A general purpose feature extractor for light detection and ranging data.

Area of Science:

• Robotics
• Computer Vision
• Geospatial Data Processing

Background:

• Feature extraction is crucial for processing Light Detection and Ranging (LIDAR) data.
• Current LIDAR feature detectors are often environment-specific, leading to poor performance in novel settings.
• Existing methods struggle with generalizability across indoor and outdoor environments.

Purpose of the Study:

• To develop a general-purpose feature detector for 2D and 3D LIDAR data.
• To create a method applicable to virtually any environment without prior knowledge.
• To enhance the stability and repeatability of feature detection in LIDAR datasets.

Main Methods:

• Adapted classic image processing feature detection methods, specifically the multi-scale Kanade-Tomasi corner detector.
• Applied the adapted method to both 2D and 3D LIDAR data.
• Evaluated performance across various spatial scales and environments.

Main Results:

• The developed detector identifies highly stable and repeatable features across different spatial scales.
• The method demonstrates robustness and generalizability across diverse environments.
• Principled uncertainty estimates and corner descriptors are generated concurrently with feature detection.

Conclusions:

• The proposed general-purpose feature detector significantly improves LIDAR data processing capabilities.
• This approach overcomes the limitations of environment-specific detectors.
• The method shows promise for applications requiring reliable feature extraction from varied LIDAR data.