Environmental Perception in Autonomous Vehicles Using Edge Level Situational Awareness

Abstract

Currently, assisted vehicles depend on GPS to deliver accurate navigation during their drive, making cellular network access essential to their function. Absence of network connectivity leads to navigation failure on vehicles that will require understanding the path ahead without GPS. Such a scenario is not a practical option when considering fully autonomous vehicles. A video analytics framework powered by edge computing can help tackle the challenge efficiently. In this paper, a general overview of the recent progresses and challenges in Autonomous Vehicles (AVs) is presented and an idea for new generation of assisted framework for AVs is discussed where the perception about the environment is achieved through drone level imagery. In this new technology, a fleet of drone provides situational awareness for autonomous vehicles and communicate with control system for better perception and more accurate decision. The main characteristic of the new assisted framework is to (i) provide drone level camera for image acquisition for a view of the area of interest of the autonomous vehicle, (ii) deliver Edge analytics using deep learning for on-board GPU based training model to provide situational awareness about the route with available geo-tagged images and landmarks, and; (iii) integrate perception and prediction with the autonomous vehicle decision making system for reliable and precise navigation.