Feature-based object tracking with an active camera

Abstract

This paper describes a new feature-based tracking system that can track moving objects with a pan-tilt camera. After eliminating the global motion of the camera movement, the proposed tracking system traces multiple corner features in the scene and segments foreground objects by clustering the motion trajectories of the corner features. We propose an efficient algorithm for clustering the motion trajectories. Key attributes for classifying the global and local motions are positions, average moving directions, and average moving magnitude of each corner feature. We command the pan-tilt controller to position the moving object at the center of the camera. The proposed tracking system has demonstrated good performance for several test video sequences.