Project 5: Recognition of specular, transparent
objects illuminated by multiple light sources
Description
Extend specular recognition
algorithm described in “Using specularities
for recognition”, M. Osadchy, D.W. Jacobs and R. Ramamoorthi to multiple
light sources.
Collect a
test set of several rotationally symmetric glass objects (at least 5):
- Take images of these objects filled with
opaque liquid for 3D model construction
from 2D object boundary.
- Take 3 images of each object with 2 and 3
light sources and different backgrounds.
Test the
algorithm on these objects.
Results
The detection results should
be presented as a confusion matrix:
|
|
|
Object
2 |
Object
… |
||
|
Image
1 |
Object with the best
score |
|
|
||
|
Image
2 |
|
|
|
||
|
Image
3 |
|
|
|
Each cell corresponds to the algorithm
decision regarding image j of object i. If the algorithm recognizes the object
correctly in all the images the object’s column will all be filled with
object’s number. If algorithm makes an error in some image, the corresponding
cell will show the number of the object that it recognized in the image.
Data sets
You should photograph the
images in the following settings:
1.
The
room should be dark.
2.
The
objects should be from thin glass and
rotationally symmetric about the vertical axis.
3.
Place
2,3 lamps around (not behind) the object and not too close to the object. Place
the lamps in such a way that you can observe the highlights from all the lamps
on the object. (I suggest showing the images to me before you proceed further
with the experiments).
4.
Place
the camera far enough from the object and use camera zooming to make the object
larger.
5.
Don’t
move the camera and the object between taking images, to avoid alignment work.
6.
Manually
extract the boundary of the object from the image with the opaque liquid.
7.
I
will provide the MATLAB script for constructing the 3D shape of the object and
alignment tool. (Contact me to get the scripts).
8.
All
three images should be taken with different textured background containing
white areas (in order to make it harder for the recognition algorithm).
Useful Links
Some
extra slides on reflection
Status
Claimed by Michal and Roni