This paper describes a method of realtime person tracking in a cluttered background by integrating optical flow and uniform brightness regions. Optical flow is extracted at points with enough contrast in an image. Assuming that the target person moves at a sufficiently large angle to the optical axis, the target can be detected in the image as a region where flow vectors are nearly uniform. Uniform brightness regions are extracted as connected points where optical flow cannot be obtained due to lack of contrast. At each frame, tracking is performed by continuously updating the flow and the uniform brightness regions of the target based on the prediction of those regions by using the target motion in the image. As long as, at least, one of these visual cues is effective to distinguished from the background and other objects, the target person can be tracked. Our proposed method was implemented on a realtime image processor with multiple DSPs and successfully tracked a target person in realtime.
