Research On Improved Moving Target Tracking Forecast Algorithm Based On Camshift

Dynamic visual information is becoming an important part of environment information received by people, which also has become a crucial research topic of computer vision. Moving object tracking in dynamic video is a key problem in applications of visual and image encoding.In this thesis, we mainly study tracing algorithms of dynamic objects under fixed camera and an improved Camshift tracking algorithm of moving target detection is proposed. First, analysis of existing target detection method and tracking algorithm and conclusions of their advantages and disadvantages are made. Also we make conclusions of coming challenges in tracking algorithm fields and find the way to overcome their defects. Several matured target detection methods and target tracking methods are introduced. Then comprehensive analysis of classic Camshift tracking algorithm is made. Based on the analysis above our improved Camshift tracking algorithm is proposed. In this method, based on colorful background update, by using principal orientation which is obtained by edge gradient method, moving direction in Kalman model is modified. So combining prediction of moving direction in advance and Camshift tracking algorithm, the task of moving target tracking can be fully achieved.At last with two videos (one is shot by author herself and the other is downloaded from internet) dealt by MATLAB software under the proposed algorithm in this thesis the validity of this method can be deduced. By comparing with classic Camshift tracking algorithm the tracking performance is much better than classic Camshift. No matter how fast the object moves and how large deflection angle the object has, the algorithm always performs perfect.