Дослідження використання нейронних мереж з функцією локалізації на платформі Raspberry PI

Abstract

The paper provides a review of computer vision methods that perform classification and localization of objects in an image. Neural network architectures have been selected for embedding in the Raspberry PI single-board computer. The performance of SSDLite and YOLOv8n was tested on two versions of Raspberry PI with the Google Coral hardware accelerator. Practical recommendations are given for choosing the image size at the input of a neural network and the type of network to ensure real time when processing video camera frames.