Probabilistic Smoothing Based Generation of a Reliable Lane Geometry Map With Uncertainty of a Lane Detector

Since lane geometry information can be used for controlling the pose of an intelligent vehicle, a lane geometry map that contains the lane geometry information should have reliable accuracy. For generating the reliable lane geometry map, lane curve which is detected from a lane detector is an useful information because the lane geometry information can be obtained directly. However, since the detected lane curve contains an uncertainty caused by the noise of the lane detector, the accuracy of the lane geometry map can be degraded. In previous studies, a near point on each detected lane is sampled at each time stamp and accumulated for reducing the noise effects of the lane detector. However, these sampled points also contain the sensing noise of the lane detector and the density of accumulated points depends on the distance interval of data acquisition. In this article, we proposed the probabilistic lane smoothing-based generation method for the reliable lane geometry map. In the probabilistic lane smoothing, the lane geometry map is modeled as the nodes with the uncertainty of its position obtained from the sensor error model. Each node of the lane geometry map is smoothed based on the Bayesian filtering scheme. The evaluation results show that the lane geometry map can be generated by reducing the noise of the detected lane curve. Additionally, the generated lane geometry map is more reliable than the sampling point-based generated map in terms of the accuracy of the distance and heading angle.