多层平面单应约束车载街景影像匹配算法研究

A Matching Algorithm of Vehicle-Mounted Street View Images Based on Multiplanar Homographic Constraint

车载街景影像匹配中含有较多难以剔除的误匹配,建议采用一种多层平面单应约束的车载街景影像匹配方法进行剔除。车载街景影像的拍摄对象主要为多层次的建筑物立面,物方平面在立体像对中的投影满足单应约束,所以建筑物立面在影像中满足多层平面单应约束。采用Harris算法提取角点特征,采用SIFT算子描述并用邻近距离比约束匹配获取初始同名点。首先,采用核线约束初步剔除误匹配;然后,结合RANSAC估计立体像对间的多层平面单应删除外点。该方法可删除纹理重复造成的误匹配,提高立体像对间匹配的正确率。

Aimed at the fact that vehicle-mounted street view images contains many mismatches which are difficult to eiim inate, this paper recommends a kind of matching algorithm of vehicle-mounted street view images hased on the multi- planar homographic constraint. The objects of vehicle-mount ed street view images are multi-level building facades and the projections of a plane in ohiect side satisfy homograph in a stereo pair, so the images of building facades satisfy multi- planar and homographic constraint. This paper adopts Harris algorithm to extracted angles and point features, describes with SIFT descriptors, constraints and matches initial points by computing the near distance ratio. The mismatches are firstly eliminated by the epipolar constraint and then elimina- ted by estimating multiplanar homograph units combined with RANSAC. The matching algorithm can lessen mismat- ches caused by repeated textures and improve matching accu- racy.