Personal Information...

_____

Office Address(es)

*Please note that ing. defines a local Belgian title for special types of engineers and has no proper global acknowledged equivalent title.

Nationality: Belgian
Date and Place of Birth: March 23th, 1983 at Lier, Belgium
Gender: Male

Education and Training...

2008--20xx	Hasselt University -- Campus Diepenbeek transnational University of Limburg	PhD: Philosophiæ Doctor in Computer Sciences, in progress
2006--2007	Hogeschool voor Wetenschap en Kunst Campus De Nayer	MSc: Master of Science (Engineer ICT), magna cum laude
2005--2006	Hogeschool voor Wetenschap en Kunst Campus De Nayer	BSc: Bachelor of Science (Electronics -- ICT), cum laude
2003--2005	Hogeschool voor Wetenschap en Kunst Campus De Nayer	2nd Bachelor Year (Electronics -- ICT), cum laude
2002--2003	Hogeschool Gent Campus Schoonmeersen	1st Bachelor Year (Computer Sciences), passed

Acquired Driving Licences...

European Type A3 and B.

Work Experience...

Jul.2007--Dec.2007	IMEC	Multimedia Application Engineer and joint PhD preparation.
Oct.2006--Dec.2006	De Nayer	Professor assistant in the computer network lab.
Sep.2006	Labo De Nayer	Disassembling and re-assembling Faraday cages for Electro-Magnetic Compatibility (EMC) measurements.
Jul.2006--Jun.2007	IMEC	Internship and Master thesis on General-Purpose GPU computing.
Jul.2004--Aug.2004	Ice Concept	Xbox game room development.
Jul.2002--Dec.2002	EPYC	Webdesigner of e-learning courses in Macromedia Studio.
Jul.1999--Aug.1999	Fashion Com	Assembling and selling PCs.
Jul.1998--Aug.1998	Computerboer	GUI developer for costum software applications.

PhD Topic...

Parallelization of Multi-Camera Video Algorithms for Intelligent Supercomputing on Hybrid Multiprocessor Environments

ABSTRACT: Nowadays a strong progression is noticeable in applications that revolve around realistic intuitive 3D, creating the need of reliable real-time depth acquisition and synthesis of real images, rather than predefined synthetic 3D models. There are several diametrically opposed methods or algorithms for acquiring depth, but there is a clear distinctable class based on triangulation. Since high-quality depth acquisition and 3D reconstruction use multicamera viewpoints more and more, they exhibit a progressive complexity and require loads of computations. It is thereby very challenging to implement this class of multi-view video algorithms -- which revolve around triangulation-based depth acquisition -- in real-time. Supercomputing can bring the solution, but it demands the use of massive parallelism. Unfortunately the algorithmic designer thinks ‘sequentially’, which makes finding tranformation methodologies for the parallelization a priority to use modern (fast) computer architectures.

Recently the research in this domain has become attractive because the Graphics Processing Unit (GPU) gets more and more acknowledged as the first commodity hardware that brings generic massive parallelism to the common consumer. The core goal of this PhD thereby focusses on finding a methodology that could lead to the autotransformations of this class of ‘sequentially’ designed multi-view video algorithms, to the generic hybrid mutiprocessor architecture of a GPU. Drawing up the methodology in this PhD. occurs with very pragmatic modular approach in the scope of a tripartite collaboration between the research entities of EDM, IMEC and NVIDIA. The greatest innovativity of this PhD lies in the fact that we form the bridge between algorithmic development and efficient implementation, as a kind of trade-off between algorithm, compiler and architecture.

The proposed modular parallelization can be divided in three levels, where each level refines the parallelization to a more optimal implementation, and a specific expertise of EDM (algorithmic/mapping), IMEC (mapping/compiler optimization), respectively NVIDIA (architectural) is being exploited. The PhD will thereby be paired with an intesive study in the head quarters of NVIDIA in the US. After the conversion of several closely related algorithms of the same class, the parallelization cen be iteratively formalized to what we call the ‘artificial intelligent rule set’. This rule set is shaped by an empirical methodology where the transformations are described from within the knowledge of the algorithm, or the ‘intelligence’ of the algorithmic designer. This term has no further relationship with the computer scientifical domain of artificial intelligence. Except for an optimal parallelization, this PhD also wants to take the next step by acquiring a maximal strength (and thereby the involved complexity) in the transformation processes. This demands a minimal complexity in the described algorithm and hardware, by which the transformations become scalable in the future to ‘more massive’ parallel computer systems.

As a result of this PhD, which will make an effective implementation on the generic NVIDIA Compute Unified Device Achitecture (CUDA), is expected that practical applications of multi-view video could be brought closer to the consumer. Applications like the ‘immersion’ of a person with 3D information like e.g. teleconference systems, or the Office of the Future, can be implemented in real-time on commodity hardware. There are further applications like the eye-gaze corrected video chat by image interpolation of multiple webcams, the post-production of digital 3D cinema, inserting synthetic models into a real image, freely choosing a fictitious camera viewpoint, and watching a collection of digital pictures in a 3D reconstructed environment like the technology conceived within Microsoft Photo Tourism and Google Earth/Maps. As an additional result, NVIDIA can upgrade their technology, especially the NVCC compiler, ‘quid pro quo’ by means of the optimal architectural and compiler definitions that follow out of this PhD.

[Download Dutch PhD Proposal (pdf, 905kB)]

Master Thesis Subject...

A Generic Framework for Implementing Real-Time Stereo Matching Algorithms on the Graphics Processing Unit

ABSTRACT: Free Viewpoint Video is a revolutionary video technology that enables the user to freely roam a custom viewpoint without the need of a physical camera to capture the requested image. The image is thereby synthesized by means of a number of fixed-point cameras. Analog to depth perception of the human eye, two neighboring cameras of the requested viewpoint are used as a stereo vision to determine the depth of the captured scene. The depth information, together with the static captured images, can be used to interpolate an intermediate camera viewpoint.

In this thesis we formally study stereo matching, which is the drive behind Free Viewpoint Video technology. Stereo matching algorithms make it possible to acquire depth from a stereo vision, and exist in general out of a tremendous amount of computations and a high complexity. It is therefore hard to use those algorithms in real-time. We present a generic framework that makes it possible to easily implement real-time stereo matching algorithms by exploiting the parallel computational horsepower within Graphics Processing Units (GPUs) on commodity graphics cards. This enables us to obtain very high-speed matching performances with up to 478 frames per second, nonetheless we do not lose a great deal of quality. Middlebury quality evaluation ranks our proposed stereo algorithm even over a number of algorithms that use global optimization. As the framework permits easy low-level expandability, it can also be used to perform state-of-the-art stereo corresponding research. Thereby we have enhanced the framework with a hybrid CPU/GPU application time benchmarking mechanism, to provide a formal way of evaluating implementations that concurrently run on both the central processor and the graphics card.

The generic framework that is codenamed SaJi, is implemented as a static library that uses the object oriented programming language C++ as a host language, and the High-Level Shading Language to program the underlying graphics hardware.

[Download Master Thesis (pdf, 6.13MB)]
[Download 10p Barco Summary (pdf, 5.06MB)]

Scientific Awards...

Barco/VIK Awards 2007

The Barco Awards are one of the most prestigious awards in the field of computer vision and vision technology in Belgium. Barco is a global technology company, which designs and develops visualization products for a variety of selected professional markets. Together with the VIK, i.e. the Flemish Chamber of Engineers, they award the three best Belgian dissertations in the field of computer vision and vision technology.

[See Event Pictures]
[Read Official Press Release (Dutch)]

Baudouin Elleboudt Awards 2007

Previously known as the NIT-ITP Awards, the Belgian Journalists on IT (B-JIT) reward the four most innovative and well explained theses with the Baudouin Elleboudt Awards.

[See Event Pictures]

Informatie Scriptieprijs 2007

Each year the professional ICT magazine Informatie selects the three best dissertations from engineering schools and universities from Belgium and the Netherlands. These three are then set to compete in an international public live event, where the authors present the novelties and work reflected in their thesis. The jury ultimately decides a final ranking from first to third place.

• 1st: Sammy Rogmans
• 2nd: Thijs van der Schroeff
• 3rd: Sofie De Cooman

[Read Official Press Release (Dutch)]
[Download Nominee and Event Invitation (Dutch, pdf, 100kB)]
[Download Article (Dutch, pdf, 671kB)]

Personal Skills and Competences...

Mother tongue:	Nederlands (Dutch)	Fluently in all aspects.
Other languages:	English Français (French) Deutsch (German)	Reading and listening fluently, speaking and writing very good. Reading and writing average, listening and speaking limited. Listening average, otherwise very limited.

Programming:	[Advanced] [Intermediate] [Beginner]	Cg/HLSL, DirectX, C++, C, Java, Strict XHTML CUDA, VHDL, PHP, Perl, SQL, CSS, JavaScript, B-Prolog, ... OpenGL, Qt, PL/SQL, Pascal, VB(A), ActionScript, XML/XSLT, ...
OS:	[Advanced] [Intermediate]	MS Windows Vista and Older, MSDOS, SuSE Linux, Gentoo Linux HP-UX System V UNIX, Fedora Core, CDE, ...
Administration:	[Intermediate]	General networking, Apache, TomCat (J2EE), Pure-FTPd, Oracle/PostgreSQL/MySQL, Postfix, OpenSSH, Courier, Samba, Vixie-cron, OpenLDAP, ...
Publishing:	[Advanced]	Adobe Photoshop, Macromedia Studio, MS Visio, vi(m), MS Office, LaTeX2e, OpenOffice.org

Important References...

IMEC:	Gauthier Lafruit, PhD Lu Jiangbo Klaas Tack, PhD Kristof Denolf, PhD	gauthier.lafruit@imec.be jiangbo@imec.be klaas.tack@imec.be kristof.denolf@imec.be
EDM:	Philippe Bekaert, PhD Frank Van Reeth, PhD	philippe.bekaert@uhasselt.be frank.vanreeth@uhasselt.be
De Nayer:	Herman Crauwels, PhD Jan Meel Toon Goedemé, PhD	herman.crauwels@denayer.wenk.be jan.meel@denayer.wenk.be toon.goedeme@denayer.wenk.be

Hobbies...

I have a numerous amount of hobbies. For the non-technical area, I like bowling, making long walks through nature and having a peaceful drink with friends at the Hasselaarse and Leuvense terraskes, i.e. the outside facilities of the local bars. As a model citizen of Hasselt, I'm also very fond of everything related to cooking ang food. For the more technical area, I tutor people experiencing difficulties in working with computers and multimedia hardware. Occasionaly I like playing survival horror and epic RPGs such as Resident Evil, Silent Hill and Devil May Cry, as well as watching a movie on a well built home cinema theater. Most of all, I like reverse-engineering, developing homebrew software and hardware, and `modding' game consoles and PCs.