Stereo depth maps reconstruction using structured light

Abstract

In recent years, stereo vision has experienced rapid development and plays a crucial role in various fields, such as autonomous driving, obstacle avoidance, virtual reality, and 3D reconstruction. The computer stereo vision system simulates human binocular vision, using two cameras to capture images. Through camera calibration, stereo rectification, and stereo matching, depth information of spatial points is obtained, facilitating 3D reconstruction of the space. However, the current passive stereo vision systems are prone to occlusion and sensitive to varying lighting conditions, which can affect the accuracy of the reconstructed depth map. The project focuses on resolving the problems using active stereo vision, which projects maximum min Stripe-Width gray codes on the object and compares the differences with the conventional passive stereo vision. This report analyses the principles and procedures required for passive and active stereo vision, such as calibration, stereo rectification, obtaining disparities and removing noise. It provides the computed disparity map of both passive and active methods, results comparison, error maps and the reconstructed 3D images. Experimental results show that passive disparity maps display errors mostly from occlusion and texture-less regions in the scene, whereas active disparity maps resolve those issues. In this project, a high-accuracy depth reconstruction system was designed and developed using a stereo camera and structured light. It builds upon an analysis of the theoretical foundations of stereo vision and camera imaging principles to eliminate the lack of features and susceptibility to lighting conditions, ultimately increasing the accuracy of the reconstructed depth maps.