双目立体视觉研究进展与应用

Research Progress on Binocular Stereo Vision Applications

双目立体视觉模仿人类视觉系统对环境进行三维感知,通过对校正后的左右图像进行立体匹配获取两幅图像的视差,再根据三角测量原理计算出场景的深度,在近几十年中一直是计算机视觉领域的研究热点,取得了一系列的进展。传统的立体匹配方法采用手工设计的特征进行立体匹配,研究表明,这类方法对弱纹理或重复纹理区域以及遮挡区域表现不佳。近年来,基于深度学习的立体匹配方法取得了显著的进展,表现出了强大的应用潜力。本综述对这一不断发展的领域进行全面的调研,讨论不同方法之间的优点和局限性,介绍市面上的双目产品,并展望该领域的研究与应用前景。

Binocular stereo vision simulates the human visual system to perceive the environment in three dimensions.The parallax between two images can be obtained by stereo-matching the corrected left and right images and calculating the scene depth based on the triangulation principle.This area has been identified as a research hotspot in computer vision and has made significant progress in the past few decades.Traditional stereo matching methods use hand-designed features that perform poorly in weak or repeated texture and occlusion areas.Recently,stereo matching methods based on deep learning have significantly progressed,showing strong application potential.In this review,we conducted a comprehensive survey on this developing field,discussed the advantages and limitations of the different methods,introduced the binocular products currently on the market,and assessed the research and application prospects in this field.