Research And Implementation Of Abnormal Behavior Detection Method In Examination Room Video

The video surveillance system has been widely used in various examinations in universities,middle schools and primary schools.However,as far as the present situation is concerned,the video surveillance system in the examination room can only play the role of recording examinations.How to use this system to detect the abnormal behavior of candidates in examinations and thus create a fair and just examination environment is a great challenge in the field of video surveillance and even in the field of computer vision.This paper studies how to effectively identify abnormal movements of candidates,and the main work is summarized as follows.Firstly,this paper analyzes the environment of the examination room and the abnormal behavior categories of candidates,and constructs a data set with 8 kinds of abnormal actions of candidates;In order to solve the problem of missing joint points and shaking joint points in candidates’ postures extracted by OpenPose,a posture completion algorithm based on Lagrange and Newton interpolation polynomials is presented in chapter 3.Different interpolation methods are used to solve the problem of missing posture to some extent.Aiming at the problem that the current motion recognition method based on graph convolution can’t capture the association features between limbs,this paper designs a combined limb subgraph to represent the spatial structure between two limbs.The spatial features of limb subgraphs and combined limb subgraphs are extracted by using the spacetime graph convolution network,and the spatial features and relevance features of each frame skeleton are aggregated according to the weight,and then the aggregated features are convoluted in time.In the training stage,in order to optimize the model,the residual structure is added to each spatio-temporal unit,which avoids the problem of subgraph eigenvalue loss caused by gradient disappearance to some extent.In order to verify the motion recognition method based on limb subgraphs,experiments were carried out on SAA data set constructed in this paper with four limb subgraphs and six limb subgraphs respectively.The results show that Topl’s results are 41.5%,which is 3.1%higher than the model using six limb subgraphs,which verifies the effectiveness of this model for the first time.Secondly,this paper verifies on NTURGB-D and Kinectics data sets,and the experimental results in CS are 6.4%higher than ST-GCN and 0.4%higher than PB-GCN,which verifies the availability and universality of this model.Finally,this paper designs and implements the examinee abnormal behavior detection system.Its functional modules include attitude annotation subsystem,attitude estimation subsystem,examinee abnormal behavior recognition subsystem and manual judgment subsystem.The labeling subsystem can label the outline and attitude information of candidates in order to train the attitude estimation model suitable for the examination room;The attitude estimation subsystem is used to extract the pose of candidates;The examinee’s abnormal behavior recognition subsystem integrates the LSG-GCN algorithm given in this paper to recognize the examinee’s abnormal action.The manual judgment subsystem realizes the zoom function of the image,and can further confirm the suspicious video segment.