Research On Detection Of Campus Irregular Behavior Based On Video

In recent years,due to the lack of school supervision,irregular behaviors have frequently appeared,which shows that the campus management method that only relies on rules and regulations to restrain students’ behavior has limited effect in practice.As a political institution with a strict management system,public security colleges and universities have continuously strengthened their internal affairs and strictly regulated student behavior since the promulgation and implementation of the new version of the "Public Security Organs and People’s Police Internal Affairs Regulations" in October 2021.Based on this,this paper takes the policing management of public security colleges as the background,based on deep learning action recognition method,automatically monitor students’ irregular behaviors in campus surveillance videos,improve the detection rate of irregular behaviors,and provide new ideas for optimizing college campus management methods.The main research contents of this paper are as follows.(1)Through literature research,the advantages and disadvantages of current action recognition methods are summarized,and an appropriate algorithm model is selected for the campus irregular behavior studied in this paper.(2)A action recognition method fused with target detection is studied.At present,the background environment of public data sets in the field of action recognition tends to be complex,and most action recognition algorithms do not pay enough attention to the core areas of data samples.The interference of background noise leads to low model recognition accuracy.Therefore,reducing the feature extraction area of the action recognition model to the core area of the human body can reduce the computational complexity and improve the performance of the model.Based on this,this paper introduces a target detection algorithm into the temporal difference network to automatically detect and crop out the human body region in the video image.The experimental results on the public dataset show that the recognition accuracy of the improved time difference network is 24.25% higher than that before the improvement.(3)A dataset of campus irregularities in the context of policing management is constructed.In different scenarios,the recognition results of action recognition algorithms will have different degrees of deviation.Therefore,the construction of data sets for specific scenarios plays an important role in the application research of action recognition algorithms.This paper selects public security colleges with strict and meticulous rules and regulations and strict requirements on student behavior as the research object,investigates typical irregular behaviors,and constructs a campus irregular behavior dataset.After data processing,5519 video data samples were finally obtained,which provided data support for the research of this paper.(4)The application scenarios of the action recognition method are expanded.At present,scholars at home and abroad have carried out application research based on action recognition algorithms in campuses.The background is mostly limited to indoors.In this paper,the action recognition method is combined with the policing management of public security colleges for the first time,and the improved time difference network is used to detect irregular behaviors in the surveillance video of the roads beside the campus of public security colleges.The results show that the recognition accuracy of the method proposed in this paper on the campus irregular behavior dataset reaches 96.91%,which proves that the method proposed in this paper is practical in practical scenarios.The research of this paper is expected to provide technical support for campus management,better improve and optimize the practical practice of policing management,and give full play to the police education function of the police academy.