基于相对位置不变性的SIFT高效匹配算法研究

为了实现全自主机器人立体视觉导航，图像匹配的准确性和快速性成为了研究热点和难点。通过移动机器人工作环境研究，提出图像匹配相对位置不变性的原理，基于这种原理，对最近邻域和次近邻域的SIFT特征点匹配算法进行了改进。先将待匹配图像（前后帧）所有特征点按Y方向像素值大小排序，再从对应位置关系的局部区域搜索SIFT特征点，如果最近邻和次最近邻的比值满足一定阈值T，则该点为匹配点，然后再通过相对位置不变性去除误匹配点。改进算法在最近邻匹配点和次近邻匹配点搜索时避免全局搜索而大大提高实时性，通过相对位置不变性基本去除所有误匹配点。通过实验验证，匹配速度和正确率大大提高，是平面移动工作环境下高效实用的匹配算法，同时，该算法稍加改进对复杂环境也是适用的。

Efficient Matching Algorithm of SIFT Based on Relative Position Invariance

In order to realize the stereo vision navigation of the autonomous robot, the accuracy and rapidity of image matching become the hotspot and difficulty of the research. From the study of the work environment of the mobile robot, the relative position invariance of image matching are put forward. Based on the above principle, the matching algorithm based on the nearest neighbor and the second-nearest neighbor SIFT feature point is improved. The first, the all features of the pair matching images (the before and after frame images) are sorted by pixel coordinate value of the Y axis. Then, the SIFT feature points are searched from the local area of corresponding relation. If the ratio of the nearest and second nearest neighbor meets certain threshold T, the point is the pair matching point. And then, the error matching points are eliminated by the relative position invariance. The improved algorithm avoids to search the nearest neighbor and the second-nearest neighbor SIFT within the global area, so the real-time performance and is improved greatly and the error matching points are eliminated mainly. The experiment result shows that the matching speed and accuracy are greatly improved. it is an efficient and practical matching algorithm in planar mobile work environment, meanwhile it is fit for the complex work environment .