2013 COSC 426 Lecture 3 on AR Tracking. Taught by Mark Billinghurst from the HIT Lab NZ at the University of Canterbury. Taught on July 26th, 2013.

1. COSC 426: Augmented Reality
Mark Billinghurst
mark.billinghurst@hitlabnz.org
July 26th 2013
Lecture 3: AR Tracking

2. Key Points from Lecture 2

3. “The product is no longer
the basis of value.The
experience is.”
Venkat Ramaswamy
The Future of Competition.

4. experiences
services
products
components
Value
Sony CSL © 2004
Gilmore + Pine: Experience Economy
Function
Emotion

5. Interaction Design is All About You
 Users should be
involved throughout
the Design Process
 Consider all the needs
of the user

6. Interaction Design Process

7. experiences
applications
tools
components
Building Compelling AR Experiences
Tracking, Display
Authoring
Interaction
Usability

8. Optical see-through head-mounted display
Virtual images
from monitors
Real
World
Optical
Combiners

9. Video see-through HMD
Video
cameras
Monitors
Graphics
Combiner
Video

10. Video Monitor AR
Video
cameras Monitor
Graphics Combiner
Video
Stereo
glasses

11. AR Tracking and Registration

12.  Registration
 Positioning virtual object wrt real world
 Tracking
 Continually locating the users viewpoint
- Position (x,y,z)
- Orientation (r,p,y)

13. Tracking

14. Tracking Requirements
 Augmented Reality Information Display
 World Stabilized
 Body Stabilized
 Head Stabilized
Increasing Tracking
Requirements
Head Stabilized Body Stabilized World Stabilized

15. Tracking Technologies
 Active
• Mechanical, Magnetic, Ultrasonic
• GPS, Wifi, cell location
 Passive
• Inertial sensors (compass, accelerometer, gyro)
• Computer Vision
• Marker based, Natural feature tracking
 Hybrid Tracking
• Combined sensors (eg Vision + Inertial)

16. AR Tracking Taxonomy
e.g. AR Toolkit
Low Accuracy
at 15-60 Hz
e.g. IVRD
High Accuracy
& High Speed
Hybrid
Tracking
Limited Range
e.g. HiBall
Many Fiducials
in space/time
but
no GPS
Extended Range
Indoor
Environment
e.g. WLVA
Not Hybridized
GPS or
Camera or
Compass
Low Accuracy &
Not Robust
e.g. BARS
Hybrid Tracking
GPS and
Camera and
Compass
High Accuracy
& Robust
Outdoor
Environment
AR
TRACKING

17. Tracking Types
Magnetic
Tracker
Inertial
Tracker
Ultrasonic
Tracker
Optical
Tracker
Marker-Based
Tracking
Markerless
Tracking
Specialized
Tracking
Edge-Based
Tracking
Template-Based
Tracking
Interest Point
Tracking
Mechanical
Tracker

18. Mechanical Tracker
 Idea: mechanical arms with joint sensors
 ++: high accuracy, haptic feedback
 -- : cumbersome, expensive
Microscribe

19. Magnetic Tracker
 Idea: difference between a magnetic transmitter
and a receiver
 ++: 6DOF, robust
 -- : wired, sensible to metal, noisy, expensive
Flock of Birds (Ascension)

20. Magnetic Tracking Error

21. Ultrasonics Tracker
 Idea: Time of Flight or Phase-Coherence Sound Waves
 ++: Small, Cheap
 -- : 3DOF, Line of Sight, Low resolution, Affected
Environment Conditon (pressure, temperature)
Ultrasonic
Logitech IS600

22. Inertial Tracker
 Idea: measuring linear and angular orientation rates
(accelerometer/gyroscope)
 ++: no transmitter, cheap, small, high frequency, wireless
 -- : drift, hysteris only 3DOF
IS300 (Intersense)
Wii Remote

23. Mobile Sensors
 Inertial compass
 Earth’s magnetic field
 Measures absolute orientation
 Accelerometers
 Measures acceleration about axis
 Used for tilt, relative rotation
 Can drift over time

24. Global Positioning System (GPS)
 Created by US in 1978
 Currently 29 satellites
 Satellites send position + time
 GPS Receiver positioning
 4 satellites need to be visible
 Differential time of arrival
 Triangulation
 Accuracy
 5-30m+, blocked by weather, buildings etc

26. Problems with GPS
 Takes time to get satellite fix
 Satellites moving around
 Earths atmosphere affects signal
 Assumes consistent speed (the speed of light).
 Delay depends where you are on Earth
 Weather effects
 Signal reflection
 Multi-path reflection off buildings
 Signal blocking
 Trees, buildings, mountains
 Satellites send out bad data
 Misreport their own position

27. Accurate to < 5cm close to base station (22m/100 km)
Expensive - $20-40,000 USD

28. Assisted-GPS (A-GPS)
 Use external location server to send GPS signal
 GPS receivers on cell towers, etc
 Sends precise satellite position (Ephemeris)
 Speeds up GPS Tracking
 Makes it faster to search for satellites
 Provides navigation data (don’t decode on phone)
 Other benefits
 Provides support for indoor positioning
 Can use cheaper GPS hardware
 Uses less battery power on device

29. Assisted GPS

30. Cell Tower Triangulation
 Calculate phone position
from signal strength
 < 50 m in cities
 > 1 km in rural

31. WiFi Positioning
 Estimate location by using WiFi access points
 Can use know locations of WiFi access points
 Triangulate through signal strength
 Eg. PlaceEngine (www.placeengine.com)
 Client software for PC and mobiles
 SDK returns position
 Accuracy
 5 – 100m (depends on WiFi density)

32. WiFi Hotspots in New York

34. Indoor WiFi Location Sensing
 Indoor Location
 Asset, people tracking
 Aeroscout
 http://aeroscout.com/
 WiFi + RFID
 Ekahau
 http://www.ekahau.com/
 WiFi + LED tracking

35. Integrated Systems
 Combine GPS, Cell tower, WiFi signals
 Skyhook (www.skyhookwireless.com)
 Core Engine
 Database of known locations
 700 million Wi-Fi access points and cellular towers.

37. Comparative Accuracies
 Study testing iPhone 3GS cf. low cost GPS
 A-GPS
 8 m error
 WiFi
 74 m error
 Cell Tower Positioning
 600 m error
Accuracy of iPhone Locations: A Comparison of Assisted GPS, WiFi, and
Cellular Positioning
In GIScience on July 15, 2009 at 8:11 pm By Paul A Zandbergen
Transactions in GIS, Volume 13 Issue s1, Pages 5 - 25

38. Optical Tracking

39. Optical Tracker
 Idea: Image Processing and Computer Vision
 Specialized
 Infrared, Retro-Reflective, Stereoscopic
 Monocular Based Vision Tracking
ART Hi-Ball

40. Outside-In vs. Inside-Out Tracking

41. Optical Tracking Technologies
 Scalable active trackers
 InterSense IS-900, 3rd Tech HiBall
 Passive optical computer vision
 Line of sight, may require landmarks
 Can be brittle.
 Computer vision is computationally-intensive
3rd Tech, Inc.

42. HiBall Tracking System (3rd Tech)
 Inside-Out Tracker
 $50K USD
 Scalable over large area
 Fast update (2000Hz)
 Latency Less than 1 ms.
 Accurate
 Position 0.4mm RMS
 Orientation 0.02° RMS

44. Starting simple: Marker tracking
 Has been done for more than 10 years
 A square marker provides 4 corners
 Enough for pose estimation!
 Several open source solutions exist
 Fairly simple to implement
 Standard computer vision methods

45. Marker Based Tracking: ARToolKit
http://artoolkit.sourceforge.net/

46. Tracking Range with Pattern Size
Rule of thumb – range = 10 x pattern width

47. Tracking Error with Range

48. Tracking Error with Angle

49. Tracking challenges in ARToolKit
False positives and inter-marker confusion
(image by M. Fiala)
Image noise
(e.g. poor lens, block coding /
compression, neon tube)
Unfocused camera,
motion blur
Dark/unevenly lit
scene, vignetting
Jittering
(Photoshop illustration)
Occlusion
(image by M. Fiala)

50. Limitations of ARToolKit
 Partial occlusions cause tracking failure
 Affected by lighting and shadows
 Tracking range depends on marker size
 Performance depends on number of markers
 cf artTag, ARToolKitPlus
 Pose accuracy depends on distance to marker
 Pose accuracy depends on angle to marker

51. Tracking, Tracking, Tracking

52. Other Marker Tracking Libraries
 arTag
 http://www.artag.net/
 ARToolKitPlus [Discontinued]
 http://studierstube.icg.tu-graz.ac.at/handheld_ar/
artoolkitplus.php
 stbTracker
 http://studierstube.icg.tu-graz.ac.at/handheld_ar/
stbtracker.php
 MXRToolKit
 http://sourceforge.net/projects/mxrtoolkit/

53. Markerless Tracking

55. Markerless Tracking
Magnetic Tracker Inertial
Tracker
Ultrasonic
Tracker
Optical
Tracker
Marker-Based
Tracking
Markerless
Tracking
Specialized
Tracking
Edge-Based
Tracking
Template-Based
Tracking
Interest Point
Tracking
 No more Markers! Markerless Tracking

56. Natural feature tracking
 Tracking from features of the surrounding
environment
 Corners, edges, blobs, ...
 Generally more difficult than marker tracking
 Markers are designed for their purpose
 The natural environment is not…
 Less well-established methods
 Usually much slower than marker tracking

57. Natural Feature Tracking
 Use Natural Cues of Real Elements
 Edges
 Surface Texture
 Interest Points
 Model or Model-Free
 ++: no visual pollution
Contours
Features Points
Surfaces

58. Texture Tracking

59. Edge Based Tracking
 RAPiD [Drummond et al. 02]
 Initialization, Control Points, Pose Prediction (Global Method)

60. Line Based Tracking
 Visual Servoing [Comport et al. 2004]

61. Model Based Tracking
 Track from 3D model
 Eg OpenTL - www.opentl.org
 General purpose library for model based visual tracking

62. Marker vs. natural feature tracking
 Marker tracking
 + Can require no image database to be stored
 + Markers can be an eye-catcher
 + Tracking is less demanding
 - The environment must be instrumented with markers
 - Markers usually work only when fully in view
 Natural feature tracking
 - A database of keypoints must be stored/downloaded
 + Natural feature targets might catch the attention less
 + Natural feature targets are potentially everywhere
 + Natural feature targets work also if partially in view

63. Hybrid Tracking

64. Sensor tracking
 Used by many “AR browsers”
 GPS, Compass, Accelerometer, (Gyroscope)
 Not sufficient alone (drift, interference)

65. Outdoor Hybrid Tracking
 Combines
 computer vision
- natural feature tracking
 inertial gyroscope sensors
 Both correct for each other
 Inertial gyro - provides frame to frame
prediction of camera orientation
 Computer vision - correct for gyro drift

66. Combining Sensors and Vision
 Sensors
- Produce noisy output (= jittering augmentations)
- Are not sufficiently accurate (= wrongly placed augmentations)
- Gives us first information on where we are in the world,
and what we are looking at
 Vision
- Is more accurate (= stable and correct augmentations)
- Requires choosing the correct keypoint database to track from
- Requires registering our local coordinate frame (online-
generated model) to the global one (world)

67. Outdoor AR Tracking System
You, Neumann, Azuma outdoor AR system (1999)

68. Robust Outdoor Tracking
 Hybrid Tracking
 Computer Vision, GPS, inertial
 Going Out
 Reitmayer & Drummond (Univ. Cambridge)

69. Handheld Display

70. Registration

71. Spatial Registration

72. The Registration Problem
 Virtual and Real must stay properly aligned
 If not:
 Breaks the illusion that the two coexist
 Prevents acceptance of many serious applications

73. Sources of registration errors
 Static errors
 Optical distortions
 Mechanical misalignments
 Tracker errors
 Incorrect viewing parameters
 Dynamic errors
 System delays (largest source of error)
- 1 ms delay = 1/3 mm registration error

74. Reducing static errors
 Distortion compensation
 Manual adjustments
 View-based or direct measurements
 Camera calibration (video)

75. View Based Calibration (Azuma 94)

76. Dynamic errors
 Total Delay = 50 + 2 + 33 + 17 = 102 ms
 1 ms delay = 1/3 mm = 33mm error
Tracking Calculate
Viewpoint
Simulation
Render
Scene
Draw to
Display
x,y,z
r,p,y
Application Loop
20 Hz = 50ms 500 Hz = 2ms 30 Hz = 33ms 60 Hz = 17ms

77. Reducing dynamic errors (1)
 Reduce system lag
 Faster components/system modules
 Reduce apparent lag
 Image deflection
 Image warping

78. Reducing System Lag
Tracking Calculate
Viewpoint
Simulation
Render
Scene
Draw to
Display
x,y,z
r,p,y
Application Loop
Faster Tracker Faster CPU Faster GPU Faster Display

79. Reducing Apparent Lag
Tracking
Update
x,y,z
r,p,y
Virtual Display
Physical
Display
(640x480)
1280 x 960
Last known position
Virtual Display
Physical
Display
(640x480)
1280 x 960
Latest position
Tracking Calculate
Viewpoint
Simulation
Render
Scene
Draw to
Display
x,y,z
r,p,y
Application Loop

80. Reducing dynamic errors (2)
 Match input streams (video)
 Delay video of real world to match system lag
 Predictive Tracking
 Inertial sensors helpful
Azuma / Bishop 1994

81. Predictive Tracking
Time
Position
Past Future
Can predict up to 80 ms in future (Holloway)
Now

82. Predictive Tracking (Azuma 94)

83. Wrap-up
 Tracking and Registration are key problems
 Registration error
 Measures against static error
 Measures against dynamic error
 AR typically requires multiple tracking technologies
 Research Areas: Hybrid Markerless Techniques,
Deformable Surface, Mobile, Outdoors

84. Project List
 Mobile
 Hybrid Tracking for Outdoor AR
 City Scale AR Visualization
 Outdoor AR Authoring Tool
 Outdoor AR collaborative game
 AR interaction for Google Glass
 Non-Mobile
 AR Face Painting
 AR Authoring Tool
 Tangible AR puppeteer studio
 Gesture based interaction with AR content

85. More Information
• Mark Billinghurst
– mark.billinghurst@hitlabnz.org
• Websites
– www.hitlabnz.org