MATLAB Based Drowsiness Detection System Using an Array of Sensors and Fuzzy Logic Paul Russel B, Agustin, Rochelle C, Lazaro, Marl Dobn L, Mazo, Alvin Cedrie P. Vidal, Evelyn Q. Raguindin, Mark Angelo C. Purio, Electronics Engineering Department ‘Adamson University ‘Manila, Philippines usselagustin@ hotmail.com, rochellelazaro2893@iyahoo,com, [email protected], alvine [email protected], ‘[email protected], [email protected] Abstract This paper presents a hybrid driver’s drowsiness etection system comprising of three measurement methods: Pulse Rate Monitoring, Eye Monitoring, and Head Movement Monitoring to increase the overall accuracy and reliability of drowsiness system using Fuzzy Logic simulation in MATLAB Keywords— fuzzy logic, drowsiness, pulse oximetry, image processing, PERCLOS 1, IstRopucrion Sleep is « Biological Enigma, About one-third of our lives (200,000 hours) is spent on sleeping." Sleeping habits vary among individuals!) While a person needs a sleep time of & hours for a great day's performance and mood, another of the same age and gender may need 9 hours of sleep or more forthe same output level ‘The National Sleep Foundation summarized the daily sleep need of # person in accordance to their age as follows: new- ‘porns (0-2 months old) need 12 to 18 hours of sleep a day, infants (3 — 11 months) need 14 to 15 hours, toddlers (1-3 years) need 12 to14 hours, pre-schoolers (3-5 years) need 11 to 13 hours, school-age children (5-10 years) need 10 to IL hours, teens (10-17 need 8.5 to 9.25 hours, and adults need 7 to 9 hours of sleep a day. These sleep needs are determined by the Basal Sleep need of an individual that changes as one ages"! Sleeping time also varies. A teenager and a young adult cean become more active at night because this is what their bodies tell them, so they often sleep very late at night but their lifestyle as a student tells them to wake up early for morning class. As a result, they would show symptoms of being sleepy during classes and will accumulate sleep debt over time.” For others, like night shift drivers and business professionals, sleeping fulltime at night just so happens to be not an option as their occupational demands and other responsibilities prohibit them to do so. As a consequence, they accurmulate Sleep debt duc to physical conditions and will certainly become sleep deprived. These sleep debt must be paid at one point unexpected and most likely unwanted, and this sometimes happens behind the wheels, 978+1-4799-4075-2/14/831.00 © 2014 Fatigue or sleep deprived drivers show slow reaction time, poor judgement ofthe situation and lack of awareness ~ skills that are of utmost importance while driving.” Dozing off behind the wheels while driving is proven to be equally dangerous and catastrophic as drunk driving Actually, it offers no difference in claiming lives of both the innocent and the offender. According to the survey conducted by the proponents, 1 out of 10 respondents, who are bus drivers, drive an average of 4-8 hours daily shift; 3 drive for 8-12 hhours; and the remaining 6 drivers drive for more than 12 hours. Filly percent (50%) of the respondents honestly admitted that they always feel sleepy during long drives and resolved to various activities to keep them awake. The risk of ‘occurrence of car accidents increases to four up to six times ‘more when the driver is sleepy compared to when driver is awake, TL. Construction oF ti: Prosecr A. Mechanism ‘The proponents constructed a system that is able to detect the drowsiness level of the driver. The fuzzy logic in MATLAB is used to matex all the values that was detected ‘from the driver B. Hardware ‘The system is composed of a eamera forthe eye movement and head position monitoring that is mounted on the ‘dashboard of the car, anda pulse oximeter for the detection of the pulse rate of the driver. The alerting system is composed fof a buzzer, LED array, vibrator, and water sprinkler. From the prior aris and existing documents related to drowsiness dotection, and also with the advice of professionals in the field, the project should have the following devices: A camera to capture and record the facial region of the driver, @ ive pulse rate sensor fo acquire the changes in the pulse ‘the driver, a processor to analyze and process the data given by the sensors, active device, such as a vibrator, to alert the driver.ig. .Concepul amewoe ofthe drowsiness detection system, Figure 1 shows the inputted parameters to be used in determining a drowsy driver such as eye movement, head position, and the pulse rate, The eye movement and head Position are obtained from the camera and the pulse rate from the oximeter. IL also shows how the inputted values are integrated to the MATLAB software and how it is used in decision-making by a drowsy driver using its fuzzy logic controller with fuzzy logic algorithm. Figure 2 shows the block diagram of the “MATLAB based drowsiness detection system using an array of sensors and fuzzy logic”. The parameters are used to detect the drowsiness of the driver. These parameters include the pulse rate detected using oximeter and the eye movement and hhead position detected using @ camera, The camera will record the head position and the eye movement of the driver and then relay it to a processing unit to process the video using MATLAB software. Each parameter has predetermined threshold values, that when exceeded would indicate a drowsy state, If it is measured that the driver is Gozing off, 2 response will be triggered depending on drowsiness intensity level Figure 3 shows the function of each component and how will i integrate to the main device and main computer. a ee es ven | stn Fy [HEAD FOSITION ae = - Fig. 2. Blok digram of he drowsiness dvecton system, igure. Hardware inerice of the tem ©. Software Considerations & Implementation ‘The proponents use the fuzzy logic in MATLAB which ‘computes for the values of three parameters used to detect the driver's drowsiness. Also, itis the one responsible to twigger the alarm system once the values are computed. Drowsiness Level for Bach Parameter For determining the overall drowsiness level, each parameter contributes their own level of drowsiness according to their individual methods of measurement; these values are combined to achieve a final single level, fof drowsiness. The tables below show the designated ‘values in each parameter ‘TABLE I, DROWSINESS LEVEL FoR PULSE RATE MONTORIG PRR VALUE Pulse Rate Rati (PRR) ~ NINO : TSPRR> 098 ° (4% TaeEPRRS 08s | [Ni Instantaneous Pose Rate D9SEPRR> 092 [No Norma Pulse Rate OOS PRR 3 ‘Table | shows the assigned values for each level in the pulse rate data. Dt. Krauchi (2001) in his study detected an average drop of 64 to 52 beats per minute, about 8% decrease fon the normal pulse rate (No) after reaching a light and continuous sleep. Since there's a gradual decrease, lower Pulse Rate Ratio rating means higher level of drowsiness. On the table presented above, Drowsiness Level contributed by Pulse Rate Monitoring (40% of the final value) is evaluated with specific pulse rate drops, A pulse rate drop of 2% will sive a pulse rate ratio of 0.98 which concludes a drowsiness level of 1. Increasing pulse rate drop will yield lower Pulse Rate Ratio (PRR) and results to higher Drowsiness Level ‘TABLE TL DROWSINESS VEL FoR EVE MOVEMENT MoNIrORING ‘PERCLOS Pacenage (Pm) -CS*100 | PERCLOS | VALUE os) eP=Dw ° ‘oualchadtine(o)-28) _ [fos0zP<0H0] tT DMP 3‘Table 2 shows the assigned percentage values of eye closure over the sampling time on the different level of Growsiness. The drowsiness level contributed by Fye ‘Movernent Monitoring (35% of the final value) is evaluated through the percentage of the total Blinking time. The total Blinking Time or Total “Eye” Closed Time is the summation ‘of blink counts throughout the sampling count. If the PERCLOS (P) percentage fits on the range of 0% to 20%, the O8D] 1 Region oaeF>om | 2 a-FullFace Arca as0SF 3 ‘Table 3 shows the number of decrease in the area of the face will determine the drowsiness level. It explain that 10% to 30% decrease on the facial area will conclude drowsiness detection, In terms of Head Position Monitoring, ‘The drowsiness level (25% of the final value) is determined by the ratio of the Facial Area present at accepted region and the Full Face Area. Lesser percentage of face area present in the accepted region dictates a higher level of drowsiness level, "TABLE IV. Ovekals Daowsintss LEVEL AN OUTPUT Reston al 0% + BERCLOSSl 350 + EaSNALNAL2S90~ A | Bova Leet oo THERGA [Teva | + Barr a PONTO > Buzzer and La PWN IOON) Table 4 indicates the allocation of output values of {drowsiness level fr all the three responses combined and the ‘corresponding system response on cach level of drowsiness, Using the formula to come up with the final ouput, the proponents has given each parameters a different weighting ‘based on their effectiveness in detecting drowsiness, ‘The researcher will use an application called MATLAB to integrate the device and form an algorithm, fuzzy logic, to detect drowsiness. In addition, the researcher will make also a Graphical User Interface (GUD to the MATLAB for it to be a user-friendly device User defined rules must be established and specified at the rule editor part of the Fuzzy Inference System, These rales will be the building block of the whole system and will be the basis of the fuzzified output. The first procedure in defining the set of rules isto identify how many are needed by the system. In this paper, the system combines the level of «drowsiness from each single parameters and generate a single fuzzified output; this is analogous to binary truth tables in which every combination of the inputs must be considered. ‘The system has three inputs and four levels of drowsiness for ceach input, thus No.of Rules=4”3=64, Weights were specifically allocated, as one (I) to bring that each rules regardless of its input parameters used would entail equal implementation throughout the system. Some fuzzy sets of rules are as follows: 1, If (Pulse Rate is level 0) and (Eye Movement is level 0) and (Head Position is Tevel_0) then (Drowsiness_Level is level 0) (1) 4. If (Pulse Rate is level 0) and (Eye Movement is level_0) “and (Head Postion is. Tevel 3) then (Drowsiness_Level is level_1) (1) 60, If (Pulse_Rate is level_$) and (Fye_Movement is level 3) and (Head Position is evel 0) then (Drowsiness_Level i level 2)(1) 63. If (Pulse Rate is level 3) and (Eye Movement is level 3) and (Head Position is Tevel 3) then (Drowsiness_Level is level 3) (1) D. Presentation, Analysis and Interpretation of Results Procedure for Data Collection ‘+ Data are to be gathered fiom the results of the testing process Information and test resulis from related works should be examined and is then compared to the ‘actual study in order to seta ground truth, + Testing different apparatus/instruments , such 4s sensors and camera, is also a concer $0 that ‘the devices chosen forthe final assembly of the project can handle what it needs to do and can sve accurate results ‘© The trials are to be implemented inside a car oF in an environment which would induce drowsiness in drivers during real driving operation, + Different factors such as light intensity, skin ‘color, driver's behavior, ee. are also considered for it affects the result ofthe system, + A variety of alert mechanism should also be tested and afterward analyzed to determine ‘what alarm works for most subjects Punctionality Test The accuracy of detection of the device will be the device main functionality. By having three parameters for detection of drowsiness, the device will provide a‘more accurate and improved results compared to others ‘Since it has several evel of detection, each level must be properly tested using different values from the three input parameters that can trigger that specific level of election, To est for the range of distance that the system ‘would recognize the face and track the eye, the device must be operated on different positions with respect to the driver, Under normal conditions, the minimum and. ‘maximum distance would be established, ‘To make the functionality and reliability of the device even more defined, response time delay introduced by the system ilself will be theoretically computed and expected to be small in value and negligible as possible to give a real-time system response, Using the rule viewer of MATLAB, possible input and ‘output combinations were determined from the set of rules implemented. The input parameters such as pulse rate, eye ‘movement, and head position can be change such that various input parameters with the assign output parameters ccan be verified il Fig. 4. Rules Viewer for each putamte of inputs nd ouput Figure 4 shows the rules determined by the researchers using the Rule Viewer, Using the rule viewer, the values of each inputs can be tested by dragging the red vertical lines. The values of each parameter is displayed on {op of the column that represents them, also, input variables can be specified by putting their values in an array format at the input panel located at the bottom of the window. After defining the rules of the fuzzy system, Surface plots can be viewed to graphically represent the system, The surface viewer selects two inputs and one ‘output ftom the system and generate a 3-D graphical representation oft, I is important to display possible output waveform combinations to deteet probable faults in the implementation of fuzzy rule sets. The figure that follows shows the surface plot of three possible combination of two inputs and the graphical representation of it in determining the fuzzitied output, org Seen ig. S. Mlotedsstace or Bye Movernon and Pale Rate with its coesponding rowsines Hv ouput Fig. 6 Potod surfs for Heed Poston nd Eye Movement with ts comesponding dowsing vel ouput Fig 7. Plo src fread Poon and Palic Rate with ts coresponding rosie level oupatAs shown in Figure 5, as the values of pulse rate creases with increasing eye movement data values, the drowsiness level also increases, It is similar withthe Figure 7 with decreasing pulse rate and inereasing head movement values, higher output level values will conclude. Also, Figure 6 shows the inereasing values of head position and «eye movement will lead a higher level of detection, The proponents have the testing done and able to have the values for detecting the accuracy, distance and time response, Fig. The GU ofthe Drowsiness Detection Device Figure & shows the GUI to be used by the Drowsiness Detection Device to easily understand and manipulate testing results to verify output action inthe electronic device prior to detection, MI, Test RESULIS A. Establishing Distance In order to identify the capabilities and limitations of the device, the proponents performed various testing to determine the maximum and minimum distance that the evice can detect the user's head under two specific conditions, with and without glasses, TABLE V. ONE-F00% DISTANCE TAL WriioUr GLASSES Nase When the head is not detected, the area where the eye is displayed will not refresh, Thus, no eye detection as Well Table § shows that for a distance of 1 foot, 27 out of 30 trials (90 %) detects the head of the user while 25 out of 30 (83.33%) detects the eye ofthe user at 1 f. without glasses en En Table 6 shows that 17 out of 30 (56.67%) detects the head and the eye of the user at 1 foot distance with lasses. Head Detection Without Glasses VS. With Glasses ———S—_— Ih 28 GR oaks Head Detection without Gases Head Detection with Gnee Fig 9, Head Detection ComparisonEyeDetection Without Glasses VS. With Glasses —. 10 Fray er ST —tyeDutection without Gasset —FyeDereation wth Glasses Fig. 10, Bye Deection Comparison Figures 9 and 10 presents a graphical comparison of eye detection and head detection for users with and without glasses respectively B. Accuracy Testing Accuracy testing of the project is defined as the number fof correct response over the number of samples, The researchers have designated any values of cach parameter and inputted in the GUL of the device in which drowsiness ‘output level shown in the GUI, It will then compare to the computed values by the researchers to determine the accuracy of the program itself ©. Time Response Testing Time response of the device happens to be a real time response. With that, the researchers conducted some ‘computations to verify its real time basis response of alarms {rom computer to the main device IV, CONCLUSION AND FUTURE WORK ‘The proponents conclude that developing a hybrid system using an array of sensors that is a MATLAB besed is more efficient and reliable rather than using the existing vices, which measures one or two parameters only For the measurement of the pulse rate, a pulse oximeter is being used while for the eye closure and head movement, ‘ camera mounted in the ca isto be used. The pulse rate is fone of the parameters considered in measuring a driver's drowsy state simply because it has @ medical basis, In terms of the camera usage, the eye closure and the head movement are to be measured since it can also be seen in the behavioural aspect of a person if he/she is in the state of dozing off. A MATLAB software is used 10 ulilize the three parameters that will be inputted into the system, A very ‘good skill in programming is a must in using fuzzy logie in MATLAB. Fuzzy logic has a very wide coverage when programming is the topic. You should be familiar with the environment of MATLAB software as well, ‘There are criteria to be consider to know if the system is good and effective one. In the case of determining the drowsy state ofa driver, distance from the mounted camera in the ear is one ofthe factors to be considered in testing the device. Also, the time it will take for the system to respond in the measured value of ether pulse rate or the detection of feye closure or head movement. The accuracy is the main target of the proponents and safety is the main goal of this project implementation ACKNOWLEDGMENT ‘The authors would like to thank Engr. B. Raguindin, Engr. A. Ganggangan, and Engr. J.D. Ramos for their support and assistance throughout the study and sharing their knowledge, which gave way toils completion, [REFERENCES MMe Steep unt, Rewieved fom Iatpsirsivdssi com tay seep deb [2] Healthy: Sleep Habis and Deprsion 2013). Rereved fom psi beatlinecomfealh depression babisfrsealty seep [5], How Mock Slesp Do. We realy Need? (ad) Reviewed fom np slepfoundaencighhow slexp-worhow-nuebslespdowe realy-neet [A] How Much Slep Do Adult Neal? (ad). Revived fom ‘ntptslopfoundaten eg seep plata whitpapestowuch sleepdocataleneed (5) Blooraeld, J, Harder, KA, and Chiba, BJ. (2008), The Ett of| ‘Sleep Deprivation on Brving Perera (8) Drowsy Driving Jus as Dangerous as Drak Desig: Study 2012. Reureved from hip:/twwo aulTiogeaps. om 20120831 drowsye ving dangerous unk teopy 91597200 ad (7) Kine 86, Digus, TA, Neale, V.L, Sudiweeks, 1D. and Ramsey, DJ (206), The Impacto Driver Iatention on Newer Risks At Analysis Using the 100 Car Nualiste Driving Std Dat,