Transparent AR Processing Acceleration at the Edge

Abstract

Mobile devices are increasingly capable of supporting advanced functionalities but still face fundamental resource limitations. While the development of custom accelerators for compute-intensive functions is progressing, precious battery life and quality vs. latency trade-offs are limiting the potential of applications relying on processing real-time, computational-intensive functions, such as Augmented Reality. Transparent network support for on-the-fly media processing at the edge can significantly extend the capabilities of mobile devices without the need for API changes. In this paper we introduce NEAR, a framework for transparent live video processing and augmentation at the network edge, along with its architecture and preliminary performance evaluation in an object detection use case.