Research On The Recognition Method Of Student Status And Abnormal Behavior In Campus Environment Based On Deep Learning

With the rapid development of education and the growth of population,people pay more and more attention to education,and with it,the population capacity on campus is increasing day by day.Meanwhile,the frequency of abnormal behaviors and dangerous actions by students will increase,and the pressure on campus security will increase.It is cumbersome and inefficient to mange colleges and universities only by artificial way.Campus monitoring has become a common means for colleges and universities to manage campuses.The continuously improved intelligent video monitoring and analysis system has played a certain role in the campus management department.While saving a lot of labor,it can monitor the large-scale campus environment in real time to prevent the occurrence of accidents makes up for the lack of manpower and material resources,assists managers in discovering dangerous behaviors and reduces the occurrence of emergencies.Based on the above considerations,this paper studies the recognition of student actions and abnormal behaviors in a campus environment based on deep learning.First of all,this paper aims at the problem of slow update iteration of hardware equipment on campus and the problem of large flow of people.It conducts research on multi-face recognition algorithms with low hardware requirements,fast training speed and high accuracy,and proposes a lightweight multi-face recognition algorithm based on ABASNet.On this basis,H-softmax is used to improve the real-time performance of the algorithm,which solves the problem that the identity of the school personnel can still be quickly and accurately realized under the condition of low hardware conditions.And verify the real-time performance and average accuracy of the proposed algorithm in the public data set.The adaptability of the algorithm is verified in the classroom environment and the dark light environment in the campus,and its light weight is verified in the embedded device.At the same time,to solve the problem that it is difficult to identify students without facing the camera,a multi-angle identification algorithm based on multi-angle ID is proposed,which integrates one stage target detection algorithm,Deep Sort algorithm,and ABASNet algorithm.In the process of forward inference,the convolutional layer and the BN layer are merged,and the performance of the algorithm is verified through ablation experiments and comparative experiments,and finally multi-angle identity recognition on campus is realized.Then,aiming at the problem that students' psychological pressure mainly comes from learning pressure,we carried out research on student status recognition and psychological pressure analysis in the classroom,and proposed a method of students' classroom behavior recognition based on MECN head,and combined with Multi-angele ID to count each student's classroom status and analysis of students' psychological pressure.Finally,in view of the slow early warning of abnormal behaviors and dangerous states by traditional campus monitoring platforms,the research on the identification methods of dangerous states and dangerous behaviors on campus is carried out.The Bisenet algorithm is used to identify the dangerous area,and the FL-RFCN-based small target human detection algorithm in the dangerous area is proposed.At the same time,in response to the problem of higher accidents for students with greater psychological pressure,detection and multi-level early warning are carried out for students entering dangerous areas.Finally,the MECN head and Multi-angle ID algorithm are applied to the abnormal recognition in the campus environment to prevent accidents on the campus and provide better protection for campus safety and management.