3D image reconstruction based on current stereo vision techniques

• Main Author: Teo, Jun Wei
• Other Authors: Qian Kemao
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Physics; Stereo vision; Camera calibration; Stereo matching; Disparity; Image rectification
• Online Access: https://hdl.handle.net/10356/175065
• Institution: Nanyang Technological University

Stereo vision is a computer vision technique used that enables the computer to perceive depth in a scene or object called 3D reconstruction. This is done by capturing and analysing images taken from two cameras at a fixed distance apart. This final year project will serve as a comprehensive guide for setting up and creating a stereo vision system which includes camera calibration, stereo calibration, undistortion of images and lastly 3D reconstruction which is all incorporated within a desktop application developed by the author for the ease of use. This report will also dive into the theories behind stereo vision such as epipolar geometry, the essential and fundamental matrices that is used to relates the points in the two camera views. This report will discuss principles such as the pinhole camera model which this project is based after and the basic geometric projection. Additionally, this report will explain the various distortion that may appear the images and how it will be rectified. Lastly, this report will explore stereo calibration and rectification and the various block matching methods. This report will also include findings from experiments done during this project to improve the results and quality of 3D reconstruction. One method to improve result is to improve camera calibration result by using a circle grid board and detecting the center of the circle to high accuracy instead of the commonly used chessboard which aims to improve results for calibration. This proposed method proved to be an improvement as compared to the traditional methods at camera calibration.