IEEE Consumer Electronics Magazine - September/October 2022 - 28

Video Technology
Table 1. Encoders Used in This Paper, Classified by Type and Embedded System.
Rapsberry Pi 4
Hardware
H.264/AVC
HEVC
AV1
omxh264enc
-
-
Software
x264
-
rav1e
Hardware
nvv412h264enc
nvv412h265enc
-
Jetson Nano
Software
x264
x265
rav1e
AOMedia Video 1
In contrast to AVC and HEVC, AV1 is a royalty-free,
nonstandard, and open-source video
compression format mainly designed for streaming
services. It was initially released in March
2018, and the latest version dates from January
2019. That implies that it is the most recent
video compression format of those discussed in
this article. It has been developed by the Alliance
for Open Media (AOMedia) consortium,
which was founded in 2015 and brings together
companies that provide video on demand, such
as Netflix and Amazon, web browsers, such as
Mozilla, content producers, and other top-technology
companies such as Intel and Nvidia. AV1
is considered as the successor to the VP9 compression
format.13
At the moment, there are not many encoders
of this standard due to its recent development.
Two of the available encoders are Scalable Video
Technology for AV1 (SVT-AV1) and rav1e. For
our assessments, rav1e was chosen since SVTAV1
can only be run on systems with a fifth-generation
Intel processor onward. However, rav1e
supports the embedded systems under study.
As of the date of publication of this article,
the rav1e encoder is still under development, so
the version used (version 0.1.1) is not the final
one. It is cross-platform software and released
under the Berkeley Software Distribution (BSD
2) license. Finally, rav1e can run on both Raspberry
Pi 4 and Jetson Nano, so it has been chosen
for our study.
Table 1 shows all the encoders used in this
article, classified by type (hardware or software)
and the embedded system on which they run.
INTRAPREDICTION
Intraprediction is characterized by constructing
the prediction samples using information
28
present in the image itself, unlike interprediction,
which uses information from other images
in the sequence. In this way, the temporal dependence
on the rest of the images that make up the
sequence is eliminated, something that cannot
occur in interprediction. In other words, only
the spatial redundancy of the image in question
is specifically exploited. Below, the characteristics
of this type of prediction are analyzed for
the video compression standards listed in the
previous section.
In the H.264/AVC standard, the image is
divided into 1616 pixel MacroBlocks (MBs) to
carry out intraprediction. There are three possible
partitions of the MB in luminance, whose
sizes are: 1616, 88, and 44 pixels. Depending
on the size of the block, different prediction
modes have been defined7:
1616 (luma): a single 1616 prediction
block is generated using four possible predictionmodes:
vertical, horizontal, dc, and plane.
88 (luma) and 44 (luma): 9 prediction
modes have been defined, including 8 directional,
and 1 nondirectional mode (DC).
On the other hand, in the case of chrominance,
a prediction block is generated for each
chrominance component, using the same prediction
mode for both chrominance components,
Cb and Cr. There are four possible prediction
modes available for both chrominances: DC, horizontal,
vertical, and plane.
When applying these directional modes, the
high spatial correlation between a block and its
neighboring pixels belonging to the upper, upperleft,
upper-right, and left blocks is exploited. Once
the block size has been determined using a cost
function, the encoder chooses the optimal predictionmode
to use.
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - September/October 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2022
