Augmented reality is gaining popularity in numerous fields such as healthcare, visualization, entertainment and education. Most of the commercially available AR devices like Atheer AiR, Microsoft Hololens and Google Glass have limited power, storage and on-chip computation capabilities for example currently Hololens has storage ~64 GB and RAM ~2GB. In turn, these devices often rely upon offloading storage as well as compute to an architecturally centralized cloud server while ensuring application response time. Mobile Edge computing (MEC) does not replace but complements the cloud infrastructure as edge clouds are resource limited in terms of bandwidth and compute. Thus, for a resource constrained system it is required to allocate resources per request while taking system capacity into consideration.
Researchers at WINLAB, Rutgers University created a low-latency AR application using MEC. Fig. (a) shows the process flow of AR implementation at WINLAB for a demo application using Microsoft Hololens. A client sends a continuous video stream to the edge server which processes the information based upon application type and returns output to the client. The video stream (30 fps) is processed for each frame. The edge server is connected to the client in two hops: (i) edge to first hop router and (ii) router to Hololens. The following use-cases are evaluated.
Smart Navigation: A user enters a building. The edge cloud in the building has her contextual information from calender entries and GPS. As shown in Fig. (b) the user is navigated to meet a person in the building using a set of cubes appearing on the device as she moves. Achievable latency is critical here because real-time activities of the user can be disrupted by late arrival of AR information.
Find out what's happening in Middlesexfor free with the latest updates from Patch.
Annotation based assistance: In this scenario, a user looks at an object having a set marker through Hololens with an intention to get supplementary information about the object. In Fig. (c), user looks at the printer and the status, ink level, number and current jobs are annotated on the user's display.
A thorough evaluation is done for these two use-cases and it is concluded that core cloud only system outperforms the edge-only system when inter-edge front-haul bandwidth is low. The study shows that adding capacity to an existing edge resource without increasing inter-network bandwidth may actually increase network-wide congestion and can result in reduced system capacity. Therefore, the reality of Augmented Reality is still far away until the inter-network bandwidth are upgraded.
