Main Menu
About us
Project Description
Quantitative Results
Research Lines
Research Results
Impact on Society
Press room
Contact us
Secure Login
Events Calendar
« < October 2017 > »
25 26 27 28 29 30 1
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29
30 31 1 2 3 4 5

Task 3 (UM): Artificial perception algorithms for advanced architectures

Leader: Pedro López de Teruel; Researchers: Antonio Leonardo Rodríguez, Antonio Ortuño, Ginés García Mateos, Alberto Ruiz.

subTask 3.1 Development of a fast prototyping tool for efficient computer vision applications

subTask 3.2 Real time simultaneous camera movement estimation and three dimensional reconstruction of scenes

subTask 3.3 Face, place and object localization, tracking and recognition

subTask 3.4 Application of the developed techniques to real world problems

1. Brief Description of the Goals

New multicore architectures need new parallel programming paradigms and environments. This is especially true in the computer vision field, where massive potential parallelism inherent to image processing must be adequately adapted to be exploited in these new architectures. Thus, development of libraries, algorithms, environments and other supporting tools that aid in efficient computer vision software development is currently a very active area of research. In this project we have developed the multithreaded QVision framework for fast prototyping of multithreaded computer vision applications.

The computing power of these new architectures makes them ideal platforms to try real-time solutions for classical computer vision problems that, due to its intensive computational load, were traditionally solved only in an off-line manner. Thus, on top of our QVision library, during the project period we worked in the development of an automated vision system able to perform an online 3D reconstruction of the environment. In order to do that, one must develop and test feature extraction, description and tracking algorithms, all of them needed to deal with the classical structure from motion problem in real-time. 

Other applications of these feature extraction and description techniques include the implementation of visually guided GUIs. In particular, our subgroup also worked on the design and implementation of real-time face detection and tracking in real time. Finally, and with all these tools at hand, a number of computer vision applications in other fields such as medicine and biology were also possible.

2. Scientific and Technical Developed Activities

In the first years of the project we developed a real-time 3D tracking algorithm for human faces, based on integral projections, and useful for the implementation of visual GUIs. The main advances in this line of work were published in workshops of the well known international conferences ICCV by García-Mateos et al. in 2007, as well as in CVPR 2008 and the International Journal Pattern Recognition and Image Analysis in 2008. All these works are among the main research results of the PhD thesis by Dr. Garcia Mateos.

During the project we have also advanced in the development of Qvision (http://qvision.sourceforge.net), a GNU tool for fast prototyping of multithreaded real-time computer vision applications. An overall description of the library was published in the international conference IPCV by Rodriguez et al. in 2008. Also related to this work, we have developed a parallel programming design pattern for multithreaded applications, which lies at the core of the aforementioned library, and which has been published by Rodriguez et alt. in the Europar conference in 2008.

In the field of multiple view geometry and structure from motion (SfM) we have also achieved important results regarding the use of tensorial and diagrammatic techniques, as well as proposing new efficient cost functions for large three dimensional reconstructions, which have been tested successfully in both real time and off-line applications. The main results of this line of work have been published by Ruiz et al. and Rodriguez et al. in the two main world conferences on computer vision, namely, ICCV in 2009, and CVPR in 2011, respectively. Another important result regarding real time structureless motion estimation was also published in a specialized workshop in the ICCV conference of year 2011.

Finally, a number of techniques developed by the group in the during the project period have found applications in several fields of biology and medicine. The most relevant contributions in this area have been published by García-Mateos et al. in Aquaculture, Vol. 307, 2009, the Bioinformatics conference in 2010, and the ICPR conference in 2010.

Publications: [Rodr08a] [Rodr08b] [Garc07a] [Garc08a] [Garc08b]

Projects funded by Public Calls: 

External collaborations Academia: --

External collaborations Industry: --

Company Agreements: --

PhD dissertations: Ginés García Mateos

Patents: --