2 / 9
allied
academies
Page 25
May 23-24, 2019 | Vienna, Austria
Robotics and Artificial Intelligence
2
nd
International Conference on
Research & Reviews: Journal of Engineering and Technology | ISSN: 2319-9873 | Volume 8
Physiological signal-based detection of driver hypovigilance
Arun Sahayadhas
Vels Institute of Science, Technology and Advanced Studies (VISTAS), India
D
river hypovigilance which includes drowsiness,
inattention and fatigue are the major reason for
road accidents. To detect the driver hypovigilance, the
physiological signals needs to be collected and analyzed. In
case of hypovigilance, the driver has to be alerted on time
so that loss can be avoided. The physiological signals are
the graphical representation of human physical condition.
Electrocardiogram (ECG), Electrooculogram (EOG) and
Electromyogram (EMG) are some of the signals that are used
here to provide the state of driver’s abnormal behaviour.
Ten subjects participated in the data collection experiment
and were asked to drive for two hours at three different
timings of the day (00:00 – 02:00 hrs, 03:00 – 05:00 hrs
and 14:00 – 16:00 hrs) when their circadian rhythm was
low. The five classes namely – normal, visual inattention,
cognitive inattention, fatigue and drowsy were analyzed. The
Butterworth 6th order filter is applied to remove the noise
from the signals. The features that are extracted from the
signals can be linear and non-linear. Sixteen Linear features
such as mean, median, minimum, maximum, standard
deviation, power, skewness, kurtosis, Energy, correlation
coefficient, central frequency, peak frequency, first quartile
frequency, third quartile frequency, Interquartile Range and
Root Mean Square were extracted. Likewise, eight Non-linear
features such as Spatial filling index (SFI), Central tendency
measure (CTM), Correlation dimension, Approximate Entropy
(ApEn), HURST exponent, Largest Lyapunov exponent, Non-
linear Predication error (NLPE) and stoppage criteria were
extracted. These extracted features were given as input to
the different classifiers (Support Vector Machine (SVM),
K-Nearest Neighbour (KNN), Convolutional Neural Networks
(CNN)) to obtain the accuracy, sensitivity and scalability. The
results show that the features from ECG can be embedded in
a smart watchwhich can alert the driver during hypovigilance.
e
:
arurun@gmail.comJET, Volume 8 | ISSN: 2319-9873