Three-dimensional reconstruction

Three-dimensional reconstruction in computer vision is to reconstruct the three-dimensional environment in which the image is located by processing the image and analyzing the information implied in the image. Three-dimensional reconstruction technology is one of the important technologies for environment perception, and three-dimensional reconstruction exists in a variety of practical applications of computer vision such as automatic driving, virtual reality technology, augmented reality technology, motion target detection, behavioral analysis, and so on.

Three-dimensional reconstruction is an important part of computer vision, the task of target recognition is only a relatively shallow surface technology in computer vision, human vision can truly perceive the three-dimensional world, so computer vision will eventually also be based on recognition towards the three-dimensional world.

Three-dimensional reconstruction is generally a process of three-dimensional information reduction of the current environment through a single view or a multi-angle view. Multi-angle views contain more sufficient information about the conditions, so 3D reconstruction is less difficult, while 3D reconstruction of a single view is more difficult.

Four representations are commonly used for 3D reconstruction: depth map, point cloud, volume elements and mesh.

Depth map

Depth map is used to represent the distance between points in the scene and the computer. Each pixel in the depth map represents the distance between the corresponding scene in the image and the camera.

Volume Element

Volume elements are also known as voxels. Like pixels, voxels are the smallest unit of division in three-dimensional space, and represent a three-dimensional area with a constant scalar or vector.

Point cloud

A point cloud is a collection of data about the surface of an object in an image obtained by measuring together. Point cloud can be divided into coefficient point cloud and dense point cloud, the use of three-dimensional coordinate measuring machine obtained by the spacing of the point cloud is called sparse point cloud, the use of three-dimensional laser scanner obtained by the more dense point cloud is called dense point cloud.

In the figure below, (A), (B) and (C) are the front view, top view and left view respectively, and (D) is the original image of the scene.

Grid

Mesh is the mesh simulation of the surfaces that make up a three-dimensional three-dimensional object, and the meshes commonly used in computer vision are triangular mesh and quadrangular mesh.

3D reconstruction has different directions in practical applications, for example, 3D reconstruction in the field of autonomous driving and robotics is called Simultaneous Localization And Mapping (SLAM), and there is also 3D reconstruction based on deep learning in computer vision, as well as 3D reconstruction of the human body, 3D reconstruction of the face, 3D reconstruction of various objects, 3D reconstruction of the human body, 3D reconstruction of a variety of objects, 3D reconstruction of the human face, 3D reconstruction of the human face, and 3D reconstruction of the human body. 3D reconstruction of human body, 3D reconstruction of human face, 3D reconstruction of various objects, 3D reconstruction of indoor scenes, etc.

Your task

What are the common representations of 3D reconstruction?

Share your thoughts and ideas in the comments below.