Course contents

• Theory
  • Definitions
  • History, 3 Types of AR
  • Technologies, Calibration, Registration, Tracking
  • Rendering, Interactions
• Practice
  • Overview of existing apps and tools
  • AR Web Programming
    • WebXR, AR.js and more!

IRCAD (2002 - 2003)

• Institut de Recherche contre les Cancers de l'Appareil Digestif
• Startup
  • Virtual-Surg team
• Augmented Reality Research Engineer

Dassault Systèmes (2003+)

• 3D Visualization Engineer
  • Scenegraph, Materials
  • Geometry, Tessellation
• Virtual and Augmented Reality (XR) Engineer
• XR Research Engineer
• XR Research Manager

Dassault Systèmes

From Shape to Life

Course audience

• Anyone looking for a simple introduction to Augmented Reality
• Computer Science students and engineers looking for a way to create cross-platform XR prototypes or even full-fledged apps

Feel free to skim through technical sections and use this course as future reference

Course prerequisites

• Math
  • 3D vectors and matrices
• Programming
  • JavaScript notions, or any similar language (HTML kept minimal)
• 3D Web API
  • THREE.js notions strongly recommended (see Web 3D course)
  • alternatives: Babylon.js, WebGL, WebGPU
• Desktop + Smartphone (or, even better, a XR HMD!)
  • VSCode

Planning

• Session 1 (2 hours)

  • Theory (45 min)
  • WebXR Theory + Exercises (1h)
  • Explore examples (15 min)

• Session 2 (4 hours)

  • Project kick-off (must be finished at home )
    • upgrade your existing THREE.js project to AR
    • 2 people per project: clear responsibilities (who does what)
    • send git repo link: source + live testing

Project evaluation criteria

• originality
• interactions
• physics / animations / sounds / eye-candy
• healthcare
• code quality , tricks , performance
• fun
• clever use of AR
  • tangible interactions, body tracking
  • real world interactions, spatial design

Grading system

• 20 points maximum
• choose features from previous slide
• for each implemented feature:
  • not done: 0 pt
  • nice try / buggy: 1 pt
  • basic / good enough: 2 pts
  • great / polished: 3 pts
  • impressive: 4 pts

Consumer applications

• Entertainment

  • Games: Pokemon GO AR+, Minecraft Earth (RIP )
  • Social Networks: SnapChat (City Painter), Facebook

• Interior Design

  • IKEA Place
  • HomeByMe AR

• IGN: Time Machine

Professional applications

• Industry
• Healthcare
• Marketing

Definitions

• Milgram's Reality-Virtuality Continuum

Milgram, Paul; H. Takemura; A. Utsumi; F. Kishino (1994). "Augmented Reality: A class of displays on the reality-virtuality continuum".

Google's version

Properties of an AR system

(according to Azuma, 2001)

• combines real and virtual objects in a real environment
• runs interactively, and in real time
• registers (aligns) real and virtual objects with each other.

Not AR:

• special effects in movies
  • technology close to AR
  • not real time
  • not in a real environment
• Google Glass
  • combines real and virtual objects in a real environment
  • no registration
  • it's a HUD (Head-Up Display)
    • can still be useful! (maintenance, sports etc.)

Google Glass concept video (2012)

Definitions

• VR : Virtual Reality Jaron Lanier, 1987

• AR : Augmented Reality Thomas P. Caudell, 1990

• MR : Mixed Reality

  • marketing term used by Microsoft
  • no clear definition! Term must be defined before use!
    • cf. What is Mixed Reality?

• XR : X = { eXtended / Cross (+) / Any (*) / A+V } Reality

  • recent generic term which encompasses AR and VR

AR | VR

AR or VR?

• Similar technologies

  • 3D rendering
  • Tracking
  • Immersive interactions

• Different effects on the user

Effects of VR

• Isolates the user from the real world
• Teleports the user to another world, which is entirely virtual

Tiltbrush

ITI

The limits of VR

• Reminder: continuum!
  • No clear boundaries
• When the whole world is modeled and registered in 3D,
  is it still VR?
• Photogrammetry / Lightfields / NeRF / Gaussian Splats
  • VR but immersion in a world entirely rebuilt in 3D

Greg Madison @ Unity

VersaillesVR

Effects of AR

• The user stays in the real world
• AR enhances the real world with contextual information
• Augmented user: acquires new senses!
• Information becomes visible
  • spatialized information overlaid on top of the real world

Audi AR manual (Metaio, 2012)

The limits of AR

• Reminder: continuum!
  • No clear boundaries
• When more virtual elements than real ones: Augmented Virtuality
  • Window to the real world
  • Real users visible

Dangers of AR

• Information overload: Hyper-reality
  • Diminished Reality desirable?
• Excessive assistance, altered behaviors, surveillance
  • Black Mirror: Nosedive
• Digital divide
  • Some people will feel handicaped, missing a sense, daltonians
• Privacy: Cloud Wars
  • MAMAA* vs Open AR Cloud

Takeaways

• VR immerses the user in a virtual wolrd
• AR brings virtual objects into the real world

Choosing the right paradigm

• Immersion useful ?

  • Yes VR
  • No 3D

• Immersion and real environment useful ?

  • Yes AR
  • No VR

• Keep in mind continuum to pick the right paradigm to create the best possible experience

Choosing the right paradigm

Prehistory (1966)

• Ivan Sutherland invents the first AR headset
• Can display a cube
• Follows the movements of the head (6 DOF)
• Nickname : The Sword of Damocles
  
• paper: The Ultimate Display

Markers (1999)

• Monochrome markers
• ARToolkit created by Hirokazu Kato
  
  
• Alternatives: ARTag, ArUco
• PC + Webcam

NFT, GPS (2005)

• NFT: Natural Feature Tracking
  • Color photo tracking
• Wikitude, Layar (GPS)
  • no image processing needed with GPS!
• Vuforia
• Marketing use-cases
• PC, mobile phones, tablets

SLAM, 3D (2015)

• 3D environment tracking
• SLAM: Simultaneous Localization And Mapping
• 3D object tracking
• Deep Learning
• Occlusion 3D
• ARKit, ARCore
• Smartphones, HoloLens, Azure Kinect

Azure Kinect + HoloLens 2

Near future

• Form-factor: glasses
• AI
  • contextual assistance
  • understands both environment and user
• 5G
  • application and information streaming (Edge Computing)
• Spatialized Web: AR Cloud

Far future

• AR will replace or complement smartphones
  • users will raise their heads again
    • but will they see better?
• Contact Lens (Mojo Vision)
  • RIP (2023)
• Ambient Computing
• Ubiquitous Computing
• Smart Cities

Gartner Hype Cycle

• Technology trends evolution
• Gartner
  • 2005 - 2017 period

Where do we stand now?

• We are getting close to productivity for AR and VR
• Or AR is nows considered as "productive" since 2020?
• Future trends:
  • AR Cloud
  • "Metaverse" and "Digital Humans"

2023

AR [is for] adding shared meaning in the interaction between people.
Johnny Lee, Google I/O 2017

• Dropped mobile VR (Cardboard , Daydream )
• Dropped Tango , to reach more devices: rely on RGB camera + AI
• API ARCore, competes with Apple's ARKit
• Google wants to provide cross-platform AR services
• Google + Qualcomm + Samsung XR Headset coming in 2025

The Web connects the world's information, and AR connects information with the physical world. So together they can be applied to solve real life problems.
Andrey Doronichev, Google I/O 2017

Gorillaz mobile AR app

Hardware

• Adds LiDAR for a robust SLAM (e.g. white walls scenario)

• Extends its 'wearables' category

  • AirPods
  • Apple Watch
  • Apple Vision Pro
    • unveiled in June 2023
    • released in February 2024

• Created Horizon Worlds social network for virtual encounters, and Horizon Workrooms for remote collaboration
• Believes that AR will replace smartphones and every screen in 10 years, does not want to miss this revolution: Project Aria / Orion
• Interested in personal data, centers of interest of their users (eye-tracking, scanned environment), although they claim the opposite
• VR with Oculus, but they now focus on AR, cf. Infinite Office
• AR filters Facebook and Instagram via Spark AR RIP
• Great WebXR supporter

Michael Abrash in 2019

Project Aria

Next?

I might get myself in trouble for saying this; I think it might be the most advanced piece of technology on the planet in its domain. In the domain of consumer electronics, it might be the most advanced thing that we’ve ever produced as a species.
Andrew 'Boz' Bosworth, Meta CTO, January 2024

• Amazon focuses on e-commerce and its Web Services
• AR View to see a product at home before buying it
• Offers Sumerian as a paid tool via AWS (Amazon Web Services) to create XR experiences
• Pushes machine learning, smart assistants (Alexa)
• Bets on AR on demand via 5G with its Wavelength Project
  • 5G + Edge computing
  • AWS

Other players

• Hardware
  • Magic Leap, Lynx, XReal, Qualcomm, Snap
• SDK
  • Wikitude, Kudan
  • PTC Vuforia: IoT
• Web
  • Firefox Reality Wolvic
  • Chrome

Takeaways

• Big tech companies invest massively in AR, which they see as a promising technology evolving fast
  • hardware
  • algorithms
  • services, data
• Many players try to bring their users into their closed ecosystem (hardware, app store, cloud)
• Others focus ont the openness of the Web to create and share open AR experiences
  • ultimate goal of this course!

Further reading

• History and future of Web AR

  • Web AR: A Promising Future for Mobile Augmented Reality - State of the Art, Challenges, and Insights.

  Qiao, Xiuquan & Pei, Ren & Dustdar, Schahram & Liu, Ling & Ma, Huadong & Junliang, Chen. (2019).

  Proceedings of the IEEE. 107. 1-16. 10.1109/JPROC.2019.2895105.

3 Types of AR

• Video
  • e.g.: smartphone,
    Meta Quest 3, Apple Vision Pro, Lynx-R1*
• Optical
  • e.g.: HoloLens
• Projective
  • e.g.: DIOTA

Lynx-R1 (video)

Optical AR calibration

• very complex
• hardware dependent
  • projection and image formation systems
• depends on the body metrics of the user
• made and provided by the AR hardware manufacturer
  • possible adjustments for each user, cf. eye calibration in HoloLens

Video camera calibration

• Goal: compute the optical parameters of the real camera
  • focal length
  • radial distortion, lens imperfections
• Method:
  • capture images of known patterns (grids, calibration patterns) with a real camera
• the focal length may be variable (autofocus)
  • update calibration data for each frame
  • calibration data is provided by the API (ARKit, ARCore, WebXR)

Video camera calibration method

Pinhole camera model

Extrinsic and intrinsic parameters

3D coordinates Camera 3D coordinates Image coordinates

Projection

$$s \; p = A [R|t] P$$

$$s \begin{bmatrix} u \\ v \\ 1 \end{bmatrix} = \begin{bmatrix} {f_x} & {0} & {c_x} \\ {0} & {f_y} & {c_y} & \\ {0} & {0} & {1} \end{bmatrix} \begin{bmatrix} r_{11} & r_{12} & r_{13} & t_x \\ r_{21} & r_{22} & r_{23} & t_y \\ r_{31} & r_{32} & r_{33} & t_z \end{bmatrix} \begin{bmatrix} X_w \\ Y_w \\ Z_w \\ 1 \end{bmatrix}$$

$(X_w, Y_w, Z_w)$ 3D world coordinates $O_w$
$(u, v)$ projected coordinates (pixels)
$[R|t]$ extrinsic matrix, $A$ intrinsic matrix
$(cx, cy)$ principal point (pixels), center of the image in the ideal case
$f_x$ et $f_y$ focals along x and y (pixels), equal in the ideal case

Non linear radial distortion

• due to the lens,approximated by a polynomial expression
  $x_{distorted} = x(1 + k_1 r^2 + k_2 r^4 + k_3 r^6)$
  $y_{distorted} = y(1 + k_1 r^2 + k_2 r^4 + k_3 r^6)$

Pose estimation

• Computed from 2D/3D pairs of points
• Optimization: projection error minmization between transformed 3D points $V_i$ et image 2D points $v_i$

$$\argmin_{R,t}\displaystyle\sum_{i} ||P(R V_i + t) - v_i ||$$

$P$: projection function
$R$: rotation matrix
$t$: translation vector

Tracking

• Degrees Of Freedom (DOF):
  • 0 DOF
    • no tracking!
    • simple information overlay, cf. HUD
  • 3 DOF
    • rotation only (gyroscope, accelerometer, compass)
      • limited experience (can be good enough, cf. planetarium)
  • 6 DOF
    • rotation + position

3 DOF

6 DOF

GPS

• global, satellite based, no network connectivity required
• no image processing
• outdoors only
• slow
• not very accurate

Marker

• accurate, fast
• tangible, printable
• need to display a marker to enable AR
• non-aesthetic
• can be hard to detect (low lighting, motion blur, occlusions)

Natural Feature Tracking

Same as marker but

• more aesthetic, easier to embed in the real world (ads)
• more robust to occlusions

SLAM

NFT evolution + reconstruction

• more natural markerless experience
• partial scene reconstruction
  • allows advanced functionalities (occlusions, collisions etc.)
• not very accurate (not ideal for medical uses)
  • drift, loop closure
  • scene reconstructed and refined in real-time
  • difficult to define the origin of the scene
    • stable anchor points required

3D object detection in a real scene

• using computer vision (lighting, edges, silhouette)
  • generic algorithm
  • but slow, especially during initial registration
• using Deep Learning
  • faster initial detection
  • more robust regarding occlusions and lighting changes
  • not generic: requires per model training

Tracking techniques : Conclusion

• No tracking technique is ideal
• Keep them all in mind and choose the right one according to:
  • the scenario of the AR experience
    • industrial context, consumer, generic or specific
  • constraints
    • indoor, outdoor, mobile

Rendering

• Realistic or not
• Lighting
  • detect the direction and intensity of real lights
  • fast environment reconstruction to simulate reflections (SLAM + AI)
• Occlusions
  • people, objects

Interactions

• The missing part of the equation
• Often neglected (cf. NReal)
• Myth of the dying mouse (p. 17)
  • each form factor has an optimal interaction technique
  • most headsets handle hand tracking, but also offer controller, keyboard and mouse support!
• The XR equivalent of the mouse has not been invented yet!

Interaction techniques

• Screen, when using a smartphone
  • not very immersive but accurate, and provides tactile feedback
• Controllers with buttons
  • great haptic feedback but not immersive
• HoloLens GGV : Gaze, Gesture, Voice
  • natural interactions, with no external hardware
  • great but tiring, lacks privacy ("hey Cortana!"), and accuracy
• Tangible interactions
  • markers or accessories to add some tactile feedback

Interactions

Conclusion

• Immersive AR interactions have yet to be invented!
• No interaction paradigm has become a standard yet
• We must guide the users and try to understand their intent

Reconstruction 3D

• Colmap
• AliceVision
• Capturing Reality
  • RealityScan for iOS (link)
• MicMac
• Scaniverse

Links

http://www.ign.fr/institut/innovation/minecraft-a-carte

http://lsc.univ-evry.fr/~didier/home/lib/exe/fetch.php?media=cours:ra:ra.pdf

Photo Credits

https://unsplash.com/photos/RgPVZvA4wBM

https://unsplash.com/photos/r2CAjGQ0gSI