Motion Estimation Using A Myrinet-connected Cluster System

T.S. Gunnawan and W. Cai (Singapore)


cluster systems, Myrinet, data-parallel, HDTV, motion estimation.


Motion estimation (ME) is central to many applications where time-varying image data are involved. Motion in formation is obtained by finding the motion field between the reference frame and the current frame. ME applied to high definition television (HDTV) video sequences requires high amount of processing time due to its large image size. Therefore, parallel processing becomes a natural choice to meet the large computation requirement. In this paper, we present several parallel motion estimation algorithms and their performance study on a Myrinet connected cluster system. Data-parallel approach is used to distribute each frame across the processors. The performance results include comparisons of execution time and speedups on various HDTV sequences. Results indicate that the performance scales according to the avail able number of processors.

Important Links:

Go Back