Riassunto analitico
The field of autonomous driving has emerged as a significant area of research in information engineering, with the potential to revolutionize the automotive industry. Providing an accurate 3D mapping of the environment, LiDARs are crucial to enable important tasks underlying many functions of ADAS systems such as distance measurement, object detection and recognition, and environmental awareness. These capabilities are crucial for enabling autonomous driving features and enhancing overall road safety. In the agricultural sector, they are widely used for measuring topography, tree characteristics, crop biomass estimation, and crop growth. However, manufacturers often provide limited information and metrics, making it difficult to comprehensively evaluate LiDARs performance. To allow reaching the required measuring interval, to date, most of the LiDARs used in the automotive and agriculture sectors are in scanning technology. Such implies that the acquisition of the surrounding environment takes place sequentially. If there is relative motion between the vehicle — the LiDAR — and the surrounding environment, the acquired 3D image is distorted. Theoretically, knowing the scanning frequency and the displacement vectors, such a distortion could be compensated. Nonetheless, as experienced by anyone who has analyzed point clouds (PCs) acquired from moving LiDARs, the distortion observed is often more severe and seemingly unpredictable than expected from the LiDAR scanning frequency and the displacement vectors. Nonetheless, the performances provided by the manufacturers generally refer to those obtained in the absence of relative motion. For automotive and agricultural applications, the most interesting performances are those relating to in-motion acquisitions. Such a type of characterization is extremely challenging. Characterization activities, which evaluate LiDAR sensors' performance in real-world scenarios, are thus particularly important for optimizing them for specific applications. However, Dynamic (in-motion) characterization of LiDAR is less developed compared to static characterization due to the complexity of dynamic environments. In fact, it involves dealing with constantly changing conditions, moving objects, and occlusions, requiring more sophisticated experimental design to ensure accurate and meaningful results. This thesis, in collaboration with CNH Industrial, aims to develop standard procedures for the dynamic (in-motion) characterization of LiDAR. To validate the proposed measurement methods, the test system was exploited for the analysis of an extremely widespread LiDAR, the Velodyne VLP16 model. The results obtained thanks to the developed test system have allowed highlighting significant measurement artifacts that are not present if the acquisitions are carried out in the absence of relative motion, confirming the importance of the developed test system.
|