Abstract
Vehicles are assets of great value that allow us the ability to transport goods and
people or render services in the modern world. Given their value, they continue
to be the target of various forms of crime which result in either the loss of the
vehicle or items contained within them. This study sets out to create a multifactor
vehicle access control model that focuses on the digital aspects of the
access control mechanism. We make use of the design science research methodology
supplemented with three primary methods: The design and distribution
of a survey intended to gather information on the perceptions and preferences of
South African vehicle owners/operators on the topic of multi-factor vehicle access
control systems; the creation of a multi-factor vehicle access control model,
VSec; and lastly the creation of a prototype which realises the VSec model,
which will allow us to evaluate whether VSec is fit for purpose as a multi-factor
vehicle access control model. A survey showed that respondents had specific
preferences for their implementation. Respondents showed that their most preferred
vehicle access control system would make use of fingerprint authentication
to secure the vehicle’s locking mechanisms and ignition with scanners located
on the vehicle’s smart key remote, door handles and ignition. The survey now
serves as a step toward closing the gap in knowledge of biometric acceptance in a
developing country. The results of the literature review and survey were used to
create VSec: A model that utilises fingerprint recognition and driver’s license
data to secure vehicles. VSec secures the ignition and locking/unlocking mechanisms
of a vehicle through fingerprint recognition with scanners located on the
vehicle steering wheel, door handles and remote. Additionally, VSec further
secures the vehicle’s ignition from unauthorised use by ensuring the identified
user has a valid driver’s license. After determining that VSec will make use of
fingerprint recognition, ConDense, a novel approach to optimising the accuracy
of a Convolutional Neural Network (CNN) was created for use in the VSec
prototype. When evaluated against FVC 2006 databases, ConDense showed
that it improved model accuracy for Inception V3 and Xception (DB1); Xception
and Deep Residual Network (ResNet) 152 V2 (DB2); and Xception and
ResNet 152 V2 (DB3). The greatest improvement was shown for the ResNet
152 V2 pre-trained model, where ConDense improved its equal error rate from
0.493% to 0.193% (61% improvement) when evaluated against FVC 2006’s DB2
dataset. Our results for ConDense show that it can now be used as an alternative
strategy for classification-based authentication. VSec was evaluated
through the creation of a Raspberry Pi-based prototype that was supplemented
with ConDense to improve its fingerprint recognition accuracy. Upon evaluation,
the VSec prototype showed that it correctly classified all enrolled users
and prevented unauthorised access to both the secured vehicle’s ignition and
locking mechanisms. This means that only persons who are authorised to use
a vehicle and possess a valid driver’s license may operate a vehicle secured by
VSec. With the successful evaluation of our model, VSec now serves to expand
available options for multi-factor vehicle access control.