Improved Tracking Algorithm For Multiple Targets Based On Camshift Algorithm Combined With Kalman Fliter

Moving object target detection and tracking are main issues of computer vision, which combine image processing, pattern recognition, artificial intelligence, automatic control and other fields of advanced technology. Visual guidance in the military, medical counseling diagnosis, video surveillance, intelligent transportation and other fields have wide application ,therefore the research of this topic has the important theory significance and actual value.The diversity of moving target and the complexity of their environment make the detecting and tracking become too difficulty. It becomes a classic problem to be solved within the field of computer vision and image processing.So, this paper focuses on the practicality of a specific algorithm to carry out the exploration of key issues. Therefore, this paper's work mainly covers the following :Firstly,in detection and label of multi-moving target, the characteristics of the various detection and label algorithms are studied, the advantages and disadvantages are analyze.As object detected is not full by the temporal difference algorithm ,so three differential combined with Canny edge detection algorithm are adpoted.This mathod uses Pixel Calibration to get target number, size, location. It can achieve accurate extraction of multiple targets, and make a good preparation for the tracking algorithm.Secondly,in object tracking aspect, the tracking algorithm of multiple targets,which based on Camshift and Kalman filtering algorithm, is studied. When target and background have a large area of similar color, or objectives are occlude, the quondam algorithm exists the serious missing track problems. So an advanced method is proposed. On the one hand, when it happens to the large area of similar color between the object and the background , ROI (region of interest) frame difference is adopted. Frame difference is used only in Kalman forecast region. On the other hand, when the target is severely blocked, Kalman prediction is used instead of the optimal location value calculated by Camshift, and the predictive value of Kalman Kalman filter is used as the observed value updated by Kalman filter update, it can effectively overcome the failure of Kalman filter when the severe block happens .Thirdly,because Camshift algorithm is the tracking methods based on objective color characteristics ,and can not handle grayscale images. So a multi-target tracking algorithm combined Kalman filtering with S (Smoothness) value is proposed. After comparing the trajectory of different object, the direction and location of the moving boject are used to determine the movement of objects in the current frame by target match.Finally, a moving target automatic recognition and tracking system is established. The system can test moving target detection and tracking algorithm .The moving target detection and tracking algorithm processing were realized on PC.In a word, this paper proposes the advanced algorithm based on Kalman filter and Camshift and effective strategy to deal with background interference. Experiment results showed that the algorithm can effectively reduce the background interference that leads to loss of target tracking, achieve real-time tracking ,and obtain better tracking performance.