A robust and real-time full 3D reconstruction method based on multiple kinect

Abstract

Although 3D reconstruction of objects has been extensively studied, the robust and fast approach still remains challenging. In this paper, we present a VR (Virtual Reality) based social system that can produce realistic full 3D reconstruction of moving objects in real-time. In this system, we propose a novel method that can refine the point clouds from multiple Kinect streams and therefore generate accurate 3D reconstruction. Specifically, the original point clouds are first filtered to remove the edge noise by using optimal triangulation algorithm. Afterwards, the refined point clouds are registered by optimal registration method. In order to elevate the visual quality of the reconstruction result, RANSAC linear regression is used to adjust the color difference between the corresponding points in adjacent point clouds. The experimental results verify the effectiveness of our 3D reconstruction method in visual quality and time efficiency.