Drowsiness monitoring based on driver and driving data fusion

Abstract

This paper presents a non-intrusive approach for monitoring driver drowsiness, based on driver and driving data fusion. The Percentage of Eye Closure (PERCLOS) is used to estimate the driver's state. The PERCLOS is computed on real time using a stereo vision-based system. The driving information used is the lateral position, the steering wheel angle and the heading error provided by the CAN bus. These three signals have been studied in the time and frequency domain. A multilayer perceptron neural network has been trained to fetch an optimal performance score. This system was installed in a naturalistic driving simulator. For evaluation purposes, several experiments were designed by psychologists and carried out with professional drivers. As ground truth, subjective experts' manual annotation of the driver video sequences and driving signals was used. A detection rate of 70% using individual indicators was raised up to 94% with the combination of indicators. An explanation about these results and some conclusion are presented.

Real-time driver drowsiness estimation by multi-source information fusion with Dempster–Shafer theory

“…After some time, the driver recovered to an alert state with an acceptable lateral offset of the vehicle. Comparing with several other studies, Daza et al (2011) proposed a two-layered feed-forward artificial neural network to fuse the driver and driving data, obtaining a 98.65% drowsiness detection rate but only in a binary classification (awake and drowsy). Friedrichs and Yang (2010b) employed the Bayes classifier with the sequential floating forward selection (SFFS) algorithm and neural network to classify the driver's awake or drowsy states based on lane data and CAN data with test errors of about 36.6% and 16.6%, respectively.…”

“…LATSD reflects the volatility of the car's position at intervals. It has been proven to detect driver drowsiness at a rate of more than 80% in the study by Daza et al (2011). Features fusion. Rombaut (1998) has stated that the fusion of different features is based on the concept that a source cannot simultaneously affect mass on a hypothesis and its complementary, thus giving antagonistic information.…”

A practical method for driver sleepiness detection by processing the EEG signals stimulated with external flickering light

“…It showed that the SVM classifier could differentiate the drowsiness and alertness states in an accuracy value of 99.3 %, so the ITR would be 5.64 bits/min. In [21], different features extracted from the states of the driver or the driving such as PERcentage of Eye Closure (PERCLOS), standard deviation and mean of the lateral position truck and wheel angle and heading error using the artificial neural network for classification are considered. It showed that the detection rate of drowsiness with the combination of these indicators (features) was 94 % in a temporal window of 20 s, so the ITR is 2.05 bits/min.…”

Detection of drowsy driving based on driving information

“…They suggested that a good strategy is to combine the detection of drowsiness with giving the driver feedback about his driving performance. In [4][5][6][7], they studied various analyses and methods of vehicle's behavior for detecting driver drowsiness using steering-related information and lane-related information.…”