Artificial Intelligence reduces waiting time in restaurants
No one likes waiting in queues to be charged, although there are times of day when for certain businesses a build-up of customers is inevitable. Lunchtime is one of those times and especially in fast food outlets, where customers must flow in a permanent and constant manner.
Nostrum (130 establishments in Spain) is a chain of restaurants founded in Barcelona in 1998, with a lot of creativity and innovative solutions for fast, healthy and traditional food to be consumed on the establishment itself or to take away.
Some of Nostrum’s restaurants, especially those located in office areas or with a high density of students, have a large number of people at midday, at specific times, and this sometimes leads to queues to pay.
Based on the idea of speeding up Nostrum’s own checkout, Boira Digital and Pervasive Technologies have developed a rapid automated payment solution for their restaurants, based on Artificial Intelligence.
Boira Digital is a company dedicated to developing retail solutions to improve the shopping experience and Pervasive Technologies is a company specialized in developing predictive models based on Machine Learning and Deep Learning, especially in vision and images.
The Artificial Intelligence system
The AI system developed consists of capturing images of food trays inside an automatic box, processing the images through a module composed of several neural networks and extracting the list of products present in the tray.
This system simultaneously performs the location of the elements present in the tray and their recognition. The process consists of two steps. First, the location is made from a model based on a class activation map and its corresponding location boxes. In a second phase, for each location box the product is predicted by a classification model.
The module, a set of models, can be executed in Google Cloud, on a computer connected to the restaurant network or even at the checkout itself by using a smart camera that combines image capture and processing in a compact device.
For cloud processing we use Google Cloud (GCP) where we have virtualized or docked machine environments, or even Google Cloud AI for mass execution and model training.
For local execution in the restaurant’s management computer we used Tensorflow that runs in a docked environment, with the same functionalities as if it were in the cloud.
If you want to run the integrated model on a smart camera device, we have a camera that combines image capture and processing on a specialized NVIDIA Jetson TX1 device. This card is a small supercomputer, with 256 CUDA cores, 64 bit CPU, processing power of more than 1 TeraFLOP with an interface for cameras up to 1400 MPix/s.
The system will incorporate Nostrum’s entire range of products in the coming months and is expected to be in production in the new restaurants in Spain and abroad.
FastPay can be used in restaurants of other chains and even in other sectors, especially in retail, and can be adapted to the recognition of other types of products.