Applying neural network analysis on heart rate variability data to assess driver fatigue

https://doi.org/10.1016/j.eswa.2010.12.028Get rights and content

Abstract

Long duration driving is a significant cause of fatigue related accidents on motorways. Fatigue caused by driving for extended hours can acutely impair driver’s alertness and performance. This papers presents an artificial intelligence based system which could detect early onset of fatigue in drivers using heart rate variability (HRV) as the human physiological measure. The detection performance of neural network was tested using a set of electrocardiogram (ECG) data recorded under laboratory conditions. The neural network gave an accuracy of 90%. This HRV based fatigue detection technique can be used as a fatigue countermeasure.

Research highlights

► In this study we show heart rate variability which can be used as a passive means to quantify drowsiness. Frequency domain components of HRV is used for early detection of fatigue. ► We also designed a neural network based artificial intelligent algorithm which detects whether the given HRV signal is in alert of fatigue state. ► The frequency domain components of HRV were also used to distinguish between parasympathetic (HF) and sympathetic (LF) activities using LF/HF ratio. Statistical analysis was also performed to identify the difference between LF/HF ratio during alert and fatigue states.

Introduction

In recent years fatigue has been considered as one of the major cause of road accidents and has become a feature on the agenda of road safety. Most people think of fatigue as a problem of falling asleep while driving however studies indicate that fatigue causes problems for driving performance well before the driver actually falls asleep. Brown in 1994 defined fatigue as a disinclination to continue performing the task, and that it involves an impairment of human efficiency when work continued after the person had become aware of their fatigued state. Brown also distinguished physical fatigue from mental fatigue (Brown, 1994). Driver related fatigue is defined as a state of reduced mental alertness, which impairs performance of a range of cognitive and psychomotor tasks including driving (Williamson, Feyer, & Friswell, 1996). Driver fatigue depends on factors like time of the day and sleep debt. Lenne, Triggs and Redman in 1997 found that performance on the driving task was poorest at 0600 h and 1400 h. Sleep debt also contributes to driver fatigue and consequently reduced performance. Fell and Black in 1997 investigated driver fatigue incidents in cities and found that 57% of drivers who had a fatigue-related incident reported insufficient sleep on the night before the incident happened.

Fatigue in drivers is a contributing factor for motor vehicle crashes. A survey done in USA by The National Highway Traffic Safety Administration (NHTSA) in 1996 estimates that there are 56,000 sleep related road crashes annually in USA, resulting in 40,000 injuries and 1550 fatalities. In the year 2007 in USA fatigue was implicated in at least 18% of fatal accidents and accounts for about 7% of all accidents (Smart Motorist, 2008). In Great Britain up to 20% of serious road accidents is caused due to fatigue (The Royal Society of the Prevention of Accidents, 2001). A study in New-Zealand in 1997 found that driver fatigue was a responsible factor in 7% of the accidents (The Royal Society of the Prevention of Accidents, 2001). As per the survey done by Road and Traffic Authority (Australia) in the year 2002, 20% of vehicle crashes had fatigue as a contributing factor (Australian Transport Safety Bureau, 2002).

Numerous physiological parameters such as electroencephalography (EEG) (Lal and Craig, 2001, Lal and Craig, 2002a) and electrooculogram (EOG) (Wierwille & Ellsworth, 1994), video monitoring can be used to measure the level of fatigue. However, very little research has been conducted in the area of applying HRV as a measure of driver fatigue and most of this area remains unexplored. The importance of developing driver countermeasures for fatigue has been discussed previous (Lal, Craig, Boord, Kirkup, & Nguyen, 2003). The importance of utilizing HRV as a physiological parameter to assess fatigue for purposes of developing fatigue countermeasure system is explored in this paper.

Heart rate variability provides a passive means to quantify drowsiness physiologically (Mulder, 1992). It is defined as the measure of variation in heart beats and is calculated by analyzing the time series of beat to beat intervals (i.e., the R–R intervals) (Institute of HeartMath, 2007). HRV has been previously used to examine mental workload (Hancock & Verwey, 1997), stress (Bornas et al., 2004, Steyvers and De Waards, 2000), and driver fatigue (Egelund, 1982). In relation to driver fatigue, HRV can provide useful data of when fatigue becomes an issue during driving. An increase in HRV is an indication of decrease in mental workload, which can occur in sleepy drivers over prolonged monotonous driving (Horne & Reyner, 1995). But lower workload may also be related to lower vigilance, which can negatively impact driver performance.

HRV can be evaluated using time-based measures or frequency domain measures. Time domain measures are common and the simplest to perform and can be assessed with calculation of the standard deviation of R–R (inter-beat) intervals (European Heart Journal, 1996). Frequency domain analysis is based on mathematical transformations (i.e., Fast Fourier Transforms) of the signals from time domain to frequency domain (expressed in cycles per beat with varying amplitudes and frequencies). Power spectral density (PSD) analysis provides the basic information of how power (i.e., variance) distributes as a function of frequency (American Heart Association, 1996). Methods for calculating PSD can be classified as nonparametric and parametric. Nonparametric (which is normally conducted using FFT) has the advantage of high processing speed and simplicity of algorithm whereas parametric (which is normally conducted using autoregressive model) has the advantage of smoother spectral components that can be distinguished independent of preselected frequency bands (American Heart Association, 1996).

Four main components which are derived from heart rate power spectrum are described in Table 1 (American College of Cardiology/American Heart Association, 1999). The low frequency (LF) component (0.04–0.15 Hz) of the HRV power spectrum is influenced by both the parasympathetic and sympathetic activity whereas the high frequency (HF) component (0.15–0.4 Hz) is influenced by the parasympathetic activity (Bezerianos, Papadimitriou, & Alexopoulos, 1999). The physiological correlates of ultra low frequency (ULF) and very low frequency (VLF) are still unknown and needs further research (Batchinsky et al., 2007). The present study will investigate LF and HF components of HRV and fatigue effects. The LF/HF ratio is considered to be a measure of sympathovagal balance (American College of Cardiology/American Heart Association, 1999). The advantage of the frequency domain analysis is its ability to break HRV time series data into different spectrum viz. ULF, VLF, LF and HF and thereby providing information regarding each spectrum individually. Information related to fatigue can be obtained by analyzing LF and HF bands and deriving LF/HF ratio. Hence the aim of the present study is to assess HRV as an indicator of driver fatigue.

Section snippets

Participants and study protocol

Data collected previously (Lal & Craig, 2002b) was used in this study. However, the heart rate data from this study (Lal & Craig, 2002b) and HRV analysis on the data has not been reported previously. Data of twelve participants were analyzed. Subjects were recruited from a large tertiary institution and the local community and were randomly assigned to the study. All the participants were truck drivers and had a valid drivers’ license when the test was conducted. They had a mean age of 47 ± 11 

Results

The spectral image plotted from the PSD was the input given to the neural network, yielded an accuracy of 90% when tested with the six data sets. Different learning constants ranging from 0.001 to 1 were used to train the neural network so as to increase the accuracy level. The highest accuracy of 90% was obtained by using a learning constant of 0.5. The cyclic error curve of the neural network which was the difference between the desired output and the actual output is shown in Fig. 8. It can

Discussion

This study examined the use of HRV and neural network to detect fatigue in drivers. The accuracy of the neural network was substantially high at 90%. As we had a limited amount of data set the accuracy of the neural network cannot be validated entirely. In order to validate the accuracy we need to have more data sets for training, validating and testing.

The HRV spectrum analysis gives a direct relationship between fatigue and the HRV. Egelund (1982) has also reported that there were significant

Acknowledgement

The analysis was supported under an Australian Research Council grant (ARC Linkage Grant No. LP0562407). The authors wish to acknowledge Salahiddin Altahat for designing algorithm to extract the heart data. Marc Carmichael’s contribution towards developing the algorithm for neural network is also acknowledged.

References (32)

  • American Heart Association

    Heart rate variability: Standards of measurement, physiological interpretation, and clinical use

    Circulation

    (1996)
  • Artaud, T., Planque, S., Lavergne, C., Cara, H., de Lepine, P., Tarriere, C., et al. (1994). An On-Board System for...
  • Australian Transport Safety Bureau. (2002). Driver fatigue, road and traffic authority, NSW, Australia....
  • X. Bornas et al.

    Self-implication and heart rate variability during simulated exposure to flight-related stimuli

    Anxiety, Stress, and Coping

    (2004)
  • I. Brown

    Driver fatigue

    Human Factors

    (1994)
  • Byeon, M., Hans, S., Min, H., Wo, Y., & Park, Y. (2006). A study of HRV analysis to detect drowsiness states of...
  • Cited by (355)

    View all citing articles on Scopus
    1

    Data Analyst and Software Developer.

    2

    Principal Academic Supervisor. Tel.: +61 02 9514 1592; fax: +61 02 9514 8206.

    3

    Ph.D. Candidate. Tel.: +61 02 9514 8311; fax: +61 02 9514 8206.

    4

    Principal Industry Supervisor. Tel.: +61 02 8307 5100; fax: +61 02 9660 9429.

    View full text