Developing IoT Devices Empowered by Artificial Intelligence: Experimental Study

Abstract

The number of real-world Internet of Things (IoT) deployments continuously and steadily increases while the capabilities of single IoT devices cannot yet be exploited for the purpose of Artificial Intelligence. Indeed, computation complexity and energy consumption are the constraining requirements for the development and implementation of truly intelligent IoT devices with AI. In this paper we present an experimental study where we perform the observation of time of execution, overheating, Central Processing Unit (CPU) load and power consumption using an embedded system by itself and embedded system with an external low-power Graphics Processing Unit (GPU) able to run the pre-trained neural networks for object detection. We report on the series of experiments conducted on the neural networks with different depth and input size. Experimental results demonstrate high potential of AI application on the IoT devices.