The Research Of Motion Object Detection And Tracking Technology In Serial Images

Motion object detection and tracking in serial images is the main research field in computer vision and image coding, which has been widely applied in military, medicine and scientific research etc. The accuracy and stability of tracking effect depend on the design of algorithms to a great extent. It is still a hot point to research an accurate and robust algorithm in computer vision.First, the thesis studies the tracking algorithm based on difference. It makes use of strong correlations in serial images and obtains motion information from difference images. It is high-speed and convient to operate. Hower, it's disadvantageous in suppressing noise. Difference method can also be used for initial segmentation in multi-object detection and tracking or be integrated with other methods.Then, the thesis analyses template matching method deeply. Though full search method is accurate, it's inefficient. So the thesis provides an improved method, which combines with difference method and multi-resolution image matching and introduces prediction technique to reduce search branches and enhance the processing speed. The simulation results show that the improved method is accurate, high-speed and real-time when template and object are quite similar.Finally, in order to solve the problem which template matching method doesn't adapt for the changing of object's motion state, the thesis analyses the characteristics of serial images and puts forward another matching method based on colour features. It combines with difference method to enhance the speed. The simulation results show that the method is fast and stable.