Estimating Surface and Spatial Structure from 3-D Line Segments using Geometrical & Heuristic Relation.

Abstract

A number of researchers have investigated vision-based navigation of autonomous mobile robot, using passive sensing systems; stereo camera or motion camera system. In general, by using the passive sensing system, the robot can obtain three-dimensional data of objects. This 3-D data we have been using, come from an edge-based algorithm that yields a wire frame description of the scene, i.e, a set of 3-D coordinates of line segments in 3-D space. For navigation, however, we need the surface-based information to determine free space where the robot can move, and to identify objects or walls. This paper presents an approach for understanding surfaces and their spatial structures from 3-D line segments. We can discover both convex and concave object's surfaces using basic geometrical knowledge of the 3-D world. The occluded surfaces can be also reconstructed by using some heuristic relationships between neighboring surfaces. Our system can recognize the surfaces and spatial configuration of the environment even if the obtained location data have noise and lose some boundary edges. We present experimental results on some simulated data and a natural scene data.