Depth Estimation | Chapter 12 of CV book

3d 视觉笔记 01 ｜ Depth Estimation

此篇笔记是「三维视觉笔记」系列的第 1 篇, 记录了 Richard Szeliski 教授的计算机视觉教材 —— 「Computer Vision: Algorithms and Applications」第二版的第 12 章 —— 深度估计

0. Introduction

Stereo matching (立体视觉匹配) is the process of taking two or more images and building a 3D model of the scene

• by finding matching pixels in the images
• and converting their 2D positions into 3D depths.

The word stereo comes from the Greek for solid; stereo vision is how we perceive solid shape (Koenderink 1990).

Question:How to build a more complete 3D model?

• Example: a sparse or dense depth map that assigns relative depths to pixels in the input images.
• Solution: Multi-view stereo algorithms (多视图立体视觉) produce complete 3D volumetric or surface-based object models, as well as monocular depth recovery algorithms that infer plausible depths from just a single image.