Depth-prior-based LiDAR point cloud de-noising method leveraging range-gated imaging.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Data Processing

Background:

• Light Detection and Ranging (LiDAR) is crucial for 3D scene understanding in self-driving vehicles and robotics.
• LiDAR data is susceptible to noise, degrading performance in critical applications.
• Current denoising methods are often slow or ineffective against specific noise sources like occlusions.

Purpose of the Study:

• To develop a real-time LiDAR point cloud denoising method.
• To address limitations of existing methods in handling diverse noise types.
• To improve the reliability of LiDAR data for downstream tasks.

Main Methods:

• A depth-prior-based approach utilizing principles of range-gated imaging.
• Synchronized acquisition of LiDAR point clouds and gated images.
• Projection of LiDAR data into a depth map for noise identification and removal based on depth inconsistencies.

Main Results:

• The proposed method effectively removes various types of noise from LiDAR point clouds.
• Achieved superior performance across all evaluated metrics compared to existing methods.
• Demonstrated real-time processing capabilities for LiDAR denoising.

Conclusions:

• The depth-prior method offers a robust and efficient solution for LiDAR point cloud denoising.
• This technique enhances the quality of LiDAR data, particularly in challenging conditions.
• The method shows significant potential for improving the safety and performance of autonomous systems.