A Lightweight Computer Vision-based Electronic Travel Aid

1
A Lightweight Computer Vision-based Electronic
Travel Aid

• Andrew B. Raij (raij_at_cs.unc.edu)
• Enabling Tech Project Status Report
• 3/6/2003

Department of Computer Science, UNC-Chapel Hill
2
High-Level (Long Term?) Goal

• Inexpensive, lightweight device that assists with
  orientation and mobility (OM)
• Gathers spatial data from the environment with
  one or more cameras and presents it to a blind
  user in a useful form

3
Big Questions (1)

• Why Cameras?
• NxM 2D grid of intensity / color data
• Find edges, other features
• Lots more information than ultrasound, infrared
  lasers, etc
• Wide field of view (particularly when using
  several cameras)
• Can extract long-range depth from the world (as
  opposed to 3-6 ft. of a cane)
• No need to scan the world like a cane
• Cameras are getting cheaper and cheaper

4
Big Questions (2)

• What data to gather?
• Intensity / Color
• Depth
• Cane is too short
• Cane does not help much for high objects
• Patterns (brick, tile, marble, etc?)
• Faces
• Objects

5
Big Questions (3)

• How to transform visual data to useful
  information?
• Auditory
• Dont want to affect ability to listen to the
  world
• Tactile
• Other?

6
Big Questions (4)

• How to do it fast on a small, lightweight device?
• Hardware Requirements
• Floating point ?
• Low heat, power
• Inexpensive
• Cameras and Camera Interface (Firewire? USB?)
• Possibilities
• PDA no floating point, useless video display
• Mini PC

7
This Semester

• Do a lot of Learning / Reading
• Perception of the world by the blind
• Existing OM Devices
• Existing OM work using computer vision
• Experiment with Multiple View Geometry concepts
• Two-View Geometry
• Mirror Stereo
• Look into inexpensive, lightweight hardware
  platforms
• Begin implementation