3D-laser-based visual odometry for autonomous mobile robot in outdoor environments

In this paper, we describe a visual odometry algorithm based on a novel 3D range data registration approach for mobile robot. Considering traditional range image´s limited adaptability in outdoor environment representation, a Bearing Angle model is used in this paper so that local SIFT (Scale-invariant feature transform) features can be extracted and matched effectively. According to the one-to-one correspondence between Bearing Angle image and raw 3D laser data, we can obtain a group of matching pairs between any two successive laser scans, which can be used to accomplish the 3D range data registration. Compared to the conventional ICP (Iterative Closest Point) algorithm, this approach can greatly improve the performance of scan registration both in computation time and accuracy. Furthermore, no initial rough pose estimation is needed in the registration. The frame-to-frame motion estimate can be performed by using the relative pose estimations resulting from the registration. Finally, experimental results are provided to demonstrate the effectiveness of the proposed visual odometry algorithm during the wheel odometry is unavailable or GPS (Global Positioning System) signal outages.