视觉SLAM中ORB配准算法的研究

Research on ORB Registration Algorithm in Visual SLAM

针对ORB配准算法在视觉同时定位与地图构建(Visual Simultaneous Localization And Mapping,V-SLAM)中存在的效率低下及误匹配问题,提出了改进的ORB配准算法。首先,在提取特征点时,采用二分区域法对图像进行预处理,以缩短特征点提取的时间,同时通过构建图像金字塔以增加特征点的尺度不变性。其次,在配准特征点时,使用汉明距离结合改进的RANSAC算法完成特征点的配准,以获得更加精准的配准点对。最后,使用OpenCV对采集的图像进行验证。实验结果表明,改进的ORB配准算法可以缩短特征点的提取及配准时间,同时有效地剔除误匹配点对,提高了特征点配准的成功率,给V-SLAM中的位姿估计算法提供了良好的初始值。

An improved ORB registration algorithm is proposed to solve the inefficiency and mismatch of ORB registration algorithm in Visual Simultaneous Localization and Mapping(V-SLAM).Firstly,when extracting feature points,in order to shorten the time of extracting feature points,bipartite region method is used to preprocess the image.At the same time,the scale invariance of feature points is increased by constructing image Pyramid.Secondly,when registering feature points,in order to obtain more accurate registration point pairs,hamming distance and improved RANSAC algorithm are used to register feature points.Finally,the captured image is verified by using OpenCV.The experimental results show that the improved ORB registration algorithm can shorten the extraction and registration time of feature points,effectively eliminate mismatched point pairs as well,also improve the success rate of feature point registration,and provide a good initial value for the position and attitude estimation algorithm in V-SLAM.