Product Detection System for Home Refrigerators implemented though a Region-based Convolutional Neural Network

摘要

This paper presents the development of a product detection system in a refrigerator based on pattern recognition by the Region-based Convolutional Neural Network (R-CNN) technique. For this case, five types of products were selected, corresponding to butter, juice, milk, sauce and soda, as the objects of interest to be detected and, in addition, if any of the products is not detected, the user can check in a graphic interface which are the ones that are missing and are necessary to buy. A test scenario was implemented in order to observe the behavior of the network, obtaining a 94% of accuracy in the recognition of the objects, while in the final environment, the overall accuracy increase to 96.3%, with detection times between 0.68 and 0.96 seconds and, additionally, a RoI detection between 92% and 100% accuracy was achieved by the network.