The Prospect of 3D Reconstruction
- CV two major tasks object recognition, 3D reconstruction
- Autonomous driving, VR, AR core
- A representative application of 3D reconstruction in restoration and reconstruction of ancient buildings
- Harder than autonomous driving (three-dimensional perception)
, computer vision, which is usually mentioned, thinks of object recognition, (recognizing whether a picture is a cat or a dog) (behavior classification task: whether a person is sitting or standing) are all objects of computer vision identify.
The goal of computer vision: to understand the image
Research purpose: to obtain visual features from the image.
The development process of 3D reconstruction
- The first international computer vision conference ICCV held in London in 1987
- The goal of computer vision is to understand images
- The purpose of the research is to obtain visual features from images
- In the 1980s, everyone was doing edge extraction based on edge. In 1986, Canny edge detection
- In the 1990s, people paid more and more attention to 3D reconstruction. One-dimensional edge is not suitable for geometric calculation. The most essential element of geometry is point
- In 1992, Olive Faugeras and Richard Hartle independently solved the problem of three-dimensional reconstruction under two images of uncalibrated cameras
3D reconstruction classification (active, passive)
- Stereo vision (binocular stereo vision reconstruction)
- depth image
learning steps
- Image processing related knowledge: image feature extraction (Hough transform, D2-Net, SIFT)
- Camera model learning (mapping from 3D space to 2D): camera calibration, optimization acceleration
- 3D reconstruction: binocular system, stereo matching (SIFT, pathmatch, structured light)
- Structure from motion (polar geometry, essential matrix)
- Rigid body transformation
learning direction
- High-precision 3D reconstruction
- Large-Scale 3D Reconstruction
- Non-rigid Structure from motion
- Dynamic 3D reconstruction
This article is my own study notes, which is convenient for myself to view, and draws on the videos of bilibili and many materials found on the Internet.