Positional space solution to the next best view problem

摘要

The "next best view" (NBV) problem encountered while acquiring the surface geometry of an object using a range scanner is solved by determining the next position for the range scanner given its previous scans of the object. A new representation, positional space, is used as a unified data structure for representing what must be and what can be scanned. The image of the range scanner in positional space is computed off-line for particular positions of the range scanner in its work space and stored in a file. Then, when the next best view is to be determined during the analysis of the object, each scanner image is simply retrieved from the file instead of being recreated by translating the scanner image calculated for some other position on the work space of the range scanner. A linear objective function of the stored scanner images and images of a seen and an unseen surface of the object is then maximized over all scanner positions to determine the next best view. This approach accommodates all range based sensors and allows arbitrarily shaped positional space surfaces and work spaces to be used while also saving in on-line computation time.