A Configurable Motion Estimation Accelerator For Video Compression

Abstract:

The design of motion estimation accelerator (ME ACC), a dedicated-purpose processing element, for AsAP2 platform is presented. AsAP2 platform consists of a 2-dimensional array of processing elements with a small amount of data storage. By its inherent nature, motion estimation is one of the most computationally intensive tasks in video encoding, and it also requires a signficant amount of memory resources. The ME ACC provides 4 KB dual ported SRAM to support the memory resources needed. It accelerates the motion estimation process by computing up to 16 absolute differences in one clock cycle and minimizing memory access overhead. The highly configurable ME ACC supports programmable motion estimation parameters and search algorithms. The independently-clocked ME ACC is the fastest programmable architecture ever fabricated (that we know of) for motion estimation, that operates at a maximum frequency of 938 MHz and occupies 0.67 mm² in 65 nm CMOS technology. Multiple search algorithms are run on five benchmark video sequences using the ME ACC. It is shown that the ME ACC allows user to adapt search algorithm and other motion estimation parameters depending on the video characteristics in order to achieve the best trade off between the motion estimation quality and the computational complexity.

Paper

  PDF (1.6 MB),   (c) Copyright, 2009

Reference

Gouri Landge, "A Configurable Motion Estimation Accelerator For Video Compression" Technical Report ECE-VCL-2009-4, VLSI Computation Laboratory, ECE Department, University of California, Davis, 2009.

BibTeX entry

@mastersthesis{gslandge:msthesis,
   author      = {Gouri Landge},
   title       = {A Configurable Motion Estimation Accelerator
                  For Video Compression},
   school      = {University of California},
   year        = 2009,
   address     = {Davis, CA, USA},
   month       = dec,
   note        = {\url{http://www.ece.ucdavis.edu/vcl/pubs/theses/2009-4}}
   }