Low-complexity adaptive radius outlier removal filter based on PCA for lidar point cloud denoising

Area of Science:

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

Background:

• Autonomous vehicles utilize light detection and ranging (lidar) for environmental perception.
• Environmental interference poses challenges for accurate lidar data interpretation.
• Existing noise reduction methods may lack efficiency for distant point cloud data.

Purpose of the Study:

• To develop an advanced noise reduction technique for lidar point clouds.
• To enhance the reliability of sensor data in autonomous driving systems.
• To improve the performance of lidar data processing for long-range applications.

Main Methods:

• Development of an adaptive radius outlier removal filter.
• Application of principal component analysis for noise filtering.
• Comparative analysis against existing clustering algorithms.

Main Results:

• The proposed filter demonstrates superior performance in precision and recall.
• Achieved an F-score of up to 0.876.
• Reduced computational complexity by at least 50% compared to traditional methods.

Conclusions:

• The adaptive radius outlier removal filter effectively reduces noise in lidar point clouds.
• The method shows significant advantages for long-distance point cloud data.
• This advancement contributes to more robust and efficient autonomous driving perception systems.