Research on Multi-Image Panorama Stitching Method Based on Improved ORB+GMS Algorithm

Panoramic image stitching is one of the pivotal tasks involved in merging multiple overlapping images into a panoramic view. The ORB (Oriented FAST and Rotated BRIEF) algorithm and the GMS (Grid-based Motion Statistics) algorithm are crucial techniques for image feature extraction and matching. This paper proposes an improved ORB+GMS method, employing the AGAST (Adaptive and Generic Accelerated Segment Test) corner detector to replace the FAST corner detector in the original ORB algorithm to enhance corner detection performance and robustness. As the GMS algorithm utilizes a 3x3 grid-based feature statistic and grid partitioning, this study employs the Descending Grid Scale Method for feature statistic computation to enhance matching speed. The improved ORB algorithm is integrated into the enhanced GMS algorithm, wherein a homography matrix is computed based on high-quality matched point pairs, and mismatched feature points are eliminated to improve matching accuracy. The proposed improved ORB+GMS method is applied to various sets of different image samples to validate its matching effectiveness. Experimental results demonstrate that the matching performance, image panorama stitching effect, matching time, and matching rate of the proposed fusion of GMS and ORB feature extraction and matching algorithm are significantly superior to those of the ORB algorithm and other similar algorithms. The matching rate can reach over 85%, with an increase in correctly matched points relative to the ORB and SIFT algorithms by 8.11% and 23.3%, respectively. The matching time is reduced to the level of 1.995 milliseconds, which is decreased by 2.468 seconds and 0.35 seconds compared to the ORB algorithm and the SIFT algorithm, respectively.