Automatic Registration of Large-Scale Range Datasets

This material is based upon work supported by the National Science Foundation under Grant No. 0215962 and by CUNY Institute of Software Design and Development. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Ioannis Stamos - Assistant Professor, Hunter College and Graduate Center of CUNY
Marius Leordeanu - Senior undergraduate student, Hunter College


We are building a system that can automatically acquire 3D range scans and 2D images to build geometrically correct, texture mapped 3D models of urban environments.  One of the main bottlenecks in the process is the automatic registration of a large number of 3D range scans in a common frame of reference. In this work we provide a novel solution to the problem. The method utilizes range segmentation and feature extraction algorithms and is able to very efficiently and accurately register a large number of scans. Our algorithm automatically computes pairwise registrations between individual scans. It then establishes global topological relations between the scans by utilizing a Dijkstra-type graph algorithm.

We work with Columbia University's Robotics laboratory on the project of reconstructing the model of the St. Pierre Cathedral in Beauvais, France (for more information on the reconstruction of the Cathedral visit this link). Our automated registration algorithm has been tested on the Beauvais dataset. We also gathered data from buildings in Hunter College's neighborhood.

Publications related to the project:

Pairwise Registration

The flow-chart of the pairwise registration algorithm is shown. An improved version of our segmentation algorithm (published in Computer Vision and Pattern Recognition, 2000) is applied to each scan. The output of the segmentation algorithm is fed into the registration modules. The goal of the algorithm is to extract a valid match between lines and planes without having to consider all possible combinations of pairs.
The pairwise registration step is followed by a global registration step.

Results: Registration of range scans gathered in Beauvais, France

Photograph of the Beauvais Cathedral.

One range scan of the Cathedral. It consists of one million 3D points.

The segmentation algorithm extracts major planes and 3D lines from the raw range scan.

More than one scan is needed in order to represent the whole Cathedral. Our algorithm registers automatically 27 scans. The top view of the cathedral is shown. The local coordinate systems of each scan are also displayed.


Detail of the registration. South-West part of the cathedral.

Detail of the registration. Close-up view. Different data sets are displayed with different colors. Note the accuracy in the alignment.