Human Gait Classification Using Compressed Video Data

• Main Author: 林永欽
• Other Authors: 韓欽銓
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/23731005245097845935

碩士 === 中華大學 === 資訊工程學系碩士班 === 91 === In the thesis, human gait classification in compressed domain has been proposed by using the compressed video data in MPEG-1 format. Detection and tracking of moving objects are the primary steps in video surveillance. They are frequently implemented in spatial domain. It needs lots of redundant time to uncompress the video data when the compressed video data are received. The detection and tracking algorithms in compressed domain must be developed to solve this problem. First, moving objects are detected by subtracting the DC values in I frames from those in background. In addition, the DC values of background are also adapted to avoid the noise and illumination change. Second, the tracking process is performed among the consecutive frames. Motion vector information extracted from P frames is used to predict the next position of moving objects. The overlapping table is constructed to determine the relationship among moving objects. In analyzing the behavior of human, the motion vectors and the velocities of human extracted from P and B frames are utilized to be the feature vectors for gait classification. These features are inputted to train the hidden Markov models of gait classifiers. Four kinds of gait behaviors, including hopping, limping, running, and walking, are identified from the video sequences. The experimental results are given to demonstrate the effectiveness of our proposed method.