Towards Multimodal Driver’s Stress Detection - Digital Signal Processing for In-Vehicle Systems and Safety - page 16

80

WDE-SWA Distracted

70

WDE-SWA Neutral

60

Distracted

50 Neutral

40

30

20

10

0

0 0 0 0 0 0 0 0 0 0 0
Number of comparison case

Fig. 1.10 Wavelet decomposition details signal energy for SWA calculated for 96 comparison
cases of lane keeping

Sample entropy (SampEnt), which is used as a measure to quantify regularity
and complexity of the signal, is a perfect match measuring the regularity of SWA
signal. It is known that the measures based on entropy have long been employed in
biosignal processing such as EEG, ECG, and EMG to measure regularity and detect
abnormality. The method to calculate the sample entropy follows the work described
in [ 26 ]. The standard deviation is calculated in a canonical form with statistics.

1.4.2 Distraction Detection Performance

Using the algorithm flow depicted in Fig. 1.9 and feature vectors explained in
Table 1.4 , 96 comparison cases for lane keeping and 113 cases for curve negotiation
were examined using 14 drivers' (20 sessions, seven female and seven male drivers)
data. As an insight, WDE_SWA feature member is given for lane keeping
maneuvers in Fig. 1.10 . It can be easily seen that the distracted sessions are
generally greater than the baseline for this metric. The accuracy of the distraction
detection is given in Table 1.5 using seven-dimensional feature vector (LKS) and
using four-dimensional feature vector subset containing only SWA-related features
(LKC) with threshold values of 0.2, 0.1, and 0 for the final classification result.
From Table 1.5 , it can be seen that if any probability value higher than zero is
taken into account, the distraction can be detected with 98% accuracy using lane
keeping segments (LKS) and by 84% accuracy using curve negotiation segments
(LKC) during Tell-Me/AA conversations.