Automated Point Cloud Registration Approach Optimized for a Stop-and-Go Scanning System.

Area of Science:

• Robotics and Automation
• Geomatics Engineering
• Computer Vision

Background:

• Terrestrial laser scanning (TLS) advances enable automatic data acquisition, but registration remains a manual bottleneck.
• Mobile robots, particularly quadruped walking robots, offer platforms for stop-and-go scanning systems.
• Automated registration is crucial for efficient processing of large-scale point cloud data.

Purpose of the Study:

• To develop and validate an automated point cloud registration approach for stop-and-go scanning systems.
• To overcome the limitations of manual registration in terrestrial laser scanning workflows.
• To enhance the efficiency and reliability of processing scan data from mobile robotic platforms.

Main Methods:

• Perpendicular constrained wall-plane extraction for initial feature identification.
• Coarse registration using plane matching and point-to-point displacement calculation.
• Fine registration employing horizontality constrained Iterative Closest Point (ICP) algorithm.

Main Results:

• Achieved automated registration with high accuracy (0.044 m).
• Demonstrated a 100% successful scan rate (SSR) for 18 scan datasets.
• Completed registration within 424.2 seconds in a real-world indoor environment.
• Outperformed conventional methods, especially for point cloud pairs with low overlap.

Conclusions:

• The proposed automated registration method is effective for stop-and-go scanning systems on quadruped robots.
• This approach significantly reduces manual intervention and processing time in point cloud data.
• It offers reliable registration performance under challenging indoor conditions with limited data overlap.