Compensation of iToF-AMCW Scanning LiDAR Depth Errors Caused by Transparent Obstacles Using CMA-ES-Optimized LightGBM

This paper proposes a novel depth error compensation method utilizing a Light Gradient Boosting Machine (LightGBM) optimized via the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The proposed method is designed to correct depth errors induced by transparent obstacles located in front of an indirect time-of-flight (iToF) amplitude-modulated continuous wave (AMCW) scanning light detection and ranging (LiDAR). Specifically, the presence of a transparent obstacle generates multiple laser signal interference, wherein the sensor receives a superposition of laser signals reflected from both the target object and the transparent obstacle. This complex signal superposition directly results in distorted target depth measurements, the magnitude of which varies as a function of the modulation frequency. To compensate for such errors, this study proposes training a CMA-ES-optimized LightGBM using pointwise scanning data derived from multi-frequency physical indices, generated via a high-fidelity iToF scanning LiDAR simulation. The validity and performance of the proposed correction method were extensively evaluated through various scenarios involving depth distortions caused by transparent obstacles. Validation results demonstrate that the proposed method significantly reduces depth error from the order of several meters to lower than 0.03 m scale in terms of mean absolute error (MAE).