Designing a Self-Payment Cashier For Bakeries Using YOLO V4

This paper introduces an automation project for bakery stores, using the YOLOv4 object detector. A self-checkout cashier was developed where the client puts a tray with the products underneath a webcam, and the system returns the total price of the purchase. This project focuses on the software and aims to build a low-cost, high accuracy, fast response, and easy to use application. To achieve this goal, it was developed some process and protocols, a user interface, a segmentation algorithm, and a server to integrate it all. The menu defined for this research is composed by 10 items - coke, guaraná, coxinha, banana, apple, ice tea, choux cream, eclair, cheese bread, and bread - and the mAP obtained for our test set is 91.52% and 96.25%, when all requisites were respected. Despite the high precision obtained, two important problems were found during the validation process: the system has some problems recognizing products when they are too close and when an unknown object is put on the trail. Throughout the text, it is provided how this project was developed, the system requirements, and operating protocols.