针对三维图像特征点的匹配问题, 研究了在Kinect三维图像中, 分别基于SIFT、SURF和ORB算法建立的三维特征描述子经RANSAC算法优化后匹配精度和匹配速度的差异。首先, 使用一组Kinect拍摄的测试集进行测试, 三种特征匹配算法经过RANSAC算法优化之后, 都表现出良好的匹配精度,ORB算法在匹配速度上稍有优势。其次, 使用Kinect实际拍摄的室内样张进行重复性测试, 实验结果表明, ORB算法在匹配精度和匹配速度上较SIFT和SURF算法更优秀。因此, 将ORB特征匹配算法用做Kinect图像的特征检测器效果最佳。

Aiming at the problem of feature matching in Kinect pictures,differences in matching accuracy and matching speed of the 3D feature descriptors were studied,which were respectively simulated with SIFT, SURF and ORB algorithm and optimized with RANSAC algorithm. Firstly,a test set photographed with Kinect was used for testing, in which the three feature matching algorithms all showed fine matching accuracy after optimization with RANSAC, and in the meanwhile, the ORB algorithm proved to show higher matching speed. Then, repeated tests were carried out with samples taken in real situations with Kinect, and the results showed that ORB did better in both matching accuracy and matching speed than SIFT and SURF. Therefore, the ORB algorithm is the best as feature matching process for Kinect pictures.