Modelling and validation of LiDAR noise distribution in fog and rain

Abstract

Rain-and-fog-induced noises deteriorate the imaging capability of LiDAR applied for autonomous driving. To explore the noise generation mechanism, a comprehensive model considering the LiDAR systematic attributes and environmental characteristics is established for predicting the noise distribution probability based on the photon motion simulation. The analytical formulations are innovatively combined with the Monte Carlo method to enhance the computational efficiency while ensuring accuracy, by utilizing a kernel regression method to link the theoretical model with practical measurements. Experiments are conducted via computer simulation and a serial of adjustable fog-and-rain laboratory simulation tests, with two novel statistical indicators proposed for model validation. The experimental results show that the maximum relative error between the model and measurements is less than 6%, outperforming the contrasting methods. The presented model and results can provide insights for reducing aerosol particle noise and evaluating LiDAR imaging in adverse weathers.