3D structure recovery from stereo using synchronous optimization processes

Abstract

This paper presents a novel system that uses two synchronous optimization processes to recover 3D structure from rectified stereo image pairs. The synchronization of the processes are done by energy terms that inform the optimization processes about the recovered positions of each other. This information is used to direct the optimizations towards a better direction. The system is initialization insensitive and it is very robust against local minima. We performed experiments on real and synthetic images with ground truth that showed the effectiveness and the robustness of our system. We also compared our system to other systems for further validation. © 2025 Elsevier B.V., All rights reserved.