Advanced Signal Processing Algorithms for Security and Safety in Vehicular Applications
Abstract
Abstract
Advanced driver assistance systems (ADAS) have witnessed a significant growth in the last
decade. To complement the ADAS technologies, wireless vehicular communications emerged
as a promising solution providing vehicles with an all-directions non-line of sight view of their
surroundings. This dissertation focuses on the security aspects of vehicular communications
in addition to in-vehicle safety aspects of ADAS applications.
Wireless communications are vulnerable to potential security attacks due to their broadcast
nature. Vehicles exchange critical safety information about their speed, direction and location which necessities efficient algorithms to secure the ongoing communications. This
motivates us to investigate the physical layer (PHY) security for wireless systems having fast fading channels, few transmit antennas and stringent transmit power constraints
as imposed by the vehicular communication standards. We study the PHY security in a
single-user orthogonal frequency-division multiplexing (OFDM) and multi-user single-carrier
frequency-division multiple-access (SC-FDMA) systems where transmitter(s), denoted by Alice(s), transmit confidential data messages to a neighbouring receiver, which we refer to as
Bob, in the presence of eavesdropping node(s), denoted by Eve(s). Alice(s) send a temporal
artificial-noise (AN) signal superimposed over their information signal to secure the ongoing communications. We investigate the impact of the channel delay spread, cyclic prefix,
information/AN signals power allocations, channel state information (CSI) knowledge and
information and AN precoders design on the achievable average secrecy rate in single-user
and multi-users systems. We highlight the trade-offs between the computational complexity
and the average secrecy rate performance for different detection strategies. In addition, we
enhance the system’s security by proposing a hybrid scheme that combines temporal AN
with channel-based secret-key extraction. Furthermore, we propose a novel secure scheme
that utilizes the sub-channel orthogonality of OFDM systems to simultaneously transmit
data and secret key symbols in two opposite directions. Bob, shares secret key symbols with
Alice using wiretap coding over a portion of the sub-channels. Simultaneously, Alice uses the
accumulated secret keys in her secret-key queue to encrypt data symbols using a one time
pad (OTP) cipher and transmits them to Bob over the remaining sub-channels. We drive
closed-form expressions for the average secrecy rate and the secrecy outage probabilities for
the different proposed schemes.
In the second part of this dissertation, we focus on the in-vehicle safety aspects of ADAS
applications with emphasis on the driver’s gaze estimation which is essential to asses the
visual attention of drivers allowing alarming drivers in different distraction scenarios. We
present a 3D head based coordinate system whose center is the driver’s head center of mass
(CoM). Then, we utilize fiducial markers to precisely annotate the driver’s gaze into elevation
and azimuth angles. We increase the realism of the discrete driving data by augmenting it
to continuous gaze data collected while the car is parked. Then, we propose deep-learning based gaze estimation algorithms that utilizes all features on the driver’s face to produce new
statistical customizable gaze regions estimates. Our statistical gaze estimations provide the
level of confidence in estimation which can be used a useful parameter for different ADAS
applications. We perform large scale cross-driver testing utilizing the data from 54 different
subjects collected in naturalistic driving environments.