Faculty of Informatics Vienna University of Technology Institute of Computer Aided Automation PRIP Home PRIP Home
Personal tools
You are here: Home Teaching PRIP-Talks Oliver Hörbinger: Object Identification in non overlapping views

Oliver Hörbinger: Object Identification in non overlapping views

Spezifikationsvortrag DA

  • Presentation
When Jul 07, 2009
from 04:15 pm to 04:35 pm
Where Sem 183/2
Add event to calendar vCal
The goal of my master thesis will be to identify objects in several non overlapping views. The
identification of the object should be achieved with little input, a few frames of a video with
the object, which means no complex learning can be done.
Due to the complexity of this task it will be necessary to make several constraints. Some of
them are obvious, others will result during implementation.
After some literature research I realized that there is a lot of research going on as well for
object recognition/classification as for object tracking. Both topics have more or less strong
overlaps but do not perfectly match my task.
For object classification the identity of the objects is not important, but only the class,
whereas my task will be to find one special object again (although one constraint will be that
objects have to be different in some way to each other to keep the complexity within a limit
– however object classification tasks are solved with complex learning algorithms which
require a lot of input data which is not available in my case).
The main goal of object tracking is to keep track of the object all the time and not losing it
whereas my goal is to find the object again over different views even if the object position is
lost because of occlusions.
To achieve my goal I will try to combine the strengths of these two research topics to get
robust object description, searching and matching algorithms.
Document Actions