A Rain Noise Removal Method for LiDAR Based on Comprehensive Confidence Score Decision

Rain-induced noise in LiDAR degrades target recognition, but conventional filtering methods lack effective information fusion and theoretical support. To address this, a rain noise removal method for LiDAR is developed, identifying noises based on their comprehensive confidence scores. These scores are derived from individual confidence scores (range, intensity, and neighbor count) through modeling LiDAR transmission-reception and rain scattering characteristics. A score threshold solved by Otsu method is used to accurately segment noise and target points. Noise removal performance is enhanced by dynamically adjusting individual score contributions using the weight entropy method, while computational efficiency is improved over ROR-based methods by restricting the neighbor search area based on LiDAR scanning patterns. Validation experiments were conducted in a controllable rain laboratory, using the lab-built full-waveform LiDAR. Results demonstrate that the proposed method achieves high F1 scores (0.9437@6m and 0.9759@12m) and fast processing speed (0.1s@15,400 points), outperforming existing state-of-the-art techniques. The presented method enables interference-resistant LiDAR imaging in rainy conditions and can be extended to handle noise in other large-aerosol-induced environments, such as snow and dust, broadening LiDAR applicability to diverse weathers.