Documentation
Pipeline
How a frame becomes structured intelligence, in thirteen stages. This is the summary; the full breakdown is on the Pipeline page.
Every analysed frame flows through the same stages:
- Input · acquire a source (camera, stream, video, image).
- Frame Capture · bounded, drop-oldest buffering.
- Preprocessing · normalise, low-light enhance, background plate.
- Detection · YOLO11 + ByteTrack with class gating.
- Segmentation · mask movers vs the static plate.
- Depth Estimation · Depth Anything V2, temporally fused.
- Pose Estimation · keypoints for intent and gait.
- Geometry Fusion · fuse depth, masks, background completion.
- 3D Reconstruction · back-project to a point cloud.
- Mesh Optimization · triangulate + edge culling.
- Texture Mapping · project RGB and inpainted texture.
- Spatial Analysis · foresight, identity, alerts.
- Output · stream, export, persist.
Full detail
Expand every stage with its inputs, outputs and internals on the Pipeline page.