Abstract
Edge AI enables machine learning models to operate directly on resource-constrained devices such as microcontrollers, allowing for real-time data processing and enhanced data privacy. This paper benchmarks multiple Edge AI frameworks, the evaluation spans five key metrics: accuracy, memory footprint, inference time, power consumption and usability. The platforms are tested using a human-motion classification dataset and deployed on an Arduino Nano 33 BLE board. Results indicate significant variations in framework performance across these metrics, providing insights into selecting the most suitable frameworks based on specific application requirements. This research highlights the strengths and weaknesses of each framework, helping developers make more informed decisions when choosing Edge AI solutions.
