Topological Feature Based Object Matching for Traffic Light Detection Using Multi-view Camera System
Liyang Gao, Yeqiang Qian, Chunxiang Wang, Ming Zhu Yang · 2021 IEEE International Conference on Robotics and Biomimetics (ROBIO) · 2021
A telephoto image with a high object resolution can help detect long-distance traffic light objects earlier, which is of great importance to improve traffic efficiency and safety of automated vehicles. However, it usually cannot tackle close-range detection because of its narrow field of view, and vice versa. For this reason, a multi-view camera system composed of a telephoto camera and a wide-angle camera is proposed in this work. Furthermore, we propose a strategy for two cameras to work alternately. And the proposed topological feature based object matching algorithm and a post-fusion algorithm are introduced during the switching process. The matching and fusion algorithms ensure the information interaction and consistency of the results of the two cameras. The proposed camera switching strategy and the object matching algorithm are evaluated on a real-world dataset.