[robotics-worldwide] PhD in dense visual mapping and optical flow at Universidad de Zaragoza

josemari josemari at unizar.es
Sat Aug 24 02:37:24 PDT 2013

The Perception Robotics and Real Time group at the I3A Institute at 
Universidad de Zaragoza in Spain, is looking for a highly qualified PhD 
candidate for the project "Semantic Visual Mapping for Rigid and 
Non-Rigid Scenes" in the area of Visual Mapping.  4 years position 
16440€/year before taxes.

The specific goal for the PhD is dense mapping and optical flow: 
developing algorithms to estimate —just from the images— the dense depth 
map and optical flow maps of the scene. It is a research goal to achieve 
real time
performance by using GPGPU processing. Find a project summary below.

- Master degree in engineering, computer science, physics, or 
- European work permit (compulsory to sign the contract)
- Motivation to work in an interdisciplinary project and international 
- Strong background in at least one of the following areas:
--- Computer Vision
--- Robotics perception
- Excellent skills in software development, Matlab, C++, GPGPU
- Starting Nov13-Feb14
- Fluent English.

Please send your application to: svmap.phd.position at gmail.com

Deadline for the application: 5 September 2013

Application should include:
- Statement of motivation and relevant experience
- Curriculum vitae
- Copies of university certificates
- List of attended classes and grades
- Names and addresses of two references

Project summary

Semantic Visual Mapping for Rigid and Non-Rigid Scenes.
Main Researcher J.M. Martínez Montiel

Recent research is showing that computing medium-scale visual maps for 
rigid scenes in real time is
feasible. Current maps, composed of a sparse set of points, are adequate 
for accurate camera
location, but quite poor for performing high-level tasks such as robot 
navigation, object manipulation
or human-computer interaction. Furthermore, most available techniques 
are intensive in computer
requirements, being beyond the reach of small mobile devices.
Our fundamental goal is to conduct cutting-edge research to push the 
limits of current visual mapping
techniques. More specifically, we propose to:
1. Develop more efficient and robust techniques, able to build 
semi-dense maps (in the order of
thousands of points) for medium-scale environments. We will build a 
working demonstrator of a
distributed keyframe-based mapping and relocation system, where a light 
tracking process runs
on a small mobile computer, while the more expensive map optimization 
and storage can be
allocated on a separate machine, or as a service on the cloud.
2. Develop scene understanding methods that boost the semantic contents 
of the map: object
recognition and insertion in the map, scene layout estimation, and scene 
recognition. We also
plan to address long-term mapping and apply learning techniques to 
determine which pieces of
information are obsolete or ephemeral and which are trustworthy for 
place recognition and
camera relocation.
3. Research new computer vision methods able to deal with sequences of 
non-rigid scenes. Our
specific approach is to embed Finite Element Methods (FEM) within the 
mapping algorithms in
order to exploit the rich models available from solid mechanics. We will 
target semi-dense
feature maps but also dense surface maps by exploiting variational 
methods and General-
Purpose computing on Graphics Processing Units (GPGPU).
4. Perform extensive validation of all developed algorithms with real 
imagery in indoor and outdoor
environments. In the case of non-rigid methods, medical imagery coming 
from real live tissue is
targeted. A close interaction with surgeons and researchers in elastic 
models for biological
tissues is planned to achieve results that are significant in real 
surgical applications.

This is an ambitious proposal rooted on the strong published 
contributions of the research team to
the current state of the art in visual mapping, and the solid experience 
in building real-time mapping
demonstrators, which have fostered an intense technology transfer.

Research team:
Unizar: J.M.M.Montiel, J.D. Tardós, J. Civera, M. Hernandez.
Hospital Clínico Lozano Blesa: I. Gil (Surgeon)
Research collaboration with A.J.Davison (Imperial College 
London),Martial Hebert (The Robotics Institute. Carnegie Mellon 

More information about the robotics-worldwide mailing list