Study Of Real-Time Object Tracking Algorithms In Image Sequences

Real-Time Object Tracking in Image Sequences is a classic problem in Computer-Vision, it means that you need to decide the real time position of the object according to a particular model in continuous image sequences. It attracts military attention initially, and it has been used in missile guiding and fire control systems etc. In recent years, with the development in computer-vision and artificial intelligence, it has wide applications in intelligent robots, electronic monitor systems, traffic control, and medical appliance etc. This paper focuses on three object tracking algorithms to solve some current problems for the further study.The object tracking based on correlation-based gray level template matching algorithm and kernel tracking based on Mean-Shift algorithm are discussed. A template scale-adaptive method is proposed and it leads to a better result in the experiment. And an adaptive noise-model is introduced to improve the performance of Kalman prediction filter. Then the kernel bandwidth adaptive method is improved and a new kernel function is given to get a better similarity curved surface. At last, a new tracking algorithm based on PCA and edge detection is proposed, experiments indicate that it can increase the number of frames in tracking greatly and has a good anti-noise ability.