3D Gaussian Splatting Volumetric Reconstruction Pipeline.
Discover how Vora transforms standard 360° smartphone video into centimeter-accurate (±1.5 cm) 3D volumetric models with terrain slope isolation and trunk diameter fitting.
Adaptive Frame Extraction & Blur Filtering
Uploaded 360° videos are processed using OpenCV with Laplacian variance detection. Frames with excessive motion blur or rapid angular acceleration are filtered out, maintaining 48–72 optimal high-overlap views.
Camera Pose Estimation & GLOMAP Global SfM
To guarantee commercial licensing compliance and carbon project integrity, Vora implements GLOMAP Global SfM (BSD-3). The system globally and deterministically reconstructs camera poses and sparse tie points, replacing non-commercial research models to ensure enterprise-grade commercial readiness.
RANSAC Ground Plane Isolation & DBH Cylinder Fitting
A 3D RANSAC algorithm extracts the ground terrain plane equation ax + by + cz + d = 0 to normalize slope angles in forestry plots. The system slices a trunk cross-section at standard breast height (1.30m above ground) and fits an algebraic circle to extract the true DBH.
3D Gaussian Splatting (gsplat) & .ksplat Format
The 3D model is trained using the gsplat engine (Nerfstudio / UC Berkeley, Apache 2.0) built on clean-room CUDA kernels. The 3D Gaussian ellipsoids are then quantized and compressed into .ksplat bundles for zero-install 60 FPS WebGL rendering directly in modern browsers.