Tunnel & Highway Incident Detection

A vision system watching highway and tunnel camera feeds for the events that matter — stopped vehicles, wrong-way driving, debris, pedestrians where there should be none — and alerting operators fast enough to act.

Deployed on the highway network around İzmir and in the road tunnels of Trabzon, two very different environments: open highway with weather and changing daylight, versus tunnels with constant sodium lighting and no GPS.

The binding constraint was never model accuracy. It was latency on hardware already installed at the roadside. Models were optimized through OpenVINO to run inference on-site rather than streaming video to a datacenter, and that choice is what made real-time alerting possible at all.

The other constraint was false positives. An operator alerted twenty times an hour stops looking, so precision mattered more than recall in a way benchmark leaderboards don’t capture.

Dependability

The work ran under FED4SAE, an EU Horizon 2020 programme for accelerating cyber-physical systems to market, in collaboration with Intel and fortiss GmbH. That partnership is where the dependability side came from — how you argue a neural network is fit for a safety-relevant deployment, not just accurate on a test set.

Both published outputs came from this work:

• Incident Detection on Junction Using Image Processing
• Application of the Neural Network Dependability Kit in Real-World Environments

ISSD Bilişim Elektronik A.Ş. · 2018–2021