Multisensor Information Fusion for People Tracking With a Mobile Robot: A Particle Filtering Approach

People tracking based on multisensor information fusion is addressed. A framework is presented for fusing the laser range finder (LRF) data and the monocular camera data. Based on this framework, an LRF-based detection algorithm is proposed to identify the pairs of human legs, by combining motion information and metric features. Moreover, a geometric observation model is developed for the camera to extract both the range and bearing measurements of the target person by focusing the target's shoes with the camera. Then, a near-optimal particle filter is designed to fuse the measurements from the LRF and the camera. To prevent the sample impoverishment, a procedure of sample diversity improvement is used after the resampling step. The full occlusion problem is solved using image matching based on speeded up robust feature. Note that either of the LRF and the camera can work independently, since both the range and bearing are simultaneously acquired from the LRF or the camera. As a result, flexible and robust tracking can be achieved. Extensive experiments demonstrate that the proposed approach achieves high tracking accuracy and robustness. Especially, only a very small number of particles suffice to maintain good tracking performance.