改进的SURF算法在彩色车载影像匹配中的应用

在分析了具有尺度不变特征的鲁棒特征加速算法(SURF算法)的基础上,提出了一种基于高斯颜色模型的增维彩色SURF算法.该算法将RGB颜色模型转换到高斯颜色模型,使用SURF算法(增加了48维颜色特征描述向量)进行匹配,再使用核线约束剔除误匹配.结果表明:尽管相对于原有的SURF算法略微增加了计算量,但是在匹配点对数、匹配正确率、匹配点分布均匀性上都具有明显的优势.

Vehicle based Images Matching by Improved SURF Algorithm

Compared with the traditional Photogrammetry, the images obtained by vehicle-based Mobile Mapping System have more complex deformation and shield. These factors make images matching become more difficult, and there are more mismatched corresponding points in the Mobile Mapping System. SURF algorithm, which is robust in rotation and resizing, can solve these problems. But if the light condition is poor in images, the SURF method could not detect enough interesting points. So a dimension-incremented color SURF algorithm, which based on Gaussian color model, is proposed in this paper. In this improved method, RGB color model is transformed to Gaussian color model firstly, then the new SURF method is used to make matching experiment, which is improved by additional 48 dimentional colour features vectors. At last epiplor constrain is applied to eliminate false matches. The experimental results show that our proposed algorithm performs favorably on matching points number, matching accuracy, and distribution uniformity of matching point , although it increased the amount of calculation slightly relatived to the original SURF algorithm. It can meet the matching demand of MMS.