Abstract
A novel extraction method of the characteristic points for automatic tracking of human motion is proposed. The characteristic points are the points that represent the position and attitude of a human in each frame and can be employed to track the human motion. The extraction is performed in two steps as follows. First, the variance of color image data around each pixel of the human figure is calculated and the pixels that exceed a threshold variance value are extracted as candidate characteristic points. Second, the points are distributed throughout the human figure by prohibiting the existence of characteristic points in certain areas around the determined characteristic points. The proposed method has the advantage that the density of characteristic points can be controlled by adjusting the prohibited area. Human motion is confirmed to be traced using these characteristic points. Each characteristic point is extended to form a square block in order to calculate the degree of matching for the various position of the human figure in the following frames. These characteristic points can trace the human motion using only a few points.
