Research on point cloud denoising based on multifeature fusion in adverse weather

Abstract

In recent years, autonomous driving technology has entered a new phase of rapid development. Accurate positioning information, as the core premise for path planning, obstacle avoidance decision-making and stable control of autonomous vehicles, directly determines the safety of system operation. LiDAR has emerged as a crucial positioning sensor in autonomous driving perception systems due to its high-precision ability for 3D environment modeling. However, under complex meteorological conditions such as snowfall, rainfall and fog, the laser radar point cloud is prone to being disturbed by noise, which will lead to a decrease in positioning accuracy and thus become a key bottleneck that severely restricts the reliable deployment of autonomous driving systems in all weather conditions. To address these issues, this paper proposes a point cloud denoising and restoration method based on multi-feature fusion. This method utilizes the temporal information of multiple frames of point clouds, distinguishes dynamic objects from noise through motion trajectory analysis, restores the environment feature points that were wrongly deleted, and achieves precise denoising through multi-feature weighted fusion. To verify its effectiveness, this paper conducted data collection and experimental analysis under various harsh weather conditions such as snowfall, rainfall and simulated fog in both hardware-in-the-loop simulation and real scenarios. The results showed that compared with traditional methods, the proposed method's comprehensive denoising metric F-score increased by more than 15%, significantly improving the point cloud quality under adverse weather conditions, and providing effective technical support for the stable positioning of autonomous driving in complicated meteorological environments.