中文摘要：

通过在双目立体测量系统中引入辅助标记点的方法来实现多视角深度图的配准，配准算法包括两部分：初配准和全局优化配准。提出一种递增式初配准算法，将多次测量结果逐次配准到同一坐标系中，然后对初配准结果进行全局优化。对基于奇异值分解的全局配准迭代算法进行改进：一是对重建误差较大的标记点进行自动加权，可有效降低全局配准误差；二是对全局配准的迭代顺序进行规划，在一些情形下可减少迭代收敛的次数。对多个实际模型的测量实验验证了算法的有效性。

英文摘要：

A multi- view range image registration algorithm based on reference points for binocular stereo vision system was introduced herein. The algorithm included two phases： crude registration and global registration. Firstly, an incremental crude registration method was proposed for registering each newly scanned range image into a common coordinate frame. Then an iterative algorithm using singular value decomposition for global registration was adapted to optimize the crude registration result. Two improvements for the global registration were carried out. One was to assign the weights to reference points automatically, which allowed the algorithm to decrease the global registration error effectively. Another one was to classify the types of registration network and did the iteration through a planned sequence so that the number of iterations can be decreased in some special cases. Experiments on real scanned data demonstrate the validity and efficiency of the algorithm.