7.pdf This presentation captures many uses and the significance of the number...
Visual Nouns for Indoor/Outdoor Navigation
1. ICCHP 2012
Visual Nouns for Indoor/Outdoor
Navigation
Edgardo Molina, Dr. Zhigang Zhu, Dr. Yingli Tian
Dept. of Computer Science and Dept. of Electrical Engineering
Grove School of Engineering, The City College of New York
ICCHP 2012 – July 11, 2012
1
2. ICCHP 2012
Motivation
Assist the visually impaired in navigating new locations
Provide guidance to facilities and destinations
Restrooms
Exits
Elevators
Water fountains
How do the sighted recall a location? Visual cues?
2
4. ICCHP 2012
Using Visual Nouns
Provide contextual information
What is it telling us?
Provide landmark markers
Have I been here before?
Camera pose estimation => User Orientation
4
5. ICCHP 2012
Technology Alternatives
RFID
Embedded info
Costly
QR Codes / Bar Codes / Etc
Embedded info
Only meaningful to machines
GPS
Outdoors only
Signs and Text are already in our environment
5
8. ICCHP 2012
Localizing a user
1. Capture video of surroundings
2. Register video image frames to the first frame (will serve
as reference)
3. Generate a wide field-of-view panorama
4. Extract and match visual-noun features
5. Localize the user in 3D space relative to the visual nouns.
8
12. ICCHP 2012
Visual Noun Extraction
Maximally Stable Extremal Regions (MSER)
Performed on a sharpened image
Use of EE-MSER could be better [Chen et. al. - ICIP’11]
12
15. ICCHP 2012
Summary
We propose to use the signage that’s already in our
environment to provide orientation and context
information to visually impaired users.
Low cost to print out more signage
Also beneficial to sighted users
15
16. ICCHP 2012
Thank you!
Questions?
Work is supported by US NSF Emerging Frontiers in Research and Innovation
Program under Award No. EFRI-1137172, and City SEEDs: City College 2011
President Grant for Interdisciplinary Scientific Research Collaborations.
16
Editor's Notes
Visual cues – signs, visual markers, indoor/outdoorExample – store in Times square, recognizing logos around it…
Icons come from US DOTSigns can be general – CAFÉ can be in different style/font
Most markers are static => have I been here before? => estimate camera pose
Pros/consSigns are readily understandable by everyone.
AIGA = American Institute of Graphic Arts
Huge repository
Input: sequence of video framesFeature extraction
Mention signs…Natural occurring signs – fire, door
Stitched panorama
EE-MSER = edge enhancedDetection is from panorama
Set of 2D points and corresponding 3D points Perspective N-Point (PnP) problem for extrinsic camera parameter estimationCompare single vs. panorama (for localization purposes)Red = visual nounGreen = additional sign provided
Single view => bad pose, 4 featuresPanorama => 4 degrees off, 4 inches offPreliminary result/work
Generated a lot of false positiveFuture work – providing direction based on a query Main work is NOT to do sign detection, but for navigation and orientation.