A Comprehensive Survey on Stereo Matching and Depth Estimation: From Classical Pipelines to Deep Learning Architectures

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Authors: Lin, Yida, Xue, Bing, Zhang, Mengjie, Schofield, Sam, Green, Richard
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902271191678976
author Lin, Yida
Xue, Bing
Zhang, Mengjie
Schofield, Sam
Green, Richard
author_facet Lin, Yida
Xue, Bing
Zhang, Mengjie
Schofield, Sam
Green, Richard
contents <div>Stereo matching and depth estimation are funda</div> <div>mental components for 3D scene understanding in applications</div> <div>ranging from autonomous driving and robotics to augmented</div> <div>reality and medical imaging. This comprehensive survey unifies</div> <div>classical stereo pipelines with modern deep learning architec</div> <div>tures, providing systematic analysis of cost volume construction,</div> <div>aggregation strategies, and optimization techniques. We present</div> <div>verifiable benchmark comparisons across major datasets (KITTI,</div> <div>Middlebury, ETH3D, Scene Flow), analyze the evolution from</div> <div>hand-crafted features to learned representations, and examine</div> <div>training paradigms including supervised, self-supervised, and</div> <div>cross-domain approaches. Our analysis covers architectural</div> <div>innovations including 3D CNNs, recurrent refinement networks,</div> <div>transformer-based methods, and hybrid approaches. We provide</div> <div>practical guidelines for reproducible evaluation, discuss deploy</div> <div>ment considerations for real-world applications, and identify key</div> <div>challenges including occlusion handling, domain adaptation, and</div> <div>computational efficiency. This survey serves as both a historical</div> <div>perspective on the field’s evolution and a practical guide for</div> <div>researchers and practitioners working on stereo vision systems.</div>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17070403
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle A Comprehensive Survey on Stereo Matching and Depth Estimation: From Classical Pipelines to Deep Learning Architectures
Lin, Yida
Xue, Bing
Zhang, Mengjie
Schofield, Sam
Green, Richard
<div>Stereo matching and depth estimation are funda</div> <div>mental components for 3D scene understanding in applications</div> <div>ranging from autonomous driving and robotics to augmented</div> <div>reality and medical imaging. This comprehensive survey unifies</div> <div>classical stereo pipelines with modern deep learning architec</div> <div>tures, providing systematic analysis of cost volume construction,</div> <div>aggregation strategies, and optimization techniques. We present</div> <div>verifiable benchmark comparisons across major datasets (KITTI,</div> <div>Middlebury, ETH3D, Scene Flow), analyze the evolution from</div> <div>hand-crafted features to learned representations, and examine</div> <div>training paradigms including supervised, self-supervised, and</div> <div>cross-domain approaches. Our analysis covers architectural</div> <div>innovations including 3D CNNs, recurrent refinement networks,</div> <div>transformer-based methods, and hybrid approaches. We provide</div> <div>practical guidelines for reproducible evaluation, discuss deploy</div> <div>ment considerations for real-world applications, and identify key</div> <div>challenges including occlusion handling, domain adaptation, and</div> <div>computational efficiency. This survey serves as both a historical</div> <div>perspective on the field’s evolution and a practical guide for</div> <div>researchers and practitioners working on stereo vision systems.</div>
title A Comprehensive Survey on Stereo Matching and Depth Estimation: From Classical Pipelines to Deep Learning Architectures
url https://doi.org/10.5281/zenodo.17070403