• Customers get a damage assessment in a maximum of 5 minutes.

• Insurers get a quick overview of the total claims volume in an affected region.

• Modular AI algorithm allows rapid expansion of the business model.

Zühlke supports the German company PDR-Team in the development of a mobile hail damage scanner and uses Deep Learning to create a flexible, future-proof and scalable solution.

For a hail damage assessment including valuation, an assessor needs about 40 minutes. PDR-Team wants to change that: using a system that automatically and reproducibly determines the number and size of hail dents on a car – with 90 % accuracy. The solution has to work on the go and without an internet connection, and is intended to expand to include other types of claims in future, resulting in high requirements where the scalability of the hardware is concerned. In addition, the solution needs to be cost-effective to ensure that vehicle insurers can use it for a reasonable price while still achieving a profit.

Building on the concept and hardware from PDR, Zühlke developed a computer vision solution based on CNN and completed a finished demonstrator in just four months. The solution is fully integrated into the existing system architecture. This saves PDR from having to make major investments in additional hardware. The flexible solution uses different methods of object tracking, segmentation, and the detection and measurement of dents and dent patterns. The modular structure of the AI algorithm allows flexible expansion to include other types of claims in future development steps.

Benefit

More speed, accuracy, flexibility and less costs

The solution offers speed, accuracy and cost benefits for customers and insurers. The finished assessment is available in less than five minutes – with an accuracy significantly higher than the 90 % initially specified. Insurers get a quick overview of the total claims volume in a region. By applying the principle to other types of damage, the business model can be expanded quickly.