Research Of Visual Object Tracking In Image Sequences

Object detecting and visual tracking is an important direction in computer vision field. It's a great challenge to investigate a robust, precise, stable, and applicable object tracking algorithm, which is both theoretically and practically valuable.In the thesis, the algorithm of background subtracting, frame difference algorithm, motion estimation based on optical flow field and object tracking based on active shape contour are investigated. On the basis of the former work, we improve the background subtraction algorithm. Also an algorithm which combines both model matching and feature matching is put forward. The algorithm uses the object contour in previous frame as the reference template of current frame. Based on the fact that object has a continuous track in movement, object's current position can be predicted based on previous position and then match the reference template around the predicted position. The contour which matches the reference template best is the object contour of current frame. The experiment results show that the algorithm which combines the model matching and feature matching can fulfill the object matching and tracking assignment in complex background with relatively good stability and reliability.