IEEE Consumer Electronics Magazine - November/December 2021 - 79

Figure 1. Conceptual system architecture of BEVA.
video streams processed by IoVT service are fed
to information visualization, which provides video
records for authorized users. To prevent visual
layer attacks, ENF-security is implemented as an
edge service by estimating ENF signals as environmental
fingerprint from video streams. The ENF
signals associated with the video are saved into
blockchain as authenticator, and the video
streams generated by each camera can be verified
by computing the correlation coefficient between
the recordedENF signals.
The blockchained network fabric provides a
fundamental networking and security infrastructure
to ensure decentralized security enforcement
for the IoVT system. All authorized edge
devices can work as validators, which execute
an efficient consensus protocol to ensure consistence
and total order of data recorded on distributed
ledger. In addition, recording transactions
and ENF fingerprint data on an immutable, traceable,
and auditable distributed ledger ensures
integrity and provenance of data in use. Therefore,
all participants in the IoVT system can rely
on the blockchain as a trust and security authority
to self-authenticate video streams among
each other instead of any centralized third-party
intermediate. Furthermore, implementing predefined
state-response rules into self-executing SCs
brings interoperability and programmability for
highly autonomous IoVT application scenarios.
ENF-SECURITY AS AN EDGE
SERVICE
A) ENF as an Environmental Fingerprint: The
term environmental fingerprint is coined to
November/December 2021
describe the reliability and accessibility of a digital
fingerprint available in digital multimedia
recordings. ENF has a nominal frequency of 50 Hz
or 60 Hz depending on the location of the power
grid and is widely available in both audio-video
recordings. The instantaneous fluctuations in the
nominal frequency are caused by power supply
demand throughout the power grid interconnect.
The randomly generated ENF signals from multiple
recordings vary at the same time period; however,
they have similar fluctuations throughout
the power grid. For example, the ENF fluctuations
throughout the eastern interconnect is similar.2
Because of the timeand location sensitive, ENFfingerprint
is adopted in our BEVA system as an
authenticationmethod in joint with the consensus
mechanism.
In BEVA system, the primary source of ENF
comes from video recordings where the ENF signals
are in the form of illumination frequency as
120 Hz. The video recordings made under indoor
artificial light include ENF traces, and the ENF
estimation depends on the type of imaging sensor
used in a camera, such as CMOS and CCD. In
our BEVA testbed, we assume the incoming
video recordings are made by cameras with
CMOS imaging sensors in an indoor setting with
artificial light.
The estimation of ENF involves various signal
processing techniques, such as power spectral
analysis and spectrogram-based techniques. Due
to the limited space, interested readers are referred
to our previous work on ENF estimation techniques
regardingmultiple types of recordings.5
B) Detecting Forgeries Using ENF Fingerprint:
To inspect the incoming recording, a sliding window
approach is utilized to analyze the recording
in batches and shift to the next set. This
technique allows for a lower time delay between
inspecting subsequent frames and locate the
modified samples. A correlation coefficient is
used to compare the ground truth ENF and the
estimated ENF from the video-audio recordings,
and a threshold detects the anomaly in estimated
ENF. Figure 2 shows the detection of ENF
mismatch in two parallel recordings, where one
recording is genuine and the other one is forged
with frame duplication attack.
The ENF fingerprints estimated from the
recordings within the interconnect reflect similar
79
IEEE Consumer Electronics Magazine - November/December 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - November/December 2021
