A Kinect-Based System for 3D Reconstruction of Sewer Manholes

Abstract

A system for automatic 3D modeling of sewer manholes using two Microsoft Kinect sensors is presented. The hardware design is based on a hand-held scanner that includes two Kinect sensors, a mini-PC and LED-based illumination bars. The scanning and reconstruction software has been implemented using open-source software toolkits. The information from the two Kinect sensors is merged into a single 3D model. Reconstruction is performed by adapting an existing RGB-D SLAM-based algorithm by filtering out nonconsistent correspondences and an alternative graph-based optimization, which results in reconstruction errors of the order of 1 cm. The final result obtained is a complete system with a hand-held scanner and management software which is highly competitive with respect to present commercial systems, using much cheaper technology and open-source toolkits with low reconstruction errors.