基于GMS和块匹配的运动估计特征匹配算法研究

针对运动估计中传统特征匹配算法存在匹配时效性差、精度不高等问题,提出一种基于块匹配搜索的改进网格运动统计算法.首先提取ORB特征点并将图像划分网格;然后计算网格内各特征点的运动平滑约束度,并以此作为准则确保匹配精度;最后采用块匹配菱形搜索算法进行特征匹配筛选以提高匹配速度.仿真试验结果表明:相较于随机抽样一致算法,特征点保持数量为500时匹配效率提升24.6%,匹配速度提高42.9%;与ORB-SLAM2算法相结合用于连续运动估计时,单帧耗时0.13 s,实时性较好.

Pose Estimation Technology Research Based on GMS and Block Matching

In response to issues like low matching precision, slow speed in traditional feature matching algorithms, and inadequate real-time performance estimation, this paper introduces an improved algorithm that combines Grid-based Motion Statistics and block matching search. Firstly, ORB feature points are extracted and the image is partitioned into a mesh. Then the motion smoothing constraint for each feature point in the mesh is calculated as the criterion to ensure the matching precision. Finally, the block-matching diamond search algorithm is used for perform feature matching screening to enhance the matching speed. The experimental results show that, compared with Random Sample Consensus algorithm, the proposed algorithm improves the matching rate by 24.6% and the speed by 42.9% with 500 feature points, When combined with the ORB-SLAM2 algorithm for continuous motion estimation, the single-frame processing time is 0.13 s, demonstrating good real-time performance.