Stereo Object Detection and Tracking Using Clustering and Bayesian Filtering
Mobile robots equipped with sensors are increasingly being deployed in real-world application domains. Vision is a high-fidelity source of information for robots in comparison to more popular range sensors. Despite the extensive research in computer vision literature, 3D range maps obtained from stereo cameras are still not fully exploited for robot applications. The research described in this paper focuses on using stereo depth maps to detect and track objects. First, a clustering algorithm is applied to cluster image points based on the 3D range information, thereby detecting candidate objects. Next, Bayesian filtering techniques are used to incorporate motion cues, filter spurious objects and track moving objects in the scene. The algorithms are implemented and evaluated on data obtained from a stereo camera mounted on a wheeled robot platform.