Road-network-map-assisted vehicle localization under multidimensional constraints

Abstract Global navigation satellite system (GNSS) is widely used in autonomous vehicle navigation systems. Unfortunately, satellite signals are very weak or even lost in some areas, making it impossible to provide stable and high-precision localization data continuously. In GNSS-challenged regions, localization methods based on odometry can serve as effective supplements. However, they rely on dead reckoning, and the localization error will accumulate as the distance traveled increases. Thus, this paper proposes a road-network-map-assisted vehicle localization method under multidimensional constraints. This method extracts the features of the initial trajectory output by odometry and matches them with the point-line road network map to accurately locate the initial location of the vehicle. In the continuous localization stage, we view the road network map as the global sensor, introducing multidimensional constraints to correct odometry drift error and predict vehicle locations in real time. The experiments carried out on the KITTI and campus datasets demonstrate that the proposed method can locate the vehicle’s initial location in the road network map with a 5 km range and continuously suppresses odometry drift error. Compared with similar map-assisted localization methods, this approach achieves higher localization accuracy, reducing the average localization error by 14.67%.