Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?

arnumber=8643424

IEEE.org IEEE Xplore IEEE-SA IEEE Spectrum More Sites SUBSCRIBE SUBSCRIBE Cart  Create Account
Contents
Personal Sign In !

& Browse ' My Settings ' Help ' Institutional Sign In

Institutional Sign In

All % $

ADVANCED SEARCH

Journals & Magazines > Proceedings of the IEEE > Volume: 107 Issue: 4

Web AR: A Promising Future for Mobile Augmented Reality

—State of the Art, Challenges, and Insights
Publisher: IEEE Cite This Cite This  PDF

6 Author(s) Xiuquan Qiao ; Pei Ren ; Schahram Dustdar ; Ling Liu ; Huadong Ma ; Junliang C… All Authors

7 7234 More Like This

Paper Full
Citations Text Views Mobile Augmented Reality Survey: From
Where We Are to Where We Go
IEEE Access
Open Access Published: 2017

Pervasive Personal Computing in an

Internet Suspend/Resume System
Abstract Abstract: Mobile augmented reality (Mobile AR) is gaining increasing attention from IEEE Internet Computing
Published: 2007
both academia and industry. Hardware-based Mobile AR and App-based Mobile AR are
Document Sections the two dominan... View more Show More

I. Introduction Metadata
Top Organizations with Patents
Abstract:
II. Background: on Technologies Mentioned in
Mobile AR Mobile augmented reality (Mobile AR) is gaining increasing attention from both academia
This Article
Principles and and industry. Hardware-based Mobile AR and App-based Mobile AR are the two
Typical dominant platforms for Mobile AR applications. However, hardware-based Mobile AR
Implementation implementation is known to be costly and lacks flexibility, while the App-based one
Mechanisms requires additional downloading and installation in advance and is inconvenient for
cross-platform deployment. In comparison, Web-based AR (Web AR) implementation
III. Different Web AR
can provide a pervasive Mobile AR experience to users thanks to the many successful
Implementation
Approaches deployments of the Web as a lightweight and cross-platform service provisioning
platform. Furthermore, the emergence of 5G mobile communication networks has the #
IV. Open Research potential to enhance the communication efficiency of Mobile AR dense computing in the
Challenges Web-based approach. We conjecture that Web AR will deliver an innovative technology
V. Conclusion to enrich our ways of interacting with the physical (and cyber) world around us. This
paper reviews the state-of-the-art technology and existing implementations of Mobile AR,
as well as enabling technologies and challenges when AR meets the Web. Furthermore,
Authors
we elaborate on the different potential Web AR provisioning approaches, especially the
adaptive and scalable collaborative distributed solution which adopts the osmotic
Figures
computing paradigm to provide Web AR services. We conclude this paper with the
discussions of open challenges and research directions under current 3G/4G networks
References
and the future 5G networks. We hope that this paper will help researchers and
developers to gain a better understanding of the state of the research and development
Citations
in Web AR and at the same time stimulate more research interest and effort on
delivering life-enriching Web AR experiences to the fast-growing mobile and wireless
Keywords
business and consumer industry of the 21st century.

Metrics
Published in: Proceedings of the IEEE ( Volume: 107 , Issue: 4 , April 2019 )
More Like This
Page(s): 651 - 666 INSPEC Accession Number: 18547593
Footnotes
Date of Publication: 18 February 2019 DOI: 10.1109/JPROC.2019.2895105

1 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

ISSN Information: Publisher: IEEE

Contents
Funding Agency:

SECTION I.
Introduction

The phenomenal growth of augmented reality (AR) [1]–[2][3] over the

past decade has attracted significant research and development efforts
from both academia and industry. By seamlessly integrating virtual
contents with the real world, AR makes it possible to provide users with a
sensory experience beyond reality. Especially, in recent years, the
advances in the following three technologies have further fueled the
research and development of AR: the emergence of dedicated AR devices
(e.g., Google Glass, Microsoft Hololens and Epson Moverio BT-300,1 and
Magic Leap) and powerful development kits (e.g., ARCore and ARKit),
the improvements in the performance of mobile devices and sensor
integration, and advances in computer vision (CV) technologies. AR has
offered tangible benefits in many areas, such as entertainment,
advertisement, education, navigation, maintenance, and so on [4]–[5]
[6][7][8]. For example, Pokémon GO, a location-based AR game, has
reached over 500 million downloads in more than 100 countries within
just eight weeks of its public release [9]. Both AR and virtual reality (VR)
can alter the perception of our presence in the world. However, AR,
unlike VR, which transposes our presence in the world to elsewhere,
allows users to be present in the world and simply “augments” our
perception of the world by adding the ability to provide users with
contextually relevant information beyond our current perceived state of
presence. Digi-Capital [10] forecasted that the global VR/AR market
would reach 108 billion dollars by 2021, and Cisco [11] reported that the
global AR traffic would increase sevenfold between 2016 and 2021.

The historical evolution of AR is shown in Fig. 1. Beginning from the first

time Sportvision applied AR to live TV (1st & Ten, 1998), and then with
the first dedicated AR device (Google Glass, 2012) and smartphones
(Project Tango, 2014), and on to the first phenomenal AR App (Pokémon
GO, 2016), it has become clear that both AR technologies and devices
tend to be powerful, mobile, and lightweight. However, the current
mobile augmented reality (Mobile AR) hardware and operating systems
(e.g., Embedded Linux, Android, iOS) present a complex diversity. Most
Mobile AR applications or solutions are designed based on a specific
platform and lack cross-platform support. To reach more users, an AR
application needs to go through repeated development cycles to
accommodate different platforms [12], which undoubtedly increases the
cost of development and deployment. Although there are already some
preliminary attempts toward Web-based AR (Web AR), the limited
networking and computing capability greatly hinder its practical
application. Since 2017, the Web AR provisioning solution has gradually
attracted developers’ attention again due to the ever-increasing
development of user device and mobile network and has emerged as a
promising direction for Mobile AR.

Fig. 1.
Historical evolution of AR.

The invention of the World Wide Web marked the beginning of a new
era, with a Web-based service provisioning paradigm. The native cross-
platform and lightweight features of the Web simplify service access for
users, thereby facilitating the large-scale promotion of Web-based

2 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

applications. Besides Web browsers, many mobile Apps (e.g., Facebook

and Snapchat) nowadays are also designed in aContents
hybrid (Native + Web)
way, which has both the advantages of good interaction experience and
cross-platform support. All of these infrastructures provide a common
platform for the pervasive promotion of Web AR. Here, we define Web
AR as a type of Web AR implementation approach.

Although the technology of the Web offers a promising approach for the
cross-platform, lightweight, and pervasive service provisioning of Mobile
AR, there are still several challenges to applying Web AR in real cases.
Computational efficiency, energy efficiency, and networking are three
important challenges. AR is a computation- and data-intensive
application. The limited computing and rendering capabilities on the
Web make it more challenging to achieve a high-performance and
energy-efficient Web AR. First, the limited performance of a Web AR
application will significantly degrade the user’s experience. Second, the
battery on the mobile device will face tremendous pressure caused by the
complex computation tasks, as it is only designed for common
functionalities. To achieve better performance, Web AR applications
usually take advantage of a way to off-load computation (e.g., cloud
computing) to accelerate the process. However, computation offloading
may introduce an additional communication delay, which will impact the
user experience and limit its application under the current mobile
networks.

The good news is that several technological advances have started to

enter the landscape of Mobile AR. First, the upcoming 5G networks [13]
bring new opportunities for Mobile AR, especially Web AR. They provide
higher bandwidth (0.1~1 Gb/s) and lower network delay (1~10 ms),
which improves the data transmission on mobile networks. Second, the
introduction of new characteristics, such as multiaccess edge computing2
(MEC), device-to-device (D2D) communication, and network slicing,
provides an adaptive and scalable communication mechanism that
further provides efficient infrastructures for the deployment and
promotion of Web AR. The soon-to-be-available 5G networks and the
rapid performance improvement of mobile devices, therefore, have laid a
solid foundation for the practical deployment and application of Web AR
on a large scale.

The rest of this paper is organized as follows. Section II presents the

principles of Mobile AR and three typical implementation mechanisms,
as well as the challenges and enabling technologies for when AR meets
the Web. Section III summarizes the different Web AR implementation
approaches based on experience from real-world use cases and
experiments. Section IV discusses the challenges ahead and some future
research directions. We conclude this paper in Section V.

SECTION II.
Background: Mobile AR Principles
and Typical Implementation
Mechanisms

AR was defined as a technology that integrates virtual objects with the

3-D real environment in real time and supports interaction by Azuma [2].
In this section, we give an overview of the principles of Mobile AR and
summarize the three typical Mobile AR implementation mechanisms,
followed by the challenges the Web AR will face when applying it to real
cases. Finally, we detail some enabling Web technologies that are
necessary or recommended for the implementation of Web AR.

A. Mobile AR Principles
AR is a visual technology between VR and real reality. By superimposing
computer-generated virtual content over the real world, AR can easily
help users to better understand their ambient environment.

A typical AR process is shown in Fig. 3. The camera and other types of

3 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

sensors are used to continually gather user ambient information. The

environment perception analyzes the capturedContents
information (e.g.,
image/video, location, and orientation) for real-world recognition and
perception. In the meantime, the user’s interaction information is also
gathered by the sensors and then analyzed for the purpose of tracking
objects. Both the results of the perception of the environment and the
interaction are used for a seamless integration of virtual contents with
the real world, i.e., a rendering operation is performed, after which the
AR will be presented to the user.

Fig. 2.
Current and future application areas of Web AR.

Fig. 3.
Typical AR process.

B. Typical Implementation Mechanisms

The advances in mobile devices, including computing and display
platforms, provide more choices for the implementation of AR
applications. In accordance with the tracking technologies, we detail the
typical Mobile AR implementation mechanisms in terms of three aspects,
that is, sensor-based, vision-based, and hybrid tracking methods.

Different implementation mechanisms are naturally with different

complexities regarding computing, networking, and storage. The sensor-
based method is a relatively lightweight Mobile AR implementation
approach, while in contrast, the vision-based approach places high
demands on the computing and storage capabilities of the runtime
platform, as well as network capability. As shown in Fig. 4, the hybrid
tracking mechanism is obviously a compromise solution.

Fig. 4.
Computational/storage/networking complexities for the three typical
implementation mechanisms.

1) Sensor-Based Mechanisms:
Mobile devices nowadays already support a variety of sensors, such as
accelerometers, gyroscopes, compasses, magnetometers, GPS, and so on.

4 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

Much effort has been devoted to this type of implementation mechanism

Contents
[16]–[17][18]. An obvious example of this is Pokémon GO, which
provides AR experience by leveraging the location-based service
technology, has launched an unprecedented revolution in the field of
mobile AR. Note that the camera can be enabled to capture the
surrounding environment, but only for the display of the environment as
the background. In addition to Mobile AR implementation mechanisms
based on a single sensor, combining different sensors allows many
applications to achieve more accurate tracking results [19]–[20][21]. The
increasing of sensor category, as well as the continuous enhancement of
sensor functionality, provides the basis and opportunities for the
diversification of Web AR applications. Considering the complexity of
computation, storage, and networking, this lightweight Web AR
implementation mechanism is currently the lowest option for users to get
started. However, this method works in an open-loop way, which will
result in an unavoidable cumulative error, since the tracking error cannot
be evaluated and corrected in real time.

2) Vision-Based Mechanisms:
Similarly, the camera on the device captures the surrounding
environment, but it further provides the basis for vision-based object
recognition, detection, and tracking. This type of mechanism uses feature
correspondences to estimate pose information to align the virtual content
with real-world objects and is analogous to a closed-loop system.
Depending on different features, it can be divided into two methods, as
discussed in the following. The frame-by-frame tracking approach avoids
the aforementioned error accumulation. However, it introduces heavy
computational pressure on mobile devices, especially for natural feature
tracking methods. Besides the improvement of device capability, the
advances in network (e.g., the upcoming 5G networks) will provide
another approach to the problem of inefficient Web AR application
performance, i.e., computation outsourcing (see Section III-B).

The marker-based method uses a predefined marker to meet the tracking

requirement, including two ways as follows.

1. The fiducial method has predefined shape, size, color, and

properties, as shown in Fig. 5. It can achieve superior accuracy and
robustness in changing environmental conditions. Its easily
identified features made it popular in the early stages of
development [23]–[24][25]. However, the difficulty of deploying
and maintaining fiducials in an unknown or outdoor environment
has limited the scope of its application.

2. The natural feature tracking method avoids the difficulties of the

fiducial tracking method mentioned earlier and thus has a broader
range of applications.

Fig. 5.
Several planar pattern marker systems [22] used in AR. (a) Intersense. (b)
ARSTudio. (c) ARToolKit. (d) ARTag.

The two approaches are listed in the following.

1. 2-D Image: The real-world image method (e.g., based on a

photograph or a poster) is an alternative to the fiducial-based
method. Note that this method requires a more powerful object
detection and recognition algorithm; not all 2-D images can be
used for AR pose estimation, for example, a solid color image
without any pattern.

5 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

2. 3-D Object: It is natural to extend the tracking of objects from a

2-D image to a 3-D object. Many algorithms Contents
are already available
for specific nonregular objects, such as human faces, but this is still
challenging for general recognition. Although this type of method is
in its early stages, its potential value still merits attention.

The markerless method detects and understands an unknown or outdoor

real-world environment (e.g., the locations of walls), and no
preknowledge of the environment is required, which will efficiently
promote large-scale Mobile AR. However, it is more challenging to adopt
simultaneous localization and mapping (SLAM) [26], the core part of
markerless environment perception, to Mobile AR, mainly because of the
computational inefficiency and limitations of the resources of mobile
devices. Current solutions mostly rely on a collaboration between SLAM
and other sensors.

3) Hybrid Tracking Mechanism:

The hybrid Mobile AR implementation mechanism is a compromise,
taking into consideration the computational inefficiency of mobile
devices. It overcomes the weaknesses and limitations of the individual
methods mentioned earlier by combining different methods. Many
applications have demonstrated the suitability of this approach
[27]–[28][29]. It not only provides Mobile AR applications with
convincing precise and robust results but also reduces the computational
complexity. Considering the limited computing capability and network
performance, this hybrid scheme will play an important role in the
promotion of Mobile AR on a large scale at present.

C. Challenges When AR Meets the Web

In recent years, with the rapid development of hardware, especially with
the emergence of artificial intelligence (AI) chips, the computing
capability of mobile devices has been greatly improved, which basically
satisfies the intensive computational requirements of Mobile AR
applications (i.e., App-based Mobile AR implementations). However,
there are still several challenges that cannot be ignored when applying
AR applications on the Web in real cases.

1) Limited Computing Capability Versus the Requirement of

Powerful Computing:
Tracking and registration are the two core parts of the AR system and
also the most computationally intensive parts. The fiducial tracking
method provides an accurate and robust tracking approach for Web AR
applications since only simple matrix operations are required. The
natural feature tracking method has a broader range of applications than
the fiducial tracking method. However, the inefficient computing
capability on the Web makes it hard to apply it in real cases. For example,
ORB [30], as a lightweight CV algorithm, still cannot meet the computing
requirements of AR on the Web. For the markerless Mobile AR
implementation method, although there have already been some efforts
to reduce the computational complexity of SLAM (e.g., ORB-SLAM [31]),
it is still challenging to port it to the Web. In addition to the difficulty of
computing on the Web, rendering is another challenge for Web AR, due
to the limited rendering capability of Web browsers. Moreover, diverse
and inefficient computing platforms (e.g., built-in browser) also result in
a degradation of the performance of Web AR. As mentioned earlier, it
will be an efficient and promising approach for Web AR by offloading
computation-intensive tasks to the edge or remote cloud for acceleration
(see Section III-B), especially in the upcoming 5G networks.

2) Network Delay Versus the Requirement of Real-Time

Performance:
Cloud servers always have a more powerful computing capability and
thus provide a performance improvement for Web AR applications.
However, AR is a computation- and data-intensive application. Large
communication delays are introduced by offloading computing tasks to
the cloud. It is therefore difficult for current mobile networks to support
real-time operations (e.g., tracking and interaction), due to the limited
data rate and unacceptable network delay. Web AR applications are more
dependent on mobile networks. The advanced technologies, such as
software-defined network and network function virtualization, which

6 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

provide us opportunities for the adaptive and intelligent network

resource scheduling (e.g., network slicing in theContents
5G era) and further
differentiated Web service provisioning according to the application
characteristics, are also worth our attention. Besides the computation
outsourcing, the self-contained solutions are also important. With the
development of the AI chip technologies, computations can be finished at
the user equipment, thereby avoiding the extra communication delays
caused by network transmissions.

3) Limited Battery Capability Versus Extreme Energy Consumption:

Web AR is a power-hungry application, but most Web AR applications
nowadays suffer from a limited energy supply. The need for the sensors
to cooperate over a long period of time, the analysis of the information,
computing, communication, and display, puts tremendous pressure on
the battery of the mobile device. However, current batteries on mobile
devices are only designed for common functionalities, such as telephone
and Internet access. The extreme energy consumption referred to will
significantly hinder the deployment of Web AR on common mobile
devices. The computation outsourcing mechanism can alleviate the
energy consumption of the end device by offloading computing pressure
to the cloud, but it also subjects to network conditions.

4) Diverse Enabling Infrastructures Versus the Requirement of

Pervasive Promotion:
The diversity of computing and display platforms, operating systems, and
even data formats gives rise to a serious compatibility challenge. As
mentioned earlier, many mobile applications are designed in a hybrid
way, where the built-in browsers are simplified for the purpose of being
lightweight. However, the diversity of computing platforms hinders the
pervasive promotion of Web AR. Moreover, supporting different sensors
and display platforms, as well as operating systems, also makes the
development of Web AR challenging. Note that the virtual contents
created by different tools face compatibility challenges as well for use on
the Web. However, all these compatibility challenges require the joint
efforts of academia, industry, and standards organizations.

D. Enabling Technologies for Web AR

Some advanced Web technologies nowadays are emerging to meet the
basic requirements of Web AR, and, moreover, also provide performance
improvement approaches. Fig. 6 shows four major Web-enabling
technologies.

Fig. 6.
Browser support tables of enabling Web technologies (i.e., WebAssembly,
WebGL, WebRTC, and Web Workers) for Web AR application up to December 1,
2018 (Source: https://caniuse.com).

1) WebRTC [32]:
This technology provides browsers with real-time communications and is
one of the most important and basic technologies for Web AR. The
camera captures the ambient environment in the form of a video stream
by using the WebRTC technology, which provides the basis for further
perception of the environment, rendering, and other operations in a Web
AR application. A large number of browsers nowadays have already
supported this technology. Besides video capture, the WebRTC
technology currently also supports video coding, encryption, rendering,
processing, and so on. However, considering the limited capability of
mobile Web platforms, an efficient WebRTC solution for Web AR is still
worth our attention.

7 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

2) WebAssembly [33]:
Contents
To simplify the programming process and achieve native speed, the
recently emerged WebAssembly is designed as a computational
acceleration approach on the Web by encoding procedures (e.g., C, C++,
Rust, and Go) into a size- and load-time-efficient binary format, which
can be executed on the Web directly [34]. Mainstream browsers (e.g.,
Chrome, Firefox, and Safari) have also started to support this Web
technology. WebAssembly solves the bottleneck problem of JavaScript
and has therefore caused wide concern. It not only improves Web AR
application performance but also makes the development process easier
bringing it into a close relation with current mature CV algorithm, for
example, OpenCV.js [35], [36], the WebAssembly version of OpenCV.
The emergence of WebAssembly will bring a revolution to the Web
platform [37].

3) Web Workers [38]:

This introduces the multithread technology to JavaScript. It utilizes
worker threads to achieve parallelized computing, rendering, and
resource loading in an asynchronous way, and moreover, it has already
been the part of HTML5 specification. As another computational
acceleration approach, Web Workers provide a simple method for
program parallelization of Web AR applications, such as 3-D model
predownloading and parallelized feature points’ matching. By scheduling
and balancing the time- and resource-consuming operations in Web AR
applications, it can provide users with a better experience, especially
under the current mobile networks.

4) WebGL [39]:
This provides a hardware-based (GPU) rendering acceleration approach
on the Web. Since image processing has a strict requirement of the
computing resources, an efficient computing platform is, therefore,
important for computation-intensive applications. A set of efficient
JavaScript APIs for interactive 2-D and 3-D graphics rendering is
available in this library. The use of a GPU in the mobile device makes the
presentation of AR smoother and more realistic on the Web. Also, worth
mentioning is Three.js [40], a WebGL-based JavaScript library, which
helps developers work with 2-D and 3-D graphics on a browser using
WebGL in a simpler and more intuitive way. WebGL 2 specification
finished in January 2017 and this technology has been widely supported
in modern browsers.

The continuous development of these technologies mentioned earlier

provides a basis for the Web AR applications and, more generally, will
also motivate the innovation of Web-based applications. In the
meantime, these applications will further spawn new Web technologies.

SECTION III.
Different Web AR Implementation
Approaches

To explore the potential of AR on mobile devices, both academia and

industry are now seeking more efficient implementation approaches to
compensate for the gap between the user experience of Mobile AR
application and the limited capability of the Web browser. Web AR, as a
branch of Mobile AR, has recently attracted a great deal of attention due
to its lightweight and cross-platform features. Depending on the different
computing paradigms, we can classify the Web AR implementation
approaches into two types as follows.

1. Self-contained method executes all tasks on the mobile device

locally (i.e., offline approach). The advantage of this method is that
it is less dependent on mobile networks, so the real-time tracking
performance will not be degraded by additional communication
delay. However, the inefficient computing capability of the mobile
device becomes its fatal flaw; current mobile devices still cannot
carry out these tasks very well, especially on the Web.

8 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

2. Computation outsourcing method leverages the computation and

Contents
storage capabilities of the cloud servers, and it can usually provide
a better user experience than the aforementioned self-contained
one. However, this method has a strong dependence on the mobile
networks, and therefore, the performance of Web AR applications
is easily affected by network conditions.

A. Self-Contained Method
There are two main implementation approaches for the self-contained
method. One is to develop pure JavaScript-based libraries or plug-ins to
provide Mobile AR services on the Web. The other is to extend the
browser kernel to achieve better Web AR application performance. We
will now present these two approaches in detail.

1) Pure JavaScript Library/Plug-In:

As mentioned earlier, Mobile AR implementation methods based on
fiducial tracking can always provide an accurate and robust identifying
and tracking performance due to their low computational complexity.
Many dedicated JavaScript libraries/plug-ins are already available to
support AR services on the Web, including JS-ArUco [43] (a port to
JavaScript of the ArUco), JSARToolkit [44] (based on the original
ARToolKit [45]), JSARToolKit5 (an emscripten port of ARToolKit), and
so on. The state of the art is the newly (in 2017) proposed AR.js [41], a
Web AR solution based on Three.js and JSARToolKit5, which can work
on all platforms and any browser with WebRTC and WebGL; it achieves
even 60 frames/s, stable on Nexus 6P. However, currently, AR.js can only
support the fiducial marker, as it involves only simple matrix operations.
It is still challenging for AR.js to support natural feature objects. Awe.js
[42] is another Web AR implementation, i.e., one based on natural
feature tracking (2-D image), and some experimental attempts have
demonstrated its suitability. However, the inefficient computing
capability of the Web results in significant tracking error due to the
complex computational requirements of Web AR applications, not to
mention the 3-D object and markerless Web AR implementation
methods. Moreover, the aforementioned compatibility challenge
regarding different computing capabilities of browsers, including built-in
browsers, also makes the pervasive and large-scale promotion of Web AR
difficult.

In addition to traditional CV methods [46]–[47][48], algorithms based

on deep learning have also received a lot of attention and development
effort in recent years. Web-based neural network algorithms, such as
ConvNetJS [49], CaffeJS [50], WebDNN [51], deeplearn.js [52], and
TensorFlow.js [53], provide a novel and cross-platform approach for
image analysis and processing on the Web. Specifically, these enabling
technologies that leverage the convolutional neural networks can be
further designed for generalized object detection, recognition, and
tracking in a variety of Web AR applications, which provide intelligent
context-aware ability and accurate vision-based tracking ability as well,
and therefore greatly enhance the capabilities of Web AR. However, the
models’ size is still an obstacle for its pervasive application. For example,
the model size of GoogLeNet [54] in CaffeJS is even up to 28 MB, which
is unacceptable for Web users. Moreover, the time of forward pass3 is
also a challenge for the real-time requirements of Mobile AR
applications, especially on the Web. The question of model compression
and inference acceleration is, therefore, arise [55]–[56][57][58] for the
purpose of Web AR practical application.

2) Extending the Browser Kernel:

The Web browser is nowadays an important entrance for users to connect
the Internet. By extending the browser kernel to support AR, Web AR
applications can often get near-native performance on mobile devices
and thus a better user experience. There have already been some efforts
from academia and industry to explore the potential of this Web AR
implementation approach, such as RWWW browser [59], Wikitude [60],
and the Argon project [61]. The state of the art from Mozilla and Google
is Project WebXR Viewer [62], and WebARonARKit and
WebARonARCore [63]. These efforts aim to provide a standard
environment for Web AR developers. However, they are still in their

9 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

infancy and have not been applied in practice on a large scale. In addition
Contents
to Mozilla and Google, there are also other companies making an effort to
bridge the gap between the Web and the AR. Both Baidu and Tencent
proposed their Web AR solutions in 2017, namely, DuMix AR [64] and
TBS AR [65], respectively. Fig. 8 shows the TBS AR system architecture
as an example. The browser-kernel extension solution presents a
promising and powerful self-contained Web AR implementation solution
compared with the pure JavaScript library/plug-in method. However,
before the standardization of AR-supported browsers is finished, the
diversity of APIs proposed by different browser kernel-extension
solutions will, in contrast, limit the large-scale promotion of Web AR
applications. Fortunately, some standardization efforts have already
begun (WebXR Editor’s Draft, W3C, March 7, 2018).

Fig. 7.
State of the art of (a) fiducial-based (AR.js [41]) and (b) natural feature-based
(awe.js [42]) Web AR JavaScript library/Plug-in implementations.

Fig. 8.
TBS AR system architecture. By extending Web browsers to support AR, Web AR
applications can provide users with near-native application performance.

B. Computation Outsourcing
Although the browser-kernel extension method achieves a skip-type
performance improvement compared with another self-contained Web
AR implementation solution (i.e., pure JavaScript library/plug-in), it is
still challenging for the perception of complex environments by mobile
devices due to their limited computational capability. Another type of
Web AR implementation mechanism is outsourcing the computations. By
outsourcing computationally intensive tasks to cloud servers, Web users
can get a better AR experience, which benefits from the stronger
computing capability of the servers. Meanwhile, it also reduces the
computing capability requirement for the mobile device and, thus, the
threshold of the promotion of Web AR. However, the additional
communication delay and deployment cost are two important issues that
deserve our attention at the same time.

Advances in network technology make it possible not only to outsource

computationally intensive tasks to cloud servers but also to achieve
collaborative computing for a better AR experience and savings of
energy. The emerging 5G networks can achieve even a 1-Gb/s data rate as
well as millisecond end-to-end delay, and moreover, the D2D technology
supports short-distance communication. All these features provide
opportunities to Web AR for its pervasive promotion and performance
improvement as well.

Another important issue lies in the offloading strategy for the

10 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

computation outsourcing method. Considering the high monetary cost of

Contents
the deployment of cloud servers, a reasonable service deployment and
computation offloading method is therefore necessary. Nowadays, a
variety of offloading frameworks [66]–[67][68][69] and approaches (e.g.,
game theory, integer linear programming, and multicriteria decision
theory, and reinforcement learning) are available, which can be used for
the deployment of Web AR to fulfill the adaptive computing paradigm
and thus optimize the resource utilization on the Internet.

A computation outsourcing mechanism provides an alternative service

provisioning paradigm for Web AR. In this section, two kinds of
computation outsourcing-based Web AR implementation methods will
be discussed: back end and collaborative.

1) Back-End Solutions:
Compared with mobile devices (e.g., smartphone or AR glass), servers in
the remote/edge cloud always have more powerful computing, rendering,
as well as storage capabilities, so complex tasks can be processed more
quickly and efficiently. In accordance with different deployment methods
of servers, the back-end solutions can be classified as cloud computing-
based and edge computing-based solutions.

a) Mobile cloud computing-based solutions:

This type of solution offloads computing or rendering tasks to the remote
cloud servers for process acceleration. It therefore not only alleviates the
computational pressure on the mobile device but also improves the
performance of the Web AR applications. Many Mobile AR applications
[71]–[72][73][74][75] have benefited a lot from this computation
outsourcing paradigm. However, there are still some issues that cannot
be ignored.

1. Bandwidth Challenge: The continuous image/video transmission

occupies a large part of the network bandwidth, which has a bad
impact on core networks.

2. Latency Challenge: An additional communication delay is added

due to the data transmission, and an unstable wireless
environment also aggravates the performance of Web AR
applications.

The upcoming 5G networks will provide higher bandwidth and lower

network delay, and therefore, they will efficiently optimize the
performance of Web AR applications in the case of computation
outsourcing. Another important issue lies in the processing and cost
pressures caused by the high concurrence in the case of centralized
service provisioning. An example is Kurento [76], one of the typical
WebRTC media server implementations, which can be used for Web AR.
Even just the encoding and decoding (640 × 480 pixels) processes in the
system will occupy about 20% of the CPU,4 not to mention the cost
pressures incurred by the concurrence requirement.

b) Promising solutions based on multiaccess edge computing:

The MEC paradigm in 5G networks provides an alternative Web AR
service provisioning mechanism considering the bandwidth and latency
challenges faced by the aforementioned centralized solution. On the one
hand, real-time pose estimation in Web AR application, either through
positioning techniques or through the camera view, or both, imposes
strict requirements for computing platforms. On the other hand, the
“virtual contents” are always relevant to user surroundings (i.e.,
“reality”). Hosting the AR service on an MEC platform instead of in the
cloud is advantageous as follows.

1. From Computing Aspect: Pose tracking and even rendering can be

performed on an MEC platform for quality-of-service
improvement.

2. From Caching Aspect: Highly localized supplementary information

on edge nodes improves the overall storage efficiency of the system.

By migrating Web AR services from the remote cloud to the network

11 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

edge, where it is closer to the users, this type of solution not only reduces
the communication delay but also alleviates theContents
bandwidth usage of core
networks at the same time. Additionally, this MEC-based solution also
has the advantages of collecting metrics, anonymized metadata, and so
on, which provide a basis for further user experience optimization.
Nowadays, some efforts have already started to explore the potential of
edge computing for Mobile AR applications and have achieved positive
results, for example, remote live support [77]. For simplicity, here, we
categorize both Cloudlets [78] and newly emerged fog computing
paradigm [79]–[80][81] as a specific type of MEC method in this paper.
The ETSI has already sketched an MEC-based AR service provisioning
scenario [15] in 2015, and a general distributed network topology [70] in
5G networks is shown in Fig. 9. In the meantime, there are already some
efforts on this promising computing paradigm [82]–[83][84][85]
[86][87][88]. Because of the native support of MEC technology in 5G
networks, the development of Web AR services will become easy and
convenient.

Fig. 9.
Network topology significantly affects the content delivery latency and, thus, the
user experience. 5G latency can be broken down into three different network
topologies, and their associated latencies in the diagram are only from an
assumption of a round-trip ping scenario (Source: ABI Research [70]).

2) Promising Collaborative Solutions:

The aforementioned self-contained Web AR implementation method
faces limited computing capability, which limits its broad application.
Moreover, the diversity of computing platforms also hinders the
pervasive promotion of Web AR due to the lack of an adaptive computing
resource scheduling mechanism. AR is a computation- and data-
intensive application, as mentioned earlier, although mobile cloud
computing- and MEC-based implementation methods provide more
powerful computing capability, and thus, a Web AR performance
improvement, the bandwidth usage, communication delay, and
deployment cost all deserve our attention in the case of high concurrence
requirement. On the other hand, advances in mobile devices also make it
possible to perform computational tasks locally, and although current
mobile devices cannot afford overly complex computational works, it still
encourages us to carry out further research to explore the potential of
collaborative distributed computing for Web AR. To take full advantage
of distributed and diverse computational and storage resources, an
adaptive and scalable collaborative computing and communication
paradigm is therefore necessary. By distributing the computational
pressure from cloud servers to mobile devices and network edges while
still satisfying the performance requirements and user experience of a
Web AR application, it can effectively gather distributed resources and
then achieve cost saving and further performance improvement.

Osmotic computing [89]–[90][91] is a novel computing paradigm that

aims to facilitate highly distributed and federated computing
environments. It enables the automatic deployment of microservices over
an interconnected cloud datacenter and an edge datacenter. In the
meantime, the proposed reverse offloading method, i.e., the movement of
functionalities from the cloud to the network edges, not only helps
latency-sensitive applications but also minimizes the amount of data that
must be transferred over the network. This adaptive and scalable
paradigm will, therefore, be a promising direction for distributed

12 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

collaborative Web AR implementations.

Contents
a) Terminal + Cloud collaborative solution:
The reverse offloading method in the osmotic computing paradigm
encourages us to offload part of the computational tasks of Web AR from
the cloud to user devices to alleviate both computing and deployment
cost pressure on the central site. We conducted a real Web AR
advertising campaign for China Mobile by WeChat (December 5–14,
2017), which is also the first time we promoted Web AR on a large scale.
It achieves 3 550 162 page views and 2 080 396 unique visitors in only 10
days. In this project, the Web AR service provisioning mechanism was
designed in two parts. All the visual operations are executed on the
mobile device locally based on the ORB algorithm in JSFeat. The remote
cloud servers are responsible for the database queries or other logical
tasks. The collaboration between the computing capability of the
terminal (mobile device) and the storage capability of the cloud server
greatly reduces the overall deployment cost of the Web AR application.
However, there is still a lot of room for performance optimization, since
the choice of JSFeat aims at reducing the threshold of the computing
capability requirement so as to reach more Web users; it compromises
the performance of the application. Yang et al. [92] discussed the
optimization of computation partitioning for a data stream application
between the terminal and the cloud. The results show that a reasonable
computation offloading method provides great benefits to the
performance of the application. Moreover, in accordance with the
differing computational capabilities of mobile devices, an adaptive and
scalable computation offloading method can schedule distributed
resources on the Internet more flexibly and intelligently, which is
important for this type of collaborative Web AR implementation solution,
since it can perform personalized computation partitioning, as shown in
Fig. 11(a), and hence maximize the individual user experience for Web
AR applications.

Fig. 10.
Basic concepts of osmotic computing as well as two osmotic scenarios. Terminal
+ Cloud (Osmotic scenario I) and Terminal + Edge + Cloud (Osmotic scenario II)
collaborative Web AR implementation approaches. (a) Osmotic concept. (b)
Osmotic computing. (c) Osmotic scenario I. (d) Osmotic scenario II.

Fig. 11.
Two promising collaborative Web AR implementation solutions and the
distribution of workload. (a) Real Web AR advertising campaign for China Mobile
by WeChat in current mobile networks and the illustration of the adaptive and
scalable computation offloading approach. (b) Collaborative computing scenario
between hierarchical platforms for Web AR and an experimental application
designed based on distributed DNNs in the 5G trial networks.

b) Terminal + Edge + Cloud collaborative solution:

Another promising, more efficient, but complex, method is to combine
the computing and storage capabilities of the mobile devices, network
edges, and remote cloud servers to explore more adaptive and scalable
collaborative Web AR implementation methods. In general, the
introduction of mobile devices alleviates the computational pressure on
the edge and cloud servers; network edges provide a temporary place for
Web AR application migration, not only supplementing the computing
capability of the mobile device but also shortening the data transmission;
a remote cloud server has stronger computing capability and adequate
storage space and is generally responsible for the database queries,

13 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

historical big data-based model training, and other tasks. In addition to

the aforementioned osmotic computing paradigm Contents
which provides
guidelinesfor this collaborative Web AR implementation solution, with
the advance of the smart city, smart home, and smart devices, the user’s
ambient environment is becoming powerful and intelligent, and
collaborative solutions help mobile devices extend their capabilities in a
more flexible manner, which therefore further facilitates the promotion
of Web AR. As shown in Fig. 11(b), devices communicate through the
cellular or WLAN technology to share supplementary information (e.g.,
3-D model, sound, and video) and collaborate to perform Web AR
applications as well. Moreover, this short-distance wireless
communication technology can also be used for collaborative multiuser
Web AR applications, such as multiplayer online AR games, but only
impose a slight performance impact on the central site. The upcoming 5G
networks promise the supports of D2D technology, and apparently, the
Web AR will benefit from it in more and more scenarios (e.g., museum,
art gallery, and city monument). As a demonstration, we implemented a
Web AR-based animal retrieval application for tourists in the zoo that
leverages distributed deep neural networks (DNNs) and evaluated in the
5G networks.5 The end device and network edge servers shoulder totally
72.2% (17.6% + 54.6%) of DNN (AlexNet [93]) computations (i.e., the
relieved computation pressure of the central site). Note that the specific
neural network computations are encoded into the Wasm format in
advance to achieve inference acceleration on the Web. In the meantime,
this collaborative solution also brings the response delay improvement
about 319.26% compared with the pure front-end (Chrome)-based
solution. Obviously, for collaborative computation outsourcing solutions,
an adaptive and scalable scheduling method will always benefit Web AR
applications as the different available computing and storage resources of
devices can be coordinated more intelligently.

C. Technology Readiness Levels

Based on the previous statements, we summarize and compare the
technology readiness levels of Web AR implementation mechanisms and
approaches, as shown in Fig. 12. Obviously, the Web AR is still in its
infancy, which requires efforts from both academia and industry.
Considering the challenges of computing capability, networking, battery
capacity, and compatibility, the pervasive promotion and application for
Web AR are still difficult in the current mobile networks as: 1) the
sensor-based implementation mechanism cannot provide the seamless
immersive experience to users due to the unavoidable cumulative error
and 2) the browser-kernel extension solution lacks cross-platform
support at present. However, the future of Web AR really deserves our
expectations. The lightweight and cross-platform Web AR technologies
have a wide range of applications in many areas, especially with the
development of networks and the improvement of mobile device
performance. The advancement of underlying technologies and the
innovation of applications are of mutual promotion, and they will
eventually propel the development of Web AR ecosystem.

Fig. 12.
Technology readiness levels for Web AR system in the current 3G/4G and
upcoming 5G era.

SECTION IV.
Open Research Challenges

14 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

The emergence of Web AR undoubtedly helps the promotion of Mobile

Contents
AR applications on a large scale. However, there are still various
obstacles waiting for the proper technologies to be available and
affordable. The practical development and deployment of Web AR
inspired us a lot. In this section, we detail these insights and provide
some further discussions as well.

A. Computation and Rendering Efficiency

AR is a computation- and data-intensive application. However, both the
computing and the rendering tasks in Web AR nowadays face an
inefficient runtime environment due to the limited computational and
storage abilities of mobile devices.

1) Computational Efficiency:
Considering the aforementioned self-contained and collaborative
computation outsourcing Web AR implementation approaches, the
computational and rendering abilities of mobile devices play an
important role in the improvement of the performance of Web AR
applications. Here are several suggestions for the improvement of their
performance.

1. WebAssembly, Web Workers, and other similar enabling Web

technologies are helpful for the improvement of the performance of
Web AR applications. WebAssembly can accelerate the Web AR
process by transcoding high-level codes into binary format in
advance, which also has the advantage of code compression at the
same time. The project WebSight [94] demonstrates that
WebAssembly can provide about a 10× performance improvement
over pure JaveScript. In addition, the introduction of multithread
technology can also effectively improve the overall efficiency of the
program by using the Web Workers technology in the Web AR
application.

2. Whether it is JSFeat, JSARToolKit, or even state-of-the-art AR.js,

there still are obvious performance weaknesses for natural feature
tracking-based and markerless Web AR implementation methods.
That is, to improve the Web AR application performance, we also
need to pay attention to a more efficient computing paradigm and
Web AR-related JavaScript library/plug-in.

3. Approximate computing is another way that is worth trying out.

The performance of Web AR applications can be improved by
reducing the complexity of the algorithms. Although some efforts
[95]–[96][97] have already proved the feasibility of this computing
paradigm, where the tolerance of imprecise operations (e.g., image
recognition and motion sensing) can help the improvement of the
Web AR user experience, there is still a lot of room for further
investigation of approximate computing, especially in terms of Web
AR.

4. Another suggestion is our aforementioned computation

outsourcing Web AR implementation approach. Both back-end and
collaborative solutions can provide a better user experience. An
adaptive and scalable collaboration strategy will benefit the whole
Web AR application provisioning framework since the computing
and storage resources can be fully scheduled and utilized in an
intelligent way.

2) Rendering Efficiency:
Rendering efficiency is another area of concern. Virtual contents (e.g.,
3-D model) generated by the computer can currently only support simple
interactions with users, such as rotating and scaling operations, on the
Web. Indeed, a complex 3-D model not only adds download time from
the cloud/network edge but also increases the computational burden on
the mobile devices. Moreover, the longer rendering time will even
degrade the user experience of Web AR applications. Zhang et al. [98]
analyzed the time and energy consumption of each part of a Mobile AR
application, noting that as the complexity of the 3-D model increases, the
proportion of the rendering part will also increase. For example, the 3-D

15 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

model that is larger than 4 or 5 MB will result in a serious lag

Contents
phenomenon in our experimental project and further degrade the Web
AR application performance. It is obvious that Mobile AR, especially Web
AR applications, needs more lightweight 3-D models or even a dedicated
lightweight 3-D model format for Web AR. A model compression
technology can only shorten the download time; the rendering operations
on mobile devices still consume a large amount of CPU, memory, and
battery resources. In addition, optimized rendering techniques or GPU-
based rendering methods can also improve the rendering efficiency on
the Web. On the other hand, by using the state-of-the-art visual attention
mechanism [99], only the part of the user’s attention in the field of view
will be augmented, which therefore also reduces the complexity of the
rendering.

B. Network Communication Efficiency

Another crucial problem for Web AR is network requirements. To
achieve a higher quality of the user’s experience, computationally
intensive tasks are usually outsourced to cloud/network edge servers for
performance improvement considering the limited computing and
rendering capability of mobile devices. The aforementioned MEC
paradigm can further lower the communication delay for Web AR
application. However, the deployment of an edge computing system has a
high monetary cost, and the infrastructures have not yet been
popularized in the current 3G/4G mobile networks. A more practical way
is to use currently available network technologies, such as content
delivery network and data centers, to compensate for the gap between
users’ ever-increasing demand for Web AR applications and the
computation outsourcing method under the current mobile networks.
The soon-to-be available 5G networks together with their newly emerging
features will bring new opportunities for the promotion of Web AR. The
network slice technology provides a more reasonable network resource
scheduling mechanism, which will therefore provide a better network
environment for Web AR. The MEC and D2D technologies will facilitate
the service provisioning on the Web in a more flexible way based on the
adaptive and scalable computing and communication paradigm.

C. Energy Efficiency
AR applications require long-time cooperation of environment
perception, interaction perception, and Internet connection. All these
power-hungry tasks place tremendous pressure on the battery in mobile
devices. However, currently, the battery is only designed for common
functionalities. To reduce the adverse impact of Web AR applications on
mobile devices, energy efficiency is also an important part that cannot be
ignored. Multicore CPUs consume less energy than single-core CPUs due
to the lower frequency and voltage, and there are already many off-the-
shelf multicore CPU processors available for mobile devices. By
parallelizing the tasks in a Web AR application to multicores, the energy
consumption can be reduced. Moreover, the upcoming 5G networks can
also help energy saving indirectly, since both the network latency and the
cost of data transmission can be optimized.

D. Compatibility
Web AR is designed as a lightweight and cross-platform Mobile AR
implementation to achieve the pervasive promotion of AR applications.
However, the compatibility issue is also one of the most serious problems
at the moment.

1) Enabling Technology Compatibility:

Various browsers, including native browsers (e.g., Chrome, Firefox, and
Safari) and built-in browsers in which the application is designed in a
hybrid way (e.g., Facebook, Twitter, and WeChat), present a great
difference in their support for and compatibility with all types of Web AR
enabling technologies, such as WebAssembly, WebGL, WebRTC, and so
on. This not only hinders the large-scale promotion of Web AR
applications but also increases the difficulty of program development.

2) Web AR Browser Compatibility:

The lack of standardization of browsers for Web AR causes another

16 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

compatibility issue. Currently, all the dedicated Web AR browsers are

Contents
isolated from each other; an AR application designed based on a specific
Web browser cannot be accessed on other platforms. The W3C group has
made some efforts [100] recently, and with the ever-growing enthusiasm
for the Web AR from users, standardization also needs attention, which
requires a joint effort from both academia and industry.

3) 3-D Model Format Compatibility:

There are also compatibility issues between Three.js and Web 3-D
models that are generated by different tools (e.g., 3DMax, MAYA, and
Blender), which cause degradation of the animation effects of Web AR
applications. For the purpose of being cross-platform, standardization of
a Web 3-D model format is also expected.

E. Privacy and Security

Social acceptance of Web AR is easily affected by privacy and security
factors. The “Stop the Cyborgs” movement has had a huge impact on
Google Glass due to the privacy leakage. Whether it is a client–server
(i.e., back end) or a collaborative Web AR implementation method, there
are various potential invasion sources. Users’ private information, such
as personal identification and location information, can possibly be
collected by third parties for other uses. To guarantee privacy safety, both
trust mechanisms for data generation and certification mechanisms for
data access, as well as a secure network environment for data
transmission, are required. Acquisti et al. [101] discussed privacy issues
of facial recognition for AR applications and also proposed several
privacy guidelines including openness, individual participation, use
limitation, purpose specification, and so on, as well as a recommended
solution: regulate usage, not collection. Besides the standard security
strategies, such as on-device and network encryption, others will need to
be rethought, especially in this new context [102], [103]. For example,
researchers have begun considering the specific AR operating system
[104] (from underlying platform perspective), the surroundings
information collection rules or retention policies [105] (from sensing
perspective), the object access governing [106], [107] (from data access
perspective), and the trusted renderer [108], [109] (from output
perspective). In addition to these technical solutions, the privacy and
security challenges for AR systems also call for social, policy, or legal
approaches [110]. Web AR is more dependent on mobile networks and
therefore more likely to be invaded. This poses a significant challenge for
the development and deployment of Web AR applications.

F. Application Deployment
Web AR has a great potential to enrich our ways of interacting with the
real world. There is a growing demand for the mobility-aware,
lightweight, and cross-platform AR applications. Google Glass was a
milestone product, which not only raised public interest but also played
an important role in the promotion of AR, especially Mobile AR.
Although most existing Web AR applications are research prototypes, the
popularity of Pokémon GO has demonstrated the attraction and the
potentially wide deployment of Web AR applications. We believe that
with more open-source software and more development platforms and
educational programs for Web AR made publicly available, more Web AR
prototype systems and applications will emerge. Similarly, Web AR also
needs killer applications to help developers and users explore its
potential value.

SECTION V.
Conclusion

We have presented a survey of Web AR in three focused subject areas.

First, we reviewed the principle of Mobile AR and three typical
implementation approaches. Second, we discussed the challenges and
enabling technologies for when AR meets the Web and described
different Web AR implementation approaches. Finally, we summarized
the ongoing challenges and future research directions of Web AR.

17 of 18 25/07/20, 13:53
Web AR: A Promising Future for Mobile Augmented Reality—... https://ieeexplore.ieee.org/document/8643424?arnumber=8643424

Although Web AR is still in its infancy, the state-of-the-art research and

development results and the different Web AR Contents
implementation
approaches discussed in this paper will provide guidelines and a
reference entry for researchers and developers to apply Web AR
technology in their Web-based mobile applications to provide a pervasive
AR experience to the users. Recently, the Web-based AR implementation
method has also received focused attention from the W3C group, and the
Web XR Editor’s draft was released in March 2018.

The upcoming 5G networks provide an efficient and powerful platform

for the pervasive promotion of Web AR. The higher data rate (0.1~1
Gb/s) and lower network delay (1~10 ms) satisfy quite well the real-time
interaction requirements of Web AR. The MEC paradigm reveals a new
trend of computing paradigms, that is, a reverse offloading mechanism
(e.g., osmotic computing). With the deployment of edge servers, an
adaptive and scalable communication and collaboration between the
cloud and network edges, as well as between edge servers and mobile
devices, will provide ubiquitous capability to leverage the distributed and
heterogeneous computing and storage resources and fulfill the high
demands of Web AR with respect to performance improvement and
energy saving. Moreover, the D2D technology provides an efficient
collaborative communication solution between the devices, and network
slicing can further optimize the data transmission for Web AR. We
conjecture that continued advances in all these computing and
networking technologies mentioned earlier will further fuel the research,
development, and deployment of the Web AR-enabled service
provisioning at a higher level.

The different Web AR implementation approaches we discussed in this

paper provide opportunities to apply Web AR applications in practice.
Authors +
Figures +
References +
Citations +
Keywords +
Metrics +
Footnotes +

IIE
EEEE
EPPeerrssoon
naall A
Accccoou
unntt P
Puurrcch
haassee D
Deettaaiillss P
Prroofi
fillee IIn
nffoorrm
maattiioon
n N
Neeeedd H
Heellp
p?? F
Foolllloow
w

CHANGE USERNAME/PASSWORD PAYMENT OPTIONS COMMUNICATIONS PREFERENCES US & CANADA: +1 800 678 4333 ( ) *
VIEW PURCHASED DOCUMENTS PROFESSION AND EDUCATION WORLDWIDE: +1 732 981 0060

TECHNICAL INTERESTS CONTACT & SUPPORT

About IEEE Xplore | Contact Us | Help | Accessibility | Terms of Use | Nondiscrimination Policy | Sitemap | Privacy & Opting Out of Cookies
A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity.

© Copyright 2020 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

IEEE Account Purchase Details Profile Information Need Help?

» Change Username/Password » Payment Options » Communications Preferences » US & Canada: +1 800 678 4333
» Update Address » Order History » Profession and Education » Worldwide: +1 732 981 0060

» View Purchased Documents » Technical Interests » Contact & Support

About IEEE Xplore | Contact Us | Help | Accessibility | Terms of Use | Nondiscrimination Policy | Sitemap | Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity.
© Copyright 2020 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

18 of 18 25/07/20, 13:53