How SBC and AI
work in the overall system

The scanner contains an optical camera-based system that detects surface irregularities based on the refraction of light. Intelligent algorithms process the camera images from three-dimensional space into usable two-dimensional images. A number of measurements are taken, such as speed, lateral distances, height and the shape of the vehicle, to create a blueprint of the vehicle with its shape and irregularities. The AI was previously trained to recognize the hail-specific dents and distinguish them from other similar-looking dents – such as those caused by water droplets. The experts classified a large number of images so that the AI could learn to distinguish between the dents. Today, it can detect almost all types of damage.
Five SBCs are installed in each scanner sheet, one for each of the five cameras connected to the camera system.
They perform each calculation independently. “This is of course a great advantage, because we can run everything in parallel, which saves a huge amount of time,” says the managing director. A further, sixth computer acts as an interface computer, coordinating communication within the system to control the individual SBCs and issue instructions. It ensures that everything is started, stopped, recorded, analyzed and evaluated at the right time. It also bundles the information and handles communication with the operating laptop or other connected computers to forward the scanner output.

The benefits of a local solution

“The AI is of course constantly being trained and improved, also based on different conditions. Hail always varies from region to region. This results in different dent shapes and intensities,” says the hail expert. Dents also look different on plastic parts. This ability to detect dents also means that fraud can be prevented: Blankenheim knows that hail events are often the subject of insurance fraud. But the scanner also detects dents made by car owners themselves.“ So if there are hundreds of dents on the roof and the scanner only detects twenty, then it gets exciting,” says Tino Blankenheim with a smile.

Data privacy guaranteed

The company deliberately opted for the SBCs rather than a cloud solution because the scanner should be able to work offline in places with poor Wi-Fi, such as an underground car park, and calculate the result locally.
Appropriate interfaces are used to import and export data. “This is an advantage for us in terms of data protection, as each customer keeps their data exclusively for themselves. It has also made it much easier for us to further develop the system from an IT security point of view,” concludes Tino Blankenheim.
Measuring process in the scanner sheet

In addition, the approximately 40 Gigabyte of data processed per vehicle do not have to be transmitted. The compression required for this would lead to a loss of quality, which the hail specialists are not prepared to accept.
The company will continue to focus on optimization in the future. They are currently working on a scanner for caravans - and, of course, on further internationalization, as the hail problem is becoming increasingly important worldwide.
 HAST hail scanner
Innovative Technology

Single-board computer
with AI and graphics functions

Kontron presents 3.5"-SBC-PTL with
Intel® CoreTM Ultra processors (Panther Lake)

Kontron will present the latest generation of its 3.5-inch SBC product family at embedded world in Nuremberg. It offers a powerful and energy-efficient solution for demanding AI applications and meets the specific requirements of industrial edge computing. For example, it enables direct processing of image data and makes autonomous decisions without the need for a cloud connection or associated latency. Samples will be available in the second quarter of 2026.
The 3.5"-SBC-PTL leverages innovative computing architecture, based on Intel® CoreTM Ultra processors (series 3) which support up to 16 cores, an integrated Neural Processing Unit (NPU) for energy-efficient AI inference, and up to 12 Xe GPU cores for high-performance AI (up to 180 TOPS) and video analytics. This hybrid architecture provides the flexibility to run complex, parallel AI pipelines efficiently — all without the need for a separate GPU. As a result, AI workloads can be processed directly at the data source, significantly reducing latency and overcoming connectivity barriers.

Enhanced Scalability

With integrated Intel® Xe Matrix Extensions (Intel® XMX) and specific OpenVINOTM optimizations the 3.5"-SBC-PTL further boosts the performance of each compute engine.
Loading... Loading...
×