Multi-Depth-Map Raytracing for Efficient Large-Scene Reconstruction.

Area of Science:

• Computer Graphics
• Geometric Modeling
• 3D Reconstruction

Background:

• Large point clouds require efficient processing and visualization.
• Traditional mesh reconstruction and texturing are impractical for massive datasets.
• Runtime stitching of depth maps offers scalability but faces performance limitations.

Purpose of the Study:

• To develop a novel method for efficient raytracing of multiple depth maps.
• To overcome the limitations of existing depth map-based scene representation methods.
• To accelerate both the texturing and rendering of large point cloud data.

Main Methods:

• Generating high-resolution textured depth maps via rendering input points.
• Employing graph-cut optimization for point-to-image subset assignment.
• Implementing efficient raytracing using point-to-image assignments for runtime intersection computation.

Main Results:

• Achieved an order of magnitude acceleration in both texturing and rendering.
• Successfully broke dependencies on the number of depth maps and their resolution.
• Enabled efficient processing and visualization of large-scale 3D scenes.

Conclusions:

• The proposed raytracing method significantly enhances the efficiency of large point cloud processing.
• This approach provides a scalable and practical solution for real-time 3D scene representation.
• The method offers substantial improvements over existing depth map-based techniques.