Object tracking based on multi information fusion
Zheng Zhao, Wcihai Chen, Xingming Wu, Jianhua Wang · 2015
Real-time object tracking is a critical task in many computer vision applications. So far, many conventional algorithms have been developed for real-time object tracking, and most of them are based on visual recognition. Unfortunately, these algorithms could easily fail in some specific circumstances when used individually. It is often an effective approach to fuse the multi information of different sensors, or to fuse the different algorithms to solve the mentioned problem. In this paper, a simple but effective algorithm that fuses the information of camera and laser range finder is proposed, in which visual recognition and laser detection are combined together. In visual recognition, Camshift (continuously adaptive mean-shift) algorithm and SURF (Speed up robust features) algorithm are both applied. In laser detection, coordinate information is used as supplement for object tracking because the detection range is limited if we only use camera. By this means, the tracking object can still be detected when out of camera's view. The proposed algorithm promotes the tracking performance by integrating camera and leaser range finder together. Meanwhile, the object's accurate position in real world coordinate can be figured out according to the visual information and laser information. In order to prove the efficiency of the proposed algorithm, vehicle tracking experiments are carried out.