Research On Dynamic Target Detection And Tracking Algorithms Based On The Regional Characteristics

Dynamic target detection and tracking technology is the main content in the area of research on the digital image and computer vision. It has a wide range of applications in the fields of traffic control system, video monitoring system, the robot system and so on. The dynamic target tracking in image sequences is determining the loacation, size or shape information of the object in each frame image according to space information of the video image.The subject mainly analyses the regional characteristics of the image sequence, which is divided into target detection and target tracking.In the dynamic target detection, first, introduced the commonly used target detection algorithm including the frame difference method, the background difference method and the optical flow method. Then, analysed each methods and pointed out their advantages and disadvantages. At last, proposed the algorithm which combines three frames difference with background difference. This algorithm can extract more complete dynamic object.In the dynamic target tracking, the subject focuses on studying the Mean Shift algorithm. First, introduced the basic ideas and principle of Mean Shift. Then, proposed two dynamic target tracking algorithm: One is target tracking algorithm based on the Kalman filter and Mean Shift. Basing on the traditional Mean Shift tracking algorithm, the new algorithm adjusted the kernel function bandwidth to change the size of tracking window for ensuring accessing to the full amount of information of moving target. Meanwhile, determined whether the block occurred by Bhattacharyya coefficient. When the target was blocked, introduced the Kalman filter and taked the target location of the previous frame as observations of the filter, predicted target location of the current frame through the filter, and then iteratively calculated the target location by Mean Shift algorithm to find the optimal position which is as observations of the next frame filter calculate the next frame. The algorithm achieved tracking the moving objects accurately in case of shelter and target deformation.The other is Mean Shift tracking algorithm based on Harris corner detection. This algorithm used Harris corner detection algorithm to extract the target corner and taked the corners as the main features of the target, and enhanced the differentiation bewteen target and background, thus restrained the interference of background on the target tracking and increase tracking performance of the Mean Shift algorithm in the complex environment. Experiments showed that the algorithm can reduce the number of iteration, and has the good real-time and accuracy.