Moving Target Detection And Tracking Algorithm Based On Image Sequences

Motion target detection and tracking is one of the most important subjects in computer vision, which combines advanced technologies in image processing, pattern recognition, automatic control, artificial intelligence, computer and other relative fields. It has broadly applied in military visual missile guidance, video surveillance, medical image analysis, intelligent transportation and other fields, so this subject research has important theoretical significance and practical value.This paper mainly aims to study algorithm of detection and tracking moving object in image sequences. First detection of moving objects in stationary scene is discussed, and then target tracking based on correlation tracking technique and mean shift algorithm are detailed studied. Considering existing problems, effective improved algorithms are presented, so tracking stability and robustness are increased. The major works of this paper are summarized as follows:In the detection of moving objects in static background, temporal difference, background subtraction, algorithm based on background image model, shadow detection and removal algorithm have been studied. Since object detection using temporal difference algorithm is not integrate, a detection algorithm based on block motion is adopted. Then binary image is obtained by adaptive threshold and through morphological processing more accurate object extraction is realized.In the study of correlation tracking algorithm, aiming to the multiple points correlation algorithm has computational complexity, which can not satisfy real time tracking, an improved SSDA algorithm using adaptive threshold sequence is proposed. Change traditional line scan template to search from the center of the template image. Only pixels whose value are one corresponding to the binary template image are taken part in the matching calculation, similarity is completely concentrated on the target pixels, so computational complexity is reduced and matching precision is increased. Then we adopt pattern size correction and dynamic template update, which ensure the accuracy of tracking. Then mean shift tracking algorithm is studied, aiming to mean shift algorithm does not use the motion information of target which may fail to track target when there are serious disturbances, an improved target tracking algorithm based on mean shift and target position prediction is proposed. According to different disturbance circumstances, adopt different scale factors to combine Kalman filter prediction result with mean shift tracking result. The improved algorithm makes good use of space position of the target, so tracking reliability is increased.Based on mean shift algorithm, continuously adaptive mean shift algorithm is then studied. This algorithm can adjust scale with object during tracking process, but only applies to track target in simple background. An improved Camshift algorithm is proposed. Three dimensional background weighted histogram is built in HSV color space, then input image is converted to color probability distribution image. According to target velocity, combine color probability distribution image and differential image adaptively. When target moves rapidly, the weights of motion cue is higher, otherwise the weight of color cue is higher when target moves slowly, which effectively overcome the disturbance of the background color.