MAS 962: Interaction Techniques for Virtual Environments
Tuesday, November 10, 1998
Sensing vs. Measuring: Any sense using computer vision?
INTRO
Let's go back to sensing, measuring, perceiving
Talked about sensing as being able to perceive with one
measurement
Joe Paradiso - direct mapping of input (measurements) to output
(music, etc.)
- measuring techniques
- electric fields
- gyros
If you can build these relatively complex systems for sensing and
can map these to desired output WHY would you want to deal with
vision?
Try to answer that today
USING COMPUTER VISION
What, How, When Might You Use Computer Vision?
- Recognition
- absolute: who you are in the whole world; can be done
unobtrusively so people don't know it's being done
- drivers licensing
- welfare fraud
- FBI
- immigration
- relative: who you are in the current interaction; keeping
track of where you've been or what you've seen
- museum
- brain opera
- tutorials
- relative easier than absolute because:
- more reliable (set is sparser)
- picture of person is current
- no giant database
- People Detection (vs. detection of movement) - can reasonably
tell if it's a person, not a dog, not something falling
- "Color" - easy to measure visual property; more accurate
depiction of environment
- Facial Expressions - are they surprised, bored, etc.
- Direction of Attention - also perceive user's attention
- Gesture Recognition
- American Sign Language
- iconic gesture interface - define gestures; design
non-confusing language
- Wall Street trades
- disabled people
- Motion
Why Would You Use Computer Vision?
- Don't want electronics on the person - desire distal
measurement, remote sensing
- intrusive/intimidating
- unencumbered
- awareness/magic/suspension of disbelief
- Variable resolution
- Non-complicity - user doesn't have to comply for system to
work
- Common "vocabulary" of sensing between system and user
Warning - High Expectations
- system should just "see"
- system that knows I'm happy sometimes; expect it to know all
the time
- brain opera not a familiar concept - no expectations
RECOGNITION OF HUMAN MOVEMENT - Prof. Bobick's slides
(Much of the information from Prof. Bobick's slides and
presentation can be found on Jim Davis' and Aaron Bobick's web page,
Appearance
Based Motion Recognition of Human Actions, additional information
on the projects can be found through links from Jim
Davis' home page)
motion - video - compare one frame to next
Building Blocks
- movement - primitive motion
- atomic, indivisible
- example: batting (single swinging motion)
- activities
- sequence of movements
- example: pitching (wind-up, leg kick, forward motion, ball
release)
- actions
- semantic labels, higher level labels
- example: tagging out a runner, a football play (requires
knowledge of context and "meaning" of movements)
Motion Energy Images - MEI
Highlight regions in which any form of motion was present
Motion History Images - MHI
Show regions where movement has occurred in such a way
that intensity is a function of recency of motion
STAR TREK DOORS
From previous weeks' discussion: how do we build Star Trek doors
that open automatically just in time for someone to pass through
- Coarse floor sensor, pressure sensor
- negative: if facing away from door and rock back, this will
open falsely)
- Add direction via Joe Paradiso
- com badges
- shoes
- negative: requires compliance, doesn't work for non-Star
Fleet
- Pressure image
- better floor sensing
- negative: long cue time
- Face facing door via vision
- negative: false alarms, "where is he"
- IR - heat proximity
- Looming detection via vision (method used by fly in landing)
- calculate time to impact
- when is as important as if
Why not just say "door open"; we're just talking about a fancy
doorknob
What's the loss to an interactive system of NOT having sentient
doors?
- there would be no sentient doors
- info is power (info from doors could be used for other
things)
- no magic
- cognitive load
Invisible technology is cool
Perception is that Star Trek doors are just next step up from
grocery store doors - easily made - not so
Back to
Course Home Page