Intrinsic and extrinsic calibration of a UAV-based multi-sensor system

Abstract The integration of multiple sensors, such as cameras and LiDARs, is increasingly employed in vehicle navigation and 3D environmental mapping. Accurate multi-sensor data fusion relies heavily on the precise estimation of extrinsic parameters, which define the geometric transformations between sensors. However, when low-cost sensors are used, systematic errors in their intrinsic characteristics often introduce inaccuracies in subsequent applications. To address this challenge, we investigate the impact of intrinsic calibration of a low-cost LiDAR on the accuracy of its extrinsic calibration parameters. Our study evaluates a generalized intrinsic calibration approach that assumes a single scale factor and bias for the LiDAR’s laser rays, in contrast to a detailed model where individual scale factors and biases are estimated for each ray. While the generalized approach exhibits noticeable deviations from the systematic parameters of some rays, it has a substantial positive impact on the estimated extrinsic parameters of the LiDAR, leading to improved mounting accuracy. This improvement highlights the critical role of intrinsic calibration in achieving reliable extrinsic calibration for multi-sensor systems. To ensure robust extrinsic calibration, we decouple the intrinsic and extrinsic calibration processes, thereby providing rectified input data for extrinsic calibration. Furthermore, we present an extrinsic calibration approach for a multi-sensor system comprising an inertial measurement unit, two cameras, and a LiDAR, where the cameras and LiDAR share only a narrow overlapping field of view, resulting in minimal shared information. By employing a high-accuracy reference instrument, we address the limited overlap issue and validate both the extrinsic calibration results and the impact of the LiDAR’s intrinsic calibration. Our findings demonstrate that untreated LiDAR intrinsic errors lead to significant deviations in the extrinsic calibration parameters, underscoring that intrinsic calibration is not merely a preliminary step but a vital factor in ensuring accurate extrinsic calibration in multi-sensor systems.