In Central America, beer is commonly sold out of fridges in small grocery shops. Beer companies don’t deliver to these shops directly, consequently dealing with a lack of stock insights. Wishing to control this part of the supply chain and making sure fridges will be refilled in time, our client asked us to build a prototype that automatically retrieves data from a fridge. Within three months we built a smart fridge prototype able to pilot in short term.
diameter of a typical can
bottles and cans counted
training pictures made
The biggest challenge for measuring fridge data, is finding the physical option to put a camera or sensor. Fridges are optimized to cool as many bottles and cans as possible, leaving little space to add any hardware. Within these constraints we reckoned making a 3D map by a camera attached to a shelf would deliver us the data needed to get real time stock info.
Prototyping is about failing and trying harder. In this case, we found out the aluminum cans disturbed the laser reflection of the 3D camera, making it impossible to make a 3D map of every shelf and recognize 50cl from 33cl cans. We switched plans, turning to a new option: using time of flight sensors, a rail and a camera sending the photographed colored caps to a server and dashboard. This way we could retrieve real time stock information, distinguishing different brands, the first stepping stone towards recognizing the entire stock. This new tech innovation is currently waiting for patent approval.
Computer Vision Model recognising different branded stock.
10% estimated sales increase based on better stock management.
Boundary object as innovation conversation piece.