3D RECONSTRUCTION USING THE CONCEPT OF STEREO VISION

Abstract

3D reconstruction is a highly valuable task within the realm of geometry-based computer vision. Its purpose is to obtain a three-dimensional representation of an object using two-dimensional images. While both deep learning-based computer vision and physics-based computer vision can be utilized for 3D reconstruction, this particular project focuses on employing the principles of geometry-based computer vision. The project's objective involves stereo vision-based 3D reconstruction utilizing stereo pairs with an overlapping rate of 91.35%, as provided by vision.middlesbury.edu. The process of obtaining a 3D reconstructed outline entails a series of algorithms and methods. These include camera calibration, feature selection, feature matching, determining essential matrices, pose recovery and triangulation, stereo rectification, disparity map generation through matching algorithms, post-filtering, and ultimately obtaining a 3D outline using cloudcompare. By sequencing and integrating all the aforementioned algorithms, a 3D outline is generated as the final resultFurthermore, evaluating the performance of this stereo-vision system is crucial for future enhancements and improvements.