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

Recently, moving object detection and tracking in video sequences has been one of the hot topics in computer vision. It has been widely used in precisely guided weapons, intelligent surveillance and other military or civil use. However, the field relates closely to many other research fields and has various applications, and so despite people have done much in the field, there are still some important issues unresolved. Based on the research about moving object detection and tracking algorithms in video sequences, this paper does some improvement to the commonly used algorithms both at home and abroad.About the moving object detection, this paper pays more attention to the algorithm based on the Gaussian mixture background model. However, when using this algorithm to detect moving objects, the moving objects'presence during the initialization often results in low convergence speed. To solve the problem, this paper presents an improved detection algorithm. The improved method uses on-line K-means clustering algorithm to initialize the model. It also saves the memory space with the improvement to the matching rule and new Gaussian distribution generation rule during the model update. The experimental results demonstrate the improved algorithm can fast and efficiently detect moving objects, and has better robustness than the traditional algorithm.About the moving object tracking, this paper does research on two methods. One is the algorithm based on Kalman filter, and the other is Camshift tracking algorithm based on color histogram. Based on the research of the Camshift algorithm, we find it can not track moving object precisely under complex background with occlusion by other objects that have the same color features as the moving object. To solve the problem, we propose a new algorithm combining Camshift and Kalman filter. The experimental results show that the new method can effectively solve large-scale color occlusion.