Abstract
This paper proposes a method of visual motion detection for vision chips which robustly detects the direction and the velocity of the motion of the objects under the real world. By using a spatial edge as a token, the method detects the motion from the correlation of the token. A computer simulation at the image processing level showed the method detects the motion in wide velocity range compared with the conventional methods. The method was implemented on an experimental vision chip using CMOS process. An experiment showed the vision chip robustly detects the direction and the velocity.