Autonomous placement of a range sensor for acquisition of optimal 3-D models

Abstract

In this paper, the authors describe the system they have implemented which automatically determines an optimized next range sensor position and orientation during the reconstruction of a three-dimensional model. The system they have developed reconstructs a model consisting of surfaces which have been viewed and volumes occluded from the camera's view. Ideally, a sensor pose determined by a "best-next-view" system will reveal the greatest quantity of previously unknown scene information. We will present results from the most intelligent of the algorithms we have developed. This algorithm attempts to intelligently cluster the occluded data and orient the sensor on the centroid of the largest cluster. A system which finds a solution to the best-next-view problem may find application in the contexts of robot navigation, manufacturing and hazardous materials handling. The methods we implement take advantage of no a priori information in finding the best-next-view position.