Research On Application Of Kalman And Mean Shift In Dynamic Target Tracking

Dynamic target tracking, as an interdisciplinary of digital image processing, pattern recognition and deep learning, is the key and difficult point of machine-vision field. It is widely used in the fields such as:Intelligent transportation, safety monitoring, human-computer interaction, military industry and medical care.The difficulty of the dynamic target tracking lies in the fact that the target shape in video is changeable while it is influenced by light intensity and occlusion of other objects. All these problems affect the tracking accuracy and real-time performance seriously. And therefore, it is urgent to seek a tracking algorithm which can overcome the difficulties above.In this paper, the kalman tracking algorithm and mean shift tracking algorithm are compared in multiple experiments through the research on dynamic target tracking under complex scene. It gave a deep analysis on the track feature of the two tracking algorithm for tracking characteristics under different light intensity, static and dynamic background and the occlusion interference. Experiments show that the two tracking algorithm can overcome most of the interference on the premise of the accuracy and real-time performance, but mean shift tracking algorithm will lead tracking failure problem when the dynamic target is severely occluded. Kalman tracking algorithm takes moving direction and the speed of the moving object as the characteristic information and it can predict the target location more accurate when the occlusion is more serious. Based on that, the Kalman filter and mean shift tracking algorithm are combined, and the size of the scale factor is changed according to the different degree of occlusion. The target positions gained from the two algorithms are linearly weighted, and the real position of the target is obtained. The feasibility of the improved algorithm is verified by experiments.