Point cloud completion in challenging indoor scenarios with human motion.

Area of Science:

• Computer Vision
• Robotics
• 3D Data Processing

Background:

• Combining point cloud data from multiple sensors with different perspectives is challenging.
• Dynamic, cluttered environments and limited scene overlap complicate sensor fusion.

Purpose of the Study:

• To develop a novel approach for registering 3D point cloud data from two sensors with unknown relative perspectives.
• To enable accurate sensor fusion in real-life scenarios with human movement.

Main Methods:

• Reduced 3D point cloud completion unknowns by aligning ground planes.
• Extracted 3D human walking sequences using a histogram-based approach.
• Matched walking paths by minimizing Fréchet distance and used 2D Iterative Closest Point (ICP) for final alignment.

Main Results:

• Successfully registered human walking paths between two camera captures.
• Estimated the transformation matrix between the two sensors.
• Demonstrated a robust method for sensor registration in challenging environments.

Conclusions:

• The proposed approach effectively handles significant perspective differences and complex scenes.
• Enables accurate sensor registration and data fusion for dynamic environments.
• Facilitates real-world applications requiring multi-sensor 3D data integration.