DETECTDUI: AN IN-CAR DETECTION SYSTEM FOR DRINK DRIVING AND BACS

Abstract

As one of the biggest contributors to road 
accidents and fatalities, drink driving is 
worthy of significant research attention. 
However, most existing systems on detecting 
or preventing drink driving either require 
special hardware or require much effort from 
the user, making these systems inapplicable 
to continuous drink driving monitoring in a 
real driving environment. In this paper, we 
present DetectDUI, a contactless, noninvasive, real-time system that yields a 
relatively highly accurate drink driving 
monitoring by combining vital signs (heart 
rate and respiration rate) extracted from incar WiFi system and driver’s psychomotor 
coordination through steering wheel 
operations. The framework consists of a 
series of signal processing algorithms for 
extracting clean and informative vital signs
and psychomotor coordination, and integrate 
the two data streams using a self-attention 
convolutional neural network (i.e., CAttention). In safe laboratory experiments 
with 15 participants, DetectDUI achieves 
drink driving detection accuracy of 96.6% 
and BAC predictions with an average mean 
error of 2 _ 5mg/dl. These promising results 
provide a highly encouraging case for 
continued development.