Progressive Streaming of Multi-Billion Point Clouds in Web Browsers
Visualizing massive LiDAR datasets and photogrammetry point clouds using hierarchical spatial Octrees and GPU point size attenuation.
Loading uncompressed LAS or LAZ point cloud files directly into browser memory causes instant crashes. Hierarchical spatial Octree structures allow WebGL applications to stream only visible point nodes based on camera proximity and viewport frustums.
1. Spatial Octree Indexing & Lod Partitioning
Subdividing 3D spatial bounding boxes recursively creates a multi-resolution hierarchy where root nodes store sparse point representations, while leaf nodes contain dense local laser scan details.
2. Screen-Space Point Budgeting
To maintain 60 FPS interaction, the rendering engine enforces strict maximum point budgets (e.g., 5 million active points per frame), dynamically prioritizing Octree nodes closest to the camera center.
3. Dynamic Point Sizing & Eye-Dome Lighting (EDL)
Scaling GPU point size based on camera distance fills visual gaps. Applying Eye-Dome Lighting (EDL) post-processing passes calculates depth gradients to highlight subtle geometric contours in un-textured scans.