3D Environmental Mapping of Mobile Robot Using a Low-cost Depth Camera

Building rich 3D maps of unknown environments is a key issue for service robots, with applications in navigation and mobile manipulation. In this paper, an application of 3D map building using a low-cost Kinect sensor equipped on a mobile robot is introduced. The application evalutes the approach that combines visual and depth information for dense point cloud alignment. During robot's continuous movement, successive frames of RGB-D data are captured and processed. Firstly, SURF features of color images are extracted and then RANSAC algorithm is employed to remove large amount of outliers. Generalized-ICP algorithm is employed to perform fine registration, which finally produces dense point cloud. The proposed method is applied to a home-care service robot for building 3D map of an office environment using a RGB-D sensor. Application of mobile robot navigation using the 2D projection of the 3D map based on Octomap format is also given. Experiment results validate the practicability and effectiveness of the approach.