Quality Evaluation for Colored Point Clouds Produced by Autonomous Vehicle Sensor Fusion Systems

Area of Science:

• Robotics and Computer Vision
• Autonomous Vehicle Perception Systems

Background:

• Autonomous vehicles (AVs) rely on sensor fusion (e.g., LiDAR, cameras) for robust environmental perception.
• Quantifiable evaluation methods are crucial for comparing diverse sensor fusion systems and design choices.

Purpose of the Study:

• To present a novel evaluation method for comparing colored point clouds generated by different sensor fusion systems.
• To assess LiDAR-camera fusion systems against a stereo camera setup using quantifiable metrics.

Main Methods:

• Developed an evaluation approach using a test artifact measured by colored point clouds.
• Metrics include point cloud spread, area coverage, and color difference.
• Compared two LiDAR-camera fusion systems and one stereo camera system.

Main Results:

• The evaluation method successfully ranked the performance of the sensor fusion systems.
• The metrics provided quantifiable data that complemented experimental observations.
• Demonstrated the suitability of the approach for comparing fused point cloud data.

Conclusions:

• The proposed evaluation methodology is effective for comparing colored point clouds from sensor fusion systems.
• This method aids in selecting and optimizing sensor configurations for autonomous driving.
• Facilitates objective assessment of perception system performance under various conditions.