Spatial occupancy recovery from a multiple-view range imaging system for path planning applications

Abstract

We strive to obtain full spatial occupancy descriptions of the workspace from multiple range images acquired from different points of view. Such representations are highly suitable in path-planning applications, for charting a collision-free course through a workspace. We propose an algorithm, called peeling, which allows us to fuse range images into a voxel array and obtain a spatial occupancy model for use by a path planner. We also take a look at another method of range image fusion, called direct mapping, which combines the images into a voxel array but does not provide a proper spatial occupancy model. We present an overview of previous work in range image fusion and look at octrees and octree generation; our final model is an octree representation, for more efficient memory use. We are building a prototype system consisting of a technical Arts 100 AT laser range scanner mounted on a PUMA 560 robot arm, both controlled from an SGI workstation. Spatial occupancy models generated by this system are fed to an off-line path-planner.