3D Environment Modeling Based on Surface Primitives

In this article we describe an algorithm for constructing a compact representation of 3D laser range data. Our approach extracts a dictionary of local scans from the scene. The words of this dictionary are used to replace recurrent local 3D structures, which leads to a substantial compression of the entire point cloud. We optimize our model in terms of complexity and accuracy by minimizing the Bayesian information criterion (BIC). Experimental evaluations on large real-world datasets show that the described method allows robots to accurately reconstruct environments with as few as 70 words. Furthermore the experiments suggest that the proposed representation gives a richer semantic description than pure occupancy based representations.