3D Object Recognition Based on Point Clouds in Underwater Environment with Global Descriptors: A Survey

Area of Science:

• Robotics
• Computer Vision
• Marine Technology

Background:

• Underwater object recognition is vital for Autonomous Underwater Vehicles (AUVs) in Inspection, Maintenance, and Repair (IMR).
• Existing global descriptors require evaluation for colorless 3D point cloud data in challenging underwater conditions.

Purpose of the Study:

• To compare the performance of state-of-the-art global descriptors for object recognition in underwater 3D point clouds.
• To evaluate descriptor performance under realistic conditions including noise, varying resolution, and platform-induced deformations.

Main Methods:

• Performance comparison of open-source global descriptors using virtual scans (CAD models) and real data from a laser sensor-equipped AUV.
• Experiments conducted with virtual and real underwater data, simulating IMR-representative objects under various environmental conditions.

Main Results:

• Descriptor performance was significantly impacted by the resolution mismatch between database and test scans for all but one method.
• Real-world testing from a moving platform introduced shape deformations affecting descriptor performance.

Conclusions:

• This research provides the first comprehensive performance evaluation of object recognition methods for underwater 3D point clouds using real data from a free-floating platform.
• Matching scan resolution is critical for reliable underwater object recognition, with implications for AUV-based IMR operations.