On-Line Detection of Weariness Using Brain and Visual Information

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 ISSN 2347 - 3983

Volume 1, No.4, December 2013

International Journal of Emerging Trends in Engineering Research Available Online at http://warse.org/pdfs/2013/ijeter01142013.pdf

On-Line Detection of Weariness Using Brain and Visual Information Vasavi dobbala1, Mr. R.V.krishnaiah2 1

PG Scholar, Dept. Of Electronics & Communication, DRK institute of science& technology.

2 Professor & Coordinator, Hyderabad, India 1 2 dobbalavasavi@gmail.com, r.v.krshnaiah@gmail.com

Abstract A weariness detection system using both brain and visual activity is presented in this paper. The activity of brain is monitored by using a electroencephalographic (EEG) channel.

An EEG-based weariness detector using characteristic techniques and fuzzy logic is recommended. Visual activity is monitored through blinking detection and characterization.

Blinking countenance is extracted from an electrooculographic (EOG) channel. Features are combined by using fuzzy logic to create an EOG-based weariness detector.

The features used by the EOG-based detector are voluntary restricted to the features that can be automatically extracted from a video analysis of the equivalent accuracy. Both detection systems are then merged using cascading decision rules according to a medical scale of weariness evaluation.

Combining brain and visual information makes it possible to detect three levels of weariness: “awake,” “drowsy,” and “very drowsy.” One main advantage of the system is that it does not have to be tuned for each driver. The detection system was tested on driving data from 20 different drivers and reached 80.6% correct classifications on three weariness levels. The execution part show that EEG and EOG detectors are redundant: EEG-based detections are used to confirm EOG-based detection and thus enable the false alarm rate to be decreased to 5% while the true positive rate is not reduced, compared with a single EOG-based detector.

Keywords— Blinking countenance, cascading rules decision,

Weariness, electroencephalographic (EEG), electrooculographic (EOG), fuzzy logic 1.INTRODUCTION Weariness can be defined as the transition between the awake state and the sleep state where one’s ability to observe and analyze are strongly reduced. In recent years, driver weariness has been one of the major causes of road accidents and can lead to severe physical injuries, deaths and powerful economic losses. Statistics demonstrate the need of a reliable driver weariness detection system which could alert the driver before a mishap occurs. According to available statistical information, over 1.3 million people die each year on the road and 20 to 50 million people suffer non-fatal injuries due to road accidents [1]. Based on police reports, the US National Highway Traffic Safety Administration (NHTSA) conservatively estimated that a total of 100,000 vehicle slam each year are the direct result of driver weariness. These slams resulted in approximately 1,550 deaths, 71,000 injuries and $12.5 billion in monetary

losses [2]. In the year 2009, the US National Sleep Foundation (NSF) reported that 54% of adult drivers have driven a vehicle while feeling drowsy and 28% of them actually fell asleep [3].

The German Road Safety Council (DVR) claims that one in four highway traffic fatalities are a result of momentary driver weariness [4]. These statistics suggest that driver weariness is one of the main causes of road accidents. Researchers have attempted to determine driver weariness using the following measures: (1) vehicle-based measures; (2) behavioural measures and (3) physiological measures. A detailed review on these measures will provide insight on the present systems, issues linked with them and the enhancements that need to be done to make a robust system. The word “drowsy” is synonymous with sleepy, which quietly means an inclination to fall asleep. The levels of sleep can be categorized as 1.

Awake, 2.Non-rapid eye movement sleep (NREM) 3.Rapid eye movements sleep (REM). The second stage, NREM, can be subdivided into the following three stages [5]. Stage I: transition from awake to asleep (drowsy) Stage II: light sleep Stages III: deep sleep In order to analyse driver weariness, researchers have mostly studied Stage I, which is the weariness phase. The crashes that occur due to driver weariness have a number of characteristics [6].      

Occur late at night (0:00 am–7:00 am) or during midafternoon (2:00 pm–4:00 pm) Involve a single vehicle running off the road Occur on high-speed roadways Driver is often alone Driver is often a young male, 16 to 25 years old No brace marks or indication of braking

In relation to these aspects, the Southwest England and the Midlands Police databases use the following criteria to identify accidents that are caused by weariness [7]:       

Blood alcohol stage below the legal driving limit Vehicle ran off the road or onto the back of one more vehicle No sign of brakes being applied Vehicle has no mechanical defect Weather conditions and clear visibility Avoidance of “speeding” or “driving too close to the vehicle in front” as potential causes The police officer at the scene imagines sleepiness as the primary cause

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 Statistics derived using these criteria cannot account weariness identification system, the strengths of the fully for accidents caused by weariness because of the various measures should be combined into a hybrid complexity involved; therefore, accidents that can be referred system. Table I shows the scale developed by Muzet to to driver weariness may be more devastating than the statistics evaluate weariness during the System for Effective reveal. Hence, in order to escape these types of accidents, it is Assessment of Driver Vigilance and Warning According to necessary to define effective measures to detect driver Traffic Risk Estimation (AWAKE)consortium[8]. weariness and alert the driver.

The electroencephalogram (EEG), which measures brain A driver who falls asleep at the wheel loses control electrical activity, is used here to detect burst of energy in α of the vehicle, an action which results in a crash with either ([8–12] Hz) and in the θ ([4–8] Hz) frequency bands. another vehicle or stationary objects. In order to stop these TABLE I devastating accidents, the state of weariness of the driver OBJECTIVE WEARINESS CRITERIA [8] should be observed. The following measures have been used Objective EEG activity in Blinking and eye widely for monitoring weariness: Weariness score the α and θ bands movements 1.

2.

3.

Vehicle-based measures—A number of metrics, including deviations from lane location, movement of the steering wheel, and pressure on the acceleration pedal are constantly monitored and any change in these that crosses a specified threshold indicates a significantly increased probability that the driver is drowsy[7],[8].

Behavioural measures—The behaviour of the driver, as well as yawning, eye closure, eye blinking, and head pose, are monitored through a camera and the driver is alerted if any of these weariness symptoms are identified [9],[10].

Physiological measures—The correlation between signals (electrocardiogram (ECG), electromyogram (EMG), electrooculogram (EOG) and electroencephalogram (EEG)) and driver weariness has been studied by many researchers.

1.1 Hybrid Measures Each method used for detecting weariness has its own advantages and limitations. Vehicle-based measures are useful in extending weariness when a lack of vigilance affects vehicle control. However, in some cases, there was no crush on vehicle-based parameters when the driver was sleepy, which makes a vehicle-based weariness detection system unreliable.

Figure 1. a. Hybrid Measures Of Weariness Other than these three, researchers have also used subjective measures where drivers are asked to rate their level of weariness either verbally or through a questionnaire. The intensity of weariness is determined based on the rating. These methods have been studied in detail and the advantages and disadvantages of each have been argued. However, in order to employ an efficient 59

0 1 2 3 4

negligible less than 5s less than 5s less than 10s more than 10s

Normal Normal Slow Slow slow

2. RELATED WORK 2.1Material The database used in this study was provided by the Centre d’Etudes de Physiologie Applique (CEPA) from Strasbourg, France. The data were obtained using the “Poste d’Analyse de la Vigilance en Conduite Automobile Simule” (PAVCAS) driving simulator. PAVCAS is a moving base driving simulator made up of a mobile base with four degrees of liberty (vertical and longitudinal movements, swaying, and pitching) and a real-time interactive visualization unit. The visualization unit reproduces the driving conditions on a freeway by day or night very well. Images are shown on five screens in front of the vehicle and are arranged in semicircle.

The database is made up of 40 recordings from 20 subjects.

Each subject was recorded while driving for 90 min, firstly while perfectly rested and secondly while suffering from sleep deprivation (the subject slept only 4 h the previous night) in diurnal conditions. The database is thus composed of 60 h of driving information. Each recording gives four EEG channels [left frontal (F3), central (C3), parietal (P3), and occipital (O1)], one EOG channel and a video of the driver’s face. Data acquisition of physiological signals was performed at 250 Hz. Objective sleepiness was evaluated on each recording by an expert doctor using the scale described in Table I. This expertise of the data represents the ground truth as the doctor expertise is actually the surest evaluation of weariness.

Moreover, self-assessment of weariness is often wrong [10].

Half the data, randomly chosen, were used to select and tune certain parameters of the weariness detectors, such as the length of the analysis windows, the optimal values for the thresholds used and the choice of different features. This dataset is named data-set 1. The different parameters are empirically tuned on these data-set using ROC curves. ROC curves display the ratio of correct detections in function of the ratio of false alarms. Hence, to tune each parameter, we made its value vary. The value corresponding to the point of the ROC curve the closest to the optimal one (100% true detections, 0% false alarms) was then selected. The other half was used to evaluate the system. This data-set is named data-

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 set 2. Each data-set contains data recorded on different drivers and the same distribution between awake and drowsy states.

2.3 Weariness Detection Based on Blinking Analysis 2.2 Weariness Detection from a Single EEG Channel

Weariness can also be monitored using visual signs (cf. Table I). Blinking is particularly a good indicator of weariness.

Expert doctors traditionally use the electrooculogram (EOG), which is the measurement of electrical eye activity, to detect and characterize blinking. Fig. 4 shows an example of the EOG signal measured when a blinking is occurring and an illustration of the EOG derivative (velocity).

Fig. 5 shows an example of 20 s of EOG signal during awakeness (a) and stage 3 of weariness (b). It can be seen in Fig. 5(a) that there are less blinks than in Fig. 5(b). Moreover, they are sharper in Fig. 5(a). On the contrary, the blinks in Fig. 5(b) are larger, long eyes closure appears (see seconds 10 to 12) and the dynamical features (slopes, etc.) are lower.

These two examples illustrates criteria used in Table I. EOG measurement is nevertheless not very practical for the driver as two electrodes must be placed on his face. Blinking can also be identified by a video analysis of the driver face which is more practical but gives rise to new problems (identification of the eyes and the driver is constantly moving).

EEG is the measurement of the electrical activity of the brain by electrodes placed on the scalp according to the international 10–20 system. EEG is analyzed in the frequency domain, where rhythmic activities are measured in several physiologically significant frequency bands. Weariness is characterized in EEG by an increase of the α ([8–12] Hz) and θ ([4–8] Hz) activities [7], which are linked to relaxed and eyes closed states, and a decrease of the β activity ([12–26] Hz), which is linked to active concentration. A few studies have also shown modifications in the δ ([0.5–4] Hz) due to weariness but they mainly appears in advanced weariness state (δ activity is linked to deep sleep in particular) and the purpose here is to detect early signs of weariness. These variations of activity mainly appear in the parietal and central areas of the brain. The red boxes in Fig. 1(b) underline α activity used by the doctor to expertise the signal. The total length of this activity is less than 5 s which corresponds effectively to level 1 of weariness according to Table I. It can be seen in Fig. 1(a) that no such activity is present and the driver is considered as awake. The general principle of this system is shown in Fig. 2.

Figure 5. Twenty seconds of EOG signal (a) during awakeness and (b) during stage 3 of weariness.

Figure 2 : 20 seconds of EEG signal during (a) awakeness and (b) during stage 1 of weariness

Previous work has shown that some of the blinking features such as D50, P80, F, and the amplitude–velocity ratio A/PCV can be extracted with the same accuracy from the analysis of a high frame rate camera (200 fps) as from the EOG. For this reason, only these features were studied to design the weariness detection system from visual signs. The system proposed could then be used with a high frame camera instead of EOG electrodes as presented. D50, P80, F, and the amplitude–velocity ratio A/PCV are automatically extracted from the blinking detected in the EOG signal according to the proposed method. This method proposes to locate blinks by analyzing the EOG velocity and characterizing the EOG behavior during an ideal blink as shown in Fig. 4. The EOG velocity (derivative of the EOG signal) is used to detect closing and opening slopes. Duration and amplitude criteria based on expert rules are then applied to validate the detection and avoid artefacts. In the future, the purpose is to automatically extract these features from the analysis of a high frame rate camera. In this paper, the system is presented using visual signs extracted from EOG. 2.4. Fusion System Using Decision Cascading

Figure 3. EEG-based weariness detection method.

The average relative power in α and β band, which is the ratio between the power computed in a specific frequency band and the total power of the EEG, is calculated from the EEG power spectrum, computed using a short time Fourier transform (STFT). They are respectively named α rel and β rel. Then, a mean comparison test (MCT) is processed on these energy signals. At the same time, a variance comparison test (VCT) is carried out to detect artefacts.

Two systems to detect weariness from different sources have been presented in the previous sections. A decision system merging the decisions made by the two detectors is now presented. In the following, to ease the notation, the

Figure 4. Windows used for the MCT. 60

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 expression epoch (j) will be omitted. Thus, DEEG, Dvisual, A 10 s delay is induced on both EEG and EOG decision. The and Dart now stand for DEEG (epoch (j)), Dvisual (epoch (j)), fact that the final decision is given on 20 s epoch add another and Dart (epoch (j)).

10 s delay. Hence, the final delay is 20 s. We assume that this 1) Principle: The fusion system is inspired by the OSS scale delay is not an issue as the system detects early stages of which is used by expert doctors to evaluate weariness. This weariness. scale is presented in Table I. It can be seen on this table that weariness appears first in brain activity and then in visual 3.RESULTS AND DISCUSSION activity. Cerebral weariness signs appear from the level one of weariness, whereas visual weariness signs appear from level In this section, the EEG-based detection system, the 2. Moreover, weariness is characterized by both cerebral and EOG based detection system, and the weariness detection visual signs from level 2 weariness. The weariness detection merging the two decisions are validated on the second part of system presented in Fig. 6 is therefore proposed. It provides a the database, data-set 2, which was not used to tune the decision for every non overlapping 20-s epoch, epoch (j) parameters of the different detectors. The results obtained are which stands for the non overlapping 20 s epoch to which the compared with the expert doctor classification. The results are jth second belongs. These epochs are the same as the expertise presented using confusion matrices. The columns refer to the epochs. When weariness is detected by the EEG-based expert’s classification and the rows to the classification detection system and not by the EOG-based detection system, performed by the method. The percentages are computed on the driver is considered to be sleepy which corresponds to each column. The number of epochs in the validation set stage 1 on the OSS scale. When weariness signs are detected present in each class is displayed in the last row. The purpose by both EEG-based and EOG-based systems, the driver is here is to find a good compromise between a large number of considered to be “very sleepy,” which corresponds to stage 2 correct detections of “drowsy” epochs and a low number of and above on the OSS scale. If else, the driver is considered to false alarms. be “awake” (level 0 on the OSS scale).

3.1 Results of the EEG-Based Detection System The weariness detection system based on the EEG analysis reaches 84.6% correct detections of “drowsy” states, which corresponds to a level of weariness greater or equal to 1 in the OSS scale (see Table I) and 17.9% false alarms (epochs classified 0 by the expert and detected as “drowsy” by the system), [5]. These results are presented in Table III.

Figure.6. General Architecture of the fusion system.

TABLE III Confusion Matrix Of The EEG-Based Design

Figure.7. Complete architecture of the weariness detection system.

If weariness signs are identified on the EOG, the driver is then considered to be “very drowsy” (level 2 and above in the OSS scale) because visual weariness signs appear only from this weariness level. The appearance of an artefact is then interpreted in the second way as weariness sign is found in the same time. This decision mechanism is summarized in Table II, where the truth table is presented. In this table, the last decision is called Dfinal. It takes the value 0 if the driver is assumed to be “awake,” 1 if he is considered to be “drowsy” and 2 if he is considered to be “very drowsy.” The method is applied on-line.

TABLE II

3.2. Results of the EOG-Based Detection System The weariness detection system based on the blinking analysis reaches 81.4% correct detections of “very sleepy” states, which corresponds to a level of weariness greater or equal to 2 in the OSS scale (see Table I) and 13.1% false alarms (epochs 0 or 1 by the expert and detected as “very drowsy” by the system). These results are presented in Table IV. Note that in this case, the number of awake states is different from the number of awake states presented in Table III. Indeed, “awake” means any state classified as 0 or 1 by the expert while in Table III, “awake” means states classified as 0 by the expert. The blinking analysis reaches 81.4% correct detections of “very drowsy” states, which corresponds to a stage of weariness greater or equal to 2.

Truth Table Of The Decision Cascading System

61

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 TABLE IV On average, the percentage of confusion in adjacent classes is about 15%, which is probably close to inter-expert Confusion Matrix Of The EOG-Based Method disagreement (percentage of epochs classified in different classes by different experts). Between the EOG-Based System and the EEG/EOG Fusion System To evaluate the interest of merging EEG information and EOG information to detect weariness, a comparison between the EOG-based weariness detection (presented in Section II-C) and the detection of the “very drowsy” states by the EEG/EOG fusion system is made.

In this case, the EEG/EOG system is considered to be a binary detector for the “very drowsy” class. An epoch is therefore 3.3 Results of the EEG/EOG System considered to be “awake” if it was classified as “awake” or The results obtained by the fusion system on the detection of “drowsy” by the EEG/EOG system. In this way, the EOG“drowsy” and “very drowsy” states are shown in Table V. based detection system and EEG/EOG detection system can The results of the weariness detection are good with 72.6% be compared because they both only detect “very drowsy” correct detections of “drowsy” states (level 1 in the OSS states (greater or equal to 2 in the OSS scale) and the driver is scale) and 79.6% correct detection of “very drowsy” states otherwise considered to be “awake” (level 0 or 1 in the OSS (level ≥ 2 in the OSS scale). The discrimination of “awake” scale). The confusion matrix gives the EEG-/EOG-based states (level 0 in the OSS scale) is also good with 82.1% system design in this case is presented in Table VI. correct detections.

TABLE VI TABLE V Confusion Matrix Of The Eeg-/Eog-Based Design Confusion Matrix The EEG/EOG Detection

The comparison between the results of the EEG/EOG system (see Table VI) and those of the EOG-based system (Table IV) shows that the number of false alarms decreases sharply when the EOG information is completed by the EEG information to detect advanced weariness. The FPrate decreases indeed from 13.1% to 5.2%. This decrease represents 255 epochs correctly classified as “awake” by the EEG/EOG system that were wrongly classified as “very drowsy” by the EOG-based system. At the same time, the number of perfect detections decreases only slightly from TPrate = 81, 4% to TPrate = 79, 6% when the EEG information is added. It represents only ten epochs wrongly labeled as “awake” by the EEG/EOG system and correctly detected as “very drowsy” by the EOG based system.

The results obtained show that the relevancy of EOG weariness detection is increased when EEG information is merged with blinking information. EEG information allows the epochs likely to be classified as “very drowsy” to be filtered when only the EOG is used. The redundancy between EEG and visual information is used to increase the relevancy of the detection. In this way, 255 epochs where non relevant visual signs of weariness were detected by the EOG-based system are correctly classified as “awake” by the EEG-based detection system. Moreover, only ten “drowsy” epochs more are wrongly classified as “awake” when the EEG information is merged. This confirms that the EEG-based detection leads to good detection of “very drowsy” states. Most of the epochs wrongly classified as “awake” by the EEG-based system are “drowsy” ones.

The results collected by this method are presented in Table VIII. The overall true detection rate is equal to 75.8% which is less good than the cascading system (80.6%). It can

The overall true decision percentage, which is the rate of epochs correctly classified in the three classes (“awake,” “sleepy,” and “very sleepy”), is 80.6%, which show that the detection system is efficient. The rate of “awake” epochs which are misclassified is the same as the one obtained with EEG-based detection (17, 9% = 14, 5% + 3, 4%). Indeed, considering the false alarm rate for the “awake” state (percentage of epochs classified 0 by the expert and detected as “drowsy” or “very drowsy” by the system), the EEG/EOG system performance is limited by the performance of the EEG-based detection system. When the EEG system identifies an epoch as “drowsy,” the decision cannot be changed into “awake” by the EOG-based system. It can only be confirmed as “drowsy” or detected as “very sleepy.” Then, any false alarm generated by the EEG detection system (“awake” epochs detected as “drowsy”) remains a false alarm for the EEG/EOG system. It can be seen in Table V that the EEG/EOG-based detection leads to very good discrimination between “awake” and “very drowsy” states since only 3.4% of “awake” states are misclassified as “very drowsy” and only 3.7% of “very drowsy” states are misclassified as “awake.”

Mix-ups mainly appear between adjacent classes. Indeed, most of the misclassified “awake” states are classified in the “drowsy” class (14.5%) and not in the “very drowsy” class (3.4%). The misclassified “drowsy” states are mixed up with both “awake” (14.0%) and “very drowsy” (13.4%) classes.

The misclassified “very drowsy” states are classified in the “drowsy” class (16.4%) and not in the “awake” class (3.7%).

62

Vasavi dobbala et al., International Journal of Emerging Trends in Engineering Research, 1(4), December 2013, 58-63 be seen that the rate of true “very drowsy” detections is the based identifier, and an EOG-based identifier. The EEG same as that obtained by the EOG-based detection system: detection system is applied by merging several features 81.4% (see Table IV). This result was predictable as only the extracted from a single EEG channel analysis using fuzzy EOG-based weariness detection is used to detect “very logic. Weariness detection using visual information of the drowsy” states. This result is very similar to the rate of true driver is carried out by using fuzzy logic merging several “very drowsy” detections obtained by the fusion system using features extracted from the blinking analysis of one EOG cascading rules (79.6%). channel. The final system merges these two systems according TABLE VII to medical rules to evaluate weariness. The method proposed Results Obtained By A “Blind” Fusion System allows weariness to be detected with three different levels (“awake,” “drowsy,” and “very drowsy”). It reaches an overall perfect detection rate of 80.6%. The interest of the approach is to use the complementary nature of EEG and visual information as well as their redundancy in detecting weariness.

ACKNOWLEDGEMENT I would like to express my gratitude to the following people for their support and guidance for the success of this paper. I would express my deep sense of gratitude to, thank Dr. R. V. Krishnaiah Principal, DRKIST, Hyderabad ,A.P, INDIA.

The detection rate of “drowsy” epochs correctly classified is the same for both fusion techniques (72.6%). Indeed, the way of detecting “drowsy” states is the same for both fusion technique: the driver is considered to be “drowsy” if weariness signs are detected in the EEG and not in the EOG. However, the detection rate of “awake” states is sharply decreased by the fusion technique using a logical “OR.” Indeed, the rate of correctly classified “awake” epochs decreases from 82.1% with the EEG/EOG system using cascading rules to 75.2% for the fusion technique using a logical “OR.” This means that 190 epochs correctly classified as “awake” by the fusion technique using cascading rules are wrongly classified as drowsy states by the fusion technique using a logical “OR.” Indeed, EEG-based detection is used by the cascading system to filter the epochs where non relevant visual signs of weariness are identified. In the case of the fusion technique using a logical “OR,” only the EOG-based identification is used to detect “very drowsy” epochs and the EEG information is not used to evaluate the relevancy of the visual signs identified. It results in an increase of the number of epochs classified as “very drowsy” instead of “awake” (from 3.4% for the cascading system to 9.9% for the “blind” system). Confronting the EOG and the EEG weariness detection allows the epochs where the visual signs detected are not relevant to be filtered. Then these false “very drowsy” detections are avoided.

REFERENCES [1] J. Connor, “The role of driver sleepiness in car crashes: A review of the epidemiological evidence,” in Drugs, Driving and Traffic Safety. New York: Springer-Verlag, 2009, pp.

187–205.

[2] J. Yang, Z. Mao, L. Tijerina, J. Coughlin, and E. Feron, “Detection of driver fatigue caused by sleep deprivation,” IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 39, no. 4, pp. 694–705, Jul. 2009.

[3] T. Pilutti and G. Ulsoy, “Identification of driver state for lane-keeping tasks,” IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 29, no. 5, pp. 486–502, Sep. 1999.

[4] Q. Ji, P. Lan, and C. Looney, “A probabilistic frame work formodeling and real-time monitoring human fatigue,” IEEE Trans. Syst., Man, Cybern. A, Syst., Humans, vol. 36, no. 5, pp. 862–875, Sep. 2006.

[5] A. Picot, S. Charbonnier, and A. Caplier, “Monitoring weariness on-line is using a single encephalographic channel,” in Recent Advances in Biomedical Engineering. Rijeka, Croatia: IN-TECH, 2009, pp. 145–164.

[6] G. Renner and S. Mehring, “Lane departure and weariness—Two major accident causes—One safety system,” Transport Res. Lab., Berkshire, U.K., Tech. Rep., 1997.

[7] M. Gillberg, G. Kecklund, and T. Akerstedt, “Sleepiness and performance of professional drivers in a truck simulator— Comparisons between day and night driving,” J. Sleep Res., vol. 5, no. 1, pp. 12–15, Mar. 1996.

[8] A. Muzet, T. Pébayle, J. Langrognet, and S. Otmani, “AWAKE Pilot Study No. 2: Testing steering grip sensor measures,” CEPA, Gatineau, QC, Canada, Tech. Rep. IST2000-28062, 2003.

[9] T. Vöhringer-Kuhnt, T. Baumgarten, K. Karrer, and S. Briest, “Wierwille’s method of driver weariness evaluation revisited,” in Proc. Int. Conf. Traffic Transp. Psychol., 2004, vol. 3, pp. 5–9.

[10] T. Akerstedt and M. Gillberg, “Subjective and objective sleepiness in the active individual,” Int. J. Neurosci., vol. 52, no. 1/2, pp. 29–37, May 1990.

4. CONCLUSION A weariness detection system using brain and visual information has been presented in this project. Weariness detection using visual information of the driver is carried out by using fuzzy logic merging several features extracted from the brain analysis of one EEG channel; Camera is also used to find out weariness of the driver. Using a video camera to extract visual information from the driver would be more comfortable for the driver. Using highly advanced ARM9 board and with the help of growing technology the project has been successfully implemented. Brain information is obtained by EEG analysis, and visual information is obtained by EOG analysis. This system uses cascading rule decision to merge two weariness detection systems, an EEG63