high performance computing on graphics processing units: hgpu.org

The intention of this thesis paper is to deploy a parallel implementation of the optical flow detection algorithm known as the Lucas-Kanade algorithm. As an important algorithm in the field of computer vision, it is believed that it holds much promise and shows much potential for benefiting from techniques used to enhance performance through parallel programming which can be executed with the use of CUDA. Though more techniques of parallel programming exist that can be used to fasten the process, Lucas-Kanade has never been implemented in parallel programming before. The result of the research has shown both serial and parallel implementation of optical flow detection using different processing units (CPUs and GPUs). The parallel implementation have lessened 2 to 13 seconds of processing time (depending on the hardware configuration) for the same database compare to serial implementation.