Human detection using motion and appearance based feature

An approach to detect moving and standing human from video is proposed in this paper. Human detection from videos is a difficult problem because of motion of human, camera and background. In order to detect moving human, the dense optical flow is calculated by two consecutive frames, to represent the motion of human. Motion based feature is extracted from optical flow field. It not only represents the global motion caused by the boundary of human body, but also contains local motion caused by the limbs. Motion based feature is combined with histogram of template feature, which is designed to detect standing human, as final feature for detection. Experiment on CAS dataset shows that this feature has more discriminative ability than other motion based feature. Besides, this feature is easier for hardware acceleration, which makes it suitable for real time application.