Ad-RMS: Adaptive Regional Motion Statistics for Feature Matching Filtering

Bin Nan, Yinghui Wang, Yanxing Liang, Min Wu, Pengjiang Qian, Gang Lin

Abstract: The Grid-based Motion Statistics (GMS) is a popular feature matching filtering method, and can effectively support 3D reconstruction systems such as ORB-SLAM and has been used effectively in many fields. However, the GMS divides the image into a certain number of grids with a fixed size, which cannot better reflect the feature information of the region. So that when large affine changes occur in the images, the grids cannot delineate a reasonable consistent region. Such a region division leads to errors and even failures in the subsequent process using the grid to reject error feature matching. As a consequence, this paper proposes an adaptive regional motion statistics method based on adaptive region division for region detection to replace the fixed grid division, which enhances the affine invariance of the feature matching filtering algorithm, and verifies the effectiveness of this paper's method through the precision experiments of feature matching and homography matrix.