Camera/Vision for Geo-Location & Geo-Identification
description
Transcript of Camera/Vision for Geo-Location & Geo-Identification
Camera/Vision for Geo-Location & Geo-Identification
John S. Zelek
Intelligent Human Machine Interface LabDept. of Systems Design Engineering
University of Waterloo
Why can’t we use GPS everywhere?
Urban canyons
Indoor navigation
1. Introduction - 2/20
What we are trying to do
CameraInertial
Altimeter, Compass+/- GPS =
Accuracy +Location +
Maps +1. Introduction – 3/20
Applications
1. Introduction – 4/20
SLAM
Given:Dead-reck.Ext. sensorWaypoints
Not Known:MapGPS
2. SLAM – 5/20
Trees as landmarks
for triangulati
on
2. SLAM - 6/20
Daniel AsmarSlide 7
Differentiating different trees
2. SLAM – 7/20
2. SLAM – 8/20
Object Category
Recognition
3. Object Detection & Recognition – 9/20
Classes of Objects vs. Instances
2 instances of an individual object(space shuttle)
2 instances of an object face class
2 instances of an
object motorcycle
class3. Object Detection & Recognition – 10/20
Visual vs. Functional classes
There is a wide variation in the
appearance of objects that are categorized
by function. We focus only on
categories related by some
visual consistency only!
3. Object Detection & Recognition – 11/20
Challenges
changes of viewpoint
transformation (translation, rotation, scaling, affine), out-of-plane (foreshortening)
illumination differences
background clutter
occlusion
intra-class variation
3. Object Detection & Recognition – 12/20
Ours
Others
Repeatability of our detector appears to be better!
3. Object Detection & Recognition – 13/20
Object Graphs
3. Object Detection & Recognition – 14/20
3. Object Detection & Recognition – 15/20
3. Object Detection & Recognition – 16/20
4. Structure from Stereo – 17/20
Structure from stereo
Structure from motion4. Structure From Motion – 18/20
5. Context Recognition – 19/20
6. Closing – 20/20
Extra. Features for Recognition & Structure – 21/20
Extra. Features for Recognition & Structure – 22/20