Research On The Algorithms Of Moving Object Detection And Tracking In Video Sequences

With the development of the intelligent algorithms nowadays, intelligent traffic monitoring system has increasingly attracted people's attention. But the technology is not mature yet, it is far from the fully intelligent, but people eagerly hope the intelligent traffic technology can not only create a smooth and orderly traffic environment for the city, but also reduce the occurrence of traffic accidents. As the basis and premise of intelligent traffic monitoring system, moving object detection and tracking has become a hot research topic in the academic field. Therefore, the subject made a study in the related technologies of moving object detection and tracking in video sequences, and a more in-depth research on object detection.The thesis made a deeper study and research on the background modeling and updating, for the classical background difference method is greatly affected by its speed and accuracy. Firstly, for background modeling, proposed a modeling method based on interval distribution density, which is mainly based on grey value classification. To begin with, classify all gray values of the current pixel in background training sequence, then select the background subclass to get the initialized background by calculating the interval distribution density. Experimental results show that this method can obtain an authentic background model, good real-time performance, and better adaptability to road environment. Secondly, for the background updating, put forward a background reconstruction method which is completed in the foreground, first of all, extract the moving object mask, then fill the dynamic area of foreground use the corresponding module of initialized background through the mask, to complete the reconstruction of the background image. The update rate of this method is higher, and is not easy to be affected by the shaking leaves or other disruptive factors in the background, lay a good foundation for the subsequent target detection.Aimed at the results of frame difference method are easy to appear cavity inside of the targets, the thesis proposed an edge detection algorithm of moving object based on eight directions translation method, which can directly extracted the edge of the dynamic objects in the foreground, and is not affected by the background objects. Then combined it with the frame difference to detect the moving targets. Results show that the method can achieve a good real-time performance, and higher detection accuracy without need of background modeling and updating.In target tracking, due to the Camshift algorithm is just based on the color feature of object to search the matching target, the rates of miss and false tracking are high; the Kalman filtering algorithm has the ability to predict the position information of the tracking target in the next few frames. Therefore, the thesis adopted the method that combines Camshift and Kalman filtering algorithm to track the target. Experimental results show that the combination of this two algorithms is more stable and reliable, and the problem of miss and false tracking is less to occur.