Towards an optimal MEC resources dimensioning for vehicle collision avoidance system: A deep learning approach

Collision detection and avoidance between vehicles is one of the key services envisioned in the Internet of Vehicles (IoV). Such services are usually deployed at the Multi-access Edge Computing (MEC) to ensure low latency communication and thus guarantee real-time reactions to avoid collisions between vehicles. In order to maximize the coverage of the road and ensure that all vehicles are connected to an optimal MEC host (in terms of geographical location), the collision avoidance application needs to be instantiated on all the MEC hosts. This may add a burden on the computing resources available at the latter. In this paper, we propose an AI-empowered framework that aims to optimize the computing resources at the MEC hosts. Our framework uses deep learning to (1) predict the vehicle density to be served by a MEC host and (2) derive the exact computing resources required by the collision detection application to run optimally. We evaluate the proposed framework using a real dataset representing vehicle mobility in a big city. Obtained results show the accuracy of our prediction model and hence, the efficiency of our resources assignment framework to exactly deduce the optimal computing resources needed by each instance of the application.