STUDY OF OCULAR INDICATORS CHARACTERISTICS TO DEVELOP FATIGUE DETECTION MODEL ON DRIVING SIMULATION

• Main Author: Arlini Puspasari, Maya
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51531
• Institution: Institut Teknologi Bandung

Fatigue is one of the main causes that increases the risk of accidents in the road transportation sector. The most dominant factors affecting fatigue are sleep related factors (which consist of time of day and homeostatic) and work related factors. Ocular indicators are physiological measurement tools based on the characteristics of potential human eye changes to measure fatigue. In general, ocular indicators consist of blinking, saccadic, and pupillary indicators. In previous studies, there were differences in the characteristics of ocular indicators as a function of the causes of fatigue. In addition, existing research still does not agree on which parameters of the ocular indicators that are reliable for detecting fatigue. Besides, it is still difficult to know when fatigue onset occurs and when fatigue becomes so severe that it endangers the driver, so it is necessary to develop an ocular indicator based fatigue detection model. This study aims to examine the characteristics of ocular indicator changes in evaluating driving fatigue. These characteristics are influenced by factors that cause fatigue (time of day, sleep duration, and work characteristics). The results of this study are in the form of ocular parameters that have the highest level of accuracy, sensitivity, and specificity in fatigue evaluation and fatigue detection models that contain awake, low-level fatigue, and heavy fatigue cutoff. This study was conducted experimentally using a driving simulator by manipulating the factors of time of day, sleep duration, and traffic density as work characteristics. Experiments carried out by a mixed design method in which there are 8 combinations of treatments that will be counterbalanced. In the time of day factor (morning and evening), the participants involved are between subjects design, while the factors of sleep duration (adequate sleep and lack of sleep) and traffic density factors (high and low) of the participants involved within subjects design. The experiment was conducted involving 24 participants (aged 31.5 ± 7.2 years with driving experience of 5.2 ± 2.7 years) who were divided into 2 groups, namely the driving group in the morning and the driving group at night. The driving session lasts for 3 hours. The evaluation in this study is based on ocular indicators, consisting of parameters of blink duration, blink frequency, Percentage of Eyelid Closure (PERCLOS), microsleep, saccadic (amplitude, velocity, and duration), Slow Eye Movement (SEM), fixation duration, and pupillary diameter. In addition, fatigue validation indicators used by this study consisted of subjective indicators (Karolinska Sleepiness Scale and Swedish Occupational Fatigue Inventory), performance indicators (line crossing and Psychomotor Vigilance Task), and EEG signal indicators (? + ? / ? ratio). There are three main findings in this study, namely: (1) The parameters that experienced significant changes in the causes of fatigue were only 8 parameters out of a total of 11 parameters, which consist of blink duration, blink frequency, PERCLOS, microsleep, saccadic (amplitude and velocity), and pupillary diameter; (2) Sleep duration is the factor that most influences changes in ocular indicators, followed by time of day; and (3) the traffic density factor does not significantly influence changes in ocular indicators. Fatigue in driving activity is characterized by an increase in the value of the blink duration, PERCLOS, and microsleep and is accompanied by a decrease in the value of the saccadic amplitude and velocity. Characteristic patterns of changes in ocular indicators occur non-linearly which are modeled against driving duration and subjective ratings. Classification of ocular indicators were conducted to develop a fatigue detection model. Parameters of blink duration, PERCLOS, microsleep, and saccadic were used as predictors, in addition, subjective indicators, performance, and EEG signals were used as prediction targets. The results of this classification are blink duration, PERCLOS, and microsleep as the best parameters in fatigue detection. The cutoff value of 189.97 ms of blink duration was established for low-level fatigue. Furthermore, the cutoff value of 360,21 ms (blink duration), 18,8% (PERCLOS), and 2,87/min (frequency of microsleep) were confirmed for heavy fatigue. The fatigue detection model has the accuracy, sensitivity, and specificity level above 80% for heavy fatigue. This indicates that the detection of heavy fatigue conditions is more accurate than the detection of mild fatigue. The characteristics of ocular indicators are described based on the pattern of changes. Under conditions of sleep deprivation, the duration of driving affects exponential changes in most ocular parameters, which indicates the effect of sleep deprived more at the end of driving. On the other hand, in a state of adequate sleep, most ocular parameters change quadratically or insignificantly to the duration of driving, which indicates an increase in fatigue not seen from the beginning to the end of driving or experiencing an increase and then a decrease. This makes the sleep duration factor as the dominant factor affecting changes in ocular indicators, followed by the time of day factor, where sleep deprivation and driving at night cause increased fatigue. On the other hand, the traffic density factor has very little effect on ocular indicators, but high levels of traffic density cause increased fatigue in sufficient sleep conditions. The implications of this study in the context of fatigue management are divided into two aspects, namely: (1) Providing input to the working duration limit for the drivers. This study confirms that the duration of sleep from the driver is a very important factor that influences the onset of fatigue. If the driver is sleep deprived, then this study does not recommend drivers to work especially at night, and only for a maximum of 1-hour in the morning.. If the sleep duration of the driver is sufficient, then the maximum working limit is 2 hours, especially in driving conditions at night. The maximum number of hours worked is determined before the appearance of severe fatigue while driving; (2) Develop a fatigue detection system based on blink duration, PERCLOS, and microsleep parameters in real-time while driving to find out when the driver feels tired, so that preventive measures can be conducted to reduce the risk of road accident. The recommended detection system is to use the blink duration, PERCLOS, and microsleep parameters which will be implemented on a device like a video camera-shaped eye tracker that is placed on the dashboard of the vehicle.