测量点云数据的多视拼合技术研究

在实物反求中，往往需要将从各个不同视角测得的点云数据进行多视定位，统一到一个全局坐标系下，并经过数据融合使之能够便于后续的模型重建。本文首先提出了一种改进ICP算法，在采用标签法进行预定位的基础上，选取两个视图重叠区域中不同位置的数据点集作为控制点集，建立控制点的名义对应关系，然后进行坐标变换迭代求解，从而方便快速地实现了两个不存在明确对应关系的点云视图之间的准确定位。在进行多视拼合时，提出一种边定位边合并的方法，减少了多视顺序拼合时所产生的积累误差。对拼合后的视图，本文根据在重叠区域的缩小包围盒内采用加权平均的方法进行了数据融合。实例表明，本文提出的方法是行之有效的。

Research of Multi-View Registration and Integration on Measured Point Cloud Data

In reverse engineering, in order to meet the requirements for model reconstruction, it is often necessary to locate and merge the different view measured cloud data in a global coordinate system. Firstly, an enhanced ICP algorithm is proposed to register precisely the pair wise views without explicit point to point matches. On the basis of pre registration with labeling method, a nominal correspondence has been created between the controlled point sets selected from the different positions in the overlapped view area, and the transformation can be achived by iterative method. Afterwards, a synchronous method is also presented to implement multi view registration and integration, which will reduce the accumulative error caused by sequential registration. Finally a weighed average algorithm is utilized in the shrunk bounding box of the overlapping area to integrate all the sets of views, and the effectiveness of the methods is validated with experimental results.