Reliable vehicle detection and classification

To ensure smooth traffic management, reliable vehicle detection and classification are key. They are vital components of modern transportation systems, as they enhance safety, efficiency, and sustainability through data-based decision-making.

In collaboration with our customer, Kapsch TrafficCom AG, we were able to support them in their vehicle classification and detection endeavors, even under complicated circumstances.?

Optical accuracy even in glaring light, fog and darkness?

Vehicle detection is a strategically important issue for our customer. Therefore, we had only two months for the proof of concept to show that we will achieve the required high recognition accuracy. We were so convincing that the customer ultimately invested a significant budget in this pioneering project. And it was worth it, as our work resulted in a patent application for the customer.?

We achieved:?

• 80% recognition accuracy after only two months?
• 99.5% detection accuracy in the end?
• Process objects with a speed of 130km/h?

?

The Challenge: Software designed for fixed camera parameters?

For the rollout on the highway, the camera and computing hardware were predefined, so the software had to be precisely designed to these fixed performance parameters. Whether day, night, backlit by sun and headlights as well as with snow and fog, and be it a passenger car being transported on a trailer – the recognition accuracy must always be right. To ensure this, we built our own annotation tool that is optimized for rapid and simultaneous detection of vehicle classes (factor 20-30). In addition to the changing visibility conditions, it was also necessary to ensure strict real-time requirements regarding object movements of > 130 km/h and to be able to process correspondingly high frame rates.?

The Solution: Successful with in-house AI training platform?

We met the challenging KPIs of >99.5% classification accuracy for 20 vehicle classes. We used our in-house AI training platform for object recognition and image classification. After the short proof of concept phase, we evaluated, optimized and extended our solution over a year, with Kapsch TrafficCom and Cloudflight teams working closely together. Now, vehicle detection can be used for intelligent traffic systems, toll services and bridge inspections, and has since become a standard product in numerous traffic monitoring systems.?

?

Our partner: Kapsch?

Kapsch TrafficCom AG develops toll and traffic management solutions. Headquartered in Vienna, the company has subsidiaries and branch offices in more than 25 countries. Around 4,220 employees worldwide work on projects in more than 50 countries.

Thirsty for knowledge and practical experience in digital transformation?

Subscribe to our newsletter below and get ready for bi-weekly insights into current trends and challenges.