| With the coming of the digital information age, video information contains a largeamount of visual information and presents the explosive development. The increasinglyhigh demand for video related processing technology is proposed. Much technology ofcomputer visual field such as human-computer interaction, video surveillance and virtualreality are based on video track. It makes video track commonly use in video processingtechnology. However, the diversity of the track target’s characteristics, the complexity ofthe environment, which the track target is in, and the limitation of the tracking algorithmsincrease the difficulty in video track. Therefore, a good effect in video tracking is alwaysthe focus of people’s attention and study on video tracking has some practical value.Mean Shift algorithm is a tracking algorithm based on estimates of the probabilitydensity gradient rising. It has attracted much attention because of its simple and fastpattern matching characteristics. But Mean Shift algorithm also has some shortcomings.The main purpose of this paper is to research and improve partial shortcomings of MeanShift algorithm.From the semi-automatic point of view that Mean Shift algorithm needs to manuallychoose the tracking region, the detection of moving targets in the initial tracking frame isproposed. In view of the fact that in the field of video tracking such track targets asvehicles and faces are the common track objects, in the initial frame it can betransformed into static target detection. This paper focuses on the study of Adaboostalgorithm, a kind of the static object detection algorithm, with the face detection as theexample. Then by combining Adaboost algorithm with skin color model and introducingFFS sample training algorithm, the efficiency of the Adaboost algorithm is improved inthe detecting face.Mean Shift algorithm uses the color feature to model the tracking target. It’s lack ofenough information about tracking target. So when the background exists regions that their color distribution is similar to the tracking target, it’s easy to failure in tracking. Butcomplex background always has this situation. To solve this problem, after analyzing thetracking principle of Mean Shift algorithm, this paper integrates color characteristics andHarris corner feature to improve the matching degree of tracking feature and reduce themismatching phenomenon in the case of complex background.This paper theoretically analyses the two major inadequacies of Mean Shiftalgorithm, and uses experiments to verify correctness of the improved algorithm. |
