Triangle-Mesh-Rasterization-Projection (TMRP): An Algorithm to Project a Point Cloud onto a Consistent, Dense and...

Area of Science:

• Computer Vision
• Robotics
• 3D Data Processing

Background:

• Point cloud projection is crucial for image analysis, multimodal processing, robotics, and generating ground truth data for deep learning.
• Current single-shot projection methods face challenges including false gaps, neighborhood inaccuracies, and sensor-specific limitations.

Purpose of the Study:

• To develop a novel, accurate, and dense 2D raster image projection method for 3D point clouds.
• To address the limitations of existing single-shot projection techniques.

Main Methods:

• Developed Triangle-Mesh-Rasterization-Projection (TMRP), a new algorithm for projecting point clouds onto 2D images.
• Achieved dense accuracy by integrating 2D neighborhood information with 3D point data for fast triangulation interpolation using sub-triangle weights.

Main Results:

• TMRP generates dense, accurate 2D raster images with only physically valid gaps.
• The method eliminates false gaps, false neighborhoods, and ambiguities, offering XYZ-independent density.
• Demonstrated TMRP's effectiveness using the KITTI-2012 dataset and diverse sensor modalities across four use cases.

Conclusions:

• TMRP offers a significant improvement over single-shot projection methods, providing accurate and dense 2D representations from 3D point clouds.
• The open-source TMRP method is versatile, applicable to various sensors and modalities, enhancing robotic perception and processing optimization, particularly for transparent objects.