Christian Heinze

Detecting fatigue from steering behaviour applying continuous wavelet transform

The aim of this paper is to develop signal processing based method to measure fatigue from motor behaviour. The advantages of this steering wheel movement approach are that obtaining steering data within driving is robust, non obtrusive, free from sensor application and calibration efforts. Applying methods of continuous wavelet transform (CWT) provides additional information regarding the dynamics and structure of steering behavior comparing to the commonly applied spectral Fourier transform features.

Applying nonlinear dynamics features for speech-based fatigue detection

This paper describes a speech signal processing method to measure fatigue from speech. The advantages of this realtime approach are that obtaining speech data is non obtrusive, free from sensor application and calibration efforts. Applying methods of Non Linear Dynamics(NLD) provides additional information regarding the dynamics and structure of fatigue speech comparing to the commonly applied speech emotion recognition feature set (e.g. fundamental frequency, intensity, pause patterns, formants, cepstral coefficients). We achieved significant correlations between fatigue and NLD features of 0.29. The validity of this approach is briefly discussed by summarizing the empirical results of a sleep deprivation study.

Heart rate measures reflect the interaction of low mental workload and fatigue during driving simulation

The objective of this study was to assess the monotonic mental workload under changing conditions of operator fatigue during a night time driver simulation study. Several cardiovascular measures were used in order to differentiate between driving and a continuous tracking task. From all of the standard cardiovascular measures, heart rate in beats per minute emerged as the most sensitive for workload discrimination. Heart rate was higher during driving than during the tracking task, pointing to a slightly higher demanding workload for the driving task. This result was stable over the course of the night and showed only a minimal fatigue influence. Heart rate variability in milliseconds, on the other hand, was on average higher for the continuous tracking task in comparison to the driving. This was especially the case for the sessions with high subjective sleepiness. It can thus be concluded that the fatigue state of the operator was more impaired during the tracking task than during driving.

Discrimination and relevance determination of heart rate variability features for the identification of congestive heart failure

We propose a machine learning framework that implements automated relevance determination in order to identify the deciding RR interval features for the discrimination between congestive heart failure and healthy condition. As a result, the most relevant features of heart rate variability (HRV) are narrowly located spectral components in the very-low and low frequency band, and specific ordinal patterns. HRV is generally reduced in comparison to the healthy condition; also the autonomic regulation of heart rate acceleration and deceleration appears to be pathlogically inversed.

Drowsiness discrimination in an overnight driving simulation on the basis of RR and QT intervals

Abstract We examined if ECG-based features are discrimi-native towards drowsiness. Twenty-five volunteers (19–32 years) completed 7×40 minutes of monotonous overnight driving simulation, designed to induce drowsiness. ECG (512 s-1) was recorded continuously; subjective ratings of drowsiness on the Karolinska sleepiness scale (KSS) were polled every five minutes. ECG recordings were divided into 5-min segments, each associated with the mean of one self- and two observer-KSS ratings. Those mean KSS values were binarized to obtain two classes not drowsy and drowsy. The Q-, R- and T-waves in the recordings were detected; R-peak positions were manually reviewed; the Q- and T-detection method was tested against the manual annotations of Physio-net’s QT database. Power spectral densities of RR intervals (RR-PSD) and quantiles of the empirical distribution of heart-rate corrected QTc intervals were estimated. Support-vector machines and random-holdout cross-validation were used for the estimation of the classification error. Using either RR-PSD or QTc features yielded mean test errors of 79.3 ± 0.3 % and 82.7 ± 0.5 %, respectively. Merging RR and QTc features improved the accuracy to 88.3 ± 0.2 %. QTc intervals of the class drowsy were generally prolonged com-pared to not drowsy. Our findings indicate that the inclusion of QT intervals contribute to the discrimination of driver sleepiness.

Circadian rhythmicity of cognitive performance

Abstract It was investigated whether cognitive performance shows a circadian rhythm during a 50 h-long forced desynchrony sleep-wake-schedule. We asked whether it would be possible to estimate the circadian period of cognitive performance under such circumstances and how strong it correlates to subjective sleepiness rating as well as body temperature. Cognitive performance was evaluated using the Bowles-Langley test (BLT), which estimates cognitive performance by capturing reaction time and counting errors in a 2-choice visual search task. Power spectral densities (PSD) were estimated by the Lomb-Scargle periodogram [1]. The circadian period τc was estimated from peak PSD. PSD of BLT scores showed lowest, yet distinct, circadian amplitude. In addition to the circadian period estimation we analyzed the correlation of the acquired variables against each other. Pearson’s correlation coefficients were significant but varied strongly at a commonly low level. Despite the obstacle of a plethora of influencing factors [2] BLT scores are sensitive to the circadian rhythm and provide correct estimates of τc compared to rectal body temperature (RBT) as reference. Subjective measures failed estimating τc.

Detection of microsleep events in a car driving simulation study using electrocardiographic features

Abstract Microsleep events (MSE) are short intrusions of sleep under the demand of sustained attention. They can impose a major threat to safety while driving a car and are considered one of the most significant causes of traffic accidents. Driver’s fatigue and MSE account for up to 20% of all car crashes in Europe and at least 100,000 accidents in the US every year. Unfortunately, there is not a standardized test developed to quantify the degree of vigilance of a driver. To account for this problem, different approaches based on biosignal analysis have been studied in the past. In this paper, we investigate an electrocardiographic-based detection of MSE using morphological and rhythmical features. 14 records from a car driving simulation study with a high incidence of MSE were analyzed and the behavior of the ECG features before and after an MSE in relation to reference baseline values (without drowsiness) were investigated. The results show that MSE cannot be detected (or predicted) using only the ECG. However, in the presence of MSE, the rhythmical and morphological features were observed to be significantly different than the ones calculated for the reference signal without sleepiness. In particular, when MSE were present, the heart rate diminished while the heart rate variability increased. Time distances between P wave and R peak, and R peak and T wave and their dispersion increased also. This demonstrates a noticeable change of the autonomous regulation of the heart. In future, the ECG parameter could be used as a surrogate measure of fatigue.

Discrimination power of spectral and nonlinear heart rate variability features for the identification of congestive heart failure

Recognizing pathological heart rhythm features remains a challenge of cardiovascular research. We adopt a machine learning framework with empirically optimized parameters to distinguish heart failure from healthy condition, emphasizing on spectral and nonlinear features of heart rate variability. Fine-grained spectral power densities of RR intervals emerged as the best discriminating group of features, yielding a classification error rate of 13.6 % when presented at a segment length of 50 minutes.

Evolutionary Simulations to Determine the Human Circadian Period Using an Extended Sleep-Wake Model

The development of the human sleep-wake cycle and the adaptation to the changing day and night condition on our planet took place over a time frame of 100000 years. As the result, a representative cross section of todays population sleeps between 4 and 11 hours, with a midsleep point between 1.00 and 9.00 am and a period length of the human circadian pacemaker between 23.5 and 25 hours. Roenneberg et al. published the distribution of midsleep point and sleep duration based on an extensive questionnaire which represents the Middle European society. Czeisler presented a normal distribution of the period length of the human circadian pacemaker as result of a forced desynchrony study. The sleep wake characteristic can be described with the well established two-process model (2PM). The adaptation of the period of our pacemaker to the 24-h day by light is best understood in terms of the phase response curve (PRC). We introduce a combination of these two well established models, called extended two-process model (E2PM). With this model, the sleep-wake behaviour and the circadian period of man under natural day (light) and night (dark) conditions can be simulated simultaneously. With this model, 250 different sleep-wake types were parametrized using evolutionary algorithms. As a breakthrough, the resulting distribution of one important parameter, the sleep-wake cycle duration, matches closely the experimentally acquired distribution of Czeisler.

SIMULATION OF THE ULTRA-SHORT SLEEP-WAKE PROTOCOL USING THE TWO PROCESS MODEL

An experimental sleep-wake protocol of 30- minute cycles with an allowed 10-minute sleep episode and an enforced 20-minute wake episode was carried out over 48 hours. The sleep and wake conditions of the experiment are simulated using the two process model. A fitness func- tion measuring the occurrence of Non-REM sleep in the 10 minute episodes between experiment and model was de- fined. Parameter optimization with sub-harmonic frequency components resulted in an average prediction accuracy of 80 %.