GPU for Surveillance
Our goal is to develop very fast image and video processing algorithms by taking advantage of Graphics processing units (GPU). We have already implemented MERL's state-of-science Bayesian background generation and foreground detection method. In comparison to the CPU version of the same algorithm, the speed of the GPU implementation is more than 20 times faster.
Background & Objective: GPU's have made their way to home computers through video games and multimedia long time ago. With the increasing programmability, GPU's are capable of performing more than the specific graphics computations for which they were designed. GPU computing with CUDATM* is a new approach to computing where hundreds of on-chip processor cores simultaneously communicate and cooperate to solve complex computing problems, especially for the time consuming video and 3D data analysis tasks.
Technical Discussion: One of the most important innovations offered by CUDA technology is the ability for threads on GPUs to cooperate when solving a problem. By enabling threads to communicate, CUDA technology allows applications to operate more efficiently. GPUs featuring CUDA technology have an on-chip Parallel Data Cache that developers can use to store frequently used information directly on the GPU. Storing information on the GPU allows computing threads to instantly share information rather than wait for data from much slower, off-chip DRAMs. This advance in technology enables users to find the answers to complex computational problems much more quickly than using traditional architectures.
* (CUDATM is a trademark of NVIDIA.)
Contacts:
Fatih Porikli
Jay Thornton
Jeroen van Baar
Technology Area: Imaging
Modification Date: September 17, 2007
