A Stochastic Cost Function for Stereo Vision
Christian Unger, Slobodan Ilić · 2014
The goal of this paper is to present a novel stochastic cost function for binocular stereo vision that delivers statistics about the most probable disparities on the pixel level. We drive these statistics by many independent stochastic processes so that robustness to outliers can be achieved. Each of these stochastic processes may be understood as an individual who is requested to deliver his opinion about the depth. Finally, the idea is to fuse all these individual measurements into one global disparity map. In this paper, we use random walks for this.