Research On Moving Object Detection Technology For Intelligent Visual Surveillance

The aim of object detection algorithm is to detect, locate and recognize foreground moving objects quickly and accurately in video sequences captured from cameras. This algorithm always runs without manual intervention in method of computer vision. The technology of object detection has a wide spectrum of applications in intelligent visual surveillance, and is also the basis of other advanced vision task. Therefore, it becomes a hot topic in academic circle. However, the actual scene is always so complex that the existing algorithms cannot meet the robust requirements for a long time monitoring.Based on the analysis of the existing object detection algorithms, the research is focusing on object detection and object classification problems in case of local illumination changes and background perturbations in this thesis. Main contents include:(1) This thesis analyzes four commonly used moving object detection models which are based on background model, frame-difference motion analyzing, optical flow field and machine learning. The thesis also elaborates the advantage and disadvantage of each model, and introduces several typical algorithms.(2) To solve the problems of illumination changing, background disturbance and long time robust monitoring, this thesis purposes a moving objects detection algorithm using Gaussian mixture model and iterative division. Experiments show that the the accuracy of moving object detection algorithm in this thsis is batter than existing algorithms in case of illumination changing and background disturbance.(3) This thesis studies the algorithm based on HOG feature and SVM classifier,and use this algorithm to classify the foreground moving objects. This method can increase the speed and robustness of human detection. Experimental results demonstrate that the method increase the speed and the robustness of human detection in the case of illumination changing and background disturbance.(4) Based on object detection algorithm which studied in this thesis, this thesis applies object detection in intelligent visual surveillance system. Finally, the prototypes for detection of illegal intrusion, classification of foreground moving objects and detection of extraordinary density are implemented.