Research On Detection Method Of Pedestrian Abnormal Behavior In Video Surveillance System

Intelligent video surveillance plays an important role in maintaining the order of social public safety.How to detect abnormal pedestrian behavior in video is an important research direction in the field of computer vision.Although many researchers have achieved certain research results in the detection of abnormal pedestrian behavior,it is difficult to extract the motion feature information of pedestrian behavior under complex environment changes,which has led to slow progress in the detection of abnormal pedestrian behavior.Based on this,this paper aims at the inaccurate extraction of pedestrian moving targets caused by the flickering pixels,shadows,and ghosts generated by the ViBe algorithm under high dynamic background changes,and studies an improved ViBe algorithm based on the original ViBe algorithm.Subsequently,on the basis of studying the structure and function of the basic convolutional neural network and the long and short-term memory network,a new pedestrian running abnormal behavior detection model based on the fusion of 3D convolutional neural network and LSTM neural network was explored.The specific summary is as follows:1)In the extraction of pedestrian moving targets,the ViBe algorithm is easily affected by the environment under high dynamic background changes.To solve this problem,an improved ViBe algorithm is proposed.In the algorithm design,the background dynamic change is used to define the matrix for detecting the degree of flicker,and the adaptive threshold radius is set to design the threshold adaptive update strategy;the RGB-HSV color space conversion is used to perform the shadow detection judgment designer;with the help of the correlation coefficient characteristic comparison histogram The method of graph similarity detects ghost images.Numerical comparison of experiments shows that the algorithm has low computational complexity,good robustness,and can effectively extract relatively complete pedestrian motion targets.2)Aiming at the inaccurate recognition of abnormal pedestrian running behavior in surveillance video,and the inability to effectively combine the characteristics of the abnormal running behavior in the space and time dimensions,the improved ViBe algorithm combined with the deep learning network structure is proposed.A 3D-CNN and LSTM fusion model method suitable for abnormal pedestrian behavior detection.The fusion model uses 3D-CNN to obtain the spatial characteristics of pedestrian moving targets.LSTM uses the memory function of time to obtain the behavioral characteristics of pedestrian moving targets in time series.Finally,Softmax is used for classification.Experimental data shows that this method can effectively improve the accuracy of detecting abnormal pedestrian behavior.