Research On Pedestrian Detection And Tracking Technology In Video Surveillance

With the continuous expansion of the country's urbanization,a large number of people pour into the city,resulting in the aggravation of the population flow,and the people put forward higher requirements for personal and social security.The application of intelligent video surveillance system has been widely concerned,pedestrian detection and tracking technology is its core.The research idea of this paper is to separate the moving area from the background by moving object detection algorithm,then detect the pedestrian object in the moving area by pedestrian detection method,and finally transfer the pedestrian target information to the tracking algorithm to track the pedestrian.The specific contents of this paper can be summarized as follows:(1)Moving object detection.In this paper,the contrast experiments of inter frame difference method,adaptive Gaussian mixture background modeling method and ViBe algorithm are carried out.From the morphological integrity of motion region extraction and detection time,ViBe algorithm achieves a better balance.In view of the ghost problem in ViBe algorithm,this paper uses the difference of detection principle between frame difference method and Vie algorithm,and combines the three frame difference image with ViBe method after hole filling algorithm and morphological processing to accelerate the elimination of ghost.The experimental results show that the combination of the improved three frame difference method and ViBe can eliminate the false detection problem caused by ghost faster,and the detection result is more accurate.(2)Pedestrian detection.After extracting the moving area,it is necessary to recognize and detect the pedestrian objects in the moving area.This part mainly introduces two different detection methods based on machine learning SVM and deep learning YOLO3-tiny network,and focuses on the research of YOLO3-tiny network model based on deep learning.This paper takes YOLO3-tiny network model as the framework,modifies the output dimension of classifier of network model,and trains and verifies it on pedestrian data set to realize the detection based on YOLO3-tiny Pedestrian detection based on iny model.The experimental results show that the pedestrian detection effect based on YOLO3-tiny model is better than that of SVM,and the detection accuracy and speed of this method can meet the requirements of pedestrian detection in the actual scene.(3)Pedestrian tracking.After the pedestrian is detected,it is necessary to track the pedestrian effectively.This part mainly introduces the Meanshift algorithm and Camshift algorithm in the generative tracking method.Aiming at the situation of short occlusion and pedestrian interference with similar color in the moving process of pedestrian target,this paper uses the combination of Kalman and Camshift to improve the tracking stability of the original Camshift algorithm.Finally,the tracking algorithm is combined with the previous detection methods to realize the automatic detection and tracking of multiple pedestrians in a simple scene.