基于光流的景象匹配定位算法研究

A Positioning Algorithm Based on Scene Matching Using Optic Flow

在景象匹配定位过程中,实时图与参考图常常因为天气、季节不同等原因差别很大,这就造成匹配概率和匹配精度下降。为此,提出了一种基于光流的景象匹配定位算法,它利用光流估计实时图序列中2个相邻帧的相对运动,并根据上1帧的匹配结果确定当前帧图像和参考图的大致位置关系,进而在1个小范围内进行搜索,完成实时图的匹配定位。仿真结果表明该算法不仅具有很高的匹配概率,还具有较好的实时性。

In the process of scene matching applied to navigation, if the real time image and the reference image differ too much, the matching probability and matching precision will decline dramatically. To meet the challenge, an algorithm of scene matching based on optic flow is presented. This algorithm uses optic flow between two adjacent frames to estimate their relative motion, so as to find the geometry translations of the two frames. Then the real time image is mapped into the coordinate of the reference image. Finally the right position can be obtained by searching in a small region of the reference image. The experiments show that the algorithm is valid and efficient even when the reference image and real time image are obtained in different seasons.