Robust point cloud registration based on semantic iterative closest point algorithm

Area of Science:

• Computer Vision
• Robotics
• Geospatial Analysis

Background:

• Point cloud registration is crucial for 3D data processing but faces challenges with noisy LiDAR data and poor initial alignment.
• Existing methods struggle with outliers and significant positional errors, limiting their real-world applicability.

Purpose of the Study:

• To develop a robust and accurate point cloud registration algorithm for LiDAR data.
• To improve registration performance in the presence of noise, outliers, and poor initial positions.

Main Methods:

• A semantic-based iterative closest point (ICP) algorithm incorporating bidirectional distance and correntropy.
• Semantic-guided correspondence establishment to narrow search ranges and enhance accuracy.
• Bidirectional semantic search point matching for improved error correction.
• Maximum correntropy criterion for noise and outlier suppression.

Main Results:

• The proposed algorithm demonstrates superior accuracy and robustness compared to existing registration methods.
• Semantic guidance effectively improves correspondence establishment and overall registration precision.
• The use of correntropy significantly enhances resilience to noise and outliers in LiDAR point clouds.

Conclusions:

• The semantic-based ICP algorithm offers a robust solution for challenging LiDAR point cloud registration tasks.
• The integration of semantic information and advanced outlier rejection techniques leads to significant performance gains.
• This method advances the state-of-the-art in 3D point cloud registration for applications requiring high accuracy and reliability.