Research And Application Of The Algorithm For Face Tracking Based On Contour-corners

This paper is aimed at discussing the tracking issue for the multi-media teaching scene. The application is required to record the real-time teaching situation, and considered the teacher as tracking object. With this special scenario, the tracking algorithm is explored with single object and single camera which are both moving.In this article, an object tracking algorithm is presented based on contour-corner using optical flow, and using color model as the assistant tracking model. At the beginning of the process, color model is adopted to eliminate the obstruction from the non-color to gain the accurate scene. Next, the face region is detected using AdaBoost classifier with Haar features. The size and position of the face also can be found for the initial parameter of the tracking. Then, object edges are extracted using canny operator, and corners would be gained by computing gradient eigenvalue of the outlines'pixels in the binary image. Afterwards, these corners are considered as the initial points of the optical flow which are updated iteratively. Finally, Kalman filter is to regard as predictor to predict the size and location of the object for avoiding to losing object. Also, the size and location of track window is real-time updated.In addition, a camera controlled by the pan and tilt is used as the video capture device with pan from 0 to 360 and tilt from 0 to 60. The moving of the camera is dominated by the target position that is renewed by the Kalman filter.A large number of experiments are made both of the test situation and application scenes. The analysis about these tests shows that, the algorithm applied in the task is available to some extent. It can implement the tracking for turning around, back-moving. It also can capture and track the object when the target comes out again. So, there is some practice worth for improving the performance of tracking.