In 2004 LCI visited one of the two biggest companies manufacturing scales worldwide. We built together one of the best optical produce recognition systems.
Before that visit we thought hard what LCI, as an AI-based company, could offer a retail scale manufacturer. There are many people who cannot remember well the codes one must tab on the scale. An embedded optical fruit and vegetables recognition system might help there. Their CTO was a first class engineer and inventor. He was pessimistic: their biggest client is french and they use specially trained personnel for weighing. No software could do that job.
We nodded in resignation: often great ideas just don’t stand the reality test. Two weeks later the competition introduced the first scale with automatic object recognition and we were in business. The project was codenamed “Seeing Scales”. It started as a modest research project, become a feasibility study, then a fully fledged test project before going into full production.
Since 2009, Carrefour, the second largest hypermarket chain in the world, is using these scales and so do other major retail companies. The French are doing a really good job at training their scales.
When started in 2004, this project was a huge technological challenge, because it stretched the boundaries of embedded systems. Using a slow industrial processor, a really small amount of RAM, and a standard web camera, the system achieved over 90% accurate classification result in about 300 milisecs.
With the latest R-EF classification engine ("Full Metal") the "Seeing Scales 5.0" object recognitionlibrary achieves a 94% recognition rate, even when the fruits and vegetables are packed into milky plastic bags.