Optimized LiDAR target for benchmarking depth measurement accuracy

Light detection and ranging (LiDAR) has been widely used in autonomous vehicles to provide 3D perception, which leads to a growing need to verify their depth measurement accuracy, besides the driving algorithms. However, the research on the explainability and optimization of LiDAR targets for such benchmarks remains limited. Therefore, we propose a LiDAR target with an optimized shape to make the return signal quality of the LiDARs under test easier to distinguish. Using a LiDAR simulator, we modeled how scanning errors vary with signal quality and developed a "disparity function" to quantify these differences. This function guided the optimization of a target shape, which we then fabricated using 3D printing. Experiments show that our prepared target increases the difference between the depth measurement errors of two candidate LiDARs by over seven times compared to a planar one, and is capable of preventing other sources of error from offsetting the performance difference.