Google Glass, The META and Co. - How to calibrate your Optical See-Through Head Mounted Displays

Post on 15-Jun-2015

1.069 views 4 download

Tags:

description

Slides from our ISMAR 2014 tutorial http://stctutorial.icg.tugraz.at/ Abstract: Head Mounted Displays such as Google Glass and the META have the potential to spur consumer-oriented Optical See-Through Augmented Reality applications. A correct spatial registration of those displays relative to a user’s eye(s) is an essential problem for any HMD-based AR application. At our ISMAR 2014 tutorial we provide an overview of established and novel approaches for the calibration of those displays (OST calibration) including hands on experience in which participants will calibrate such head mounted displays.

Transcript of Google Glass, The META and Co. - How to calibrate your Optical See-Through Head Mounted Displays

Introduction to Optical See-Through HMD Calibration

Jens Grubert (TU Graz) Yuta Itoh (TU Munich)

9th Sep 2014

jg@jensgrubert.de yuta.itoh@in.tum.de

Theory14:15 Introduction to OST Calibration15:00 coffee break15:15 Details of OST Calibration16:15 coffee break

Practice16:30 Hands on session: calibration of OST HMDs17:30 Discussion: experiences, feedback17:50 wrap-up, mailing list18:00 end of tutorial

Part 1

Theory: Introduction

Non Optical See-Through

Optical See-Through (OST)

Head Mounted Displays (HMD)

Issues on OST-HMD

Photo by Mikepanhu

ConsistencyConsistency

Photo by javier1949

The Lack of consistenciesSpatial

Visual

Temporal

Social

Temporal Inconsist. in OST-HMD

“How fast is fast enough? : a study of the effects of latency in direct-touch pointing tasks” Jota et al. CH’13

https://www.youtube.com/watch?v=PCbSTj7LjJg

“latencies down to 2.38 ms are required to alleviate user perception when dragging”

Temporal Inconsist. in OST-HMD

Digital Light Processing Projector

“Minimizing Latency for Augmented Reality Displays: Frames Considered Harmful” Zheng et al. ISMAR’14

Visual Consistency

Occlusion, Depth, Color, Shadow,

Lee and woo, 2009

Kiyokawa et al. 2003

Liu et al. 2008

Wide Field of View, etc…Visual Consistency

“Pinlight Displays: Wide Field of View Augmented Reality Eyeglassesusing Defocused Point Light Sources” Maimone et al., TOG’14

Social Consistency

Image from Google Glass: Don't Be A Glasshole | Mashable

Social Consistency

Image from http://www.thephoblographer.com/

Spatial registration Calibration

Spatial Inconsistency in OST-HMD

Spatial Inconsistency in OST-HMD

Calibration of Eye&OST-HMD

Find 3D-2D Projection :

Camera Calibration Analogy

KIntrinsic

Find 3D-2D Projection:

OST-HMD’s Screen to Camera

P = K*[R t]

P

We can not see what you see!

In the Eye of the Beholder…

P = K*[R t] ?

– Intensive user interaction–User-dependent noise

Manual alignment

[AZU97]

P is a Black Box

PPFind 3D-2D Projection:

3D 2D

–Medium user interaction

–User-dependent noise

SPAAM: Single Point Active Alignment Method

[TN00][GTN02]

N times

You got a perfect P!!!

P

Oops, sorry I touched your HMD…

P

Essential Difficulties

?1 Data acquisition

2 Dynamic parameter changes

Data collectionSPAAM, MPAAM, Stereo Calibration

Part 2: Overview

State of the art

Practical tips

Confirmation Methods

Evaluation

Data Collection: SPAAM

Data Collection: MPAAM

Data Collection: Stereo

Confirmation Methods

Keyboard

Voice

Handheld

Waiting

State of the Art

Practical Tips

Theory14:15 Introduction to OST Calibration15:00 coffee break15:15 Details of OST Calibration16:15 coffee break

Practice16:30 Hands on session: calibration of OST HMDs17:30 Discussion: experiences, feedback17:50 wrap-up, mailing list18:00 end of tutorial

Part 1

Theory: Details

Data Collection Methods: SPAAM Single Point Active Alignment Method

Eye-HMD Calibration

P3D

2D

P is a Black Box

PFind 3D-2D Projection:

P 3D2D

P

2D

Say, is a Perspective ProjectionP

2D-3D correspondences gives

3D2D P

P

Only users can see the 2D points!

–Medium user interaction

–User-dependent noise

SPAAM: Single Point Active Alignment Method

[TN00][GTN02]

N times

SPAAM: Single Point Active Alignment Method

Minimum 6 pairs

Better 16~20 pairs

3D

2D

3D

3D

Better distributed in Z axis

Data Collection Methods: Stereo Calibration

SPAAM: Calibration for a Single Display

How to calibrate stereo systems?

How to calibrate stereo systems?

Idea 1: Calibrate each eye individually

Calibrate each eye individually

How to calibrate stereo systems?

Idea 2: Calibrate both eyes simultaneously Why?Save time

Calibrate both eyes simultaneously

Idea

1. display 2D objects with disparity in left and right eye appears as single object at a certain distance

2. Align virtual with physical 3D object Get point correspondence for both eyes

[GSW00]

Challenges for Simultaneous Alignment

• Shape of the virtual object

• Occlusion of physical target

• Vergence-accomodation conflict

Simultaneous calibration can be significantly faster to calibrate

Perceptual issues might hinder quality calibration

Stereo Calibration Take Aways

Data Collection Methods: Multi-Point Collection

Idea

SPAAM:

align a single point multiple times

Multi-Point Active Alignment (MPAAM):

align several points concurently but only once

Why?

save time

Example: SPAAM

Example: MPAAM

MPAAM Variants

• Align all pointsat once

• Minimum of six points

• Vary spatial distribution

[TMX07]

MPAAM Variants

• Align all pointsat once

• Minimum of six points

• Vary spatial distribution

• Missing: tradeoff # points - # calibration steps [GTM10]

Performance• MPAAM can be conducted

significantly faster than SPAAM (in average in 84 seconds vs 154 seconds for SPAAM) [GTM10]

• MPAAM has comparable accuracy in the calibrated range

MPAAM take aways

MPAAM can be alternative to SPAAM if

• Working volume can be covered by calibration body

• Need for repeated calibration (e.g., after HMD slips)

Confirmation Methods

User has to confirm 2D-3D matching

I take the pair NOW!

How to make confirmation stable?

I take the pair NOW!

Different confirmation options

Keyboard

Voice

Handheld

Waiting

[MDW11]

Less motion is better[MDW11]

Evaluation: User in the Loop

Evaluation Questions• How accurate is the overlay given

the current calibration? [MGT01] [GTM10]

• How much do the calibration results vary between calibrations? [ASO11]

• What is the impact of individual error sources on the calibration results?– Head pointing accuracy, body sway,

confirmation methods ... [AXH11]

Evaluation Questions• How accurate is the overlay

given the current calibration? [MGT01] [GTM10]

• How much do the calibration results vary between calibrations? [ASO11]

• What is the impact of individual error sources on the calibration results?– Head pointing accuracy, body sway,

confirmation methods ... [AXH11]

How accurate is the overlay given the current calibration?

Popular Approaches

Use a camera Ask the user

User in the Loop Evaluation

Qualitative feedback„overlay looks good“

Quantitative feedback

User in the Loop Evaluation

Qualitative feedback„overlay looks good“

Quantitative feedback

Quantitative Feedback

McGarrity et al. [MGT01]:• Use a tracked evaluation

board• Ask AR system to

superimpose object on,

• Ask user to indicate where she perceives the object on the board ,

• Offset:

Quantitative Feedback

McGarrity et al. [MGT01]:• Use a tracked evaluation

board• Ask AR system to

superimpose object on,

• Ask user to indicate where she perceives the object on the board ,

• Offset:

Quantitative Feedback

• Drawback of stylus approach: evaluation only within arm‘s reach

Alternatives• Use laser pointer + human

operator instead (beware pointing accuracy) [GTM10]

• Use projector / large display + indirect pointing (e.g., mouse)

Quantitative Feedback

Benefits:• Only way to approximate how the

user herself perceives the augmentation

Drawbacks:• Only valid for current view (distance,

orientation)• Additional pointing error introduced

Take Aways

• Quantitative user feedback only way to approximate how large the registration error is for indivdual users

• Feedback methods introduce additional (pointing) errors

• Make sure to test for all relevant working distances

Evaluation: Error Measurements

OST-HMD Calibration

2DProjection

Matrix

P3D

Ideal Case

3D-2D pairs:S

Eye positions:(Camera center)

2D Projection Error

,

Wrong Projection

Reprojection Error

3D Eye Positions[m]

>10 cm

Semi-Automatic Calibration Approaches

Motivation

Can the calibration process be shortened?

User guided See-Through Calibration too tedious

https://www.flickr.com/photos/stuartncook/4613088809/in/photostream/

Observation

We have to estimate 11 parameters

--> At least 6 point correspodences needed

P2D 3D

Reminder: Collecting Correspondences

Idea

Separate certain parameters which are independent from the user?

The user would need to collect fewer point correspondences, making the task faster and easier.

Reminder: Calibratrion Parameters Pinhole Camera

TCS: Tracking Coordinate SystemEDCS: Eye-Display Coordinate System

TCS

EDCS

Rotation and Translation between Tracking Coordinate System and Eye-Display Coordinate System: 6 Parameters for center of projection

5 intrinsic parameters of Eye-Display optical system:

focal length (x,y), shear, principal point (x,y)

(+ more if you want to modell distortion)

Separate intrinsic + extrinsic parameters

[OZT04]:1. Determine ALL parameters

(including distortion) via camera without user intervention

2. Update center of projection in a user phase

State of the art: Automatic Method

Utilizes 3D Eye Localization [IK14]

– Interaction-free, thus do not bother users

–More accurate than a realistic SPAAM setup

INDICA: Interaction-free DIsplay CAlibration

1. Estimate a 2D iris ellipse– Iris detector + Fitting by

RANSAC

2. Back project it to 3D circle

[SBD12]

3D Eye Position Estimation

[NNT11]

Manual (SPAAM)

Interaction Free (INDICA Recycle)

Interaction Free (INDICA Full)

World to HMD(eye) Projection

3D

Screen

2D

P

P

P

Simple No user interaction

Accurate better than Degraded manual calibrations

Summary of INDICA

Calibration of OST-HMDs using3D eye position

Practical Tips

How many control points for SPAAM?

• Minimum of 6 can lead to unstable and innaccurate results?

• The more the better? Not neccesarily 16-20 control points sufficient if points are equally distributed in all three dimensions

16 20

Calibration Error [mm] [CAR94]

Calibration Volume

If possible calibrate the working volume you want to operate in

Working Volume

Calibration

Volume

Quality of Tracking System

Ensure the best calibration possible for your external tracking system

Ensure a low latency

Summary of Part 2

Reducing user errors:- Data-collection- Confirmation- Evaluation

Manual to automatic:State of the art

Practical tips

References 1/2[AXH11] Axholt, M. (2011). Pinhole Camera Calibration in the Presence of Human Noise.

[ASO11] Axholt, M., Skoglund, M. A., O'Connell, S. D., Cooper, M. D., Ellis, S. R., & Ynnerman, A. (2011, March). Parameter estimation variance of the single point active alignment method in optical see-through head mounted display calibration. In Virtual Reality Conference (VR), 2011 IEEE (pp. 27-34). IEEE.

[AZU97] Azuma, R. T. (1997). A survey of augmented reality. Presence, 6(4), 355-385.

[CAR94] Chen, L., Armstrong, C. W., & Raftopoulos, D. D. (1994). An investigation on the accuracy of three-dimensional space reconstruction using the direct linear transformation technique. Journal of biomechanics, 27(4), 493-500.

[CNN11] Christian, N., Atsushi, N., & Haruo, T. (2011). Image-based Eye Pose and Reflection Analysis for Advanced Interaction Techniques and Scene Understanding. CVIM,, 2011(31), 1-16.

[GTM10] Grubert, J., Tuemler, J., Mecke, R., & Schenk, M. (2010). Comparative User Study of two See-through Calibration Methods. In VR (pp. 269-270).

[GTN02] Genc, Y., Tuceryan, M., & Navab, N. (2002, September). Practical solutions for calibration of optical see-through devices. In Proceedings of the 1st International Symposium on Mixed and Augmented Reality (p. 169). IEEE Computer Society.

References 2/2[MAE14] Moser, K. R., Axholt, M., & Edward Swan, J. (2014, March). Baseline SPAAM calibration accuracy and precision in the absence of human postural sway error. In Virtual Reality (VR), 2014 iEEE (pp. 99-100). IEEE.

[MGT01] McGarrity, E., Genc, Y., Tuceryan, M., Owen, C., & Navab, N. (2001). A new system for online quantitative evaluation of optical see-through augmentation. In ISAR 2001 (pp. 157-166). IEEE.

[MDW11] P. Maier, A. Dey, C. A. Waechter, C. Sandor, M. Tönnis and G. Klinker, "An empiric evaluation of confirmation methods for optical see-through head-mounted display calibration. In International Symposium on Mixed and Augmented Reality (ISMAR), 2011 IEEE.

[OZT04] Owen, C. B., Zhou, J., Tang, A., & Xiao, F. (2004, November). Display-relative calibration for optical see-through head-mounted displays. In Mixed and Augmented Reality, 2004. ISMAR 2004. Third IEEE and ACM International Symposium on (pp. 70-78). IEEE.

[SBD12] Świrski, L., Bulling, A., & Dodgson, N. (2012, March). Robust real-time pupil tracking in highly off-axis images. In Proceedings of the Symposium on Eye Tracking Research and Applications (pp. 173-176). ACM.

[TU00] Tuceryan, M., & Navab, N. (2000). Single point active alignment method (SPAAM) for optical see-through HMD calibration for AR. In Augmented Reality, 2000.(ISAR 2000). Proceedings. IEEE and ACM International Symposium on (pp. 149-158). IEEE.