presentatie van mm x(final2)

28
Willem Mestdagh - Kevin Cordemans

Upload: gueste26e85

Post on 13-Dec-2014

396 views

Category:

Technology


0 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Presentatie Van Mm X(Final2)

Willem Mestdagh - Kevin Cordemans

Page 2: Presentatie Van Mm X(Final2)

PapersMS Research

PlayAnywhere:Compact Interactive Tabletop Projection-Vision System []

Andy Wilson

Thinsight:Integrated Optical Multi-touch Sensing through Thin Form-factor Displays []

Bill Buxton, Steve Hodges, ...

Page 3: Presentatie Van Mm X(Final2)

PlayAnywherePlayAnywhere:

A Compact Interactive Tabletop Projection-Vision System

Andy Wilson at Microsoft Research, Redmond

2005

Page 4: Presentatie Van Mm X(Final2)

Introduction

Looks great...

But... Calibration? Installation? Portability? Cost?

Page 5: Presentatie Van Mm X(Final2)

Mainstream Consumer AcceptanceMulti-touch: new sensing and display technologies

Ubiquitious computing

PortabilityEase of installation

Ease of use: any surface?Calibration

Cost

Devise an easy-to-use, low-cost multi-touch system...

Page 6: Presentatie Van Mm X(Final2)

PlayAnywhereFront-projected computer vision-based

interactive table system which uses a new projecting technology to obtain an exceptionnaly self-contained form factor

Requirements:Quick set up: no calibration beyond factoryNo “installation”Usable on any flat surfaceCompact but still displaying and

sensing over a large surface (form factor)General purpose(Low cost)

Page 7: Presentatie Van Mm X(Final2)

Future?Canesta projection keyboard

Page 8: Presentatie Van Mm X(Final2)

PlayAnywhereWhat is it?Known problems in existing systemsHow does it work?

PlayAnywhere configuration Projector Camera and IR Illuminant

Image Processing Image Rectification Touch and Hover Analysis of shadow PlayAnywhere Visual Code Flow Move

Computer Vision Pros and Cons

Page 9: Presentatie Van Mm X(Final2)

Known problems in existing systems

Page 10: Presentatie Van Mm X(Final2)

PlayAnywhereWhat is it?Known problems in existing systemsHow does it work?

PlayAnywhere configuration Projector Camera and IR Illuminant

Image Processing Image Rectification Touch and Hover Analysis of shadow PlayAnywhere Visual Code Flow Move

Computer Vision Pros and Cons

Page 11: Presentatie Van Mm X(Final2)

PlayAnywhere Configuration

All-in-One

IR illuminant

Projector (with mirrors)

Camera with IR-passing-filter

Prototype

Page 12: Presentatie Van Mm X(Final2)

ProjectorNEC WT600 DLP

40’’ diagonale afbeeldingVoordelen van de plaats van de projector

Gevaarlijke plaatsing projector is vermeden Wanneer camera en projector samen zijn geplaatst

geen nood om de camera te hercalibreren bij verplaatsing van het toestel

• Nadelen Beeldkwaliteit kan niet gegarandeerd worden,

want projectie-oppervlak is niet gekend

Page 13: Presentatie Van Mm X(Final2)

Camera en IR IlluminantDe projector belicht de scene met zichtbaar

stralingDe illuminant belicht de scene met infrarode

stralingCamera ontvangt alleen gereflecteerde

infrarode stralen dankij de infrared-passing-filter die aan de camera bevestigd is(Camera blind voor reflecties van de projectorstraling)

Page 14: Presentatie Van Mm X(Final2)

How does it work?Image Processing

Preparation: Image RectificationComputer Vision Techniques: Analysis of

ShadowHover and TouchFlow MovePlayAnywhere Visual Code

Page 15: Presentatie Van Mm X(Final2)

Image Rectification (1)

Remove distortion via standard bilinear interpolation techniques

Parameters for transformations need only be determined once (factory):Lens specs known Fixed topology device-plane (assumption)

Setup Good thing Bad thing

Wide lens Maximize usable surface Creates barrel distortion

Oblique position of camera

Optimizes form factor (compact device)

Produces projective distortion

Page 16: Presentatie Van Mm X(Final2)

Image Rectification (2)

Page 17: Presentatie Van Mm X(Final2)

Image Rectification (3)Input image and projected image are in a one-to-

one correspondence, meaning:A rectangular object on the surface appears as a

rectangular object in the input image at the same (scaled) coordinates.

Rectified image suited for computer vision techniques

Limitation: objects further away from the unit will appear at a lower resolution(Minimum effective resolution is less than that of the acquired image)

Page 18: Presentatie Van Mm X(Final2)

Hover and TouchDetect touch without relying on special

instrumentation of the surface, so that the device may operate on any available flat surface (goal)

PlayAnywhere: exploite change in appearance of shadows as an object approaches the surface

No tracking of fingers, ...

Page 19: Presentatie Van Mm X(Final2)

Hover and Touch

thresholding operation

HoverCandidate finger positions:highest on each of the distinct shadows which enter the scene from the bottom of the image

TouchAnalysis of shadow shapes:threshold width of finger shadow at a location slightly below the topmost point

Page 20: Presentatie Van Mm X(Final2)

Hover and TouchTo limit false positives, candidate finger must lie on

a shadow that extends to the bottom of the image

Objects on table can corrupt touch detection by mimicing arm shadows, meaning their shadows extends to the

bottom of the image

Presently, only one finger per hand(more fingers requires more spohisticated analysis)

Precision limited by image resolution

Page 21: Presentatie Van Mm X(Final2)

Flow MoveSupport direct manipulation of virtual objects

(doc’s, photo’s)Gestures that are natural to a specific activity

E.g. resizing and rotating photo’s as if they were lying on a table

PlayAnywhere: Optical flow techniquesBased on local texture on the movig objectNo absolute position-based tracker, no widgetsInstead simple statistics summarizing the flow

field, computing only relative motion information

Page 22: Presentatie Van Mm X(Final2)

Optical Flow (1)Optical Flow computed from the most recently

acquired rectified image I(t) and the previous image I(t-1)

Mask images with grid

Image (t-1) Image (t)

Page 23: Presentatie Van Mm X(Final2)

For each (x,y) on a regular grid, the integer vector quantity (dx, dy) is determined, such that: image patch P centered on (x, y) at time t-1 most closely matches image patch P’ centered on (x+dx, y+dy) at time t.

Image patches are compared by computing pixelwise absolute differences (low values close match).

If image patch P closely matches image patch P’, there is a good chance that the two patches are representing the same content, shifted to a different gridpoint in the interval.

(dx, dy) gives the direction of this shift.

Optical Flow (2)

Image(t)(x+dx, y+dy)Image patch P’

Image(t-1)(x, y)Image patch P

Resulting vector

Page 24: Presentatie Van Mm X(Final2)

Optical Flow (3)For a given patch in the image, select (dx, dy) that minimizes

Vector field can be used as an argument in a function to infer meaning from user’s

gesture

Resulting vector field

Page 25: Presentatie Van Mm X(Final2)

PlayAnywhere Visual CodesOok mogelijk te werken met infrarood

systemen in objecten mogelijkheid tot werken met barcodes (derde vb met barcode)

Page 26: Presentatie Van Mm X(Final2)

Resultaat http://youtube.com/watch?v=YRfWhUnYyY8

Nu : Thingsight

Page 27: Presentatie Van Mm X(Final2)

Beschrijving“Optical sensing system, fully integrated into

a thin form factor display”Scherm met multi-touch systeem2 mogelijkheden om multitouchsystemen uit

te werken : - een arrangement van electroden op het

oppervlak- werken met cameras : voor of achter het

scherm

Page 28: Presentatie Van Mm X(Final2)

VergelijkingBeide Touch-algoritme op basis van

schaduwinval van handenBeide Handig te gebruiken en te verplaatsenPrincipe werking ongeveer gelijk

verschil : thinsight werkt met elektroden op het scherm, playanywhere werkt met camera’s boven het scherm

Beiden hebben de mogelijkheid tot werken met externe objecten met een infrarode bron

Voordeel bij thinsight : geen oppervlakte voor apparaat nodig