments The present invention may be embodied in a variety of forms without departing from the spirit or scope of the invention. It is to be understood that the invention is not limited to the details of the invention. All changes and modifications that come within the scope of the claims or the equivalents thereof are intended to be covered by the appended claims.

Claims (10)

  1. 一种羽毛球挥拍动作识别方法，其特征在于，包括：A method for identifying a badminton swing motion, characterized in that it comprises:
    获取跟踪羽毛球拍产生的原始数据，所述原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到；Obtaining raw data generated by the tracking badminton racket, the raw data being continuously collected by the sensor on the badminton racket during the swinging process;
    端点检测所述原始数据得到有效数据段；The endpoint detects the original data to obtain a valid data segment;
    通过动态时间规整算法匹配所述有效数据段和预置的模板，所述预置的模板对应于羽毛球挥拍动作的动作类型；Matching the valid data segment and the preset template by a dynamic time warping algorithm, where the preset template corresponds to an action type of a badminton swing action;
    根据匹配结果得到对应的羽毛球挥拍动作识别结果。According to the matching result, the corresponding badminton swing motion recognition result is obtained.
  2. 根据权利要求1所述的方法，其特征在于，所述通过动态时间规整算法匹配所述有效数据段和预置的模板之前，所述方法还包括：The method according to claim 1, wherein before the matching the valid data segment and the preset template by the dynamic time warping algorithm, the method further comprises:
    通过预置的特征值对所述有效数据段预先进行区分；Predetermining the valid data segments by preset feature values;
    判断所述有效数据段是否能够通过特征值区分得到动作类型，若为是，则输出区分得到的动作类型为羽毛球挥拍动作识别结果，若为否，则Determining whether the valid data segment can be distinguished by the feature value, and if yes, outputting the differentiated action type is a badminton swing action recognition result, if not, then
    进入所述通过动态时间规整算法匹配所述有效数据段和预置的模板的步骤。Entering the step of matching the valid data segment and the preset template by a dynamic time warping algorithm.
  3. 根据权利要求2所述的方法，其特征在于，所述通过预置的特征值对所述有效数据段预先进行区分的步骤包括：The method according to claim 2, wherein the step of pre-distinguishing the valid data segments by preset feature values comprises:
    读取预置的特征值，所述预置的特征值用于识别部分羽毛球挥拍动作，对应于部分动作类型；Reading a preset feature value, the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
    比对所述特征值和有效数据段，在所述有效数据段与所述特征值相符时所述特征值对应的动作类型为区分得到的动作类型。Comparing the feature value and the valid data segment, when the valid data segment matches the feature value, the action type corresponding to the feature value is a differentiated action type.
  4. 根据权利要求1所述的方法，其特征在于，所述通过动态时间规整算法匹配所述有效数据段和预置的模板的步骤包括：The method according to claim 1, wherein the step of matching the valid data segment and the preset template by a dynamic time warping algorithm comprises:
    在所述有效数据段中进行姿态融合得到三轴重力分量的时间序列；Performing a pose fusion in the valid data segment to obtain a time series of three-axis gravity components;
    由动作类型模板库中预置的模板分别与所述时间序列通过动态时间规整算法进行匹配，得到所述时间序列与任一模板之间的匹配距离； The template preset by the action type template library is matched with the time series by a dynamic time warping algorithm to obtain a matching distance between the time series and any template.
    提取最小匹配距离，判断所述最小匹配距离是否小于阈值，若为是，则根据所述最小匹配距离生成匹配结果。Extracting a minimum matching distance, determining whether the minimum matching distance is smaller than a threshold, and if yes, generating a matching result according to the minimum matching distance.
  5. 根据权利要求1所述的方法，其特征在于，所述方法还包括：The method of claim 1 further comprising:
    为羽毛球挥拍中的动作类型采集预设数量的测试数据，所述测试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到；Collecting a preset number of test data for the action type in the badminton swing, the test data being collected by the sensor on the badminton racket during the swing of the same swing action;
    对同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到所述临时模板与其它测试数据之间的匹配距离之和；For a preset number of test data of the same action type, a test data is used as a temporary template matching to obtain a matching distance with other test data, and the operation obtains a sum of matching distances between the temporary template and other test data;
    选取匹配距离之和最小的测试数据作为所述动作类型的预置模板，并保存。The test data with the smallest sum of matching distances is selected as a preset template of the action type, and saved.
  6. 一种羽毛球挥拍动作识别装置，其特征在于，包括：A badminton swing action recognition device, comprising:
    原始数据获取模块，用于获取跟踪羽毛球拍产生的原始数据，所述原始数据由羽毛球拍上的传感器在挥拍过程持续采集得到；a raw data obtaining module, configured to acquire raw data generated by tracking a badminton racket, wherein the raw data is continuously collected by a sensor on the badminton racket during a swinging process;
    端点检测模块，用于端点检测所述原始数据得到有效数据段；An endpoint detection module, configured to detect, by the endpoint, the valid data segment by using the original data;
    模板识别模块，用于通过动态时间规整算法匹配所述有效数据段和预置的模板，所述预置的模板对应于羽毛球挥拍动作的动作类型；a template identification module, configured to match the valid data segment and the preset template by a dynamic time warping algorithm, where the preset template corresponds to an action type of a badminton swing action;
    结果输出模块，用于根据匹配结果得到对应的羽毛球挥拍动作识别结果。The result output module is configured to obtain a corresponding badminton swing motion recognition result according to the matching result.
  7. 根据权利要求6所述的装置，其特征在于，所述装置还包括：The device according to claim 6, wherein the device further comprises:
    特征值区分模块，用于通过预置的特征值对所述有效数据段预先进行区分；An eigenvalue distinguishing module, configured to pre-determine the valid data segment by using preset feature values;
    判断模块，用于判断所述有效数据段是否能够通过特征值区分得到动作类型，若为是，则输出区分得到的动作类型为羽毛球挥拍动作识别结果，若为否，则通知所述模板识别模块。The determining module is configured to determine whether the valid data segment can be distinguished by the feature value, and if yes, outputting the differentiated action type to the badminton swing action recognition result, and if not, notifying the template identification Module.
  8. 根据权利要求7所述的装置，其特征在于，所述特征值区分模块包括：The device according to claim 7, wherein the feature value distinguishing module comprises:
    读取单元，用于读取预置的特征值，该预置的特征值用于识别部分羽毛球挥拍动作，对应于部分动作类型； a reading unit, configured to read a preset feature value, wherein the preset feature value is used to identify a part of the badminton swing action, corresponding to a partial action type;
    比对单元，用于比对所述特征值和有效数据段，在所述有效数据段与所述特征值相符时所述特征值对应的动作类型为区分得到的动作类型。The comparison unit is configured to compare the feature value and the valid data segment, and when the valid data segment matches the feature value, the action type corresponding to the feature value is a differentiated action type.
  9. 根据权利要求6所述的装置，其特征在于，所述模板识别模块包括：The device according to claim 6, wherein the template identification module comprises:
    姿态融合单元，用于在所述有效数据段中进行姿态融合得到三轴重力分量的时间序列；a posture fusion unit, configured to perform posture fusion in the valid data segment to obtain a time series of three-axis gravity components;
    匹配单元，用于由动作类型模板库中预置的模板分别与所述时间序列通过动态时间规整算法进行匹配，得到所述时间序列与任一模板之间的匹配距离；a matching unit, configured to be matched with the time series by the dynamic time warping algorithm by the template preset in the action type template library, to obtain a matching distance between the time series and any template;
    距离判断单元，用于提取最小匹配距离，判断所述最小匹配距离是否小于阈值，若为是，则根据所述最小匹配距离生成匹配结果。The distance determining unit is configured to extract a minimum matching distance, determine whether the minimum matching distance is smaller than a threshold, and if yes, generate a matching result according to the minimum matching distance.
  10. 根据权利要求6所述的装置，其特征在于，所述装置还包括：The device according to claim 6, wherein the device further comprises:
    测试数据采集模块，用于为羽毛球挥拍中的动作类型采集预设数量的测试数据，所述测试数据由羽毛球拍上的传感器在同一挥拍动作的挥拍过程中采集得到；a test data acquisition module, configured to collect a preset number of test data for the action type in the badminton swing, wherein the test data is collected by the sensor on the badminton racket during the swing of the same swing action;
    测试数据匹配模块，用于对同一动作类型的预设数量测试数据，将以一测试数据为临时模板匹配得到与其它测试数据之间的匹配距离，运算得到所述临时模板与其它测试数据之间的匹配距离之和；The test data matching module is configured to use a test data as a temporary template matching to obtain a matching distance between the test data and the other test data, and obtain a matching distance between the temporary template and the other test data. The sum of the matching distances;
    距离选取模块，用于选取匹配距离之和最小的测试数据作为所述动作类型的预置模板，并保存。 The distance selection module is configured to select test data with the smallest sum of matching distances as a preset template of the action type, and save.
PCT/CN2016/093071 2015-12-25 2016-08-03 Method and device for recognizing badminton racket swinging motion WO2017107494A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510991323.6 2015-12-25
CN201510991323.6A CN105617638B (en) 2015-12-25 2015-12-25 Badminton racket swing action identification method and device

Publications (1)

Publication Number Publication Date
WO2017107494A1 true WO2017107494A1 (en) 2017-06-29

Family

ID=56033220

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/093071 WO2017107494A1 (en) 2015-12-25 2016-08-03 Method and device for recognizing badminton racket swinging motion

Country Status (2)

Country Link
CN (1) CN105617638B (en)
WO (1) WO2017107494A1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110135283A (en) * 2019-04-25 2019-08-16 上海大学 The signal recognition method of optical fiber perimeter defence system based on FastDTW algorithm
CN110648362A (en) * 2019-09-11 2020-01-03 商丘师范学院 Binocular stereo vision badminton positioning identification and posture calculation method
CN111881859A (en) * 2020-07-31 2020-11-03 北京融链科技有限公司 Template generation method and device
CN114237397A (en) * 2021-12-16 2022-03-25 厦门大学 Boxing action recognition method and system and electronic equipment
CN114699757A (en) * 2022-04-28 2022-07-05 闽江学院 Intelligent badminton serving device and method based on visual perception

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105617638B (en) * 2015-12-25 2019-04-05 深圳市酷浪云计算有限公司 Badminton racket swing action identification method and device
CN106310609B (en) * 2016-09-29 2019-10-29 广东小天才科技有限公司 A kind of analysis method and device of badminton
CN106778477B (en) 2016-11-21 2020-04-03 深圳市酷浪云计算有限公司 Tennis racket action recognition method and device
CN106650659B (en) * 2016-12-19 2020-06-19 深圳市酷浪云计算有限公司 Method and device for identifying motion parameters of bat
WO2018232557A1 (en) * 2017-06-19 2018-12-27 深圳市酷浪云计算有限公司 Exercise movement recognition method and apparatus, and electronic device
CN108398952A (en) * 2018-03-22 2018-08-14 上海荷福人工智能科技（集团）有限公司 Robot localization method and robot
CN109087336B (en) * 2018-08-29 2020-10-02 上海体育学院 Method and device for detecting rotation posture of sphere, storage medium and computer equipment
CN110151187B (en) * 2019-04-09 2022-07-05 缤刻普达（北京）科技有限责任公司 Body-building action recognition method and device, computer equipment and storage medium
CN110020630B (en) * 2019-04-11 2020-12-18 成都乐动信息技术有限公司 Method and device for evaluating action completion degree, storage medium and electronic equipment
CN111840948B (en) * 2019-04-29 2022-06-10 北京卡路里信息技术有限公司 Method and device for detecting number of times of badminton racket swinging and wearable equipment
CN110674683B (en) * 2019-08-15 2022-07-22 深圳供电局有限公司 Robot hand motion recognition method and system
CN110731762B (en) * 2019-09-18 2022-02-08 平安科技（深圳）有限公司 Method, device, computer system and readable storage medium for preprocessing pulse wave based on similarity
CN111530044A (en) * 2020-05-11 2020-08-14 杭州鹤鸣堂文化创意有限公司 Badminton training system based on unmanned aerial vehicle high-speed photography

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6671390B1 (en) * 1999-10-18 2003-12-30 Sport-X Inc. Automated collection, processing and use of sports movement information via information extraction from electromagnetic energy based upon multi-characteristic spatial phase processing
CN101114449A (en) * 2006-07-26 2008-01-30 大连三曦智能科技有限公司 Model training method for unspecified person alone word, recognition system and recognition method
CN103785157A (en) * 2012-10-30 2014-05-14 莫凌飞 Human body motion type identification accuracy improving method
CN105617638A (en) * 2015-12-25 2016-06-01 深圳市酷浪云计算有限公司 Badminton racket swinging movement recognizing method and device

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008212425A (en) * 2007-03-05 2008-09-18 Terumo Corp Exercise support device and data processing method
CN103310192B (en) * 2013-06-06 2016-05-11 南京邮电大学 A kind of motor behavior recognition methods based on single-axis acceleration sensors
CN103440277A (en) * 2013-08-12 2013-12-11 合肥寰景信息技术有限公司 Action model feature library and construction method thereof
CN103442114B (en) * 2013-08-16 2015-10-21 中南大学 A kind of identity identifying method based on dynamic gesture
CN104731307B (en) * 2013-12-20 2019-05-10 孙伯元 A kind of body-sensing action identification method and human-computer interaction device
CN103892840B (en) * 2014-03-06 2015-11-18 深圳市德凯瑞科技有限公司 The feature extracting method of a kind of Intelligent worn device and human motion

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6671390B1 (en) * 1999-10-18 2003-12-30 Sport-X Inc. Automated collection, processing and use of sports movement information via information extraction from electromagnetic energy based upon multi-characteristic spatial phase processing
CN101114449A (en) * 2006-07-26 2008-01-30 大连三曦智能科技有限公司 Model training method for unspecified person alone word, recognition system and recognition method
CN103785157A (en) * 2012-10-30 2014-05-14 莫凌飞 Human body motion type identification accuracy improving method
CN105617638A (en) * 2015-12-25 2016-06-01 深圳市酷浪云计算有限公司 Badminton racket swinging movement recognizing method and device

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110135283A (en) * 2019-04-25 2019-08-16 上海大学 The signal recognition method of optical fiber perimeter defence system based on FastDTW algorithm
CN110648362A (en) * 2019-09-11 2020-01-03 商丘师范学院 Binocular stereo vision badminton positioning identification and posture calculation method
CN110648362B (en) * 2019-09-11 2022-09-23 商丘师范学院 Binocular stereo vision badminton positioning identification and posture calculation method
CN111881859A (en) * 2020-07-31 2020-11-03 北京融链科技有限公司 Template generation method and device
CN114237397A (en) * 2021-12-16 2022-03-25 厦门大学 Boxing action recognition method and system and electronic equipment
CN114699757A (en) * 2022-04-28 2022-07-05 闽江学院 Intelligent badminton serving device and method based on visual perception

Also Published As

Publication number Publication date
CN105617638A (en) 2016-06-01
CN105617638B (en) 2019-04-05

Similar Documents

Publication Publication Date Title
WO2017107494A1 (en) Method and device for recognizing badminton racket swinging motion
US10737158B2 (en) Method and device for recognizing movement of tennis racket
Van Gemert et al. APT: Action localization proposals from dense trajectories.
CN105809144B (en) A kind of gesture recognition system and method using movement cutting
Yang et al. From facial parts responses to face detection: A deep learning approach
Ren et al. Robust hand gesture recognition based on finger-earth mover's distance with a commodity depth camera
Rios-Cabrera et al. Discriminatively trained templates for 3d object detection: A real time scalable approach
JP5202037B2 (en) Feature point position determination method and apparatus
CN108810620A (en) Identify method, computer equipment and the storage medium of the material time point in video
Kaâniche et al. Gesture recognition by learning local motion signatures
CN106407958B (en) Face feature detection method based on double-layer cascade
Deepa et al. Comparison of yolo, ssd, faster rcnn for real time tennis ball tracking for action decision networks
WO2013091370A1 (en) Human body part detection method based on parallel statistics learning of 3d depth image information
TW201201115A (en) Facial expression recognition systems and methods and computer program products thereof
Ahn et al. Dynamic vision sensor camera based bare hand gesture recognition
CN109583329A (en) Winding detection method based on the screening of road semanteme road sign
Almajai et al. Anomaly detection and knowledge transfer in automatic sports video annotation
WO2014183275A1 (en) Detection method and system for locally deformable object based on on-line learning
Andiani et al. Face recognition for work attendance using multitask convolutional neural network (MTCNN) and pre-trained facenet
Chen et al. Attending to distinctive moments: Weakly-supervised attention models for action localization in video
Dasari et al. Mpeg cdvs feature trajectories for action recognition in videos
JP5552946B2 (en) Face image sample collection device, face image sample collection method, program
Hammouche et al. Toward a real time view-invariant 3d action recognition
Wang et al. A convolutional neural network combined with aggregate channel feature for face detection
Nguyen et al. Action recognition using bag of features extracted from a beam of trajectories

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16877306

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC, EPO FORM 1205A DATED 29.10.18

122 Ep: pct application non-entry in european phase

Ref document number: 16877306

Country of ref document: EP

Kind code of ref document: A1