Surface Three-Dimensional Scene Acquisition and Establishment
本文于1329天之前发表，文中内容可能已经过时。如有疑问，请在评论区留言。
Binocular stereo vision is the research focus in the field of computer vision. Usually, two identical cameras are used to obtain two digital images of the surrounding objects from different angles, and the 3D geometric information of the objects is recovered based on the parallax principle to reconstruct the 3D shape and position of the surrounding objects.
Based on the binocular stereo system, this project studies the calibration and calibration, salient object extraction, stereo matching and reconstruction technology of binocular stereo camera. To obtain meaningful 3D images of ground scene provides a new technical means for virtual reality, augmented reality and other consumer electronic products.
First Step 第一步
Research on calibration of infrared binocular camera， epipolar rectification and its improvement
Second Step
Feature based stereo matching
Third Step
Construction and experiment of acquisition and establishment system of ground 3D scene
Flow Chart
Calibration Method
Zhang’s classical calibration method uses two-dimensional checkerboard as the calibration object, which simplifies the calibration process, reduces the cost of camera calibration, but also maintains high calibration accuracy. With its advantages of high precision and good flexibility, it is still the most common calibration method in camera calibration.
Epipolar rectification can be carried out with camera calibration and without camera calibration. Due to the existence of a large number of uncalibrated image pairs in stereo vision, it is difficult to match stereoscopically directly.
The self calibration method gets rid of the limitation of the calibration board, so it is more flexible and practical. However, the calibration accuracy is far different from the plane calibration method, which is the biggest restriction of its practical application.
In this research, we choosed epipolar rectification as the criterion of binocular camera optical axis parallel.
Stereo Matching
SIFT algorithm for feature points
Constraints in stereo matching:
{1.Epipolar constraints.
2.Uniqueness constraint.
3.Similarity constraints:
4.Parallax continuity constraint.
5.Order consistency constraint.
6.Parallax constraint.
7.Left and right consistency constraints}
Improved k-nearest neighbor mismatching removal method
Acquisition and Modeling of Ground 3D Scene
The principle of three-dimensional reconstruction of binocular stereo vision comes from the biological binocular vision system. The specific process is as follows: select two images of the same scene and different perspectives, and use the two-dimensional information to recover the three-dimensional information of the visible surface of the scene. The picture shows the principle model of binocular stereo vision in ideal state.
Three-Dimensional Reconstruction
Sparse Reconstruction（Visual SFM）–>>Dense Point Cloud Reconstruction. (PMVS/CMVS)–>>Poisson Algorithm for Surface Reconstruction
##Hardware System
##Software System
The 3D reconstruction algorithm was completed in visual SFM platform Dr. Chang Wu loading PMVs / CMVs environment Dr. Yasutaka Furukawa.
The surface reconstruction of Poisson algorithm is completed by meshlab.
Importing Binocular Images
Sparse Reconstruction（Visual SFM）and Dense Point Cloud Reconstruction
(PMVS/CMVS)
Poisson Algorithm for Surface Reconstruction(meshlab)
