Design Of Pan-tilt Target Tracking System By Means Of Fusion Camshift And Particle Filter Algorithm

Target tracking based on computer vision is one of the hotspot research issues in recent years. In a number of fields, such as smart surveillance, smart human-computer interation, militray reconnaissance, virtual reality and aerial photography, object tracking has important applications. However, the target tracking system universally has the problems that poor real-time and easily to be affected by the interference, which also restricts its wide application in the embedded system. So how to improve the real-time performance and robustness of the target system is the ugent to be solved. So the main contributions of this thesis are as follows:First, we design and implent a visual-guided pan-tilt target tracking system, sepecially,this sends the target position to the pan-tilt with the camera mounted on the pan-tilt. Then the pan-tilt keeps the target at the center of the image according to the position feedback, so as to achieve real-time target tracking.Second, to ensure the accuracy and rapidity of the target system, a tracking strategy which combines an improved Camshift and particle filter algorithm has been proposed. With the Bhattacharyya distance between the target feature and the candidate target feature, the strategy can identify the miss tracking happens. When the Bhattacharyya distance is less than the setting threshold, it uses the efficient Camshift algorithm to ensure real-time tracking. When the Bhattacharyya distance is more than the setting threshold, the particle filter algorithm is engaged to ensure the global optimality so that the system can recovery from miss tracking.Third, in order to reduce the observation noise of the image sensor and the delay to the feedback control system, this thesis combines Kalman algorithm and the angular velocity value of the attitude sensor to predict the optimal target position. Consequently, it effectively reduces the dynamic error and gets better tracking results.Finally, the experimental results for the vision-guided target tracking system indicate that the rapidity and accuracy of the system is greatly increased with the application of the the strategy proposed in this thesis and Kalman filter. Besides, results of tracking fast moving targets experiments show that the system can track the moving target quickly, accurately and steadily. When the verticle distance between camera and the target is 1.8 meter, the system can track the target which line speed is 1.25 meter per second steadily which indicates the superiority performance of the target track system in tracking fast moving target.