CN109800665A - A kind of Human bodys' response method, system and storage medium - Google Patents
A kind of Human bodys' response method, system and storage medium Download PDFInfo
- Publication number
- CN109800665A CN109800665A CN201811621829.8A CN201811621829A CN109800665A CN 109800665 A CN109800665 A CN 109800665A CN 201811621829 A CN201811621829 A CN 201811621829A CN 109800665 A CN109800665 A CN 109800665A
- Authority
- CN
- China
- Prior art keywords
- physical feeling
- classification
- recognition result
- human body
- image
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Human bodys' response method, system and storage medium, method includes: to be carried out physical feeling detection based on SSD target detection model to collected human body image, obtained the location information of physical feeling;According to the location information of physical feeling, Classification and Identification is carried out to each physical feeling by deep learning multi-classification algorithm;Wherein, the result of the Classification and Identification include but is not limited to head safety cap wear recognition result, body work clothes wear the safety recognition result of recognition result, number recognition result and behavior.The present invention can identify in real time human body behavior, and the accuracy and reliability of recognition result is improved by SSD target detection model, can be widely applied to depth learning technology field.
Description
Technical field
The present invention relates to depth learning technology field, especially a kind of Human bodys' response method, system and storage medium.
Background technique
The human body identification technology and Activity recognition technology used on current market is roughly divided into following several:
Optical flow field: optical flow field is the instantaneous velocity of space motion object pixel motion on observation imaging surface.It utilizes figure
The variation and scene of the gray scale of image in time are studied in time domain variation and correlation as the pixel intensity data in sequence
Object structures and its relationship of movement.Optical flow method usually assumes that the image difference of consecutive frame is very small, to obtain to true fortune
The approximate evaluation of dynamic field.It does not have to predict any priori knowledge, can provide moving region in related movement velocity and image
Succinct description, the situation suitable for camera motion.But influence of the optical flow method vulnerable to noise and illumination variation, and calculate more multiple
It is miscellaneous, it is difficult to real-time video monitoring system.
The locus of points: the motion profile of target can also be used as feature, to substantially be inferred to behavior belonging to target movement
Classification.But the track on the plane of delineation is more sensitive to transformation such as translation, rotation and scalings, therefore under most circumstances, this is special
Sign seems less reliable.Common alternative features expression has path velocity, space-time curvature etc..Dependence is compared in the acquisition of motion profile
In accurate track algorithm.Locus of points feature is extracted from original video to be equally easy by noise, block and distracting background etc.
Influence.
Body shape expression: in the case of video camera is fixed, it is assumed that background it is known that can be held very much by the background method of wiping out
It is easy to get to movement human shape.Description based on the overall situation, boundary and skeleton etc. may serve to expression body shape.Overall situation side
Method such as outline, square are to calculate to describe son in global shape region, and boundary method only considers shape contour, and skeleton technique is then
It is to represent a complicated body shape, such as Medial-Axis Transformation etc. with one group of skeleton curve.But the behavioural characteristic of this method identification
Equally exist insecure disadvantage.
Filter response: space-time filter response is a sensu lato classification.Gaussian derivative is calculated on time shaft etc. depositing,
Filter is responded into higher region as moving region.Using Gaussian derivative filter when one group of sky by Harris (Harris)
Corner Detection expanded application point of interest when three-dimensional video data is to detect sky.Such methods are all based on greatly simple volume
Product operation, operation are quickly easy.In the case of video resolution is lower, extracts light stream or outline feature is more difficult, benefit
Effective low-level image feature can be extracted from video data with filter response characteristics.But if being applied to high definition live video stream
Under conditions of, also need the processing for carrying out a series of complex that can just be truly realized Activity recognition detection.
Summary of the invention
In order to solve the above technical problems, it is an object of the invention to: provide that a kind of real-time is high and accurately and reliably human body
Activity recognition method, system and storage medium.
The technical solution that one aspect of the present invention is taken are as follows:
A kind of Human bodys' response method, comprising the following steps:
Based on SSD target detection model, physical feeling detection is carried out to collected human body image, obtains physical feeling
Location information;
According to the location information of physical feeling, classification knowledge is carried out to each physical feeling by deep learning multi-classification algorithm
Not;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work of recognition result, body
Make the safety recognition result that clothes wear recognition result, number recognition result and behavior.
Further, described to be based on SSD target detection model, physical feeling detection is carried out to collected human body image, is obtained
To physical feeling location information the step for, comprising the following steps:
Image is acquired by image capture module, described image acquisition module includes video camera;
According to acquired image, the human body image in image is confirmed;
According to the human body image confirmed, by SSD target detection model to the head and shoulder position of human body, metastomium and
Thigh position is positioned, and the location information of each physical feeling is obtained.
Further, the human body image that the basis confirms, by SSD target detection model to the head and shoulder position of human body,
The step for metastomium and thigh position are positioned, obtain the location information of each physical feeling, comprising the following steps:
Learning training is carried out to the corresponding detector in each position by multiple priori examples, obtains SSD target detection mould
Type;
The human body image that will confirm that is divided into 3 detection windows;3 detection windows respectively correspond the head and shoulder of human body
Position, metastomium and thigh position;
3 detection windows are inputted into SSD target detection model respectively, obtain corresponding response;
Response is determined, the location information at head and shoulder position, metastomium and thigh position is obtained.
Further, described that learning training is carried out to the corresponding detector in each position by multiple priori examples, obtain SSD
The step for target detection model, comprising the following steps:
By the convolutional neural networks of propagated forward, several bounding boxs are generated, the size of the bounding box is fixed;
Calculate the similarity between each bounding box;
The convolutional layer added by several predicts each position of human body image, obtains the classification of each position
Score;
Score threshold is divided according to obtained classification score;
According to score threshold, handle to obtain SSD target detection model by non-maxima suppression.
Further, the location information according to physical feeling, by deep learning multi-classification algorithm to each body
Position carries out the step for Classification and Identification, comprising the following steps:
Using Adaboost algorithm, Weak Classifier is overlapped based on preset weight, obtains strong classifier;
The number of Weak Classifier is increased, strong classifier is trained;
According to the strong classifier that training obtains, the classification recognition result of each physical feeling is obtained.
Further, further comprising the steps of:
The classification recognition result of each physical feeling is integrated, global analysis result is obtained.
Another aspect of the present invention is adopted the technical scheme that:
A kind of Human bodys' response system, comprising:
Detection module carries out physical feeling detection to collected human body image for being based on SSD target detection model,
Obtain the location information of physical feeling;
Identification module, for the location information according to physical feeling, by deep learning multi-classification algorithm to each body
Position carries out Classification and Identification;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work of recognition result, body
Make the safety recognition result that clothes wear recognition result, number recognition result and behavior.
Further, further includes:
Module is integrated, for integrating the classification recognition result of each physical feeling, obtains global analysis result.
Another aspect of the present invention is adopted the technical scheme that:
A kind of Human bodys' response system, comprising:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized
The Human bodys' response method.
Another aspect of the present invention is adopted the technical scheme that:
A kind of storage medium, wherein be stored with the executable instruction of processor, the executable instruction of the processor by
For executing the Human bodys' response method when processor executes.
The beneficial effects of the present invention are: the present invention can by SSD target detection model to collected human body image into
Then the detection of row physical feeling carries out Classification and Identification to each physical feeling by deep learning multi-classification algorithm, and then obtains
The safety cap on head wears recognition result, the work clothes of body wears the safety of recognition result, number recognition result and behavior
Property recognition result;The present invention can identify in real time human body behavior, and improve identification knot by SSD target detection model
The accuracy and reliability of fruit.
Detailed description of the invention
Fig. 1 is the step flow chart of the embodiment of the present invention.
Specific embodiment
The present invention is further explained and is illustrated with specific embodiment with reference to the accompanying drawings of the specification.For of the invention real
The step number in example is applied, is arranged only for the purposes of illustrating explanation, any restriction is not done to the sequence between step, is implemented
The execution sequence of each step in example can be adaptively adjusted according to the understanding of those skilled in the art.
Referring to Fig.1, the embodiment of the invention provides a kind of Human bodys' response methods, comprising the following steps:
Based on SSD target detection model, physical feeling detection is carried out to collected human body image, obtains physical feeling
Location information;
According to the location information of physical feeling, classification knowledge is carried out to each physical feeling by deep learning multi-classification algorithm
Not;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work of recognition result, body
Make the safety recognition result that clothes wear recognition result, number recognition result and behavior.
Be further used as preferred embodiment, it is described to be based on SSD target detection model, to collected human body image into
The step for row physical feeling detects, and obtains the location information of physical feeling, comprising the following steps:
Image is acquired by image capture module, described image acquisition module includes video camera;
According to acquired image, the human body image in image is confirmed;
According to the human body image confirmed, by SSD target detection model to the head and shoulder position of human body, metastomium and
Thigh position is positioned, and the location information of each physical feeling is obtained.
It is further used as preferred embodiment, the human body image that the basis confirms passes through SSD target detection model
The head and shoulder position of human body, metastomium and thigh position are positioned, obtain each physical feeling location information this
Step, comprising the following steps:
Learning training is carried out to the corresponding detector in each position by multiple priori examples, obtains SSD target detection mould
Type;
The human body image that will confirm that is divided into 3 detection windows;3 detection windows respectively correspond the head and shoulder of human body
Position, metastomium and thigh position;
3 detection windows are inputted into SSD target detection model respectively, obtain corresponding response;
Response is determined, the location information at head and shoulder position, metastomium and thigh position is obtained.
Be further used as preferred embodiment, it is described by multiple priori examples to the corresponding detector in each position into
Row learning training, the step for obtaining SSD target detection model, comprising the following steps:
By the convolutional neural networks of propagated forward, several bounding boxs are generated, the size of the bounding box is fixed;
Calculate the similarity between each bounding box;
The convolutional layer added by several predicts each position of human body image, obtains the classification of each position
Score;
Score threshold is divided according to obtained classification score;
According to score threshold, handle to obtain SSD target detection model by non-maxima suppression.
It is further used as preferred embodiment, the location information according to physical feeling passes through more points of deep learning
The step for class algorithm carries out Classification and Identification to each physical feeling, comprising the following steps:
Using Adaboost algorithm, Weak Classifier is overlapped based on preset weight, obtains strong classifier;
The number of Weak Classifier is increased, strong classifier is trained;
According to the strong classifier that training obtains, the classification recognition result of each physical feeling is obtained.
It is further used as preferred embodiment, further comprising the steps of:
The classification recognition result of each physical feeling is integrated, global analysis result is obtained.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of Human bodys' response systems, comprising:
Detection module carries out physical feeling detection to collected human body image for being based on SSD target detection model,
Obtain the location information of physical feeling;
Identification module, for the location information according to physical feeling, by deep learning multi-classification algorithm to each body
Position carries out Classification and Identification;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work of recognition result, body
Make the safety recognition result that clothes wear recognition result, number recognition result and behavior.
It is further used as preferred embodiment, further includes:
Module is integrated, for integrating the classification recognition result of each physical feeling, obtains global analysis result.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of Human bodys' response systems, comprising:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized
The Human bodys' response method.
Corresponding with the method for Fig. 1, the embodiment of the invention also provides a kind of storage mediums, wherein being stored with processor can
The instruction of execution, the executable instruction of the processor is when executed by the processor for executing the Human bodys' response side
Method.
The specific work process of Human bodys' response method of the present invention is described below in detail:
S1, human body image acquisition and detection;
Specifically, the present embodiment is acquired human body image by camera.Different human body images can be transferred through taking the photograph
As camera lens collects, for example still image, dynamic image, different positions, different gestures etc. can obtain very well
Acquisition.When user is in the coverage for acquiring equipment, acquisition equipment can search for automatically and shoot human body image.
S2, human testing;
Human body position is carried out to the human body image of step A acquisition to detect.
The present embodiment is using SSD (Single Shot MultiBox Detector) target detection model to the body of people
Position is detected, and the position of entire human body can be accurately positioned in picture, including head and shoulder, trunk and the positioning of leg, is protected
The picture of target is deposited first to coordinate.
Specifically, the present embodiment will test 3 positions that window is divided into 3 sections as human body, first use at each position more
The corresponding detector in each position of a priori case-based learning training, by sharing the thought of Weak Classifier, to output response value into
Row determines, converts multicategory classification problem for two-value classification problem.People when human body occurs in camera in real-time capturing pictures
Body identifies head, the upper part of the body and the leg of human body respectively.Wherein, the response is exported by detector, and output valve finally only has
One in 0,1,2 three value, what is respectively represented is head, the upper part of the body, leg.The two-value classification problem refers to, when wherein
This automatic discrimination is another (less than 50%) when the probability is relatively small by one, and multicategory classification problem be then take it is wherein general
Rate highest one as identification item.
Wherein, the thought by sharing Weak Classifier, the step for judgement output response value, including it is following
Step:
First Weak Classifier is obtained to the study of N number of training sample 1. first passing through;
2. the sample of misclassification classification and other new datas are constituted a new N number of training sample together, by right
The study of this sample obtains second Weak Classifier;
3. the sample of 1 and 2 all misclassifications, which is added other new samples, constitutes another new N number of training sample, lead to
It crosses and third Weak Classifier is obtained to the study of this sample;
4. eventually passing through the strong classifier of promotion.I.e. which kind of some data is divided into and to be determined by each classifier weight.
Wherein, the strong classifier is the classifier for referring to correctly identify rapidly classification, and the Weak Classifier refers to
It is easy to produce the classifier of classification error.
Picture is put into the strong classifier of multilayer Weak Classifier composition by the present invention, show the picture is which classification belonged to
(head, the upper part of the body, leg).
Wherein, SSD is to generate a series of fixed sizes based on propagated forward CNN (convolutional neural networks) network
(fixed-size) a possibility that including object example in bounding boxes (maximal encasing box) and each box,
That is similarity.Later, carry out non-maxima suppression (Non-maximum suppression) obtain it is final
Predictions (prediction result).
Existing basic network be by 5 layers of convolutional layer, 5 layers of pond layer and 2 layers of full articulamentums composition.Preferably, originally
Invention is added to level 2 volume lamination and pond layer in existing basic network, and redistributes to the weight of full articulamentum.
The present invention is added to additional convolutional layer after infrastructure network, and the size of these convolutional layers is successively to successively decrease
, it can be predicted under multiple dimensioned.In the present embodiment, a series of convolution kernel is can be used in each newly added layer
It is predicted.It is the characteristic layer of m*n, p-channel for a size, is predicted using the convolution kernel of 3*3, on some position
A value is predicted, which can be the score of a certain classification, and will generate a value in each position of image.Most
Afterwards, the present embodiment predicts the position of target by setting score threshold, and the target of general 0.5 or more score is available targets,
By obtaining target position target is separated to the progress more classification and Detections of deep learning from original image, achievees the purpose that behavioural analysis.
S3, human body behavioural analysis;
Since the reduction of target leak detection is largely dependent upon the learning algorithm on each position again, the present invention
The object instance of high confidence level is found by depth learning technology, to reduce complicated posture and be blocked caused by missing inspection.
Preferably, the present invention solves the problems, such as the deep learning at position using Adaboost algorithm frame, specifically will
Weak Classifier is stacked up with certain weight, obtains a strong classifier.In addition, the present embodiment, which also passes through, increases Weak Classifier
Number, to guarantee that target reduces missing inspection and wrong report;Sample weights, training classifier are updated by using two kinds of Boost methods.
Specifically, described two Boost methods include:
1.Real AdaBoost algorithm
After the Real Adaboost algorithm of the present embodiment belongs to the probability of certain class by each Weak Classifier output sample,
The corresponding probability value of sample is mapped to real number field by a logarithmic function, last classifier is all mapping functions
With.
2.Gentle AdaBoost algorithm
The present embodiment merges two parts of each iteration of Real Adaboost algorithm, generates one and is mapped to
The function of real number field, and then generate Gentle AdaBoost.Gentle AdaBoost then in each iteration, is based on minimum two
Multiply and do a weighted regression, finally the sum of all regression functions is as final classifier.
The present invention is trained using Real AdaBoost algorithm first, compares prediction result and legitimate reading.Work as appearance
When accuracy is close to 100%, using training set outside forecast set predicted, if predictablity rate does not meet expection, say
Bright trained over-fitting, can accurately identify trained sample set, can not accurately identify the forecast set outside training sample, then be tied
Fruit weighting, and use Gentle AdaBoost algorithm adjusting training model.
S4, the present invention separate the head of identification, carry out Classification and Identification, judge whether safe wearing cap.By the upper of separation
In conjunction with leg, progress work clothes classification judgement first judges whether to wear work clothes half body;Secondly behavioral value is carried out, individually
Judge whether upper part of the body limbs smoke, finally judges global behavior whether in violation of rules and regulations.
In conclusion the depth learning technology and target classification training technique of present invention combination convolutional neural networks, very greatly
Reduce the probability of wrong report and the missing inspection of Activity recognition in degree.
Furthermore the present invention can be integrated in system module, quickly analyzed by obtaining monitoring camera data flow in real time
Personnel in picture whether safe wearing cap, if smoke, if wear work clothes and other safety-related behaviors.It reduces artificial
Monitoring cost can reach timely early warning, and the working efficiency of the units in charge of construction such as building site, factory is improved while reducing cost.
In addition, the present invention can access multiple cameras, algorithm carries out calculating acceleration by using GPU, and it is real-time, accurate to accomplish
Identification.
It is to be illustrated to preferable implementation of the invention, but the present invention is not limited to the embodiment above, it is ripe
Various equivalent deformation or replacement can also be made on the premise of without prejudice to spirit of the invention by knowing those skilled in the art, this
Equivalent deformation or replacement are all included in the scope defined by the claims of the present application a bit.
Claims (10)
1. a kind of Human bodys' response method, it is characterised in that: the following steps are included:
Based on SSD target detection model, physical feeling detection is carried out to collected human body image, obtains the positioning of physical feeling
Information;
According to the location information of physical feeling, Classification and Identification is carried out to each physical feeling by deep learning multi-classification algorithm;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work clothes of recognition result, body
Wear the safety recognition result of recognition result, number recognition result and behavior.
2. a kind of Human bodys' response method according to claim 1, it is characterised in that: described to be based on SSD target detection
Model, the step for carrying out physical feeling detection to collected human body image, obtain the location information of physical feeling, including with
Lower step:
Image is acquired by image capture module, described image acquisition module includes video camera;
According to acquired image, the human body image in image is confirmed;
According to the human body image confirmed, by SSD target detection model to head and shoulder position, metastomium and the thigh of human body
Position is positioned, and the location information of each physical feeling is obtained.
3. a kind of Human bodys' response method according to claim 2, it is characterised in that: the human body that the basis confirms
Image positions the head and shoulder position of human body, metastomium and thigh position by SSD target detection model, obtains each
The step for location information of a physical feeling, comprising the following steps:
Learning training is carried out to the corresponding detector in each position by multiple priori examples, obtains SSD target detection model;
The human body image that will confirm that is divided into 3 detection windows;3 detection windows respectively correspond the head and shoulder portion of human body
Position, metastomium and thigh position;
3 detection windows are inputted into SSD target detection model respectively, obtain corresponding response;
Response is determined, the location information at head and shoulder position, metastomium and thigh position is obtained.
4. a kind of Human bodys' response method according to claim 3, it is characterised in that: described to pass through multiple priori examples
The step for is carried out by learning training, obtains SSD target detection model for the corresponding detector in each position, comprising the following steps:
By the convolutional neural networks of propagated forward, several bounding boxs are generated, the size of the bounding box is fixed;
Calculate the similarity between each bounding box;
The convolutional layer added by several predicts each position of human body image, obtains classifying for each position
Point;
Score threshold is divided according to obtained classification score;
According to score threshold, handle to obtain SSD target detection model by non-maxima suppression.
5. a kind of Human bodys' response method according to claim 1, it is characterised in that: described to be determined according to physical feeling
Position information, by deep learning multi-classification algorithm to each physical feeling carry out Classification and Identification the step for, comprising the following steps:
Using Adaboost algorithm, Weak Classifier is overlapped based on preset weight, obtains strong classifier;
The number of Weak Classifier is increased, strong classifier is trained;
According to the strong classifier that training obtains, the classification recognition result of each physical feeling is obtained.
6. a kind of Human bodys' response method according to claim 1, it is characterised in that: further comprising the steps of:
The classification recognition result of each physical feeling is integrated, global analysis result is obtained.
7. a kind of Human bodys' response system, it is characterised in that: include:
Detection module carries out physical feeling detection to collected human body image, obtains for being based on SSD target detection model
The location information of physical feeling;
Identification module, for the location information according to physical feeling, by deep learning multi-classification algorithm to each physical feeling
Carry out Classification and Identification;
Wherein, the result of the Classification and Identification includes but is not limited to that the safety cap on head wears the work clothes of recognition result, body
Wear the safety recognition result of recognition result, number recognition result and behavior.
8. a kind of Human bodys' response system according to claim 7, it is characterised in that: further include:
Module is integrated, for integrating the classification recognition result of each physical feeling, obtains global analysis result.
9. a kind of Human bodys' response system, it is characterised in that: include:
At least one processor;
At least one processor, for storing at least one program;
When at least one described program is executed by least one described processor, so that at least one described processor is realized as weighed
Benefit requires Human bodys' response method described in any one of 1-6.
10. a kind of storage medium, wherein being stored with the executable instruction of processor, it is characterised in that: the processor is executable
Instruction be used to execute when executed by the processor such as Human bodys' response method of any of claims 1-6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811621829.8A CN109800665A (en) | 2018-12-28 | 2018-12-28 | A kind of Human bodys' response method, system and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811621829.8A CN109800665A (en) | 2018-12-28 | 2018-12-28 | A kind of Human bodys' response method, system and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109800665A true CN109800665A (en) | 2019-05-24 |
Family
ID=66557817
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811621829.8A Pending CN109800665A (en) | 2018-12-28 | 2018-12-28 | A kind of Human bodys' response method, system and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109800665A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110348290A (en) * | 2019-05-27 | 2019-10-18 | 天津中科智能识别产业技术研究院有限公司 | Coke tank truck safe early warning visible detection method |
CN110414421A (en) * | 2019-07-25 | 2019-11-05 | 电子科技大学 | A kind of Activity recognition method based on sequential frame image |
CN110688893A (en) * | 2019-08-22 | 2020-01-14 | 成都通甲优博科技有限责任公司 | Detection method for wearing safety helmet, model training method and related device |
CN111898514A (en) * | 2020-07-24 | 2020-11-06 | 燕山大学 | Multi-target visual supervision method based on target detection and action recognition |
CN111931661A (en) * | 2020-08-12 | 2020-11-13 | 桂林电子科技大学 | Real-time mask wearing detection method based on convolutional neural network |
CN112101139A (en) * | 2020-08-27 | 2020-12-18 | 普联国际有限公司 | Human shape detection method, device, equipment and storage medium |
CN112926427A (en) * | 2021-02-18 | 2021-06-08 | 浙江智慧视频安防创新中心有限公司 | Target user dressing attribute identification method and device |
CN113052140A (en) * | 2021-04-25 | 2021-06-29 | 合肥中科类脑智能技术有限公司 | Video-based substation personnel and vehicle violation detection method and system |
WO2022022368A1 (en) * | 2020-07-28 | 2022-02-03 | 宁波环视信息科技有限公司 | Deep-learning-based apparatus and method for monitoring behavioral norms in jail |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105740892A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | High-accuracy human body multi-position identification method based on convolutional neural network |
CN106778609A (en) * | 2016-12-15 | 2017-05-31 | 国网浙江省电力公司杭州供电公司 | A kind of electric power construction field personnel uniform wears recognition methods |
CN108529363A (en) * | 2018-04-26 | 2018-09-14 | 珠海亿智电子科技有限公司 | A kind of intelligent control method |
-
2018
- 2018-12-28 CN CN201811621829.8A patent/CN109800665A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105740892A (en) * | 2016-01-27 | 2016-07-06 | 北京工业大学 | High-accuracy human body multi-position identification method based on convolutional neural network |
CN106778609A (en) * | 2016-12-15 | 2017-05-31 | 国网浙江省电力公司杭州供电公司 | A kind of electric power construction field personnel uniform wears recognition methods |
CN108529363A (en) * | 2018-04-26 | 2018-09-14 | 珠海亿智电子科技有限公司 | A kind of intelligent control method |
Non-Patent Citations (2)
Title |
---|
GEORGIA GKIOXARI等: "Actions and Attributes from Wholes and Parts", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
陈慧岩等: "《智能车辆理论与应用》", 31 July 2018 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110348290A (en) * | 2019-05-27 | 2019-10-18 | 天津中科智能识别产业技术研究院有限公司 | Coke tank truck safe early warning visible detection method |
CN110414421A (en) * | 2019-07-25 | 2019-11-05 | 电子科技大学 | A kind of Activity recognition method based on sequential frame image |
CN110414421B (en) * | 2019-07-25 | 2023-04-07 | 电子科技大学 | Behavior identification method based on continuous frame images |
CN110688893A (en) * | 2019-08-22 | 2020-01-14 | 成都通甲优博科技有限责任公司 | Detection method for wearing safety helmet, model training method and related device |
CN111898514A (en) * | 2020-07-24 | 2020-11-06 | 燕山大学 | Multi-target visual supervision method based on target detection and action recognition |
CN111898514B (en) * | 2020-07-24 | 2022-10-18 | 燕山大学 | Multi-target visual supervision method based on target detection and action recognition |
WO2022022368A1 (en) * | 2020-07-28 | 2022-02-03 | 宁波环视信息科技有限公司 | Deep-learning-based apparatus and method for monitoring behavioral norms in jail |
CN111931661A (en) * | 2020-08-12 | 2020-11-13 | 桂林电子科技大学 | Real-time mask wearing detection method based on convolutional neural network |
CN112101139A (en) * | 2020-08-27 | 2020-12-18 | 普联国际有限公司 | Human shape detection method, device, equipment and storage medium |
CN112101139B (en) * | 2020-08-27 | 2024-05-03 | 普联国际有限公司 | Human shape detection method, device, equipment and storage medium |
CN112926427A (en) * | 2021-02-18 | 2021-06-08 | 浙江智慧视频安防创新中心有限公司 | Target user dressing attribute identification method and device |
CN113052140A (en) * | 2021-04-25 | 2021-06-29 | 合肥中科类脑智能技术有限公司 | Video-based substation personnel and vehicle violation detection method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109800665A (en) | A kind of Human bodys' response method, system and storage medium | |
Shen et al. | Flame detection using deep learning | |
CN110765964B (en) | Method for detecting abnormal behaviors in elevator car based on computer vision | |
Rakibe et al. | Background subtraction algorithm based human motion detection | |
Sommer et al. | Flying object detection for automatic UAV recognition | |
CN109344717B (en) | Multi-threshold dynamic statistical deep sea target online detection and identification method | |
CN111932583A (en) | Space-time information integrated intelligent tracking method based on complex background | |
CN108932479A (en) | A kind of human body anomaly detection method | |
CN111881853B (en) | Method and device for identifying abnormal behaviors in oversized bridge and tunnel | |
CN108648211A (en) | A kind of small target detecting method, device, equipment and medium based on deep learning | |
CN113378649A (en) | Identity, position and action recognition method, system, electronic equipment and storage medium | |
CN110909672A (en) | Smoking action recognition method based on double-current convolutional neural network and SVM | |
KR101406334B1 (en) | System and method for tracking multiple object using reliability and delayed decision | |
Afsar et al. | Automatic human action recognition from video using hidden markov model | |
Ali et al. | Deep Learning Algorithms for Human Fighting Action Recognition. | |
Afsar et al. | Automatic human trajectory destination prediction from video | |
CN113378638B (en) | Method for identifying abnormal behavior of turbine operator based on human body joint point detection and D-GRU network | |
Pervaiz et al. | Artificial neural network for human object interaction system over Aerial images | |
Junejo et al. | Single-class SVM for dynamic scene modeling | |
Fan et al. | Video anomaly detection using CycleGan based on skeleton features | |
CN113591722A (en) | Target person following control method and system of mobile robot | |
JP7488674B2 (en) | OBJECT RECOGNITION DEVICE, OBJECT RECOGNITION METHOD, AND OBJECT RECOGNITION PROGRAM | |
CN112183287A (en) | People counting method of mobile robot under complex background | |
Kim et al. | Unsupervised moving object segmentation and recognition using clustering and a neural network | |
KR102544492B1 (en) | Apparatus and method of managing safety of swimming pool |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190524 |