Status:
✅ Finalized
⇨
Context
Trento, a city of around 119 000 residents on the Brenner transport corridor, combines local urban movement with regional through-traffic on its road network.
Intersections are widely recognised as a concentrated source of road-safety risk, particularly for pedestrians and cyclists.
Specific problem
Trento’s traffic-safety authorities need to identify risky situations at intersections as they develop, rather than only counting accidents afterwards. Detection that depends on sending raw video elsewhere for processing is slow and raises privacy concerns for the people crossing the junction.
Current situation and gap
Video analytics for intersection safety typically rely on footage being transmitted to a central or cloud platform, which adds latency and raises privacy concerns.
Low-latency detection that avoids transmitting identifiable footage was not in standard use.
Specific need
Trento needs real-time, privacy-preserving detection of road-safety risk at its intersections that does not depend on transmitting raw video to the cloud.
Expected outcome
– An edge-based video-analytics approach that processes and anonymises footage at the intersection rather than sending raw video to the cloud.
– The approach should detect risky situations in real time while keeping personally identifiable data on the local device.
– Real-time, privacy-preserving risk alerts for safety-relevant situations at intersections, with only anonymised data leaving the device.
Who would use the results
Trento’s traffic-safety authorities, to identify and respond to intersection-safety risks without transmitting identifiable video.
Expected impact
Earlier identification of risky situations could support targeted safety measures at intersections, while keeping personally identifiable video on the local device.
Space. Intersections in Trento, with scope to extend to further intersections.
Scale. Site scale, initially two intersections, with scope to extend.
Data. Camera infrastructure and video feeds at the test intersections provide the raw input. Anonymised, edge-processed risk-event data can be generated during experimentation.
Constraints. Experimentation is limited to a small number of intersections rather than city-wide deployment.
Experience. Edge-AI approaches to intersection-risk detection have previously been tested at two Trento intersections with research partners.
Get in touch to express your interest or explore how you can contribute.
