Monte Carlo variance propagation for the uncertainty modeling of a kinematic LiDAR-based multi-sensor system

Abstract

Abstract Kinematic multi-sensor systems (MSS) are widely used for various applications, like mobile mapping or for autonomous systems. Depending on the application, insufficient knowledge of a system, like wrong assumptions about the accuracy of calibrations, might lead to inaccurate maps for mapping tasks or it might endanger humans in the context of autonomous driving. Uncertainty modeling can help to gain knowledge about the data captured by a system. Usually, uncertainty estimations for MSSs are done as backward modeling based on a comparison to reference datasets. In this paper, a forward modeling approach for the uncertainty modeling of a LiDAR-based kinematic MSS is chosen to estimate the uncertainty of an acquired point cloud. The MSS consists of a Leica Absolute Tracker and a platform with a 6-DoF sensor and Velodyne VLP-16 LiDAR. Results of multiple calibrations are used as the source for the uncertainty information for a Monte Carlo (MC) variance propagation of the point uncertainties. The deviations of the acquired point clouds in comparison to a ground truth can be decreased by an ensemble referencing process using the MC samples. Furthermore, the predicted uncertainties for the point clouds are well representing the actual deviations for reference panels closer to the system. Panels farther away indicate remaining distance depending effects.