Walking pedestrian tracking among several pedestrians by dynamic camera

• Authors: K. Chamnongthai;S. Boonrat
• Venue: International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems
• Year: 2005

In tracking a walking pedestrian among multiple walking pedestrians by movable camera, the pedestrian-target determination is sometimes mistaken, and it needs real-time image processing. This paper proposes an approach for tracking a walking pedestrian among multiple walking pedestrians with dynamic camera mounted on a pan platform. The objective of this approach is to correctly select the target, and maintain target within field of view. The approach has three processes, consisting of motion extraction, target prediction and camera movement respectively. The motion extraction with background compensation is performed to determine the groups of moving edges. In the target prediction, target position is predicted by using appropriate size window to identify target among several target candidates and to increase the accuracy. The final calculated distance, and direction against target and error are cooperated to determine the predicted target position. At the final process, the camera is moved in order to locate the center of the view at the target center. The experiments were demonstrated by tracking a walking pedestrians at 1 m/s speed and five meters distance from camera. Each experiment roughly consumed 480 ms/frame processing time at 8.5 rev./s motor speed. The results showed that the system error was 5-8 % for five walking patterns of a walking pedestrian situation. In the situation of multiple walking pedestrians, the demonstration were performed by tracking a pedestrian among two and three pedestrians, the results revealed that the system succeeded in maintaining the target within the field of view excepting the cases that the movement was suddenly changed in term of direction and speed.