| Resumo: | After more than two decades of existence, the internet of things has been revolutionizing the way we interact with the world around us. Although, in its origins, the adoption of a cloud computing paradigm supported this ubiquitous computing model, the increasing complexity of IoT systems has led to the gradual fading of the traditional hierarchical model of cloud computing. The search for solutions to the problems of latency, scalability and privacy has, in recent years, driven the movement of data processing and storage, from the cloud, to the edge of the network (edge computing). Starting from the particular case of edge computing that keeps the focus on extending the boundaries of artificial intelligence to the edge of the network - Edge intelligence - a survey of the current state of the art is carried out, culminating into the specification of an architecture to support edge intelligence applications. In order to validate the proposed architecture, two scenarios are presented. In the scope of waste management and energy recycling, a system for used cooking oil classification in a national domestic collection network is presented. With the local classification of the trustworthiness of each deposit, it was possible to significantly shorten the response times, with a direct impact on energy consumption levels. Aimed at smart cities, a second application scenario, proposes an approach based on computer vision and deep learning, for local detection of pedestrians on crosswalks. In this context, an edge intelligence paradigm allowed to overcome privacy related issues, as well as reducing response times by more than 80 times, when compared to a cloud computing based solution.
|
|---|