Uncertainty Modeling for Kinematic LiDAR-based Multi-Sensor Systems

LiDAR (Light Detection and Ranging) sensors are used in a variety of applications. Their ability to directly capture 3D data opens up new possibilities and complements other sensors. The point clouds generated by LiDAR sensors are used for localization, fusion with image data, and general environmental perception, such as obstacle detection. In contrast to terrestrial laser scanners (TLS), which are high-quality LiDARs used in geodetic applications, uncertainty analysis of cost-effective multi-beam or solid-state LiDARs is still neglected, even though it is of great importance in various applications. Two examples include their use in multisensor systems (MSS) as kinematic scanning systems in surveying and as additional sensors for autonomous vehicles.

Uncertainty modeling is important because, particularly in autonomous systems, decisions based on the acquired data can lead to dangerous situations for people and the environment. Consistently predicted uncertainty information can support other algorithms for localization or object tracking in making correct decisions. Consistent means that the system’s confidence—i.e., the predicted uncertainty—should correspond to the actual accuracy. This can be verified in various ways, e.g., through hypothesis testing. Furthermore, the predicted uncertainty is required for determining integrity, which is already a standard for aircraft. The goal of this doctoral project is to investigate methods that enable a consistent estimation of uncertainties for LiDAR-based MSS while addressing the challenges arising from the uncertainties of individual sensors and their interactions within the system.

Initially, a consistent calibration of LiDAR sensors on platforms with other sensors was implemented. Better parameters can be achieved through an additional estimation of intrinsic calibration parameters. Additionally, data fusion was further developed within the framework of an error-state Kalman filter. Linear pose interpolation for calculating the update enables significantly more accurate trajectory estimates. To account for the different material properties during measurement by LiDAR sensors, a recursive variance component estimation was developed and applied to estimate the uncertainties in distance measurement.

This research was supported by the DFG through the i.c.sens Research Training Group (RTG 2159) until November 2025.