Fast relocation method for indoor mobile robots based on ORB feature matching

To address the issue of failure in localizing in similar areas using AMCL (Adaptive Monte Carlo Localization), this paper proposes a method that combines ORB (Oriented FAST and Rotated BRIEF) feature matching with AMCL for assisted relocalization. Firstly, the point cloud generated by the LiDAR sensor is transformed into an image. ORB feature matching is then performed between the local point cloud image and the map image, and the pixel mapping is utilized to obtain the vehicle's coordinates in the global map, achieving coarse localization based on ORB image features. Subsequently, this pose is used as the initial pose in the adaptive Monte Carlo localization algorithm, and particles are distributed around it. The particles are updated using motion and observation information until convergence, achieving precise localization based on particle filtering. This approach enables the robot to quickly achieve localization in similar environments, enhancing the robustness and efficiency of relocalization.