Horizontal Electrooculogram Research Articles

Application Specific Reconfigurable Processor for Eyeblink Detection from Dual-Channel EOG Signal

The electrooculogram (EOG) is one of the most significant signals carrying eye movement information, such as blinks and saccades. There are many human–computer interface (HCI) applications based on eye blinks. For example, the detection of eye blinks can be useful for paralyzed people in controlling wheelchairs. Eye blink features from EOG signals can be useful in drowsiness detection. In some applications of electroencephalograms (EEGs), eye blinks are considered noise. The accurate detection of eye blinks can help achieve denoised EEG signals. In this paper, we aimed to design an application-specific reconfigurable binary EOG signal processor to classify blinks and saccades. This work used dual-channel EOG signals containing horizontal and vertical EOG signals. At first, the EOG signals were preprocessed, and then, by extracting only two features, the root mean square (RMS) and standard deviation (STD), blink and saccades were classified. In the classification stage, 97.5% accuracy was obtained using a support vector machine (SVM) at the simulation level. Further, we implemented the system on Xilinx Zynq-7000 FPGAs by hardware/software co-design. The processing was entirely carried out using a hybrid serial–parallel technique for low-power hardware optimization. The overall hardware accuracy for detecting blinks was 95%. The on-chip power consumption for this design was 0.8 watts, whereas the dynamic power was 0.684 watts (86%), and the static power was 0.116 watts (14%).

EOG acquisition system based on ATmega AVR microcontroller

This paper presents a hardware and software of an electrooculogram (EOG) acquisition system based on ATmega AVR microcontroller for the acquisition of vertical and horizontal eye movements. The presented system is used to acquire a dataset of eye movements for volunteers. This system gives two channels representing vertical and horizontal EOG signals. The frequency range of the EOG signal is known to be 0.1 to 10 Hz, and hence this frequency range is isolated with a High-Pass Filter (HPF) with a cutoff frequency of 0.1 Hz followed by a Low-Pass Filter (LPF) with a cutoff frequency of 10 Hz. The EOG acquisition system is interfaced with an ATmega AVR microcontroller to acquire a dataset that can be used for controlling hardware such as Light Emitting Diodes (LEDs), wheelchair, and robot arm. The presented system is composed of EOG signal acquisition, Ag/AgCl electrodes, analog-to-digital converter through Arduino Mega 2560 board microcontroller unit, trainer board, laptop, keypad, and Liquid Crystal Display (LCD). The eye movement is detected by measuring the potential difference between cornea and retina using five Ag–Agcl disposable electrodes. Different volunteers of different ages at different times have been treated with the presented system to obtain data. Classified vertical and horizontal EOG signals and the basic eye movements e.g., open eye, left, right, up, and down can be used to control robots and wheelchairs for rehabilitation purposes.

DWT-BSS: Blind Source Separation applied to EEG signals by extracting wavelet transform’s approximation coefficients

The Electroencephalogram (EEG) signal is widely contaminated by a physiological artifact, such as muscle activity, heart rhythm, and eye movement. The researcher has proposed a number of methods to clean the EEG signal. A group of these methods is called Blind Source Separation (BSS). In this paper, we suggest an approach that combines the BSS methods and the Discrete Wavelet Transform (DWT) algorithm, in order to evaluate the BSS methods after applying them to the approximation coefficients extracted using the DWT. The aim of this work is to identify which BSS algorithms, using which family of wavelet and at which decomposition level, would provide excellent performance. We used the Spearman Correlation Coefficient (SCC) to rate our methods. The technique that performs the best, as evaluated by the SCC between the generated component and the approximation coefficient obtained from the Horizontal EOG results, is AMICA, which obtains a value of 0.81 for levels 2 while using the wavelet symlet at scales 7 and 11. With a value of 0.70 and the use of the wavelet Daubechies at scale 9 and Coiflets at scales 2 and 5 for level 1, AMICA also has the best SCC value calculated between the separated component and the approximation coefficient recovered from the Vertical EOG. While employing the wavelet symlets at scales 5, 7, 8, and 11. for level 2, and level 3 when using the wavelet symlets at scales 1 and 2.

KL Divergence-based transfer learning for cross-subject eye movement recognition with EOG signals.

Human-machine interfaces (HMIs) based on Electro-oculogram (EOG) signals have been widely explored. However, due to the individual variability, it is still challenging for an EOG-based eye movement recognition model to achieve favorable results among cross-subjects. The classical transfer learning methods such as CORrelation Alignment (CORAL), Transfer Component Analysis (TCA), and Joint Distribution Adaptation (JDA) are mainly based on feature transformation and distribution alignment, which do not consider similarities/dissimilarities between target subject and source subjects. In this paper, the Kullback-Leibler (KL) divergence of the log-Power Spectral Density (log-PSD) features of horizontal EOG (HEOG) between the target subject and each source subject is calculated for adaptively selecting partial subjects that suppose to have similar distribution with target subject for further training. It not only consider the similarity but also reduce computational consumption. The results show that the proposed approach is superior to the baseline and classical transfer learning methods, and significantly improves the performance of target subjects who have poor performance with the primary classifiers. The best improvement of Support Vector Machines (SVM) classifier has improved by 13.1% for subject 31 compared with baseline result. The preliminary results of this study demonstrate the effectiveness of the proposed transfer framework and provide a promising tool for implementing cross-subject eye movement recognition models in real-life scenarios.

SNOAR: a new regression approach for the removal of ocular artifact from multi-channel electroencephalogram signals.

Electroencephalogram (EEG) signals are often corrupted by undesirable sources like electrooculogram (EOG) artifacts, which have a substantial impact on the performance of EEG-based systems. This study proposes a new singular spectrum analysis (SSA)-non-negative matrix factorization (NMF)-based ocular artifact removal (SNOAR) method to suppress ocular artifacts from multi-channel EEG signals. First, SSA was used to estimate EOG artifacts using a small subset of frontal electrodes. Then, NMF was applied to decompose the estimated EOG artifacts into vertical EOG (VEOG) and horizontal EOG (HEOG) signals. Finally, a simple linear regression with estimated VEOG and HEOG signals was used to remove artifacts from multi-channel EEG signals. EEG recordings from two EEG datasets (Klados dataset and KARA ONE) were used to evaluate the efficiency of the proposed method. From the simulation results, it was observed that the proposed method achieved betters results in terms of low root-mean-square error (RMSE), low delta band energy ratio, and less power spectral density (PSD) difference between the original clean EEG signal and its filtered version of contaminated EEG signal compared to selected EOG artifact removal methods (independent component analysis (ICA), wavelet-enhanced ICA (wICA), improved wICA, and multivariate empirical mode decomposition (MEMD)).

A novel approach for detection of dyslexia using convolutional neural network with EOG signals

Dyslexia is a learning disability in acquiring reading skills, even though the individual has the appropriate learning opportunity, adequate education, and appropriate sociocultural environment. Dyslexia negatively affects children's educational development; hence, early detection is highly important. Electrooculogram (EOG) signals are one of the most frequently used physiological signals in human-computer interfaces applications. EOG is a method based on the examination of the electrical potential of eye movements. The advantages of EOG-based systems are non-invasive, affordable, easy to record, and can be processed in real time.In this paper, a novel 1D CNN approach using EOG signals is proposed for the diagnosis of dyslexia. The proposed approach aims to diagnose dyslexia using EOG signals that are recorded simultaneously during reading texts, which are prepared in different typefaces and fonts. EOG signals were recorded from both horizontal and vertical channels, thus comparing the success of vertical and horizontal EOG signals in detecting dyslexia. The proposed approach provided an effective classification without requiring any hand-crafted feature extraction techniques. The proposed method achieved classifier accuracy of 98.70% and 80.94% for horizontal and vertical channel EOG signals, respectively. The results show that the EOG signals-based approach gives successful results for the diagnosis of dyslexia.

Detecting slow eye movements with bimodal-LSTM for recognizing drivers’ sleep onset period

Slow eye movements (SEMs) indicate sleep onset period but are rarely studied in the field of driver fatigue detection. Through visual observation and statistical analysis we find that SEMs occur during eye-closed periods and have a high overall coverage rate during driving, which makes it feasible to utilize detecting SEMs to recognize drivers’ sleep onset period. To detect SEMs, we adopt a bimodal Long Short-Term Memory (LSTM) network to deal with temporal information and multimodal information in physiological signals. To extend the distinguishing information, we define a novel horizontal sum (HSUM) signal, which is sum of signals from two horizontal electrooculogram (EOG) channels. Electroencephalogram (EEG)-related features are extracted from the HSUM signal in contrast to those from the traditional O2 signal. EOG features are extracted from the horizontal EOG (HEOG) signals. The results demonstrate that features from multimodal signals (HSUM and HEOG signals, or O2 and HEOG signals) achieve better classification results than features from the single HEOG signal. And the EEG-related features extracted from self-defined HSUM signals achieve comparable results to those from traditional O2 signals, thus avoiding using extra O2 channel. The proposed method of detecting SEMs using the bimodal LSTM to classify features from HSUM and HEOG signals achieves the average F-score of 76.5%, which is higher than the classic support vector machine by 7.5%. This method of detecting SEMs using only two channels helps to build a user acceptable and feasible system for recognizing drivers’ sleep onset period.

Novel Hybrid Brain-Computer Interface for Virtual Reality Applications Using Steady-State Visual-Evoked Potential-Based Brain-Computer Interface and Electrooculogram-Based Eye Tracking for Increased Information Transfer Rate.

Brain–computer interfaces (BCIs) based on electroencephalogram (EEG) have recently attracted increasing attention in virtual reality (VR) applications as a promising tool for controlling virtual objects or generating commands in a “hands-free” manner. Video-oculography (VOG) has been frequently used as a tool to improve BCI performance by identifying the gaze location on the screen, however, current VOG devices are generally too expensive to be embedded in practical low-cost VR head-mounted display (HMD) systems. In this study, we proposed a novel calibration-free hybrid BCI system combining steady-state visual-evoked potential (SSVEP)-based BCI and electrooculogram (EOG)-based eye tracking to increase the information transfer rate (ITR) of a nine-target SSVEP-based BCI in VR environment. Experiments were repeated on three different frequency configurations of pattern-reversal checkerboard stimuli arranged in a 3 × 3 matrix. When a user was staring at one of the nine visual stimuli, the column containing the target stimulus was first identified based on the user’s horizontal eye movement direction (left, middle, or right) classified using horizontal EOG recorded from a pair of electrodes that can be readily incorporated with any existing VR-HMD systems. Note that the EOG can be recorded using the same amplifier for recording SSVEP, unlike the VOG system. Then, the target visual stimulus was identified among the three visual stimuli vertically arranged in the selected column using the extension of multivariate synchronization index (EMSI) algorithm, one of the widely used SSVEP detection algorithms. In our experiments with 20 participants wearing a commercial VR-HMD system, it was shown that both the accuracy and ITR of the proposed hybrid BCI were significantly increased compared to those of the traditional SSVEP-based BCI in VR environment.

An Electro-Oculogram Based Vision System for Grasp Assistive Devices-A Proof of Concept Study.

Humans typically fixate on objects before moving their arm to grasp the object. Patients with ALS disorder can also select the object with their intact eye movement, but are unable to move their limb due to the loss of voluntary muscle control. Though several research works have already achieved success in generating the correct grasp type from their brain measurement, we are still searching for fine controll over an object with a grasp assistive device (orthosis/exoskeleton/robotic arm). Object orientation and object width are two important parameters for controlling the wrist angle and the grasp aperture of the assistive device to replicate a human-like stable grasp. Vision systems are already evolved to measure the geometrical attributes of the object to control the grasp with a prosthetic hand. However, most of the existing vision systems are integrated with electromyography and require some amount of voluntary muscle movement to control the vision system. Due to that reason, those systems are not beneficial for the users with brain-controlled assistive devices. Here, we implemented a vision system which can be controlled through the human gaze. We measured the vertical and horizontal electrooculogram signals and controlled the pan and tilt of a cap-mounted webcam to keep the object of interest in focus and at the centre of the picture. A simple ‘signature’ extraction procedure was also utilized to reduce the algorithmic complexity and system storage capacity. The developed device has been tested with ten healthy participants. We approximated the object orientation and the size of the object and determined an appropriate wrist orientation angle and the grasp aperture size within 22 ms. The combined accuracy exceeded 75%. The integration of the proposed system with the brain-controlled grasp assistive device and increasing the number of grasps can offer more natural manoeuvring in grasp for ALS patients.

Elimination of EOG signals from raw EEG signals using step size based recursive least squares- least mean fourth adaptive algorithm

A novel Step Size based Recursive least squares- Least mean fourth(SSRL) adaptive algorithm is proposed as a noise canceller, in this paper, to eliminate the Ocular artifacts(OAs) from the recorded raw EEG signals. Here, both second and fourth-order power optimization algorithms have been used to eliminate the OAs from raw EEG signals. The Reference signals, such as vertical Electrooculogram and horizontal Electrooculogram signals, have been recorded separately.FIR filter has been used to process the reference signals by updating filter coefficients using the proposed SSRL algorithm. Thereby, true EEG signals are obtained by subtracting the processed signals from the raw signals. Moreover, a mathematical model for the mean square deviation analysis is proposed for the SSRL algorithm, which provides better results compared to the conventional methods.

Mixed step size normalized least mean fourth adaptive algorithm for artifact elimination from raw EEG signals

In this paper, a novel algorithm is proposed to eliminate the ocular artifacts (OAs) using mixed step size normalized least mean fourth adaptive algorithm from the raw EEG signals. Here, the fourth-order power optimization technique is used to eliminate the OAs effectively. The reference signals, such as vertical electrooculogram and horizontal electrooculogram, are recorded separately. Thereby, these signals are processed by the FIR filter, whose coefficients are adaptively updated at each iteration by using the proposed algorithm. The processed signals are subtracted from the raw EEG signals in order to obtain clean EEG signals. In addition, the mathematical model for the mean square deviation analysis is provided and compared with conventional methods like combined step size normalized least mean squares and variable parameter normalized mixed norm adaptive algorithm. From the simulation results, it is found that the proposed algorithm exhibits better performance in terms of minimum mean square deviation.

Impact of Stimulus Features on the Performance of a Gaze-Independent Brain-Computer Interface Based on Covert Spatial Attention Shifts.

Regaining communication abilities in patients who are unable to speak or move is one of the main goals in decoding brain waves for brain-computer interface (BCI) control. Many BCI approaches designed for communication rely on attention to visual stimuli, commonly applying an oddball paradigm, and require both eye movements and adequate visual acuity. These abilities may, however, be absent in patients who depend on BCI communication. We have therefore developed a response-based communication BCI, which is independent of gaze shifts but utilizes covert shifts of attention to the left or right visual field. We recorded the electroencephalogram (EEG) from 29 channels and coregistered the vertical and horizontal electrooculogram. Data-driven decoding of small attention-based differences between the hemispheres, also known as N2pc, was performed using 14 posterior channels, which are expected to reflect correlates of visual spatial attention. Eighteen healthy participants responded to 120 statements by covertly directing attention to one of two colored symbols (green and red crosses for “yes” and “no,” respectively), presented in the user’s left and right visual field, respectively, while maintaining central gaze fixation. On average across participants, 88.5% (std: 7.8%) of responses were correctly decoded online. In order to investigate the potential influence of stimulus features on accuracy, we presented the symbols with different visual angles, by altering symbol size and eccentricity. The offline analysis revealed that stimulus features have a minimal impact on the controllability of the BCI. Hence, we show with our novel approach that spatial attention to a colored symbol is a robust method with which to control a BCI, which has the potential to support severely paralyzed people with impaired eye movements and low visual acuity in communicating with their environment.

Driving Fatigue Classification Based on Fusion Entropy Analysis Combining EOG and EEG

The rising number of traffic accidents has become a major issue in our daily life, which has attracted the concern of society and governments. To deal with this issue, in our previous study, we have designed a real-time driving fatigue detection system using power spectrum density and sample entropy. By using the wireless technology and dry electrodes for EEG collection, we further integrated virtual reality simulated driving environment, which made our study more applicable to realistic settings. However, the high accuracy of classification for driving fatigue has not been obtained. To measure the time series complexity of the EEG signal, we proposed a fusion entropy (sample entropy, approximate entropy, and spectral entropy) analysis method of EEG and EOG. First, a sample entropy was applied for feature extraction from the horizontal and vertical EOG. Second, an approximate entropy, sample entropy, and spectral entropy features of each sub-band of EEG are extracted. Third, feature fusion for sub-band is performed by canonical correlation analysis (CCA). Finally, the features of EOG and EEG are classified using a relevant vector machine (RVM). Twenty-two subjects participated in the driving fatigue experiments for a duration of 90 min. The results demonstrated that the fusion entropy analysis combining EOG and EEG could provide an alternative method for driving fatigue detection, and the average accuracy rate was up to 99.1 ± 1.2%. The authors further analyzed the effect of feature fusion in four sub-bands (δ, α, β and θ) and compared with every single sub-band on classification performance, it is proved that the former is superior to the latter presenting the proposed method can provide effective indicators for driving fatigue detection.

Efficient sleep classification based on entropy features and a support vector machine classifier

Objective: Sleep quality helps to reflect on the physical and mental condition, and efficient sleep stage scoring promises considerable advantages to health care. The aim of this study is to propose a simple and efficient sleep classification method based on entropy features and a support vector machine classifier, named SC-En&SVM. Approach: Entropy features, including fuzzy measure entropy (FuzzMEn), fuzzy entropy, and sample entropy are applied for the analysis and classification of sleep stages. FuzzyMEn has been used for heart rate variability analysis since it was proposed, while this is the first time it has been used for sleep scoring. The three features are extracted from 6 376 730 s epochs from Fpz-Cz electroencephalogram (EEG), Pz-Oz EEG and horizontal electrooculogram (EOG) signals in the sleep-EDF database. The independent samples t-test shows that the entropy values have significant differences among six sleep stages. The multi-class support vector machine (SVM) with a one-against-all class approach is utilized in this specific application for the first time. We perform 10-fold cross-validation as well as leave-one-subject-out cross-validation for 61 subjects to test the effectiveness and reliability of SC-En&SVM. Main results: The 10-fold cross-validation shows an effective performance with high stability of SC-En&SVM. The average accuracy and standard deviation for 2–6 states are 97.02 ± 0.58, 92.74 ± 1.32, 89.08 ± 0.90, 86.02 ± 1.06 and 83.94 ± 1.61, respectively. While for a more practical evaluation, the independent scheme is further performed, and the results show that our method achieved similar or slightly better average accuracies for 2–6 states of 94.15%, 85.06%, 80.96%, 78.68% and 75.98% compared with state-of-the-art methods. The corresponding kappa coefficients (0.81, 0.74, 0.72, 0.71, 0.67) guarantee substantial agreement of the classification. Significance: We propose a novel sleep stage scoring method, SC-En&SVM, with easily accessible features and a simple classification algorithm, without reducing the classification performance compared with other approaches.

New Method for Pure-Tone Audiometry Using Electrooculogram: A Proof-of-Concept Study.

Precise and timely evaluation of an individual’s hearing loss plays an important role in determining appropriate treatment strategies, including medication and aural rehabilitation. However, currently available hearing assessment systems do not satisfy the need for an objective assessment tool with a simple and non-invasive procedure. In this paper, we propose a new method for pure-tone audiometry, which may potentially be used to assess an individual’s hearing ability objectively and quantitatively, without need for the user’s active response. The proposed method is based on the auditory oculogyric reflex, where the eyes involuntary rotate towards the source of a sound, in response to spatially moving pure-tone audio stimuli modulated at specific frequencies and intensities. We quantitatively analyzed horizontal electrooculograms (EOG) recorded with a pair of electrodes under two conditions—when pure-tone stimuli were (1) “inaudible” or (2) “audible” to a participant. Preliminary experimental results showed significantly increased EOG amplitude in the audible condition compared to the inaudible condition for all ten healthy participants. This demonstrates potential use of the proposed method as a new non-invasive hearing assessment tool.

Development of an electrooculogram-based human-computer interface using involuntary eye movement by spatially rotating sound for communication of locked-in patients

Individuals who have lost normal pathways for communication need augmentative and alternative communication (AAC) devices. In this study, we propose a new electrooculogram (EOG)-based human-computer interface (HCI) paradigm for AAC that does not require a user’s voluntary eye movement for binary yes/no communication by patients in locked-in state (LIS). The proposed HCI uses a horizontal EOG elicited by involuntary auditory oculogyric reflex, in response to a rotating sound source. In the proposed HCI paradigm, a user was asked to selectively attend to one of two sound sources rotating in directions opposite to each other, based on the user’s intention. The user’s intentions could then be recognised by quantifying EOGs. To validate its performance, a series of experiments was conducted with ten healthy subjects, and two patients with amyotrophic lateral sclerosis (ALS). The online experimental results exhibited high-classification accuracies of 94% in both healthy subjects and ALS patients in cases where decisions were made every six seconds. The ALS patients also participated in a practical yes/no communication experiment with 26 or 30 questions with known answers. The accuracy of the experiments with questionnaires was 94%, demonstrating that our paradigm could constitute an auxiliary AAC system for some LIS patients.

Attention avoidance of the threat conditioned stimulus during extinction increases physiological arousal generalisation and retention

Exposure therapy is a key component of efficacious treatment for anxiety. Biases in the allocation of attention towards versus away from threat assessed prior to exposure-based treatments such as cognitive-behavioural therapy differentially predict treatment outcomes. However, it is unclear whether monitoring versus avoiding threat stimuli influences learning during exposure. Extinction paradigms are the experimental analogue of exposure therapy. Therefore, manipulating attention towards versus away from threat during extinction trials may shed light on the role of attention during exposure therapy. This study utilised a Pavlovian fear conditioning and extinction paradigm to examine whether directing attention towards versus away from the threat conditioned stimulus (CS+) related to differences in extinction, as indexed by skin conductance responses (SCR), CS evaluations and subjective measures of anxiety. Following a fear conditioning phase in which a dog image (CS+) was paired with an aversive tone unconditioned stimulus (US) and another dog image (CS-) was presented alone, 57 participants were randomly assigned to one of three conditions during extinction in which both CSs were presented alone: monitor the CS+ (N = 19), avoid the CS+ and attend to another stimulus (N = 18), no attention manipulation control (N = 20). Eye movements were monitored for visual adherence to assigned location using horizontal electro-oculogram. In the context of the acquisition of differential conditioning and visual adherence during extinction, both active groups exhibited larger SCRs to the CS + relative to the CS- during the first extinction block compared to the control group, and the avoid group exhibited significantly larger SCRs on CS+ and CS- trials throughout the extinction phase compared to the other groups. The avoid group also exhibited less decline in SCRs to the CS+ during the extinction retest phase relative to the control group. No significant group differences were observed in between-phase CS evaluations and subjective anxiety ratings. Avoidance of threat conditioned stimuli may impair extinction learning and increase physiological arousal generalisation to safe stimuli.

Development of an electrooculogram-based eye-computer interface for communication of individuals with amyotrophic lateral sclerosis

BackgroundElectrooculogram (EOG) can be used to continuously track eye movements and can thus be considered as an alternative to conventional camera-based eye trackers. Although many EOG-based eye tracking systems have been studied with the ultimate goal of providing a new way of communication for individuals with amyotrophic lateral sclerosis (ALS), most of them were tested with healthy people only. In this paper, we investigated the feasibility of EOG-based eye-writing as a new mode of communication for individuals with ALS.MethodsWe developed an EOG-based eye-writing system and tested this system with 18 healthy participants and three participants with ALS. We also applied a new method for removing crosstalk between horizontal and vertical EOG components. All study participants were asked to eye-write specially designed patterns of 10 Arabic numbers three times after a short practice session.ResultsOur system achieved a mean recognition rates of 95.93% for healthy participants and showed recognition rates of 95.00%, 66.67%, and 93.33% for the three participants with ALS. The low recognition rates in one of the participants with ALS was mainly due to miswritten letters, the number of which decreased as the experiment proceeded.ConclusionOur proposed eye-writing system is a feasible human-computer interface (HCI) tool for enabling practical communication of individuals with ALS.

EOG feature relevance determination for microsleep detection

Abstract Automatic relevance determination (ARD) was applied to two-channel EOG recordings for microsleep event (MSE) recognition. 10 s immediately before MSE and also before counterexamples of fatigued, but attentive driving were analysed. Two type of signal features were extracted: the maximum cross correlation (MaxCC) and logarithmic power spectral densities (PSD) averaged in spectral bands of 0.5 Hz width ranging between 0 and 8 Hz. Generalised learn-ing vector quantisation (GRLVQ) was used as ARD method to show the potential of feature reduction. This is compared to support-vector machines (SVM), in which the feature reduction plays a much smaller role. Cross validation yielded mean normalised relevancies of PSD features in the range of 1.6 - 4.9 % and 1.9 - 10.4 % for horizontal and vertical EOG, respectively. MaxCC relevancies were 0.002 - 0.006 % and 0.002 - 0.06 %, respectively. This shows that PSD features of vertical EOG are indispensable, whereas MaxCC can be neglected. Mean classification accuracies were estimated at 86.6±b 1.3 % and 92.3±b 0.2 % for GRLVQ and SVM, respec-tively. GRLVQ permits objective feature reduction by inclu-sion of all processing stages, but is not as accurate as SVM.

EOG feature relevance determination for microsleep detection

Abstract Automatic relevance determination (ARD) was applied to two-channel EOG recordings for microsleep event (MSE) recognition. 10 s immediately before MSE and also before counterexamples of fatigued, but attentive driving were analysed. Two type of signal features were extracted: the maximum cross correlation (MaxCC) and logarithmic power spectral densities (PSD) averaged in spectral bands of 0.5 Hz width ranging between 0 and 8 Hz. Generalised learn-ing vector quantisation (GRLVQ) was used as ARD method to show the potential of feature reduction. This is compared to support-vector machines (SVM), in which the feature reduction plays a much smaller role. Cross validation yielded mean normalised relevancies of PSD features in the range of 1.6 – 4.9 % and 1.9 – 10.4 % for horizontal and vertical EOG, respectively. MaxCC relevancies were 0.002 – 0.006 % and 0.002 – 0.06 %, respectively. This shows that PSD features of vertical EOG are indispensable, whereas MaxCC can be neglected. Mean classification accuracies were estimated at 86.6±b 1.3 % and 92.3±b 0.2 % for GRLVQ and SVM, respectively. GRLVQ permits objective feature reduction by inclusion of all processing stages, but is not as accurate as SVM.