Fast Template Matching Method with Heuristic Initialization Oriented to Railway Patrol Robot

This paper proposes a fast target template matching method based on best buddies similarity to realize the integration of railway patrol robot and video surveillance system for railway safety assurance. Railway patrol robots are a viable way to address the challenges of blind spots in the field of vision and dependence on manual line inspection in video surveillance systems. The video surveillance system covers the whole line and can quickly identify the target. This supplements the limited vision of the robot. The multifunctional sensors carried by the patrol robot provide flexible on-site disposal methods for the monitoring system. In this process, the key issue is how to hand over the target quickly and accurately between the two systems. Existing methods are not suitable for our scenario because cross-device and cross-viewpoint targets have serious non-rigid deformation due to factors such as perspective, illumination, and occlusion. We propose a fast target template matching method based on best buddies similarity, which utilizes edgebox to screen out texture-rich regions in overlapping fields, and introduces region inevitability-guided particle swarm optimization to search for targets, thereby achieving the acceleration. The method utilizes color-feature-based best buddies similarity as the fitness function and can handle non-rigid deformations scenarios. We verify the performance of our method by carrying out comparative experiments and simulation experiments on real railway scenes. Our method achieves fast and robust target handover while maintaining the accuracy of the original BBS algorithm and improving its robustness. Our method has advantages such as reducing human labor, improving efficiency, and enhancing railway safety.