Research And Application Of Abnormal Behavior Detection In Examination Room Based On Deep Learning

With the development of computer vision and artificial intelligence,image recognition and understanding based on video surveillance has been widely studied and applied.Nowadays,video surveillance has been widely applied in the examination room abnormal behavior monitoring,but at present,the main arrangement of electronic monitoring and on-site examination room invigilator combined,after the examination video replay inspection,its human and material costs are huge,easy to happen the omission situation.The research of computer vision in video understanding provides a way to solve this problem.At present,some scholars have applied the traditional abnormal behavior recognition method to the abnormal behavior detection in the examination room,but its recognition effect needs to be improved.Based on human bone feature points,thesis improves the spatio-temporal graph convolution network(ST-GCN),and uses deep learning technology to build a multi-attention mechanism based spatio-temporal graph convolution human behavior recognition model.Verified the effect of the proposed model on the different public action recognition datasets,improved accuracy and declined the parameters.Combined with Open Pose algorithm,a set of abnormal behavior detection system in the examination room is developed to assist the invigilation,achieved good results in building a dataset of abnormal behavior in self-made dataset and real examination dataset.Thesis mainly does the following work:(1)Made an examination abnormal action dataset based on human bone feature points.The dataset contains 7 types of abnormal actions,300 normal test clips and 1104 abnormal test action clips.(2)ST-GCN human behavior recognition algorithm based on multi-attention mechanism is studied.The proposed method embedded multi-attention mechanism in STG-CN network.The method was verified on different conventional datasets.The results showed that not only scaled down the parameters,also improved accuracy.(3)Designed and developed a set of abnormal behavior detection system in the examination room assisted invigilator by using Open Pose algorithm combined with ST-GCN algorithm under the multi-attention mechanism.Marked abnormal behaviors in the examination room in the video,and saved the abnormal behavior information to the background for later reference.After testing on self-made dataset and real examination dataset,the proposed method has considerable effect.