Supermarket Shopping with the help of Deep Learning

This study presents the development of an innovative system designed to facilitate the shopping experience by assisting customers with shopping list preparation, supermarket route planning, list fulfillment, and checkout. An Android application serves as the user interface, offering functionalities such as product selection, shortest path calculation, shopping list updates, and total cost estimation. The system comprises three separate subsystems: a tablet, a cloud platform, and an embedded system attached to the shopping cart.

Deep learning was employed for product detection during list fulfillment, using a dataset of 41 supermarket products with 3,000 images. Transfer learning was applied to retrain the MobileNet SSD model using the PyTorch framework, which was subsequently deployed on an NVIDIA Jetson Nano. Two cameras mounted on the shopping cart provided image input for the deep learning model.

The system achieved an accuracy of approximately 70% in field tests. Limitations were observed in the detection of certain products, necessitating further training for items obscured by others in the cart. Future work aims to improve detection capabilities and expand the system's effectiveness in streamlining the supermarket shopping experience.