POS辅助下的车载序列影像匹配

针对车载序列影像中所包含的具有大量重复纹理的建筑立面区域给现有匹配方法带来的误匹配率高、匹配点不足等不适应性问题,提出了一种联合POS辅助及KLT(Kanade-Lucas-Tomasi)角点特征跟踪的新匹配方法。通过充分利用POS信息所能够建立的局部近似建筑立面分层平面及DEM信息,获取KLT跟踪算法中所需的初始位移信息,并将建筑立面及路面区域进行分开追踪匹配,以保证匹配点的良好分布特性,最后利用RANSAC结合核线约束进行误差剔除得到最终匹配结果。试验证明本文方法能够有效解决车载序列影像中重复纹理区域的匹配问题,可高效获取分布均匀、数量充足的匹配特征点,为后期空中三角测量提供更稳健的同名约束条件。

POS-aided Matching of Vehicle-borne Sequential Images

Aiming at the existing problem of high mismatching rate and insufficient corresponding points for vehicle-borne sequential images,as a result of the plentiful repetitive texture in the building faqade area,a new method combining auxiliary POS and the KLT corner tracking algorithm is proposed in this paper.In the first place,through taking full advantage of POS information,the local approximate stratification fa(c)ade plane and DEM are established to get the initial displacement required by the KLT algorithm.Then a new strategy tracking the fa(c)ade and road area separately is enployed to ensure a good distribution of matching points.After that,the final result is obtained by integrating the RANSAC and epipolar constraint to eliminate the matching error.The implemented experiments show that the presented method could solve the preceding mismatching problem,as well as obtaining adequate and uniformly distributed corresponding points for a more robust aerial triangulation process.