Integration of Physical Features and Machine Learning: CSF-RF Framework for Optimizing Ground Point Filtering in...

Area of Science:

• Geospatial science
• Remote sensing
• Computer science

Background:

• Point cloud processing faces challenges in vegetated areas with complex terrain.
• Accurate ground point extraction is crucial for Digital Elevation Model (DEM) generation.

Purpose of the Study:

• To develop an improved algorithm for filtering ground points in challenging environments.
• To enhance the accuracy of DEMs in areas with dense vegetation and complex terrain.

Main Methods:

• Integration of the Cloth Simulation Filter (CSF) algorithm with the Random Forest (RF) machine learning framework.
• Development of the CSF-RF fusion algorithm for classifying ground and non-ground points.
• Validation of the algorithm's performance in complex terrain and dense vegetation.

Main Results:

• The CSF-RF algorithm achieved a total error of 0.19% in dense vegetation and complex terrain.
• This represents a 79.6% relative reduction in error compared to the traditional CSF algorithm (0.93%).
• Type I and type II errors were below 0.05%, with overall error within 0.03%.

Conclusions:

• The CSF-RF fusion algorithm effectively reduces vegetation interference and improves point cloud filtering accuracy.
• The algorithm demonstrates good stability and provides effective technical support for DEM extraction.
• This method is particularly advantageous in challenging environments with dense vegetation and severe terrain undulations.