Predicting Parking Occupancy in Real-time Using Fog Layer Hosted DNN Implemented in FPGA

Abstract

This paper explores Fog Computing and presents the results of processing data streams measured continuously using IoT devices. Machine Learning at the Fog Layer is considered, specifically processing of Deep Neural Networks (DNN). Novel IoT solutions will be possible using a Distributed Platform that does not depend on the Cloud for Data Analytics or storage. This platform uses a simple architectural approach, hierarchically connecting IoT devices, gateways in the Edge Layer, and Fog Computing resources. Our platform comprises a Fog Layer divided into an upper and lower tier. The lower tier manages streams of data while the upper tier hosts Data Analytics operations and a distributed storage system. The Fog layer consists of clusters utilising low-cost devices. Each cluster uses a mix of single-board computers and FPGA-based accelerators.