Registration of Multiple Triangulated Surfaces via the ICP (Iterative Closest Point) Algorithm

Riccardo Corsi


Abstract

In this work I analyze the problem of the registration of multiple triangulated surfaces, through the implementation of the ICP (Iterative Closest Point) algorithm, in a version adapted to this specific input format. The registration of multiple data sets is necessary when a 3D model of a real object is built by merging reconstructions from several data acquitisions taken from different points of view.

The sensing devices used to acquire the data sets, need to change their relative position to the object, in order to capture different views of the object itself. Even if this motion can be tracked, *the* approximation of the relative positions has not the same accuracy as the resolution of the scan data generated. Hence, to create a high resolution 3D model of the object combining different acquisitions, a software registration step is required to align the data sets in the same coordinate system.

A well known algorithm used for the multi-view registration is ICP. In this paper I will evaluate ways to adapt the ICP algorithm to the specific data input of triangulated surfaces, which is one of the most common ways to represent a 3D surface. Besides the accuracy of the registration, I worked with the purpose to minimize the computation time, to obtain a real-time implementation of the process. This would permit, for instance, the interaction with a real-time acquisition system, in order to generate a continuously updated model of a scanned object.


Maintained by Philippos Mordohai