Friday, September 24th 2010

NVIDIA Adds GPU Acceleration for OpenCV Application Development

NVIDIA today announced CUDA support for OpenCV, the popular Computer Vision library used in developing advanced applications for the robotics, automotive, medical, consumer, security, manufacturing, and research fields.

With the addition of GPU acceleration to OpenCV, developers can run more accurate and sophisticated OpenCV algorithms in real-time on higher-resolution images while consuming less power. This will facilitate the development of scores of new, mainstream Computer Vision applications.

With thousands of developers and well over two million downloads to date, OpenCV is a popular Computer Vision library for the development of computational-intensive and powerful applications, many of which require robust real-time performance. For example, the new OpenCV depth calculation engine performs 5-10 times faster with GPU acceleration than with the equivalent CPU-only implementation.

"Computational power in Computer Vision has been a limiting factor not only for the use of recent powerful algorithms in object recognition, tracking and 3D reconstruction, but also has limited the creativity of algorithms people are willing to invent," said Gary Bradski, senior researcher at Willow Garage, and founder of OpenCV. "With CUDA GPU acceleration, many OpenCV algorithms will run five to ten times faster, making current algorithms more practical for application developers and allowing the invention and combination of more capable applications in the future."

"NVIDIA GPU acceleration of OpenCV now supplies the computational power for the sophisticated algorithms needed for advanced automotive driver assistance applications, and other popular consumer applications," said Taner Ozcelik, general manager of NVIDIA's automotive business. "OpenCV gives developers the toolbox they need to quickly unleash this power for research and development of these products without needing to recreate vision algorithms from scratch. This is a key milestone that could usher in a significant increase in the use of Computer Vision across a broad range of industries."

An initial release of OpenCV with CUDA GPU support for many common and powerful functions is expected to be available in the Spring 2011. Some initial GPU-enabled OpenCV functionality is currently available via the OpenCV source code repository.

"My research lab uses OpenCV extensively in our autonomous vehicles," said Sebastian Thrun, professor of computer science and electrical engineering at Stanford University. "CUDA GPU acceleration for OpenCV provides my research team an instant performance bump which is critical in our research. OpenCV and CUDA will dramatically increase what is possible with computer vision in our autonomous vehicles."

OpenCV Sessions at GTC (Sept. 20-23)
With more than 280 hours of GPU-focused sessions and more than 25 sessions on CUDA C/C++ development, NVIDIA's GPU Technology Conference (GTC) provides a wealth of information on GPU computing news, developments and achievements. GTC also includes presentations discussing OpenCV on GPUs, including the keynote address on Sept. 23 by Sebastian Thrun, a robotics pioneer at Stanford University, who leveraged OpenCV in the development of his DARPA award-winning robotics technology.

For more information on GTC and relevant OpenCV sessions, please visit: this page.
Add your own comment

3 Comments on NVIDIA Adds GPU Acceleration for OpenCV Application Development

#1
toyo
AMD, guys, wake up!!! Do something! It's damn obvious that the future of the GPU is as a coprocessor for parallel tasks... get to work and show that you understand it.

CUDA in Adobe, CUDA everywhere, CUDA supercomputer, CUDA, CUDA, CUDA... Where are the AMD news? The big companies supporting OpenCL, BrookGPU (is it still alive?)

Applications and games that use BrookGPU

This pretty much sums it up... I think this HD 4850 is to be my las Radeon, if things go this way... it's stupid that the most expensive piece of hardware (in better systems) is good only for games. It needs to do much more.
Posted on Reply
#2
FreedomEclipse
~Technological Technocrat~
good news for Nvidia i suppose - their CUDA is getting more PR - and their worming their way into markets that would normally buy a 1 off order of 50+ machines from IBM/Dell/HP or CompaQ, but now they can have more specialised units
Posted on Reply
#3
Cheeseball
by: toyo
AMD, guys, wake up!!! Do something! It's damn obvious that the future of the GPU is as a coprocessor for parallel tasks... get to work and show that you understand it.
Alas, because of their lackluster support for Stream and the robust libraries that CUDA/OpenCL has to offer, NVIDIA still gets my cash when it comes to hardware-dependent software development.

But my 2 HD 5770s are still kickin' it in gaming, with a GTX 460 for PhysX and CUDA (seriously 15x more speed when compiling compared to my old 8600GT) work.
Posted on Reply