State-of-the-art Image Classification using Machine Learning is becoming as good and, in certain cases, better than Human accuracy at the same tasks. The problem is these algorithms continue to require massive amounts of computational horsepower to achieve these accuracies, which couples them to desktop or server-grade hardware.
This research works to identify a possible path forward to marry these highly accurate algorithms with an ultra low-power device. The results, if successful, would have powerful implications.
Providing low-cost, highly portable hardware the ability to understand the world in which it finds itself has usefulness far outside the scope of this preliminary research.
Stage 1: Coin Recognition
Here is a quick video of the application running with video-output, so the viewer can better understand what is happening behind the scenes: