Hardware-Oriented Adaptive Multi-resolution Motion
Estimation Algorithm and Its VLSI Architecture
Guoqing Xiang, Huizhu Jia*, Jie Liu, Yuan Li, Xiaodong Xie
EECS of Peking University
Beijing, 100871
P.R. China
Email: {gqxiang, hzjia, liuzimin, yuanli, xdxie}@jdl.ac.cn
Abstract
—In this paper, we propose a hardware architecture
of an adaptive multi-resolution motion estimation algorithm
(AMMEA) for high definition video encoder to reduce hardware
cost. The texture-based search strategies are based on temporal
stationarity and spatial homogeneity with Sobel edge operator.
The proposed algorithm makes motion estimation more concise.
We also propose Sobel edge operator hardware architecture. The
four-pixel SAD unit which is the basic processing element (PE) in
our proposed architecture is used for SAD calculation and Sobel
edge operator computation. The hardware architecture achieves
very high data utilization and data throughout. Using our
proposed AMMEA with regular data flow, simulation results
show that the proposed architecture can significantly reduce the
hardware cost with a negligible PSNR loss of 0.03dB compared
with the full-search. The design is implemented with SMIC
0.18µm CMOS technology and costs 950K gates count, and it
supports the real-time encoding of 1080P@30fps with two
reference frames under a clock frequency of 150MHz.
Keywords—Multi-resolution motion estimation; adaptive
search strategies; VLSI; data re-use; Sobel
I.
I
NTRODUCTION
Motion estimation (ME) is the most complex part of most
popular video compression standards such as MPEG-1/2/4 and
H.264/AVC [1]. The goal of integer motion estimation is to
reduce temporal redundancies between the current frame and
the reference frame. These video coding standards also use new
techniques such as variable block size motion estimation
(VBSME) and multiple reference frames. Therefore, real-time
motion estimation implementation for high definition (HD)
video encoder brings great challenges for hardware resources
and power consumption.
There are many fast motion estimation algorithms
proposed for video coding, such as SEA [2] and DSA
[3].Although these software-oriented algorithms achieve time
saving, the irregular data flow makes these algorithms
unsuitable for hardware implementation. A commonly used
hardware-friendly ME algorithm is full-search block matching
algorithm (FSBMA) [4], which examines all points in search
window. Due to the large search window size requirement for
*The corresponding author, Huizhu Jia is with Peking University,
also with Cooperative Medianet Innovation Center and Beida
(Binhai) Information Research.
HD video encoder, on-chip memory consumption and
computational resource cost are huge. Multi-resolution ME
algorithm (MMEA) is a good choice for VLSI implementation
to achieve good balance between performance and complexity
in HD encoder [5][6], which is developed with a coarse-to-fine
search hierarchy. However, as analyzed in [7], traditional
MMEA only used fixed search range and the same down-
sampling rate for all sequences without discrimination is a big
problem. For some sequences with complex texture, wrong
motion vector from the coarse level will mislead the search in
the fine level which leads to performance degradation. And, for
some stationary regions such as background, MMEA also
searches in unnecessarily large search window and thus wastes
much computational resources.
In this paper, a hardware oriented adaptive multi-
resolution motion estimation algorithm (AMMEA) and its
VLSI architecture are proposed. AMMEA makes search
strategies customized for each block based on stationary and
homogeneous features of current macroblock (MB). The four-
pixel SAD unit which is the basic PE is applied to search
strategy determination and multi-resolution motion estimation.
The proposed architecture can make full use of PEs between
SAD calculation and Sobel edge operator computation. It will
achieve a better balance between the hardware resource and
performance.
The remainder of this paper is arranged as follows.
Section II gives a brief introduction of hardware oriented
AMMEA. The overall VLSI architecture is developed and a
search strategy decision block is proposed based on
reconfigurable PE array in Section III. The simulation results
and the comparisons with other previous works are given in
Section IV. Finally, conclusions are drawn in Section V.
II. A
DAPTIVE
M
ULTI
-R
ESOLUTION
M
OTION
E
STIMATION
A
LGORITHM
In this section, we will introduce the adaptive multi-
resolution motion estimation algorithm [7]. In traditional three-
level MMEA [5][6], it is developed with a coarse-to-fine
hierarchical search. In the coarsest level, potential match
candidates in the reference are obtained from the largest search
window. In the modestly coarse level, several search windows
are centered at the candidates provided from the immediate
upper coarse level. In the finest level, the search range (SR)
will be set to be very small for calculation reduction and is