Use of Consumer-grade Depth Cameras in Mobile Robot Navigation

Simultaneous Localization And Mapping (SLAM) stands as one of the core techniques used by robots for autonomous navigation. Cameras combining Red-Green-Blue (RGB) color information and depth (D) information are called RGB-D cameras or depth cameras. RGB-D cameras can provide rich information for indoor mobile robot navigation. Microsoft’s Kinect device, a representative low cost RGB-D camera product, has attracted tremendous attention from researchers in recent years, for its relatively high quality of depth measurement. By analyzing the multi-data stream of both color and depth, better 3D plane detectors, local shape registration techniques can be designed to improve the quality of mobile robot navigation. In the first part of this work, models of the Kinect’s cameras and projector are established, which can be applied for calibration and characterization of the Kinect device. Experiments show both variable depth resolution and Kinect’s own optical noises in depth values calculation. Based on Kinect’s models and characterization, this project implements an optimized 3D matching system for SLAM, from processing of RGB-D data to further algorithms design. The developed system includes the following parts: (1) raw data preprocessing and de-noising, improving the quality of integrated environment depth maps. (2) 3D planes surfaces detection and fitting with RANSAC algorithms; also providing applications and illustrative examples about multi-scale-multi-planes detections algorithms which designed for common indoor environment. The proposed approach is validated on scene and object reconstruction. RGB-D features matching under uncertainty and noise in a large scale of data, forms the basis of future application in mobile robot navigation. Experimental results have shown that system performance improvement is valid and feasible.