The Computer Vision Center, CVC-Barcelona (www.cvc.uab.es), offers 4 PhD
research positions in the field of Computer Vision and Image Understanding.

Background

The CVC is a non-profit institution whose strategic objective is to do
Research and Development on Computer Vision. From a scientific point of
view, the CVC wants to contribute to increase the knowledge in this
field. From a technological point of view, it aims to contribute to
innovation and industrial competitiveness collaborating with companies
to develop technological projects.

The CVC model plans a natural cooperation between Research and
Development towards technology transfer, and offers a modern and
enthusiastic research environment with strong interdisciplinary and
international links.

Responsibilities of successful candidates include project work and
research results, and supervision of student projects. The working
language is English.


Offered Positions

Candidates will perform high quality research to acquire the know-how
and research experience in the following areas:

PhD Student 1: Face Detection and Tracking based on Active Cameras

We will use standard body, and face-detection software (for example via
boosted classifier) to provide an attentional mechanism for controlling
zoom. When a body/face is detected, the camera's zoom will be
controlled to supply imagery at the appropriate resolution for
verification of a higher-level hypothesis (for example about emotion
recognition). The challenge is to modify the parameters of a PTZ camera
in order to obtain the best resolution and viewpoint for head tracking.
In addition, the camera's zoom will be controlled to supply imagery
at an appropriate resolution for motion analysis of the human face, thus
facilitating emotion analysis and expression recognition. Even more
interesting will be the control of zoom in response to uncertainty, and
in particular uncertainties and ambiguities during face tracking.


PhD Student 2: Body Gesture and Action Recognition based on Image
Sequences

The goal is to study current algorithms related to pose analysis of
human bodies. This could lead to a concept on how to develop a new and
robust method to capture the motion of human postures. In principle, a
hierarchical approach will be proposed; in which first, the torso is
found then the head, arms, and legs. The developed system should be able
to acknowledge the fact that sometimes not all body parts can be found.
This incomplete data problem will be addressed by a Bayesian approach in
which training data will constitute the prior information. These models
will greatly help the tasks of tracking and action recognition in image
sequences. A database of human actions will be recorded, stored and
managed to build suitable learning and test sets.



PhD Student 3: Agent Detection and Tracking on Active Cameras

We propose to track agents and note their trajectories and other coarse
scale features that will be useful for action and intention recognition.
The goal is to study the notion of learning patterns of zoom and
pre-emptive zoom. This is observed in human camera operators (e.g. at
sporting events or in nature photography) in which a particular pattern
of activity can lead to anticipation of either greater uncertainty (and
hence the need for wider zoom such as when a batsman takes a wild swing
at a cricket ball) or conversely when a pre-learned activity is
indicative of the need for higher resolution information (e.g. zooming
in response to what appears to be a threatening pose in a bank). We will
use input from multiple cameras at possibly different resolutions and
including the trajectories of nearby agents in order to predict
occlusions. Furthermore, we aim to consider how cooperating
pan-tilt-zoom sensors can enhance the process of cognition via
controlled responses to uncertain or 


PhD Student 4: Interpretation of Human Behaviour based on Image
Sequences

We propose to model human behaviors using sets of fuzzy or probabilistic
rules that encode a suitable overload goal and expected activities. Such
a generative model will be used for prediction and fused with
observation data to obtain a posteriori estimates of the agent's
internal state. Such estimates can then be used for inference and causal
reasoning by considering coupled behavior. Bayesian Networks and belief
propagation will constitute the general framework in which uncertain
information will be fused and relative to which inference can be made in
a principled fashion. Likewise, we will establish similar models for
crowd observations and behavior. Where in sparse scenes it is reasonable
to assume that reliable trajectories can be extracted (and inferences
about individuals are possible), in a crowd it seems more appropriate to
extract coarse-scale statistical properties: texture, optic flow, etc.
These low level descriptors; like trajectories, will then be used for
general i

Applications

As the CVC only works with Computer Science (particularly in Computer
Vision), PhD students who are admitted ought to have a strong background
in computer science. Evidence of this background should include either a
Bachelor's Degree or a Master's Degree in Computer Science or a closely
related field. Exceptionally, candidates with degrees in other areas
(Electrical Engineering, Physics, and Mathematics) will also be
considered.

In order to be considered for admission, grade point average of the
student during his/her undergraduate courses must be of at least 1.5.

Successful candidates are expected to do research in these fundamental
disciplines contribute to R&D deliverables such as "Video Surveillance",
"Human-Motion Modeling", Human-Computer Interfaces", and "Virtual Actors
in Synthetic Environments".


Submission of applications

Through an e-mail to: gkohatsu@cvc.uab.es, it should include the
following information:

Application letter

Curriculum Vitae and Academic Record

Letters of Reference

Gisele Kohatsu

Administration

Computer Vision Center - CVC

Campus UAB, Edifici O
08193, Bellaterra
Barcelona - Spain
Tel. +34 93 581 1228
Fax +34 93 581 1670