Evaluation of Modern Dynamic Programming Algorithms for Real-time Active Stereo Systems

Abstract

Stereo Vision has been an active research field that has produced a variety of different algorithms. Unfortunately most of the algorithms that produce superior results rely on non-linear optimization techniques that are very computationally expensive, and therefore not feasible to use for real-time applications such as tele-immersion. This paper will examine a number of real-time stereo algorithms based on dynamic programming (DP) used in conjunction with structured light in order to improve the quality and facilitate the correspondence search. We will examine some of the early DP algorithms as well as the more recent work produced by [Criminisi et al] for the purpose of Gaze manipulation for teleconferencing in the context of 3d reconstruction. We present adaptation of [Birchfield et al] DP algorithm to work with structured light. Additional we look at spatialtemporal support region for computing matching costs.