You are here: ACVT Home > Communication

News

Google Engaged

The ACVT has secured a grant from Google to investigate feature descriptors which might be used to identify objects of interest in video. This project forms part of a growing ACVT engagement with Google.

VideoTrace Impact Continues to Grow

The VideoTrace technology was presented at Siggraph (the leading computer graphics conference world-wide) in 2007, and subsequently to Google in Colorado, Sony in London and 2D3 in Oxford.  Negotiations continue with Google, Autodesk, and Sony amongst a host of others as to the commercialisation of the technology. The web site describing the technology has had over 250,000 hits since August, and has received world-wide media coverage. The following patent applications have been lodged to protect the technology

• Method and System for Generating a 3D Model, Anton van den Hengel, Anthony Dick, Thorsten Thormählen, Ben Ward, and Philip H. S. Torr, Australian Provisional Patent Application No. 2007202157 and US Provisional Patent Application No. 60/917361. Filing Date 11th May, 2007.
• Method & System for Generating a 3D Model from Images, Anton van den Hengel, Anthony Dick, Thorsten Thormählen, Ben Ward, and Philip H. S. Torr, US Provisional Patent Application No. 20070629 32341, Filing date 18 June, 2007

Bachelors Degree in Computer Graphics

A new Bachelors Degree in Computer Graphics has been created at the University of Adelaide, with the first intake of students in 2008.  The degree structure has been planned in consultation with the Industry Partners, and particularly Rising Sun Pictures. The degree initially focuses on building a general computer science background, which is essential for a technical career in software development. This is coupled with an understanding of the 3D modelling and animation technology that underlies computer graphics. Later stages of the degree specialise in the technical aspects of writing 3D graphics applications, and with software engineering skills that are required to successfully complete team based projects. Advanced courses in 3D modelling and animation are also included, giving graduates a complete grasp of all aspects of computer graphics development. The degree emphasises practical skills as well as an understanding of the issues underlying the creation of graphics software. Students will be required to complete a half-year, team based project, and a major graphics software project in their final year. The degree has received the most student first preferences of any of the new degrees within the University this year. For more information see http://www.cs.adelaide.edu.au/programs/graphics/ .  A postgraduate degree in Computer Graphics is currently being developed.

Collaboration with CMU and TAFE

A new program of collaboration between students from the University of Adelaide School of Computer Science and the Digital Media Design program of TAFE has been initiated.  The students are collaborating in designing and producing computer games.  A group of students from the new Bachelor of Computer Graphics degree have been invited to take part in Carnegie Mellon University’s GamesLab series of workshops which aims to impart skills in games production with a particular focus on the commercial aspects of successful game development.

Innovation Project Delivers

The first Innovation Project with DSTO has been completed and delivered.   The degree of interaction with DSTO has far exceeded expectations and the results have been extremely successful.  A new programme of interaction is currently being planned, with mechanisms for strengthening the relationship being examined.

Previous News

Premier's Science and Research Fund result

The ACVT, Tenix Defence, DSTO, and Rising Sun Pictures have been awarded a Premier's Scence and Research Fund grant to develop a Visual Technologies Laboratory. The PSRF will provide $750,000 over 3 years towards the project, with partner contributions brining the total grant amount to over $1.5 million.

This project will establish a state-of-the-art Visual Technologies Laboratory for research into the production and analysis of visual digital media based in South Australia. The Laboratory will provide a mechanism for generating both the world-class technologies and the skilled workforce required by the next generation of digital content-based industries. The Visual Technologies Laboratory will provide the infrastructure, focus and leadership in research required to establish South Australia as the focal point of Australian visual technologies industries. The capabilities to be provided will be invaluable to the Defence, Film and Television Production, Computer Games, Video Surveillance, Mobile Content and ICT industries amongst a host of others. These industries are unified by their dependence on visual digital media, and the fact that constant innovation is vital to their ongoing growth. These industries also represent an important opportunity for South Australia to leverage existing capabilities in order to take part in the rapidly expanding new economies surrounding digital content. This dynamic sector of the economy is largely immune to issues of geographic separation and thus offers a significant opportunity for growth in exports for South Australia.

International Workshop on Parameter Estimation for Computer Vision Problems

A series of talks were presented including the following

• Object Recognition Beyond the Visible Spectrum

  Dr. Antonio Robles-Kelly, National ICT Australia, Canberra
  With the advent and development of new sensor technology, it is now possible to capture image data in tens or hundreds of bands covering a broad spectral range. Spectral imaging provides an information-rich representation of the object under study. This representation is often high-dimensional in nature. As a result, many classical algorithms in pattern recognition and machine learning have been naturally borrowed and adapted so as to perform object recognition, feature extraction and classification. In this talk, I will commence by exploring the use of spectral absorption features for purposes of identification and recognition and show how discriminant learning can be used to perform a localised analysis of the spectra for purposes of identification and recognition. This treatment opens up the possibility of training statistical classifiers on a small sample set with a marginal detriment on its generalisation properties. If time permits, I will go on to elaborate on the use of linear discriminant analysis (LDA) methods for multiclass classification problems and will present a novel pairwise discriminant analysis algorithm for learning class categories. This approach is based upon a novel cost function which can be minimised making use of unconstrained optimisation techniques over Grassmann manifolds and solved using a projected gradient method. This approach does not require matrix inversion operations and, therefore, does not suffer of stability problems for small training sets.

• Speaker: Dr. Chunhua Shen National ICT Australia, Canberra

  Title: Trace ratio problems revisited
  We proposed a new formulation for directly solving the trace ratio problem. The original non-convex problem is relaxed and it can be solved by efficiently solving a sequence of semidefinite feasibility problems. The solution is therefore globally optimal. Besides global optimality, our algorithm naturally generates orthonormal projection matrix. Moreover it relaxes the restriction of linear discriminant analysis that the projection matrix’s rank can only be at most c-1. c is the number of classes. Our approach is more flexible.

• Statistical Optimization for Geometric Fitting: Theoretical Accuracy Bound and High Order Error Analysis

  Kenichi Kanatani Department of Computer Science, Okayama University, Japan
  A rigorous accuracy analysis is given to various techniques for estimating parameters of geometric models from noisy data for computer vision applications. First, it is pointed out that parameter estimation for vision applications is very different in nature from traditional statistical analysis and hence a different mathematical framework is necessary in such a domain. After general theories on estimation and accuracy are given, typical existing techniques are selected, and their accuracy is evaluated up to higher order terms. This leads to a ``hyperaccurate'' method that outperforms existing methods.

• Fast Projective Reconstruction: Toward Ultimate Efficiency

  Kenichi Kanatani Department of Computer Science, Okayama University, Japan
  We accelerate the time-consuming iterations for projective reconstruction, a key component of self-calibration for 3-D reconstruction from a video sequence. We first introduce the power method for eigenvalue computation. We then accelerate the power method itself. We also introduce SOR for further acceleration. Using simulated and real video images, we demonstrate that our techniques dramatically speed up the computation, in some cases about 8,000 times faster.

Seminar Series

Previous Seminars

• Cleve Moler - The first seminar in the series was by Cleve Moler. Cleve is original author of MATLAB and co-founder of The MathWorks. The seminar gave his personal insight on the "Evolution of MATLAB". Used by over 1,000,000 of the world's leading technical people, on all seven continents, MATLAB has become a fundamental tool for accelerating the pace of engineering and science.
• David Suter - Professor Suter spoke on Computer Vision - Finding Structure in Complex Data.
• Andrew P. Paplinski - Integration of visual and auditory stimuli using Multimodal Self-Organizing Networks (MuSONs)