Improving automatic incident detection in tunnels
Challenge
There are thousands of vehicular tunnels in use worldwide and that number is expected to continue to grow to alleviate the impact of congested traffic. These tunnels mainly deploy CCTV cameras which are wired back to a central control room where human operators monitor the flow of vehicles to ensure the safety of the tunnel and its users. A known problem is that human operators struggle to scrutinize large numbers of monitors over extended periods of time due to cognitive overload.
ISSD has developed an Automatic Incident Detection (AID) system and deployed this to multiple sites around Turkey to aid the human operators in their monitoring tasks. These systems analyse parallel video streams from multiple cameras in real-time and provide automated alerts to the operators in the event that an issue is detected which they may need to act upon. This makes their job easier, more accurate and efficient, and makes the tunnel safer for its users.
The current systems utilize traditional Computer Vision image processing algorithms which suffer from known limitations of this approach. Thus, the challenge is to improve the system accuracy, reliability and performance while extending its capabilities and ideally reducing the hardware costs.
Solution
Neural networks have shown excellent performance in computer vision related tasks. Thus, the new AID system design will utilise neural networks to achieve the desired goals. By augmenting the current AID solution, it is anticipated that the final system will be able to outperform the current system, detecting and tracking the movement of the various vehicle types or pedestrians with higher accuracy and better performance. To achieve this it is planned to use Vision Processing Units (VPU) that are dedicated to video image processing instead of the current reliance on CPUs. This will provide better performance, reduce the system hardware costs and contribute to a scalable architecture.
FED4SAE Support
Through the EU funded FED4SAE Program, ISSD will be able to apply its many years of experience in tunnel monitoring systems, partnering with fortiss, BLUMORPHO and Intel to deliver this new product.
Partners will provide ISSD with business innovation coaching and contribute their expertise in product development. Fortiss will contribute technically via their Neural Network Dependability toolkit to reduce the uncertainties inherent in the operation of artificial neural networks. The desired solution is targeting the Movidius Myriad X VPU hardware – the most advanced VPU from Intel which features a Neural Compute Engine with a dedicated hardware accelerator for deep neural network inference.
Impact
The prototype system will be deployed and validated at one of ISSD’s existing road tunnel customers in Turkey. The solution will enable ISSD to have a stronger product offering to improve tunnel safety solutions deployed by authorities. The solution can be deployed to both new and – as a retrofit – to existing tunnel control centre infrastructure and thus bring the benefits of AI enhanced computer vision to these markets.