Enhanced Grid Map Correction Using an Improved ORB Algorithm

This paper addresses the challenge of grid map accuracy errors in multi-map merging, focusing on developing a robust and consistent occupancy grid map. To enhance feature point detection, the oFAST algorithm usingadaptive thresholding is proposed, which amplifies edge responses and increases detection accuracy. Additionally, descriptors derived from local self-correlation matrices are integrated into the BRIEF algorithm to improve robustness in complex environments. Matched feature points are refined using a K-Nearest Neighbors strategy combined with distance threshold filtering, followed by RANSAC-based estimation of the homography matrix to further ensure matching precision. A correction algorithm is proposed, which adjusts the coordinates of the matched feature points and fits a correction curve, thereby refining the pixel structure of the entire grid map. The proposed algorithm is validated by experimental data.