Motivation For Drowsiness Detection Information Technology Essay

Motivation For Drowsiness Detection Information Technology Essay Monitoring the drivers action while driving by examining the manoeuvred of the vehicle can be a very prominent task in order to enhance safety while driving. To differentiate between unintentional and intentional car steering wheel inputs, will be the main key element to be discovered, such as a sudden large steering input could indicate the drivers level of alertness. Almost all the statistics have identified driver drowsiness as a high priority vehicle safety issue. Drowsiness has been estimated to be involved in 10-40 per cent of crashes on motorways [5, 6]. Fall-asleep crashes are very serious in terms of injury severity and more likely to occur in sleep-deprived individuals [8]. Drowsiness influences mental alertness, decreasing an individuals capability to handle a vehicle safely and expanding the possibility of a human mistakes that could lead to deaths and injuries. Furthermore, it has been indicated to slow response time, decreases awareness, and impairs judgment. A drowsy driver is unable to predict when he or she will have an uncontrolled sleep onset [9]. There is an increased interest with respect to the design and advancement of computer controlled automotive applications to overcome those problems by enhancing safety to reduce accidents, increase traffic flow, and enhance comfort for drivers. This thesis presented a way to detect drowsiness in driver non intrusively by warning the drivers, preventing accidents and to improve safety on the motorways. This method is employing Support Vector Machine (SVM) to train the classifier by using steering wheel angle, distance to outside lane and acceleration as an input to the SVM. All the parameters extracted from vehicle parameter data collected in a driving simulator. With all the features, a SVM drowsiness detection model is constructed. ACKNOWLEDGEMENTS I would also like to extend my appreciation to Mr John Mellor Dr Ping Jiang for his assistance in educating, assisting and helping me on the preparation of this thesis and who has supported the work not just financially but also provided very valuable feedback and guiding ideas for the production of this thesis. Chapter 1 This chapter illustrates a general overview of this research. Background information related to the topic of drowsiness detection and support vector machine along with research objectives are introduced. Related literature is reviewed in this section, linking relevant topics to the research presented here. Finally, an outline of the thesis and a brief description on the contents of each chapter are also presented. Introduction The proposed non intrusive drowsiness warning system uses a integration technique comprise of vision sensor to obtain road information and steering wheel angle data logger. Both parameters are taken from road simulation experiment. The system is composed of three main processes; To obtain the road information by calculating the distance of the outside lanes from vision input and extracting the steering wheel angle data. These data are used for training and testing intentions during the modelling of the SVM. To give a proper warning to the driver to eliminate false alarm. It is most important that a drowsiness warning system guarantee safety and reliance. Therefore the system must reliably as well as estimate the driver vehicle state in order to give proper warning. It must also consider driving habits and intention of the driver to be of practical use. Research Aim Objectives The aim of this thesis is to contribute to the study of driver behaviour while driving, through the development and evaluation of a drowsiness driver model system. Non-intrusive is chosen as a method due to comfort to the drivers. The result from the research will be integrated to produce the systems that can be efficient in detecting the drowsiness level at an early stage by giving a warning to them about their lack of attention due to drowsiness or other factors. In other words, they can correct the behavior or stop driving when they in the drowsiness state. This system will need to be robust against model mismatch and disturbances and comfort constraints. The objective of this research is to identify the current drowsiness detection by investigating flexible methods for studying the relationships between drivers manoeuvre performances whiles the vehicle on the move and the physiological driver drowsiness states. This thesis paper outlines the design and development of a system that focuses on drivers drowsiness detection and prediction through the following methods:-. Monitoring the driver behaviour by observing the vehicle manoeuvre stability and performance. Validate and measure the progress by using Specific algorithm. Updating the current performance by comparing with the last action stored in system database. Warning the drivers if the behaviour beyond the thresholds. To increase the detection and its reliability of the prediction, the methods stated earlier will be used. Here we will employ machine learning methods to classify the data of actual human behaviour during drowsiness. This will be done by studying and evaluating the learning phase identification of a driver driving pattern. After that we will look to evaluate the parameters comprehensively. In the detection phase on-line model adaptive identification; model error classification; drowsiness alert model will be studied. By implement a control system mechanism that integrates human and machine for classification of the dynamic model for drowsiness detection using information from various sources for achieving a probabilistic best possible alert. Scope The scope of the thesis is defined as follows: The road manoeuvre will be restricted to simulation environment only. There are no obstacles in the road lane, and thus there is no collision-avoidance aspect to manoeuvre. It is assumed the vehicle will operate with a fix velocity range of 50km/h. Two main parameters will be an indictor for the system detection consists of distance to outside lane and steering wheel angle. Motivation for drowsiness detection. Driver drowsiness is a significant factor in the increasing number of accidents on todays roads and has been extensively accepted [2]. This proof has been verified by many researchers that have demonstrated ties between driver drowsiness and road accidents. Although it is hard to decide the exact number of accidents due to drowsiness, it is much likely to be underestimated. The above statement shows the significance of a research with the objective of reducing the dangers of accidents anticipated to drowsiness. So far, researchers have tried to model the behavior by creating links between drowsiness and certain indications related to the vehicle and to the driver [2,3,4]. Previous approaches to drowsiness detection primarily make pre-assumptions about the relevant behavior, focusing on blink rate, eye closure, and yawning [29,30]. The automobile business also has tried to build several systems to predict driver drowsiness but there are only a few commercial products available today[31]. The systems do not look at driver performance and overlook driver ability and characteristics. Naturally, most people would agree that different people drive differently. The system that being develop able to adapt to the changes of the drivers behaviour. Contributions The contributions of thesis research extend to five areas. The introduction of a fully integrated drowsiness warning system with specific algorithms to detect driver condition. The main contribution of this study is it contributes an algorithm of drowsiness driver detection and tracking which based on incorporation of vision and vehicle performance parameter. The implementation of support vector machine for robust and accurate drowsiness warning system. The input incorporation from vision and data logger provides an efficient method for detecting drowsiness driver under varying mode and road conditions. Consideration of various type of driver with various conditions in order to build the system. Software tool Support Vector Machine In the way classifying things Support Vector Machine is the modern technique in the field of machine learning and has been successfully used in many fields of application. The aim of this thesis is not to give a comprehensive demonstration about the theoretical background but to reveal the fundamental functionality to get an extensive understanding how SVMs work. The thesis also summaries what has to be considered when SVMs are applied, which fields of application exist and what the fields of researches nowadays are. The machine is a learning algorithm for performing classification and regression via a hyperplane in a large virtual feature space. For classification, the SVM is given a set of inputs called the training set, and attempts to automatically determine a hyperplane in feature space that separates these inputs into two classes. The hyperplane allows the machine to make an informed classification on a test vector where the true classification is unknown. Based on the assumption that the test vector and the training set are drawn from the same source, the SVM has predictable bounds on getting the classification of the test vector correct. For regression, the SVM similarly uses training vectors but derives a hyperplane-based function that can estimate a real valued function. One of the things that set SVMs apart from more traditional linear systems is their use of what is known as a kernel function. Kernels functions which allow the SVM to classify features that are nonlinear functions of the training vector attributes. While it performs this classification in a space of very high dimensionality (the feature space), it only requires computation in the smaller dimensional space of the training vectors (attribute space or input space). The other thing that sets SVMs apart is parametrically controlling the capacity of the SVM (its VC Dimension) to avoid underfitting and overfitting. Let take some example what if you do if you have given a collection of oranges and apples, and you being asked to differentiate between the two types of fruit? Within a second, everyone can immediately separate them based on how they look and feel. Although this problem of differentiating orange and apples does not look very complex, automating this process turns out to be fairly complex. What should be the basis for the decision to call an object orange, and another object apple? This problem is called classification in order to assign a new object to one of a set of classes, which are known already. The classifier which should perform this classification operation, is based on a set of example objects. This thesis will not focus on this classification problem though, but on the next problem, the problem of one-class classification. Here an object should be classified as a genuine object (orange or apple), or an outlier object (another type of fruit). The one-class classification problem differs in one essential aspect from the conventional classification problem. In one-class classification it is assumed that only information of one of the classes, the target class, is available. This means that just example objects of the target class can be used and that no information about the other class of outlier objects is present. The boundary between the two classes has to be estimated from data of only the normal, genuine class. The task is to define a boundary around the target class, such that it accepts as much of the target objects as possible, while it minimizes the chance of accepting outlier objects. MATLAB: MATLAB is a matrix-based numerical computing environment and programming language developed by The MathWorks. Simulink was used extensively for modelling, simulating, and analysing the drowsiness detection system. By using the Simulink application such as Hough Transform, Hough Lines and Kalman Filter blocks to create a lane detection and tracking algorithm. Thesis Outline Chapter 2: describes the literature review. Chapter 3: Definitions of variables associated with this particular approach for solving the problem are discussed. Chapter 4: Summarizes the results of this research and presents findings from the parametric study. Chapter 5: Finally, the conclusion of the research and recommendation on future research are provided in Appendix contains the major experiment files used to perform the simulation. Chapter 2.0: Literature Review The initial phase of this thesis was the preparation of a literature review. This review included literature from past research projects, conferences and journals on the drowsiness detection system. A comprehensive search was studied and has been reviewed to identify key studies, reports and researches initiative addressing drowsiness toward driving issues. It is attended to investigate the available knowledge in the field and to distinguish the most encouraging indicators of drowsiness drivers. Most of these methodologies have only been developed in the laboratory or have had a limited application on-road. In the current development of the drowsiness detection system, the possible techniques can be generally divided into the several categories. This category technique includes measures of: The drivers current state, especially relating to the eye and eyelid movements and physiological state changes. Driver performance, with a focus on the vehicles behaviour including lateral position and headway. A combination of the drivers current state and driver performance. We can conclude the methodology can be separated into two sections: Intrusive methods Electroencephalography Some researchers have looked at the use of EEG as a method for detecting drowsiness. Most of these studies have used EEG to verify the existence of drowsiness when other measures are being evaluated rather than as a fatigue-detection measure [12]. For example, a study by [13] demonstrated substantial relationships between an EEG algorithm for detecting fatigue and drowsiness under simulated conditions. The biggest disadvantage associated with EEG as an on-road drowsiness detection device is the difficulty in obtaining recordings under natural driving conditions; making it a slightly unrealistic option for the detection of drowsiness. In summary the transition from wakefulness to sleep can be described as a shift towards slower frequencies in the EEG. The process different between individuals but seems to be consistent within the individual [10, 11]. EEG is widely received as a good indicator of the transition between wakefulness and sleep as well as between the different sleep stages. When a driver gets drowsy a burst of alpha activity can often be seen in the central regions of the brain. An increase in alpha activity is thus the first sign of drowsiness. As the driver gets drowsier, alpha activity is replaced by theta activity. When delta activity occurs in the EEG the driver is no longer awake, this is an indicator of deep sleep [10]. Electrooculography Electrooculography is a method used for measuring the potential difference between the front and back of the eye ball. The EOG can therefore be used for detection of eye movements and blinks. The eye is a dipole with the positive cornea in the front and the negative retina in the back and the potential between cornea and retina lies in the range 0.4 1.0 mV. When the eyes are fixated straight ahead a steady baseline potential is measured by electrodes placed around the eyes. When moving the eyes a change in potential is detected as the poles come closer or farther away from the electrodes. The sign of the change depends on the direction of the movement [10]. EOG is measured by placing electrodes around the eyes. Usually silver-silver chloride electrodes are used as they show negligible drift and develop almost no polarization potentials. The electrodes should be placed as near the eyes as possible to maximize the measured potential. Problems with EOG measurement are artefacts that arise from muscle potentials and small electromagnetic disturbances that can be induced in the cables. To reduce the impedance between skin and electrode, the skin must be cleaned carefully before measurement and electrode paste should be used [10]. When measuring blinks related characteristics, the sampling frequency should be high (at least 500 Hz) as a high resolution is required to measure small differences in for example blink duration. DC recording is preferable, while filtering the low frequency components away makes the detection of long blinks difficult. One problem with DC recording however, is the risk of slow baseline drift, which makes it important to monitor the EOG signal and adjust for the drift during the measurement [14]. Non Intrusive methods PERCLOS PERCLOS (Percent Eye Closure) is a video-based method that measures eye closure. One of the strengths of PERCLOS is that attempts have been made to establish its validity as a fatigue detection device. Satisfactory relationships were obtained between eye closure and lapses in attention, providing some convergent evidence. When a measure correlates with other tests believed to measure the same construct of the systems ability to detect the current state of the driver. Furthermore, PERCLOS showed the clearest relationship with performance on a driving simulator in comparison to a number of other potential drowsiness detection devices including two electroencephalographic (EEG) algorithms, a head tracker device, and two wearable eye-blink monitors. PERCLOS is the most reliable and valid measure of a drivers alertness level between many drowsiness detection measures. According to a study performed by [17], drivers in an automobile simulator exhibit certain characteristics when drowsy, that can be easily observed in eye and facial changes [17]. Alert drivers were reported to have normal facial tone, and fast eye blinks with short ordinary glances. Drowsy drivers were reported to have decreased facial tone and slower eyelid. Gaze Direction Other potentially good fatigue parameters include various parameters that characterize the pupil movement, which relates to the driver gaze and awareness of the happenings in surroundings area. The movement of a persons pupil (gaze) may have the potential to indicate ones intention and mental condition. For example, for a driver, the nominal gaze is frontal. Looking at other directions for an extended period of time may indicate fatigue or inattention. In addition, when people are drowsy, their visual awareness cannot cover a wide enough area, concentrating on one direction. Hence, gaze (deliberate fixation) and saccade eye movement may contain information about the ones level of alertness. Many recent efforts [18, 19] produce a computer vision system that can extract various parameters in real time to characterize an eyelid movement, gaze, head movement, and facial expression. The major benefits of the visual measures are that they can be acquired non-intrusively. Lane Departure Warning Systems (LDWS). LDWS system is used to determine the position of the vehicle on the road. It is used either to warn the driver when the vehicle is on a white line (like rumble strips) or to predict when the driver is in danger of departing from the road, which rumbles strips cannot do [20]. A vehicle lateral position or lane departure situation occurs when the vehicle runs off the road, either on the left or on the right side of the road. This kind of situation is also called Run-Off-Road (ROR) or Single Vehicle Roadway Departure (SVRD). It is defined in [21] as the crashes where the first harmful event is the vehicle leaving the road high way. The simplest system is the rumble strip in which it alerts the driver when he is in a situation of lane departure in order to avoid ROR crashes. Rumble strips are areas of grooved pavement usually situated under the white lines of the road. When the vehicle drifts to the line, its tire hits a rumble strip, which vibrates the vehicle and makes a loud noise, alerting the driver to take a corrective action. This simple system is efficient since it has been shown to reduce the number of run off road crashes by 70% [22] but requires infrastructure modification. Another approach is to use a system inside the vehicle, which detects when the driver is in danger of departing from the road, and trigger an alarm in time for the driver to react. Steering wheel algorithm. Studies indicate that the steering wheel variability increases with the amount of drowsiness [23]. The steering movements also become larger and occur less often, and the lateral position variability increases as the driver gets drowsier. Also, the speed variability increases and the minimum distance to any lead vehicle decreases. The reaction time to any unexpected events also gets longer with increased drowsiness. Different studies have shown that there is a relationship between various steering related variables and the sleepiness of the driver. The steering related variables have the advantage that they are easy to measure since they require no camera or image processing. The drawback is that these variables are dependent upon the road curvature and are therefore mostly reliable on highways. [24] Other literature review has studied drowsiness detection by using steering angle rotation as an input to detect drowsiness by tracking steering angle by using a camera [25]. It tracks the steering wheel angle by using a single camera system put on inside the car. The approach is based on the modelling of the motion of the steering wheel, as it appears perceptively distorted by the point of view of the un-calibrated camera. The system has some disadvantages such as the steering image being block by the drivers head, light beam that confuses the feature detection algorithm and camera setup that not suitable for a portable application in monitoring steering angle analysis. Another drowsiness detection algorithm is based on the steering wheel. This algorithm works with three kinds of functions [26]: Time based functions (weighting functions developed from the time variations of the angle and the angular velocity), Frequency based functions (weighting functions developed from the variations in the power spectrum) Phase based functions (weighting functions developed from the variations in the angle plotted against the angular velocity). This algorithm is interesting because it proposes new detection ideas, such as the use of the phase diagram. The algorithm was tested on a special track with really drowsy drivers and it seemed to work pretty well. However, it has been created using data from drives on straight roads, so it may only work for straight roads, similar to motorways. Head position monitoring rotation. The advantage of computer vision techniques is that they are non-invasive, and thus are more amenable to use by the general public. There are some significant previous studies about drowsiness detection using computer vision techniques. Most of the published research on computer vision approaches to detection of drowsiness has focused on the analysis of blinks and head movements. It has been studied that these drivers exhibits certain physiological patterns that are expected and detectible. The standard head bobbing movement, where the drivers head drops and then rapidly pulls back upward is one of the patterns that is frequently displayed when an individual is becoming drowsy while seated in an upright position. Head movement like nodding or inclination is a good indicator of a drivers drowsiness or the onset of drowsiness [27]. It could also indicate ones attention. Head movement parameters such as head orientation, movement speed, frequency, etc. could potentially indicate ones level of attention. Finally, facial expression may also provide information about ones attention. For example, a typical facial expression that indicates the onset of drowsiness is yawning. Head monitoring tracking is a significant process for many vision-driven interactive user interfaces. The acquired position and orientation allow for pose determination and recognition of simple gestures such as nodding and head shaking. The stabilized image obtained by perspective de-warping of the facial image according to the acquired parameters is ideal for facial expression recognition [28] or face recognition applications. There are several commercial products capable of accurate and reliable 3D head position and orientation estimation. These are either based on magnetic sensors or on special markers placed on the face; both practices causing discomfort and limiting natural motion. Also, commercial systems based on gaze tracking employing infrared illumination do guarantee reliable detection of eye location, at the cost, however of restrictions placed on head position and orientation Head monitoring system developed by Advanced Safety Concepts, Inc. is the non-contact Proximity Array Sensing System (PASS), is an apparatus designed to record the x, y and z coordinates of the head at electronic rates using three electromagnetic fields. Its development is based on research that indicates a relationship between micro-motion of the head and impairment or drowsiness. It is hypothesized by ASC that changes in the X, Y, Z coordinates of the head may be an indicator of drowsiness onset, and that PASS may detect micro-sleeps based on different head movement patterns. Advanced Safety Concepts, Inc. reports that in laboratory tests, the PASS system has detected changes in head position as little as 0.0 l, while providing absolute XYZ resolution of head position to about 0.1. Disadvantages of current system. PERCLOS Disadvantages. PERCLOS stands for Percent Eye Closure. The technical definition is the percent of time a drivers eyes are closed. Sometimes a driver who is trying to stay awake can fall asleep with his eyes open, this is the disadvantage of PERCLOS. Another problem with this system is that the curve for warning is very steep at the end, which means that no warning is given at an early stage, and then the situation is very serious quickly. LDWS Disadvantages. Lane departure warning systems (LDWS) are system that currently being use to detect drowsiness. If the driver is drowsy, sooner or later the vehicle will drift to the side of the road and when it crosses the lane boundaries a warning signal is given to alert the driver. The problem with this system is that the warning signal is given every time the driver crosses the line, it does not take into consideration that the crossing could be intentional. TLC. Disadvantages. A commonly used variable in the warning algorithm of the LDWS is the Time to Line Crossing (TLC). The Time-to-Line Crossing (TLC), is the estimated time it takes for the vehicle to cross the line, which is based on a predicted path of the vehicle and the speed. The major problem with TLC is its computation in real time while driving on the road. Moreover, the computation is different on straight roads and on curve roads. EEG Disadvantages. To measure this signal while driving causes annoyance to the driver, because multiple sensors have to be attached to the driver. This can affect the driver so much that it changes the driving behaviour, which is not good at all in traffic safety research. Eye Detection Systems Disadvantages. The eye detection systems are good but not perfect, when the driver is wearing glasses there might be errors in the detection, which in some systems lead to false warnings. Sunglasses cause problems that almost none of the systems can deal with, which makes the inattention detection almost impossible when the driver is wearing sunglasses. Different ethnical people are another problem, the eyes of Asian people differ from European people, but most manufacturers claim that it should not be a problem. Research Approach Several elements have been taken into a consideration into designing the drowsiness detection system. Some researchers have already followed this route with encouraging results. By using several hypotheses and finding transformations in vehicle and driver behaviour, three based parameters will be tested for potential to predict the vehicle behaviour characteristic. In the investigation the signal will be recorded for a various driver, therefore data recorded each of the driver will were analyzed. It is important to notice that the data, of each individual driver has his own style of driving pattern. Diameter to Lane. As we all known Lane Departure Warning System can determine the position of the vehicle on the road. This position can then be used either to warn the driver when the vehicle is on a white line or to predict when the driver is in danger of departing from the road, [4]. The technique that we plan to use is to measures the distance between the car coordination toward the road lane border. It is a relevant suggestion because LDWS normally triggered when it reaches the lane. By the way it was too late to notice the drivers. Steering wheel angle. Studies indicate that the steering wheel variability increases with the amount of drowsiness [5]. The steering movements also become larger and occur less often, and the lateral position variability increases as the driver gets drowsier. Changes of velocity. More recent research demonstrated that speed variability was higher for sleep-deprived drivers than for control drivers [6].