Automatic pyramidal intensity-based laser scan matcher for 3D modeling of large scale unstructured environments

We are developing a vision-based system for photorealistic 3D modeling of previously unknown, complex and unstructured underground environments. Nowadays, laser range finders allow us to build 3D maps of the environment by taking multiple scans from different viewpoints. The scans are usually aligned via two post-processing steps: first, a coarse alignment is provided by an operator and second, a fine solution is computed via Iterative Closest Point algorithm. In this paper we describe an automatic on line scan matcher system which replaces the two post-processing steps of the existing methods. The scan matcher is powered by a pyramidal pairwise matching of 2D intensity panoramic views using a dense correlation procedure via quaternions. The proposed method does not rely on feature extraction providing thus an environment-independent solution for the scan matching task. The pyramidal structure provides a fast and accurate scan alignment in a coarse to fine approach. The scan matcher allows us to automatically build in situ 3D mosaics by integrating multiple partially overlapping scans based on a topological inference criterion which improves the global matching consistency. Tests on real data from two prehistoric caves are presented and a performance evaluation is given.