The whole algorithm process is described as in figure 3. In the preprocessing process, we use Otsu binarization method [1] to the acquired frames and then we dilate the binary mask by a 5×5 rectangle kernel to fill up holes. After that connected areas that are larger than 5000 pixels in area are discarded and connected areas in the mask are considered as glass chip candidate areas for later use. Then we use optical flow method to track the moving glass chip in the image sequence. Jianbo Shi el al. proposed a method to find good features to track [2] and we use the method to find feature points. Glass chip candidate area is used as region of interest in the featurefinding process and can greatly increase the efficiency of the good-features-totrack algorithm. The acquired feature points are passed to a pyramidal Lucas-Kanade feature tracker [3]. Finally, we analyze the track result to determine whether there is glass chip moving in the image sequence.