Research And Implement Of Real-time Tracking System For Movine Object Based On DSP

Visual object tracking(VOT) is an important research content of computer vision. VOT has important application in the areas of military guidance, radar tracking, video surveillance and traffice monitoring. Visual object tasking can simulate the human visual perception, recognize and identify the moving object, which provides an important basis for analyzing and understanding video. However, we frequently encounter the problems in actual object tracking: the target is transformed, the object lost or shielded. These prolems bring difficulty to achieve object tracking. So the accuracy and real-time is significance to the object tracking.Basing on the former studies, this thesis applys Mean Shift, Kalman filter(KF) and particle filter(PF) to develop the object traking system based on DSP. High stability and fast in calculation ensure Mean Shift is a real-time method in object tracking. KF has a good performance in target tracking, and this method overcomes the shortcoming of the Wiener filter requiring infinite previous information. So KF can meet the requirements of real-time. PF also has been widely used in object tracking with its strong adapation in non-linear and non-Gaussian system. The mainly research work and contributions of this thesis are as follows:(1) In object detection, this thesis analysis several objetc detection algorithms, and gives the simulation and performance of the algorithms.(2) Research on the basic theory of Mean Shift, KF and PF, and gives the performance and simulation of the algorithms.(3) In order to reduce computation complexibility and improve the performance of real-time, we combine Mean Shift and KF. KF can reduce the candidate target area effectively, Mean Shift can improve the real-time and accuracy of the system.(4) Build an object real-time tracking system based on DSP, we implement several algorithms on hardware system.(5) Compelete system debugging and software optimization, and improve efficiency and real-time of the system.The results of the experiments illustrate that the above methods can meet the real-time and accuracy requirements.