Zcart: a smart cart prototype

Abstract

The recent advancements in artificial intelligence have already impacted many aspects of modern life but have yet to impact one important aspect of the consumer experience in a meaningful way: shopping in physical stores. Tech giants such as Amazon have recently deployed the só called smart carts to their physical retail stores, allowing customers to have an improved shopping experience, including better product information and a seamless checkout process. In that regard, this work analyses the current market and describes the development of a prototype that achieves similar functionality to the ones currently available, using the same building blocks such as computer vision and sensor data. A Deep Learning model was developed for product detection and deployed to a single board computer capable of running inferences at 4 FPS with an average precision higher than 80%. Finally, this work discusses the challenges and practical constraints of developing such a prototype and also presents suggestions for future work to improve the solution into a commercial product.