You are here

RECONFIGURABLE ARCHITECTURE FOR H.264/AVC VARIABLE BLOCK SIZE MOTION ESTIMATION BASED ON MOTION ACTIVITY AND ADAPTIVE SEARCH RANGE

Download pdf | Full Screen View

Date Issued:
2010
Abstract/Description:
Motion Estimation (ME) technique plays a key role in the video coding systems to achieve high compression ratios by removing temporal redundancies among video frames. Especially in the newest H.264/AVC video coding standard, ME engine demands large amount of computational capabilities due to its support for wide range of different block sizes for a given macroblock in order to increase accuracy in finding best matching block in the previous frames. We propose scalable architecture for H.264/AVC Variable Block Size (VBS) Motion Estimation with adaptive computing capability to support various search ranges, input video resolutions, and frame rates. Hardware architecture of the proposed ME consists of scalable Sum of Absolute Difference (SAD) arrays which can perform Full Search Block Matching Algorithm (FSBMA) for smaller 4x4 blocks. It is also shown that by predicting motion activity and adaptively adjusting the Search Range (SR) on the reconfigurable hardware platform, the computational cost of ME required for inter-frame encoding in H.264/AVC video coding standard can be reduced significantly. Dynamic Partial Reconfiguration is a unique feature of Field Programmable Gate Arrays (FPGAs) that makes best use of hardware resources and power by allowing adaptive algorithm to be implemented during run-time. We exploit this feature of FPGA to implement the proposed reconfigurable architecture of ME and maximize the architectural benefits through prediction of motion activities in the video sequences ,adaptation of SR during run-time, and fractional ME refinement. The implemented ME architecture can support real time applications at a maximum frequency of 90MHz with multiple reconfigurable regions. When compared to reconfiguration of complete design, partial reconfiguration process results in smaller bitstream size which allows FPGA to implement different configurations at higher speed. The proposed architecture has modular structure, regular data flow, and efficient memory organization with lower memory accesses. By increasing the number of active partial reconfigurable modules from one to four, there is a 4 fold increase in data re-use. Also, by introducing adaptive SR reduction algorithm at frame level, the computational load of ME is reduced significantly with only small degradation in PSNR (0.1dB).
Title: RECONFIGURABLE ARCHITECTURE FOR H.264/AVC VARIABLE BLOCK SIZE MOTION ESTIMATION BASED ON MOTION ACTIVITY AND ADAPTIVE SEARCH RANGE.
0 views
0 downloads
Name(s): Kodipyaka, Sumedha, Author
Lee, Jooheung, Committee Chair
University of Central Florida, Degree Grantor
Type of Resource: text
Date Issued: 2010
Publisher: University of Central Florida
Language(s): English
Abstract/Description: Motion Estimation (ME) technique plays a key role in the video coding systems to achieve high compression ratios by removing temporal redundancies among video frames. Especially in the newest H.264/AVC video coding standard, ME engine demands large amount of computational capabilities due to its support for wide range of different block sizes for a given macroblock in order to increase accuracy in finding best matching block in the previous frames. We propose scalable architecture for H.264/AVC Variable Block Size (VBS) Motion Estimation with adaptive computing capability to support various search ranges, input video resolutions, and frame rates. Hardware architecture of the proposed ME consists of scalable Sum of Absolute Difference (SAD) arrays which can perform Full Search Block Matching Algorithm (FSBMA) for smaller 4x4 blocks. It is also shown that by predicting motion activity and adaptively adjusting the Search Range (SR) on the reconfigurable hardware platform, the computational cost of ME required for inter-frame encoding in H.264/AVC video coding standard can be reduced significantly. Dynamic Partial Reconfiguration is a unique feature of Field Programmable Gate Arrays (FPGAs) that makes best use of hardware resources and power by allowing adaptive algorithm to be implemented during run-time. We exploit this feature of FPGA to implement the proposed reconfigurable architecture of ME and maximize the architectural benefits through prediction of motion activities in the video sequences ,adaptation of SR during run-time, and fractional ME refinement. The implemented ME architecture can support real time applications at a maximum frequency of 90MHz with multiple reconfigurable regions. When compared to reconfiguration of complete design, partial reconfiguration process results in smaller bitstream size which allows FPGA to implement different configurations at higher speed. The proposed architecture has modular structure, regular data flow, and efficient memory organization with lower memory accesses. By increasing the number of active partial reconfigurable modules from one to four, there is a 4 fold increase in data re-use. Also, by introducing adaptive SR reduction algorithm at frame level, the computational load of ME is reduced significantly with only small degradation in PSNR (0.1dB).
Identifier: CFE0003316 (IID), ucf:48488 (fedora)
Note(s): 2010-08-01
M.S.
Engineering and Computer Science, School of Electrical Engineering and Computer Science
Masters
This record was generated from author submitted information.
Subject(s): Partial Reconfiguration
FPGA
Digital logic design
Persistent Link to This Record: http://purl.flvc.org/ucf/fd/CFE0003316
Restrictions on Access: private 2015-07-01
Host Institution: UCF

In Collections