Data fusion of visual odometry and ranges for accurate SLAM

Kurzfassung

Autonomous vehicles are now intensively studied due to their wide range of possible applications. For autonomous maneuver, vehicles need to accurately estimate their positions and orientation. Global Navigation Satellite System (GNSS) is generally used from drones to autonomous cars to estimate their positions. When GNSS is unavailable or unreliable, Visual Simultaneous Localization and Mapping (VSLAM) can be employed. However, VSLAM is a dead-reckoning process, so the estimation errors are accumulated over time without a bound. The errors can be significantly reduced using the loop closing technique, but it requires large computing power. Moreover, loop closures cannot resolve the VSLAM's inherent issue with the scale estimation. We propose a fusion of visual odometry and range measurements for accurate SLAM. Our experimental evaluation results show that the estimation errors are mitigated by fusing visual odometry and range measurements. The absolute scale can be also accurately estimated using ranges. Furthermore, our method requires much smaller processing power than loop closures since only odometry (6DoF) and range (scalar) measurements are used for the data fusion.