Architectural Model Capture and LocationAware Applications: Geometric Models in the Real World Seth

1
Architectural Model Captureand Location-Aware
ApplicationsGeometric Models in the Real
WorldSeth TellerComputer Graphics GroupMIT
LCS/AI LabNTT Presentation, January 2002NTT
Partner Takayuki Yasuno(Some work joint with
Hari Balakrishnan, Erik Demaine)
2
We are working to

• Capture and create accurate geometric models of
  architectural spaces, rapidly and automatically
• Develop location and orientation sensors that
  work reliably both indoors and outdoors
• Demonstrate a variety of applications that use
  models of the real world, in the real world

3
Goal support these applications

• Indoor navigation
• Software compass (Oxygen hand-held device)
• Asset tracking
• Software marker
• Facilities maintenance
• Software flashlight

4
Navigation, Resource Finding

• Hand-held software compass
• Knows its position, orientation
• Enables you to
• Find people, resources in offices, museums, etc.

LCS Floor 2
LCS Floor 1
5
Asset Maintenance

• Software marker
• Model Software compass Laser pointer
• Enables you to
• Annotate and query real-world objects in database

Maintenance request Filed by Inspector Button
does not work. Date January 15, 2002
Office lock NE43-202 Occupant Leiserson Keys
Leiserson, Re-keyed Oct. 1995
Printer NE43-2 South Owner Graphics
group Subnet 18.24.2. Admin hanna_at_graphics
6
Facilities Maintenance

• Software flashlight
• 3D Model Software compass Projector
• Enables you to
• Project CAD information onto real objects

Inspect Wiring
Install Power outlet
7
How are we getting there?

• Vision-based model capture
• Develop scalable computer vision algorithms to
  acquire accurate models of architectural spaces
• CAD-based model generation
• Exploit existing 2D CAD information, and simple
  rules, to generate rich 3D models
• Pervasive location/orientation capability
• Cricket beacons (w/ Hari Balakrishnan)
• Deploy throughout LCS, rest of campus

8
How are we getting there?
Vision-based model capture
Environment model and object metadata
CAD-based model generation
Cricket pervasive location, orientation
Cricket is joint work with Hari Balakrishnan
9
Vision-based model capture Goal

• Scan 3D environment with one or more ordinary
  hand-held video cameras
• Example LCS 2nd Floor Video Sequence

10
Vision-based model capture Progress

• Rotation-stabilized video (locked to scene)
• Extracted 3D line cloud (no polygons, texture)

11
Vision-based capture Next steps

• Address scaling and complexity issues
• Extract coarse geometry and texture
• Camera excursions through multiple floors
• Multiple cameras operating in parallel
• Target capture 3D model of all LCS/AI Lab

12
CAD-based model generation Goal

• Exploit existing (poor) 2D CAD models

• Compile 2D CAD into well-formed 3Dgeometric
  models

13
CAD-based model capture Progress

• Collected 100 MIT buildings, 900 floorplans
• Registered to common 3D coordinate system
• Extracted named spaces, 2D 3D adjacencies

14
Cricket location and orientation mechanism
Beacons on ceiling
SPACENE43-510 ID34 COORD146 272 0
Ultrasound (pulse)
Cricket listeners
Software compass
Obtain distances to multiple beacons Solve for
(x, y, z, ?) of hand-held device
15
Cricket v1 Prototype
RF module (transmit)
Ultrasonic sensor
RF module (receive)
RF antennas
Listener
Beacon
Atmel processor
RS232 i/f
Host software libraries in Java Linux daemon
(in C) for Oxygen BackPaq handhelds Several
prototype applications
16
Deployment (LCS 5th floor)
Prototype applications Navigation Resource
location Next steps Deploy to all LCS/AI
Marker, flashlight proofs-of-concept
17
Summary
Pursuing three integrated activities

• Model creation
• Vision-based capture, CAD-based generation
• Pervasive location, orientation capability
• Multiple active beacons, passive listeners
• Compelling location-aware applications
• Navigation, asset management, maintenance

18
(No Transcript)