Abstract:

In order to enhance the positioning accuracy of intelligent vehicles， a method fusing LiDAR and double-layer map model is proposed， in which the double-layer map model is created by adding laser point-cloud-based sparse feature map on the top of lane map， and the sparse feature map consists of the position and azimuth of vehicles， 2D intensity features and 3D features. The sparse feature map can provide an accurate position reference for intelligent vehicle positioning， effectively reducing accumulative positioning error. In addition， the lane lines are extracted from the LiDAR intensity data to provide highly accurate and linear lateral position constraints. During positioning， a Kalman filter framework is introduced to fulfill the effective fusion of LiDAR and double-layer map， in which the process of state prediction utilizes the motion constraints of vehicle to construct the short-time and constant-speed movement model and to observe the variables including the results of laser odometer positioning， the lateral position constraints based on lane map layer and the positioning based on sparse feature map layer. Tests and measurements are conducted on both campus and urban road environment to verify the effectiveness of the proposed algorithm. The results show that the fusion positioning algorithm proposed can reduce the positioning error by 40%~60% under different environments， with a relative positioning error less than 0.3%.

Key words: intelligent vehicles, double-layer map model, point cloud processing, Kalman filtering