Research On Campus Conflict Behavior Based On Video Image Processing

From analog cameras to deep smart cameras combined with AI,video surveillance is evolving rapidly.At present,video surveillance has the characteristics of intuitive content,flexible layout and diverse scenes.Campus monitoring has been very perfect now,but as the frequent campus violence,watching monitor artificially is unable to know the situation at the first time.Therefore,abnormal behaviors can be analyzed for the video stream,and the actions of people in the video can be judged by the algorithm.If students are in a fight,the staff of the monitor center can order the staff and teachers in relevant areas to deal with it in time,so as to reduce the occurrence of campus conflicts and ensure the safety of students.This thesis analyzes abnormal fighting behavior,and the work is as follows:(1)For the video of the CASIA behavior database,the Vibe+ algorithm with better robustness and best extraction effect is selected for moving target detection.Next,the MeanShift target tracking algorithm is used to track the moving targets in the framed area frame by frame.(2)This thesis improves on the basis of the traditional optical flow method.By increasing the distance relationship of the moving target and the constraints of facial expression recognition,and comparing with the traditional optical flow method for behavior recognition,the algorithm runs faster and the recognition accuracy is better than the traditional optical flow method.(3)Based on the convolutional neural network structure model,this thesis proposes a behavior recognition algorithm that directly takes images as input without the need for a feature extractor.By building a model and training positive and negative sample data sets,the test results are compared and the recognition rate is better than the traditional optical flow method.