About the GPU Research Center at JOANNEUM RESEARCH - DIGITAL
The DIGITAL - Institute for Information and Communication Technologies (one of four institutes within the non-profit research organization JOANNEUM RESEARCH) is a leading international research partner and center of expertise in the area of information and communication technology. The institute's technological and scientific focus includes web and internet technologies, image, video and acoustic signal processing along with remote sensing, communication and navigation technologies. Several research groups within DIGITAL have successfully explored the potential of GPU computing for large-scale computer vision problems.

The main research field of the Audiovisual Media research group (AVM) within DIGITAL is the content-based analysis of multimedia data with computer vision methods. Specific application fields include real-time video quality analysis, digital film restoration (dust & dirt, flicker, noise etc.) and the real-time analysis of video for supporting broadcast production (e.g., tracking football players in ultra-HD omnidirectional video) and surveillance (e.g., detection of wrong-way drivers). In most of these application areas, the analysis has to be done in real-time. Recognizing the performance advantages of GPUs for massively parallelizable algorithms, within the group we are actively working on the GPU implementation of key computer vision algorithms using CUDA. Several important algorithms for e.g. image registration, image warping, image inpainting, feature point detection and tracking and video breakup detection have already been implemented successfully on the GPU and have been integrated into prototype applications. Typically, the speedup factor of the GPU implementation is between 3 – 10 when compared against a highly optimized multi-threaded CPU routine.

The Machine Vision Applications research group (MVA) within DIGITAL conducts computer vision related research in three areas: Human-centered image analysis, industrial inspection and mobility applications. Real world applications in all three of these areas require the processing of large amounts of data in a short amount of time. MVA uses and develops GPU implementations of computer vision and image processing algorithms to make such applications possible. Examples are the detection of pre-crack deformations in timber tensions tests, real-time integration of 2.5D data into dynamically sized occupancy grids for SLAM and 3D reconstruction and highly accurate 2D visual odometry. Typical speeds up factors are 5 to 10, reaching up to 50 for selected methods.

About the co-PIs
Hannes Fassold received a MSc in Technical Mathematics (branch: information science) from Graz University of Technology in 2004. Since then he works at JOANNEUM RESEARCH, where he is currently a senior researcher at the Audiovisual Media research group of the DIGITAL institute. His main research fields are the research and development of algorithms for digital film restoration and content-based video quality analysis and the efficient parallelization of these algorithms on the GPU. He has published several publications in these fields and coordinates the GPU-related activities for the research group.

Jakub Rosner is a PhD student at Silesian University of Technology at the Institute of Informatics where he is involved in a EU-funded project in the field of data mining. He received his MSc in Robotics in 2009 with the master thesis "Image processing algorithms in CUDA". Since 2007 he has been porting and optimizing major computer vision algorithms (feature point detection and tracking, image warping and inpainting, video quality analysis etc.) for GPUs in CUDA in collaboration with the DIGITAL institute. His research area is focused on thorough optimization of computer vision algorithms by exploiting the computing power of both GPU and CPU in heterogeneous programming. Results of his research in this field have been presented at several international conferences and workshops.

Hermann Fürntratt studied Telematics at the Graz University of Technology where he received his MSc in 1997. During study, his special focus was on medical image processing. Since then he worked for more than a year in the UK for a company focused on digital colour correction and is now a senior researcher at the Audiovisual Media research group of the DIGITAL insitute. Since high performance and real-time processing is neccessary for most of the algorithms he introduced CUDA at the Audiovisual Media group and implemented a real-time GPU-accelerated template-tracking library based on block-matching. His further research activities comprise porting all sorts of algorithms to GPU/CUDA.