Review of stereo matching based on deep learning

Abstract

In recent years, the introduction of deep learning has greatly advanced computer vision technology. Stereo matching, as a key technique, has shown impressive performance with the help of deep neural networks, leading to a significant amount of research. Compared to traditional methods, stereo matching based on deep learning differs in many ways. Therefore, this paper provides a comprehensive review of the latest developments in this field. We take a novel perspective by categorizing these algorithms into four branches based on how they handle the left and right views in stereo matching networks. We outline the evolution and prospects of each branch, summarizing and analyzing the outstanding methods in each, exploring their advantages, disadvantages, and main challenges. We also identify unresolved issues in each branch and discuss future research directions.