CN110414380A - A kind of students ' behavior detection method based on target detection - Google Patents
A kind of students ' behavior detection method based on target detection Download PDFInfo
- Publication number
- CN110414380A CN110414380A CN201910620750.1A CN201910620750A CN110414380A CN 110414380 A CN110414380 A CN 110414380A CN 201910620750 A CN201910620750 A CN 201910620750A CN 110414380 A CN110414380 A CN 110414380A
- Authority
- CN
- China
- Prior art keywords
- students
- behavior
- behavior detection
- convolution
- method based
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 23
- 230000006872 improvement Effects 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 8
- 241001269238 Data Species 0.000 claims description 4
- 238000005286 illumination Methods 0.000 claims description 4
- 230000000644 propagated effect Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims 1
- 230000006399 behavior Effects 0.000 description 52
- 230000000694 effects Effects 0.000 description 17
- 238000013527 convolutional neural network Methods 0.000 description 13
- 230000003542 behavioural effect Effects 0.000 description 9
- 230000007958 sleep Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000012800 visualization Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The students ' behavior detection method based on target detection that the present invention relates to a kind of, comprising the following steps: S1, establish the data set comprising students ' behavior information, the students ' behavior includes raising one's hand, stand and sleeping；S2, students ' behavior detection model is established, which is the improvement Faster R-CNN model based on residual error network ResNet-101；S3, collect based on the data, students ' behavior detection model is trained using online difficult sample method for digging；S4, video to be measured is detected using the students ' behavior detection model after training, obtain students ' behavior result and visualized；Wherein, the residual error network ResNet-101 uses multilayer feature convergence strategy, and the 5th convolution stage of the residual error network ResNet-101 includes multiple branches of different sizes with receptive field.Compared with prior art, the present invention has many advantages, such as that precision is high.
Description
Technical field
The present invention relates to behavioral value fields, more particularly, to a kind of students ' behavior detection method based on target detection.
Background technique
Behavioral value is an important subject of current manual's smart field, be widely used in public security protection,
The fields such as human-computer interaction.Students ' behavior detection under the scenes such as classroom is the important ring in Following course analysis, can be effective
School is helped to improve the quality of teaching.Therefore automatic detection students ' behavior can mitigate teacher's burden significantly, mention for teaching process
For lasting quality evaluation.However, there are resolution ratio in true classroom scene low, students ' behavior posture multiplicity, serious shielding
The problems such as.Meanwhile there is biggish difference in the camera shooting angle in different classrooms, shooting distance, illumination condition etc..Tradition
Behavioral value method be difficult to obtain preferable effect under true classroom scene.With the quick hair of deep learning in recent years
Exhibition, the object detection method based on convolutional neural networks (CNN) is also applied to behavioral value, and realizes preferable effect
Fruit.
The existing behavioral value method based on target detection is divided into two steps: establishing training sample；Training depth convolutional Neural
Network, the structure and training method of network determine final behavior detection effect.Under above technological frame, existing technology
Scheme mainly designs more preferable network structure, algorithm flow and training method etc..Currently, two stage object detector is mainstream
Method is primarily based on picture and proposes that several may include the region of object, then classified and return to obtain each region
Classification and coordinate.Classical algorithm has Faster R-CNN and R-FCN.When target detection is applied to the scenes such as behavioral value,
One challenge is the characteristic information learnt from the huge different objects sample of dimensional variation to Scale invariant.Also have one in recent years
A little object detection methods are suggested, for solving the problems, such as the scale invariability of network.Feature pyramid network is by establishing nerve
The feature pyramid of network different depth obtains high-resolution characteristic pattern, and according to target sizes different depth characteristic pattern
On detected, improve the detection effect of Small object；Image pyramid training method uses different scale by picture scaling
Under picture training network, allow the network to the feature for acquiring Scale invariant, improve detection effect.But the above technology all needs
It to be calculated on high-resolution characteristic pattern or picture, cause such methods to infer that speed is very slow, it is difficult to reach real-time detection；
And simply merge multiple dimensioned feature and will affect the stronger semantic information of network deep layer, the detection effect of big target is damaged,
The mode of multiple dimensioned training needs the longer training time, and convergence rate is slower.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind to be based on target detection
Students ' behavior detection method.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of students ' behavior detection method based on target detection is applied to smart classroom, comprising the following steps:
S1, the data set comprising students ' behavior information is established, the students ' behavior includes raising one's hand, stand and sleeping；
S2, students ' behavior detection model is established, which is the improvement Faster R- based on residual error network ResNet-101
CNN model；
S3, collect based on the data, students ' behavior detection model is trained using online difficult sample method for digging；
S4, video to be measured is detected using the students ' behavior detection model after training, obtains students ' behavior result simultaneously
Visualization；
Wherein, the core network of the Faster R-CNN model uses residual error network ResNet-101, the residual error net
Network ResNet-101 uses multilayer feature convergence strategy, and the 5th convolution stage of the residual error network ResNet-101 includes more
A branch of different sizes with receptive field.
Further, this method application places are place of education, and the place of education includes classroom.
Further, using the empty convolution with different voidages in the branch method particularly includes: several branches
The empty rate score used is spaced apart from small to large, and the object of different scale size is distributed to point with corresponding voidage
Branch.
Further, the multilayer feature convergence strategy is i.e. by the 3rd, 4 convolution of the residual error network ResNet-101
The feature in stage is merged with lifting feature figure resolution ratio.
Further, the feature in the 3rd, 4 convolution stage is merged specifically:
Four points neighbouring in 3rd convolution stage output are merged into channel dimension, then special with the 4th convolution stage
Sign is spliced, and the input dimension of fused characteristic pattern channel dimension and the 5th convolution stage is made finally by the convolution of 1x1
Unanimously.
Further, online difficulty sample method for digging described in step S3 specifically:
S301, random acquisition include the data of several frame image datas；
S302, data input students ' behavior detection model is subjected to propagated forward, N number of area-of-interest of generation；
S303, the loss for calculating each area-of-interest, are ranked up based on loss, and it is highest several to choose loss
A area-of-interest carries out backpropagation, undated parameter；
Further, the detailed process that data set described in step S1 is established are as follows:
S101, by the video composition data library comprising having recorded students ' behavior；
It S102, is that frame marks out frame middle school student's behavior classification and frame for each frame image by the video slicing；
S103, the data that record a behavior classification and location information are saved for each frame image, according to PASCAL VOC
Data set format arranges and stores the data, obtains data set.
Further, the video shoots under different scenes, different camera angles and different illumination conditions and obtains.
Compared with prior art, the present invention have with following the utility model has the advantages that
(1) it is multiple branches with different feeling open country size that the conv5 of residual error network ResNet-101 of the present invention, which is expanded,
Using the detection head of such scale-sensitive, in the case where not increasing additional parameter, the object of each scale can be used
Specific branch detection, improves that this kind of for place of education there are the detection effects of the biggish students ' behavior of different scale；
(2) present invention uses neural network multilayer feature convergence strategy, and the feature of conv3 and conv4 are merged,
It is not involved with that operand is biggish mutually to sum it up multilayer convolution by element, not only ensure that the semantic information of network deep layer, but also improve
The resolution ratio of characteristic pattern under the premise of not losing big target detection effect greatly improves and raises one's hand, sleeps etc. compared with Small object
Recall ratio and accuracy rate；
(3) more difficult training sample is paid close attention in the training process using the training method of online difficult sample in the present invention,
Such as the sleep sample of negligible amounts, low resolution, the sample of raising one's hand of many attitude etc. are improved unbalanced in training sample classification
In the case of e-learning effect, thus promoted to be relatively difficult to walk for detection effect.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is the conv5 schematic diagram of ResNet-101；
Fig. 3 is the target scale distribution map of students ' behavior data set；
Fig. 4 is students ' behavior detection model structure chart；
Fig. 5 is that the students ' behavior under true classroom scene detects visualization result schematic diagram.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to
Following embodiments.
The present invention provides a kind of students ' behavior detection method based on target detection, as shown in Figure 1, comprising the following steps:
S1, the data set comprising students ' behavior information is established, the students ' behavior includes raising one's hand, stand and sleeping；
S2, students ' behavior detection model is established, which is the improvement Faster R- based on residual error network ResNet-101
CNN model；
S3, collect based on the data, students ' behavior detection model is trained using online difficult sample method for digging；
S4, video to be measured is detected using the students ' behavior detection model after training, obtains students ' behavior result simultaneously
Visualization；
1, students ' behavior data set
This method establishes a large-scale students ' behavior data set, the detailed process that data set is established are as follows:
S101, it will be imaged comprising having recorded video described in the video composition data library of students ' behavior in different scenes, difference
It shoots and obtains under brilliance degree and different illumination conditions；
It S102, is that frame marks out frame middle school student's behavior classification and frame for each frame image by the video slicing；
S103, the data that record a behavior classification and location information are saved for each frame image, according to PASCAL VOC
Data set format arranges and stores the data, obtains data set.
2, the network structure design of students ' behavior detection model
(1) core network and testing process
Students ' behavior detection model is on the whole by basic network (Backbone Network) and detection head
(Detection Head) is constituted.The former provides the expression of image different size, different abstraction hierarchies as feature extractor；
The latter is then associated with according to these expressions with supervision message study classification with position.It detects the responsible class prediction in head and position is returned
Two tasks are returned often to carry out parallel, the loss for constituting multitask carries out joint training.
The basic network (core network) of the present embodiment uses residual error network ResNet-101, including 5 convolution stages, inspection
Flow gauge is divided into two steps, and the first step suggests network (Region Proposal Network, RPN) the sense of access by a region
Interest region (Region of Interests, RoI), second step carry out RoIPooling operation to these regions, i.e., will be different
The region of size is converted to the feature vector of regular length, obtains class probability and bezel locations using R-CNN network, wherein
RPN and R-CNN shares the output of residual error network ResNet-101, reduces calculation amount, while precision improves, also improves
The speed of detection.
(2) the detection head of scale-sensitive
Detection head is the portion for being responsible for carrying out class prediction to the feature extracted and position returns in target detection network
Point, the 5th convolution stage conv5 that head includes residual error network ResNet-101 is detected in students ' behavior detection model.
The present invention designs the detection head of a kind of pair of scale-sensitive, for enhancing model for different scale larger samples
Detection effect.As shown in Fig. 2, the 5th convolution stage conv5 of the residual error network ResNet-101 includes 3 branches, In
Different size of receptive field is obtained using the different empty convolution of voidage in each branch, for detecting the object of different scale
Body.
Empty convolution is one kind of convolution, on characteristic pattern in every line or one column access with convolution kernel carry out convolution.Sense
It is the area size that the pixel on the characteristic pattern of each layer of output of convolutional neural networks maps on original image by open country；By three
The parameter sharing of a branch obtains different receptive field sizes, the ginseng of model in the case where not introducing new convolution nuclear parameter
Quantity is few, and the speed of training and test is fast.
In the present embodiment, the filter that the convolution kernel size in three branches of conv5 is 3 is replaced with into voidage respectively
The object of different scale is distributed into corresponding branch, for slight ruler for different receptive fields for 1,2 and 4 filter
Degree movement is such as raised one's hand, the branch detection that the smaller i.e. voidage in use feeling open country is 1；For large scale movement as stood, use feeling
The branch detection for being 4 by wild larger i.e. voidage.By using this detection head to scale-sensitive, do not increasing additional ginseng
In the case where number, specific branch detection can be used to the object of each scale, it is larger for different scale to improve network
Sample detection effect.
The data set comprising students ' behavior information established in the present embodiment is from the place of education such as classroom, place of education
There are target resolutions it is low, scene is complicated the problems such as, as shown in figure 3, in data set horizontal axis be sample object size account for whole figure
The ratio of piece, the longitudinal axis are the ratio that different sample object sizes account for whole samples, and the scale of a half data only accounts for whole picture
0.2%, 10% target scale only accounts for the 0.13% of whole picture, and 90% target scale accounts for the 1.7% of whole picture, ruler
Degree variation has reached 14 times.The data for detecting low resolution are very challenging, therefore are merged using neural network multilayer feature
Strategy improves the detection effect of small scale movement in the case where hardly increasing calculating consumption.
(3) neural network multilayer feature convergence strategy
Residual error network ResNet-101 uses multilayer feature convergence strategy, and the multilayer feature convergence strategy i.e. will be described residual
The feature of the 4th convolution stage conv4 of the 3rd convolution stage conv3 of poor network ResNet-101 is merged to promote spy
Figure resolution ratio is levied, specifically:
Using the strategy of Re-organize, operated by reshape, by layer export in neighbouring four points be merged into it is logical
In road dimension, then spliced with C4 layers of feature, makes fused characteristic pattern channel dimension and C5 layers finally by the convolution of 1x1
Input dimension it is consistent, entire students ' behavior detection model structure chart is as shown in Figure 4.
The present embodiment middle school student's behavioral data collection includes to raise one's hand, stand, sleeping, due to really educating the sample slept in scene
This is less, so data set is there are more serious classification is unbalanced, number of samples of raising one's hand is to sleep 15 times of number of samples, influences
The result precision of final students ' behavior detection model；In addition, there is also difficulty or ease imbalanced training sets in data set, behavior of raising one's hand
Since resolution ratio is low, and there are more to block, and detection difficulty is much larger than the biggish standing behavior of target.
(4) training of students ' behavior detection model
The students ' behavior data set established in step S1 includes a variety of behavioral datas such as raise one's hand, stand, sleeping, due to true
The sample slept in the scene of classroom is fewer, so notebook data collection, there are more serious classification is unbalanced, number of samples of raising one's hand is
15 times of sleep number of samples, affect the effect of final students ' behavior detection algorithm.In addition, there is also relatively tight in the data set
Weight difficulty or ease imbalanced training sets, behavior of raising one's hand is since resolution ratio is low, and there are more to block, detection difficulty much larger than target compared with
Big standing behavior.Therefore present invention uses the instructions of Online Hard Example Mining (online difficulty sample excavates)
The mode of white silk, pays close attention to more difficult training sample, such as the sleep sample of negligible amounts, low resolution, many attitude in the training process
Sample of raising one's hand etc., improve relatively be difficult to walk for these for detection effect.
Students ' behavior detection model is trained using online difficult sample method for digging specifically:
S301, random acquisition a batch include the data of two frame image datas；
S302, data input students ' behavior detection model is subjected to propagated forward, obtains the characteristic pattern of picture；The spy
Sign figure judges that anchors belongs to prospect or background by the softmax layer in RPN network, and frame is recycled to return amendment
Anchors obtains accurate area-of-interest RoIs；The input of RoIs and characteristic pattern one as subsequent R-CNN network.
S303, the loss for calculating each area-of-interest RoIs, are ranked up based on loss, and it is highest several to choose loss
A area-of-interest carries out backpropagation, and undated parameter repeats S301, until convergence.
Table 1 illustrates accuracy rate effect of the present invention on the students ' behavior data set of classroom, and wherein Ours indicates the present invention
Method.As can be seen that the method for the present invention is compared to some existing methods (such as tradition Faster R-CNN, R- under the scene of classroom
FCN and FPN network model) accuracy rate is higher, and behavioral value effect is more preferable.
Table 1
| mAP | |
| Faster R-CNN | 54.2 |
| R-FCN | 50.6 |
| FPN | 55.8 |
| Ours | 57.6 |
Fig. 5 is students ' behavior detection visualization result of the invention under true classroom scene, and visual information includes learning
Raw behavior type and respective confidence, various student's rows in video frame can accurately and comprehensively be detected by illustrating the invention
For.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that those skilled in the art without
It needs creative work according to the present invention can conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel are available by logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Technical solution, all should be within the scope of protection determined by the claims.
Claims (8)
1. a kind of students ' behavior detection method based on target detection, which comprises the following steps:
S1, the data set comprising students ' behavior information is established, the students ' behavior includes raising one's hand, stand and sleeping；
S2, students ' behavior detection model is established, which is the improvement Faster R-CNN based on residual error network ResNet-101
Model；
S3, collect based on the data, students ' behavior detection model is trained using online difficult sample method for digging；
S4, video to be measured is detected using the students ' behavior detection model after training, obtains students ' behavior result；
Wherein, the residual error network ResNet-101 uses multilayer feature convergence strategy, the residual error network ResNet-101's
5th convolution stage includes multiple branches with different feeling open country size.
2. a kind of students ' behavior detection method based on target detection according to claim 1, which is characterized in that described more
Using the empty convolution with different voidages in a branch with different feeling open country, for detecting the object of corresponding scale.
3. a kind of students ' behavior detection method based on target detection according to claim 2, which is characterized in that described point
Using the empty convolution with different voidages in branch method particularly includes: the empty rate score that several branches use is from small to large
It is spaced apart, the object of different scale size is distributed into the branch with corresponding voidage.
4. a kind of students ' behavior detection method based on target detection according to claim 1, which is characterized in that described more
Layer Fusion Features strategy is merged the feature in the 3rd, the 4 convolution stage of the residual error network ResNet-101 to be promoted
Characteristic pattern resolution ratio.
5. a kind of students ' behavior detection method based on target detection according to claim 4, which is characterized in that by the 3rd,
The feature in 4 convolution stages is merged specifically:
Four points neighbouring in the output of 3rd convolution stage are merged into channel dimension, then with the 4th convolution phase characteristic into
Row splicing, the input dimension one of fused characteristic pattern channel dimension and the 5th convolution stage is made finally by the convolution of 1x1
It causes.
6. a kind of students ' behavior detection method based on target detection according to claim 1, which is characterized in that step S3
The online difficult sample method for digging specifically:
S301, random acquisition include the data of several frame image datas；
S302, data input students ' behavior detection model is subjected to propagated forward, N number of area-of-interest of generation；
S303, the loss for calculating each area-of-interest, are ranked up based on loss, choose several highest senses of loss
Interest region carries out backpropagation, undated parameter.
7. a kind of students ' behavior detection method based on target detection according to claim 1, which is characterized in that step S1
The detailed process that the data set is established are as follows:
S101, by the video composition data library comprising having recorded students ' behavior；
It S102, is that frame marks out frame middle school student's behavior classification and frame for each frame image by the video slicing；
S103, the data that record a behavior classification and location information are saved for each frame image, arrange and store the data,
Obtain data set.
8. a kind of students ' behavior detection method based on target detection according to claim 7, which is characterized in that the view
Frequency shoots under different scenes, different camera angles and different illumination conditions and obtains.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910620750.1A CN110414380A (en) | 2019-07-10 | 2019-07-10 | A kind of students ' behavior detection method based on target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910620750.1A CN110414380A (en) | 2019-07-10 | 2019-07-10 | A kind of students ' behavior detection method based on target detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110414380A true CN110414380A (en) | 2019-11-05 |
Family
ID=68360948
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910620750.1A Pending CN110414380A (en) | 2019-07-10 | 2019-07-10 | A kind of students ' behavior detection method based on target detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110414380A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929649A (en) * | 2019-11-24 | 2020-03-27 | 华南理工大学 | Network and difficult sample mining method for small target detection |
| CN111144368A (en) * | 2019-12-31 | 2020-05-12 | 重庆和贯科技有限公司 | Student behavior detection method based on long-time and short-time memory neural network |
| CN111275592A (en) * | 2020-01-16 | 2020-06-12 | 浙江工业大学 | Classroom behavior analysis method based on video images |
| CN111507226A (en) * | 2020-04-10 | 2020-08-07 | 北京觉非科技有限公司 | Road image recognition model modeling method, image recognition method and electronic equipment |
| CN112597977A (en) * | 2021-03-02 | 2021-04-02 | 南京泛在实境科技有限公司 | HSV-YOLOv 3-based online class student behavior identification method |
| CN112686154A (en) * | 2020-12-29 | 2021-04-20 | 杭州晨安科技股份有限公司 | Student standing detection method based on head detection and picture sequence |
| CN113052165A (en) * | 2021-01-28 | 2021-06-29 | 北京迈格威科技有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN113052200A (en) * | 2020-12-09 | 2021-06-29 | 江苏科技大学 | Sonar image target detection method based on yolov3 network |
| CN113139530A (en) * | 2021-06-21 | 2021-07-20 | 城云科技（中国）有限公司 | Method and device for detecting sleep post behavior and electronic equipment thereof |
| CN113407670A (en) * | 2021-06-21 | 2021-09-17 | 福州大学 | textCNN-based method and system for detecting online learning behaviors of students |
| WO2024020774A1 (en) * | 2022-07-26 | 2024-02-01 | 江苏树实科技有限公司 | Model generation method, object detection method, controller and electronic device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth convolutional neural networks |
| CN107808376A (en) * | 2017-10-31 | 2018-03-16 | 上海交通大学 | A kind of detection method of raising one's hand based on deep learning |
| CN108986124A (en) * | 2018-06-20 | 2018-12-11 | 天津大学 | In conjunction with Analysis On Multi-scale Features convolutional neural networks retinal vascular images dividing method |
| CN109902677A (en) * | 2019-01-30 | 2019-06-18 | 深圳北斗通信科技有限公司 | A kind of vehicle checking method based on deep learning |
| CN109961034A (en) * | 2019-03-18 | 2019-07-02 | 西安电子科技大学 | Video object detection method based on convolution gating cycle neural unit |
-
2019
- 2019-07-10 CN CN201910620750.1A patent/CN110414380A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth convolutional neural networks |
| CN107808376A (en) * | 2017-10-31 | 2018-03-16 | 上海交通大学 | A kind of detection method of raising one's hand based on deep learning |
| CN108986124A (en) * | 2018-06-20 | 2018-12-11 | 天津大学 | In conjunction with Analysis On Multi-scale Features convolutional neural networks retinal vascular images dividing method |
| CN109902677A (en) * | 2019-01-30 | 2019-06-18 | 深圳北斗通信科技有限公司 | A kind of vehicle checking method based on deep learning |
| CN109961034A (en) * | 2019-03-18 | 2019-07-02 | 西安电子科技大学 | Video object detection method based on convolution gating cycle neural unit |
Non-Patent Citations (3)
| Title |
|---|
| 寇大磊等: "基于深度学习的目标检测框架进展研究", 《计算机工程与应用》 * |
| 张超等: "残差网络下基于困难样本挖掘的目标检测", 《激光与光电子学进展》 * |
| 李文等: "SLEEP GESTURE DETECTION IN CLASSROOM MONITOR SYSTEM", 《ICASSP 2019 - 2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929649A (en) * | 2019-11-24 | 2020-03-27 | 华南理工大学 | Network and difficult sample mining method for small target detection |
| CN110929649B (en) * | 2019-11-24 | 2023-05-26 | 华南理工大学 | Network and difficult sample mining method for small target detection |
| CN111144368A (en) * | 2019-12-31 | 2020-05-12 | 重庆和贯科技有限公司 | Student behavior detection method based on long-time and short-time memory neural network |
| CN111275592A (en) * | 2020-01-16 | 2020-06-12 | 浙江工业大学 | Classroom behavior analysis method based on video images |
| CN111275592B (en) * | 2020-01-16 | 2023-04-18 | 浙江工业大学 | Classroom behavior analysis method based on video images |
| CN111507226A (en) * | 2020-04-10 | 2020-08-07 | 北京觉非科技有限公司 | Road image recognition model modeling method, image recognition method and electronic equipment |
| CN111507226B (en) * | 2020-04-10 | 2023-08-11 | 北京觉非科技有限公司 | Road image recognition model modeling method, image recognition method and electronic equipment |
| CN113052200A (en) * | 2020-12-09 | 2021-06-29 | 江苏科技大学 | Sonar image target detection method based on yolov3 network |
| CN113052200B (en) * | 2020-12-09 | 2024-03-19 | 江苏科技大学 | Sonar image target detection method based on yolov3 network |
| CN112686154B (en) * | 2020-12-29 | 2023-03-07 | 杭州晨安科技股份有限公司 | Student standing detection method based on head detection and picture sequence |
| CN112686154A (en) * | 2020-12-29 | 2021-04-20 | 杭州晨安科技股份有限公司 | Student standing detection method based on head detection and picture sequence |
| CN113052165A (en) * | 2021-01-28 | 2021-06-29 | 北京迈格威科技有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112597977A (en) * | 2021-03-02 | 2021-04-02 | 南京泛在实境科技有限公司 | HSV-YOLOv 3-based online class student behavior identification method |
| CN113407670B (en) * | 2021-06-21 | 2022-07-08 | 福州大学 | textCNN-based method and system for detecting online learning behaviors of students |
| CN113407670A (en) * | 2021-06-21 | 2021-09-17 | 福州大学 | textCNN-based method and system for detecting online learning behaviors of students |
| CN113139530B (en) * | 2021-06-21 | 2021-09-03 | 城云科技（中国）有限公司 | Method and device for detecting sleep post behavior and electronic equipment thereof |
| CN113139530A (en) * | 2021-06-21 | 2021-07-20 | 城云科技（中国）有限公司 | Method and device for detecting sleep post behavior and electronic equipment thereof |
| WO2024020774A1 (en) * | 2022-07-26 | 2024-02-01 | 江苏树实科技有限公司 | Model generation method, object detection method, controller and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414380A (en) | A kind of students ' behavior detection method based on target detection | |
| CN106127204A (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN109410242A (en) | Method for tracking target, system, equipment and medium based on double-current convolutional neural networks | |
| CN108428229A (en) | It is a kind of that apparent and geometric properties lung's Texture Recognitions are extracted based on deep neural network | |
| CN109241982A (en) | Object detection method based on depth layer convolutional neural networks | |
| CN107492095A (en) | Medical image pulmonary nodule detection method based on deep learning | |
| CN110321891A (en) | A kind of big infusion medical fluid foreign matter object detection method of combined depth neural network and clustering algorithm | |
| CN110516539A (en) | Remote sensing image building extracting method, system, storage medium and equipment based on confrontation network | |
| CN107610087A (en) | A kind of tongue fur automatic division method based on deep learning | |
| CN107609575A (en) | Calligraphy evaluation method, calligraphy evaluating apparatus and electronic equipment | |
| CN107507170A (en) | A kind of airfield runway crack detection method based on multi-scale image information fusion | |
| CN103268607B (en) | A kind of common object detection method under weak supervision condition | |
| CN104268140B (en) | Image search method based on weight self study hypergraph and multivariate information fusion | |
| CN109615604A (en) | Accessory appearance flaw detection method based on image reconstruction convolutional neural networks | |
| CN107767416A (en) | The recognition methods of pedestrian's direction in a kind of low-resolution image | |
| CN107463881A (en) | A kind of character image searching method based on depth enhancing study | |
| CN110084812A (en) | A kind of terahertz image defect inspection method, device, system and storage medium | |
| Sloan et al. | Does base map size and imagery matter in sketch mapping? | |
| Zhang | Practice Teaching of Landscape Survey Course Based on eCognition Remote Sensing Image Interpretation* Technology. | |
| He et al. | VLMAE: Vision-language masked autoencoder | |
| CN112818777B (en) | Remote sensing image target detection method based on dense connection and feature enhancement | |
| CN110188662A (en) | A kind of AI intelligent identification Method of water meter number | |
| CN107506370A (en) | Multi-medium data depth method for digging, storage medium and electronic equipment | |
| Luo et al. | Distinguishing different subclasses of water bodies for long-term and large-scale statistics of lakes: a case study of the Yangtze River basin from 2008 to 2018 | |
| Mao et al. | The greener the living environment, the better the health? Examining the effects of multiple green exposure metrics on physical activity and health among young students |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 200030 Dongchuan Road, Minhang District, Minhang District, Shanghai Applicant after: SHANGHAI JIAO TONG University Address before: 200030 Huashan Road, Shanghai, No. 1954, No. Applicant before: SHANGHAI JIAO TONG University |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191105 |