Research And Implementation Of Abnormal Behavior Detection Method For Examination Room Surveillance Video Based On Deep Learning

In recent years,with the development of science and technology,the application of surveillance cameras has become more and more extensive.Massive amounts of monitor video data have emerged in all walks of life.At present,in the field of education,due to the wide application of cameras in the examination room,large examinations such as the middle school entrance examination and the college entrance examination use the invigilator auxiliary camera to ensure the fairness and impartiality of the exam.When faced with a large amount of examination room surveillance video data,it is unrealistic to rely on manual recognition of the examination room surveillance video,which will consume a lot of manpower and material resources and is not efficient.Therefore,in order to save labor costs and achieve more efficient monitoring,video abnormal behavior detection technology came into being.By collecting real examination room surveillance video data,this paper designs an abnormal behavior detection method for examination room surveillance video by using deep learning methods,and develops a corresponding examination room monitoring system to effectively identify candidates’ behavior,so as to better ensure the fairness and impartiality of the examination.The exam room surveillance video data in this article comes from real exam room monitoring.Due to the different examination rooms,there are certain differences in the position and angle of the camera in each examination room,so the resolution of candidates in the video frame is also different,and the resolution of candidates who are closer to the camera is larger,while the resolution of candidates who are relatively far away from the camera is relatively small,which belongs to small objects.In addition,due to the large number of candidates in the examination room,coupled with the camera angle problem,there will be a certain degree of occlusion between the candidates,which undoubtedly increases the difficulty of the detection task.At present,due to its particularity and confidentiality,the industry lacks sufficient annotation datasets,and a large part of the premise of good performance based on deep learning methods is due to the support of data,which makes the construction of a dataset for examination room surveillance videos become the primary task of work.In addition,because the video data comes from the real examination room scene,the number of abnormal behavior samples of candidates is small,and the number of normal action samples of candidates is much greater than the number of abnormal action samples,that is,there is data imbalance.In order to solve the above problems,this paper proposes a method for detecting abnormal behavior of examination room surveillance video based on deep learning,which is mainly as follows:In this paper,the data for the monitoring video of the examination room are sorted out,the labeling rules are formulated,and the Examination_Object dataset used in the object detection stage and the Examination_Behavior dataset for the detection of abnormal behavior of candidates are constructed according to the labeling rules,which provides strong data support for the model.This paper proposes a two-stage method for detecting abnormal behavior in examination room video,including the object detection stage and the behavior detection stage.In the object detection phase,in order to distinguish the invigilator from the test taker,this paper examines the candidate and the desk he or she is on as a whole,so as to focus on the action video clips formed by each candidate.In order to solve the multiscale problem in object detection and improve the detection ability of the detector for small objects,the feature pyramid structure is added in the feature extraction process to integrate the feature layer information of different scales,so as to realize the prediction of objects of different sizes on different feature layers,and effectively solve the problem of multi-scale object detection in the examination room environment.Aiming at the serious problem of occlusion between candidates in the examination room scene,the classification loss is optimized on the basis of the cascade network,and PolyLoss is introduced to improve the model training effect,so that the model achieves a high recognition accuracy rate for invigilators and candidates on the examination room object detection dataset.In the behavior detection stage,a novel feature extraction module is proposed,which combines multi-scale visual transformer with feature pyramid,and adopts pooled attention mechanism to realize the downsampling operation.In order to improve the recognition ability of the behavior detection model to the behavior of small objects and increase the sensitivity of the model to the behavior of small objects,3D convolution is innovatively used in the feature pyramid network,and video feature extraction is carried out with multi-scale visual transformer.In addition,this paper proposes a multiobject classification head module that can be used for object-level action classification to effectively classify object-level actions.At the same time,by adopting the weighted cross-entropy loss function and weighted sampling strategy,the corresponding weight coefficient is set according to the number of samples of each action category in training,which effectively alleviates the problem of data imbalance and improves the recognition ability of the model for abnormal behavior.A variety of data augmentation methods are adopted to increase data diversity and effectively alleviate the overfitting phenomenon.In this paper,an examination room monitoring system for examination room surveillance video is designed.By obtaining video stream data through monitoring equipment,the model can process the video data to obtain the candidate’s behavior category and location information,and return the visualization results,which helps the relevant staff to find possible abnormal behaviors of the candidates,and assist the invigilator in proctoring,and is of great significance for better ensuring the fairness and impartiality of the exam.