Rozpoznání objektů na malých mikropočítačových platformách

Abstract

This bachelor’s thesis focuses on the design, optimization, and experimental evaluation of the Rasp- berry Pi 5 and Raspberry Pi Zero microcomputer platforms for image-based object recognition using neural networks. The primary goal is to identify a suitable combination of hardware and software that enables on-device inference without the need for powerful graphical accelerators. The thesis compares both platforms in terms of computational performance, latency, throughput, and memory requirements, and evaluates frameworks suitable for IoT deployment, particularly TensorFlow Lite and TensorFlow Lite Micro. The outcome is a functional system integrating a camera module and a selected microcomputer capable of detecting target objects in real time or near real time. The system’s functionality is validated through a series of experimental tests, and the measured results are compared with existing published studies.