A fast 3D map building method for indoor robots based on point-line features extraction

This paper reports a real-time 3D environmental reconstruction algorithm under semi-structure environment using low-cost Kinect2 sensor. By minimizing the sum of reprojection errors, the changing matrix is solved by Gauss-Newton iteration method. After that, the closed loop detection is performed by using the extracted point-line features, and the closed loop error is reduced by the closed loop optimization, so that a large scale global uniform 3D environment model is obtained. It is converted into the octree structure based on probability. Through off-line and real time on-line data acquire from robot to test the reliability of the algorithm, the accuracy of the calculation time cost and output point cloud model is compared, which verified the reliability and robustness of the reconstruction method.