Project

Description

This group project for the MAAI course focused on computer vision. I built an application for self-checkout systems in supermarkets that automatically estimates customer age, reducing the time and effort employees spend on manual checks.

When a product requiring age verification is scanned, the system captures an image of the customer and estimates their age using a pre-trained model, DeepFace, optimized on the UTKFace dataset. If the estimated age is below the required threshold, an employee is notified to verify the customer’s ID. The checkout system communicates with the employee’s monitoring device through WebSockets.

The experiments (see research paper) showed that the optimized model significantly improved recall, correctly identifying over 200 more customers above the age of 25 while still maintaining accuracy for underage detection. Although this came with a slight drop in precision, the improvement in recall reduces the risk of illegal alcohol sales, which was our primary objective.