Research Of Moving Targets Detection And Tracking Based On Omnidirectional Vision

With the development of computer technology and image processing technology, video surveillance has become an important research topic of the field of computer vision. Moving targets detection and tracking is an important component of video surveillance. The information of the moving targets is obtained based on the omnidirectional vision in this paper. The moving targets detection is implemented through the change of pixel value in the region of sequence frame image. The tracking of moving targets is realized by the characteristics of the moving targets.The background subtraction technique is adopted for detection in this paper. Firstly background model is established through statistic method and updated in real-time. In this case background can be influenced by less disturbs and the background model can be initialized precisely. In the processing of image binarization, the threshold of difference image is obtained by self-adaptive algorithm. The region of moving targets can be divided integrally in case of the changes of light conditions. So the accuracy of targets detection can be improved.The traditional algorithm of connecting domain in binary images needs to label the equivalent labels with each other of the non-zero pixel points in neighborhood of current pixel point. The improved algorithm in this paper only needs to consider part of the situations of equivalent labels, which reduces the number of equivalent label obviously. The experiment results testify that the method can reduce the time of processing and satisfy the needs of real-time.Aiming at the existence of background interference in the region of moving objects, the geometric operators are designed for filtering in this paper. The experiment results testify that the method can filter the background interference effectively and improve the accuracy of targets detection. Also, it can reduce the computational complexity to improve the capability of moving targets detection and tracking in real-time.Aiming at the distorted moving targets in omnidirectional vision, the method is presented to improve the robustness of tracking targets by the feature fusion of the area of pixel of moving area, center coordinate, external rectangular and the color model of moving targets in this paper. Kalman filter is applied to predict position of pixel point in next frame image to improve matching precision and speed.The moving targets detection and tracking is implemented in this paper. The experiment results testify that the brought up method is practicable and accurate in targets detection and tracking in real-time.