Fog Computing for Real-Time Accident Identification and Related Congestion Control

This paper focuses on developing (i) a benchmark application for Real-Time traffic incidence identification and related traffic management, using Real-Time congestion-aware navigation of smart vehicles (Edge nodes) with video feeds, (ii) an image database for Deep Learning used for recognition and classification of traffic incidences such as accidents and congestions, (iii) the System Level Software (or Middleware) required for Distributed Computing in such a heterogeneous Real-Time constrained system with Rapid Mobility - today's Internet-of-Everything (IoE), and (iv) a hardware prototype of the distributed computing and storage infrastructure. The video bandwidth requirement of 10-100 GigaBytes of data per minute per vehicular camera makes it a Big Data problem. With millions of smart vehicles projected to be deployed within the next 5 years, BigData from a single vehicle, multiplied with the large number of vehicles, presents a Big-Squared-Data computing space which will easily overwhelm any Cloud infrastructure with its Real-Time or near Real-Time demands. Hence the need for a Fog tier between the Edge nodes and the Cloud to bring distributed computation (servers) and storage closer to the Edge nodes. Such a Fog consists of multiple Fog instances, each one of which services cells or Virtual Clusters of Edge nodes. Results show that Fog-Cloud computing framework outperforms a Cloud-only platform by 79.7% reduction in total latency or response time.