A Multithreaded Denoising Algorithm for LiDAR Point Clouds Based on Moving Least Squares

LiDAR point cloud data is essential in autonomous driving, robotic navigation, and 3D modeling. However, noise caused by sensor errors and environmental factors degrades data quality and affects downstream processing. Denoising has thus become a critical preprocessing step. This paper presents a comparative study of classical point cloud filtering algorithms and introduces a multi-threaded denoising approach based on Moving Least Squares (MLS) smoothing. Experimental results show that the proposed method significantly reduces noise while preserving geometric details. Evaluation using cosine similarity and angular error metrics confirms its effectiveness and superior performance in both accuracy and speed.