From analogue to digital with a camera and AI
Lenka built a modular system with a Raspberry Pi, camera module, lighting and image recognition software that can recognise multiple types of meters: dials, digital displays and rolling numbers. The camera is pointed at the existing meter, after which the system automatically takes and reads photos periodically.
A specially trained image recognition algorithm converts these into readable data. This took into account the varying readability of the gauges due to light or dirt, for example. The challenge there was mainly in developing an algorithm that could be used as a ‘single shot inference’. So a machine learning model that reads the correct numbers despite pollution, poor lighting and numbers that are not easily readable. To this end, Lenka developed a model that first reads the digits and then reconstructs digits that are not easily readable with a subsequent model. The model already has high accuracy but needs some further development for implementation.
No previous research had been done on this and Lenka graduated cum laude. The information can then be stored or forwarded to a dashboard. The system runs entirely locally, ensuring privacy and reliability.
This provided edge ai
Automatic reading of old meters, less chance of human error, real-time insight without manual work, scalable and affordable for hundreds of locations and the real-time data can be further used for process optimisation.
This solution uses existing infrastructure and lowers the threshold for making production data available digitally. Moreover, it opens the door to further applications with Edge AI, such as predictive analytics based on historical measurement data.
Many SMEs still have machines with analogue meters, thinking that digitisation can only be done via costly replacement. This use case shows that smart image recognition, combined with Edge AI, offers a viable alternative.
more info or register?
Want to make your existing meters smart too? Contact the Fieldlab Edge AI at Andre Gerver for an exploratory interview.
CONTACT
Andre Gerver
📧 a.gerver@techport.nl