One of the prominent methods for accurate self-localization for autonomous vehicles is map-matching with light detection and ranging (LiDAR) based on Normal distribution transform (NDT). In NDT, map space is divided into the grids, and for each grid, normal distribution (ND) of the points are calculated, and LiDAR scan is matched to these NDs. Bigger grid sizes (lower resolution) are more favorable because it can abstract more points in each grid and reduce map size. However, if the resolution is low, many details of the environment are ignored, and the localization accuracy degrades. This information loss and localization error is different from place to place on the map and can be evaluated beforehand for each resolution. In this work, ten map factors are used to evaluate the localization ability of the map in a specific position for each resolution. Using the evaluation result, for each position of the map, a lower resolution that can preserve the required localization accuracy are determined. In this method, NDT map is generated by adaptively selecting the resolution for each position of the map. Experimental results in Shinjuku, Tokyo, show that by using this strategy, map size can be reduced by up to 32% of the original size while the mean localization error remains less than 0.141m.
Adaptive Resolution Refinement of NDT Map Based on Localization Error Modeled by Map Factors
2018-11-01
1226270 byte
Conference paper
Electronic Resource
English
Data refinement reduces error factors
Tema Archive | 1979
Error Estimation and Adaptive Mesh Refinement for Aerodynamic Flows
Springer Verlag | 2010
|Cartesian Adaptive Mesh Refinement for Rotorcraft Wake Resolution
British Library Conference Proceedings | 2010
|