Examination of a tracking and detection method using compressed domain information

In this paper, we propose a moving object tracking method in the compressed domain. Because video data are large in size, it is usually compressed by H.264/AVC or MPEG for storage or transmission. Most image processing systems track targets in the pixel domain after decoding compressed data. Objects in video data are tracked based on their color and luminance information in the pixel domain. However, tracking methods in the pixel domain have higher computational complexity than in the compressed domain. We propose a method that directly employs moving object tracking in the compressed domain without a decoding process. Tracking algorithms in the compressed domain are usually based on motion information, or motion vectors. However, motion vectors are not generated when objects stop moving, and their accuracy is low. To overcome these problems, we propose a method that uses motion vectors and other information in macroblocks to track moving objects in a compressed H.264/AVC video stream. The proposed method can reduce computational complexity compared to a pixel domain algorithm. In addition, experimental results show that tracking accuracy is improved by the proposed method. Therefore, this method can be effectively used for video surveillance systems.