LiftDepth Risk Demo

A camera-only perception pipeline: RGB → DA2 metric depth → road segmentation → tracked road users → BEV occupancy → object-aware risk.
DA2 metric depth
Road segmentation
YOLO tracking
BEV occupancy
Risk heatmap
Frame 0 / 0

RGB Input + Live Tracks drawn in browser

RGB frame

DA2 Metric Depth geometry source

Depth map

Road / Sidewalk Segmentation context gate

Road overlay

Projected Road BEV green = road + projected objects

Projected road BEV

Object Occupancy Grid rectangular road users

Object occupancy grid

Object-Aware Risk Heatmap distance × center × motion

Risk heatmap
Purpose: this demo shows how a monocular driving frame becomes a structured BEV scene representation. Depth supplies geometry, road segmentation filters context, YOLO tracking gives object identity, and the final occupancy/risk maps focus attention on road-relevant objects.