Technology
ADSS mines continuous video from static roadside cameras and turns every observed road user into a candidate ego vehicle. The result is a scenario database with precise trajectories, criticality information and ASAM-standard output.
The solution
One static camera. Six steps. A searchable scenario database instead of a folder of video files.
Live video from crossings and highways, 24/7.
Every road user detected, classified and tracked.
Image tracks become road-level trajectories.
Trajectories combined with the road model into OpenSCENARIO.
Manoeuvre types and critical-event tags.
Filter by road user, manoeuvre, criticality and location.
Static elevated cameras see traffic very differently from an ego-vehicle camera. Our detection, tracking and ground-projection models are tuned for that fixed viewpoint, so the whole interaction is captured — including road users hidden from any single vehicle.
For every track we derive time-dependent features — trajectory curvature, distance to critical objects, minimum time-to-collision. They drive manoeuvre classification and automatic tagging of critical situations, such as unprotected left turns or an occluded pedestrian or cyclist. Similar cases are clustered into scenario families; outliers are flagged as corner cases.
Road model in ASAM OpenDRIVE, scenarios in ASAM OpenSCENARIO, plus event tags and metadata such as time and location. Open formats mean no vendor lock-in: scenarios replay in esmini and import into IPG CarMaker.
Observed data does not replace generated scenarios — it grounds them. From each site we estimate distributions for speed, time gap, yielding, pedestrian and cyclist crossing timing, lane discipline and traffic composition. Use them to vary logical scenario families, set boundaries for AI-generated variants, and compare a generated portfolio against what actually occurs.
Two intersections can share geometry and road-user classes and still behave very differently. Drivers accept different gaps, yield and brake differently, and interact differently with pedestrians and cyclists. ADSS extracts site-specific distributions — and measures how other road users react to a manoeuvre of the ego.
Beyond replay
Replay answers “what happened”. Validation asks “what if the ego brakes harder, arrives earlier or swerves?” For that, surrounding traffic must react. We are building local traffic behaviour models from observed data — from polite NPCs to a traffic reaction model.
We run pilot projects with OEMs, Tier-1 suppliers, test houses and infrastructure owners. Tell us about your camera, your function under test, or the gap in your validation.