A Method of Fast and Accurate Occupancy Grid Maps Merging in a Large-Scale Scene

In large-scale environments, efficient and accurate map merging is essential for the collaborative operations of multiple autonomous robots. This paper proposes a novel featurebased methodology for merging Occupancy Grid Maps that addresses key challenges such as computational efficiency, map quality, and robustness. The approach leverages local feature extraction using the AKAZE and ORB algorithms to detect and describe feature points. These features are matched across maps, and the Random Sample Consensus algorithm is employed to eliminate outliers, ensuring precise map alignment. Additionally, a deep learning-based map quality evaluation method, utilizing the VGG16 convolutional neural network, provides a robust mechanism to assess the quality of the merged maps, outperforming traditional pixel comparison techniques. Experimental results conducted in large-scale industrial settings demonstrate that the proposed method significantly improves the accuracy and efficiency of map merging compared to conventional methods, such as SIFT, KAZE, and BRISK. Furthermore, the paper introduces a reversible distributed framework for map merging, which enhances computational resource optimization and error correction capabilities. These contributions present a substantial advancement in multi-robot mapping, offering a reliable solution for autonomous exploration in complex environments.