A METHOD OF PEDESTRIAN DETECTION BASED ON RHYTHM OF WALKING

It is important for mobile robots and security purposes to discriminat e a pedestrian among moving objects in a video. This paper proposes a method which discriminates a pedestrian from other objects by using pe riodic motions of the feet, and the characteristics of the rhythm of p ace and stride. This method uses brightness changes in sequential imag es of a moving object contacting with the ground, and extracts the mai n component of the frequency using a power spectrum estimation. Theref ore, the method is independent of pedestrian's appearance such as clot hes, build, and hair style. The system realizing this method consists of three processes: moving object detection which uses the difference between successive image frames; tracking of the moving object using a moving model with an extended Kahman filter and an observation model; and detection of the pedestrian based on rhythm of walking using the difference of brightness between images on successive frames. The whol e process is performed in realtime. Experiments using real scenes conf irm the effectiveness of the proposed method: 446 pedestrians out of 4 70 were successfully detected (detection rate 94.9%), and 101 persons out of 106 were successfully detected as non-pedestrians (96.2%).