Optimized projection patterns for stereo systems

Abstract

This paper describes how to generate optimal projection patterns to supplement general stereo camera systems. In contrast to structured light, the active stereo systems utilize the projected patterns only as auxiliary information in correspondence search, whereas the structured light systems have to detect the patterns and decode them to compute depth. The concept of non-recurring De Bruijn sequences is introduced, and a few algorithms based on the non-recurring De Bruijn sequence are designed to build optimized projection patterns for several stereo parameters. When only the search window size of a stereo system is given, we show that a non-recurring De Bruijn sequence with corresponding parameters makes the longest functional pattern, and presents experimental results using real scenes to show the effectiveness of the proposed projection patterns. Additionally if the pattern length is given in the form of maximum disparity search range, the algorithm using branch-and-bound search scheme to find an optimal sub-sequence of a non-recurring De Bruijn sequence is proposed.