An Iterative Approach to Determine the Lower Bound of Depth Resolution in Stereo Vision Systems

• Guilherme Toledo UNICAMP
• Lorenzo Carrera de Oliveira UFCG
• Eanes Torres Pereira UFCG

Resumo

We present a novel method to calculate the lowest expected depth resolution for non-parallel and heterogeneous stereo vision systems automatically based on a simulated environment. Our algorithm iterates over an arbitrary 3D counterpart to an epipolar line and uses a modified binary search to find the depths in which the disparity changes. We show how, given the parameters of the stereo system, our method can calculate what is the best resolution one algorithm might achieve if they do not use sub-pixel information. Throughout this paper, we show how changes in the scene might affect the resulting lower bound, such as increasing camera resolution or changing camera angles around the subject. Furthermore, we propose a method to compare and plot stereo vision algorithms against the generated lower bound. We anticipate that our approach may reduce development costs for stereo vision solutions by minimizing the necessity to physically construct stereo systems prior to assessing their potential accuracy.

Palavras-chave: Graphics, Lower bound, Accuracy, Costs, Three-dimensional displays, Machine learning algorithms, Machine learning, Cameras, Stereo vision, Iterative methods