Map Fusion Method Based on Image Stitching for Multi-robot SLAM

摘要

Compared with the single-robot SLAM, the SLAM task completed by a multi-robot system in cooperation has the advantages of more accuracy, more efficiency and more robustness. This study focuses on the map fusion problem in the multi-robot SLAM task, which is to fuse the local maps created by multiple independent robots into an integrated map. A multi-robot SLAM map fusion method based on image stitching is therefore proposed. A single robot uses lidar SLAM to build a local environment map and upload it to a central node. The central node then maps each local map from a two-dimensional occupancy grid map to a grayscale image. The SuperPoint network is used to extract the depth features from the grayscale images, and the transformation relationships between the local maps are calculated via the feature matching. The matching topology graph is used to realize the final map fusion. It carries out experimental verification in the indoor environment on three mobile robots, which were developed by our own, and the experiment proved that the method has good real-time performance and robustness. After obtaining the global map, some new robots were placed in the environment, and realized the task of multi-robot target search by using the relocalization function.