17-02-2011, 02:09 PM
Summary
International statistics shows that a large number of roadaccidents are caused by driver fatigue. Therefore, a systemthat can detect oncoming driver fatigue and issue timelywarning could help to prevent many accidents, andconsequently save money and reduce personal suffering. Inthis proposed system security camera can be used that pointsdirectly towards the driver’s face and monitors the driver’seyes & mouth in order to detect fatigue. If the fatigue isdetected a warning signal is issued to alert the driver. The skincolor based algorithm is used to detect the face of the driver.Once the face area is found, the eyes and mouth can be foundby thresholding & segmentation process. The propose systemuses neural network along with genetic algorithm for detectingthe Driver’s Fatigue, named as Neuron-Genetic system.
1. Introduction
The ever increasing numbers of traffic accidents all over theworld are due to diminished driver’s vigilance level. Driverswith a diminished vigilance level suffer from a marked declinein their perception; recognition and vehicle control abilitiesand therefore pose a serious danger to their own lives and thelives of the other people. For this reason, developing systemsthat actively monitors the driver’s level of vigilance andalerting the driver of any insecure driving condition isessential for accident prevention.Many efforts have been reported in the literature fordeveloping an active safety system for reducing the number ofautomobiles accidents due to reduced vigilance. Drowsiness indrivers can be generally divided into the following categories
ensing of physiological characteristics, sensing of driveroperation, sensing of vehicle response, monitoring theresponse of driver. Among these methods, the techniques based on human physiological phenomena are the mostaccurate. This technique is implemented in two ways:measuring changes in physiological signals, such as brainwaves, heart rate, and eye blinking; and measuring physicalchanges such as sagging posture, leaning of the driver’s headand the open/closed states of the eyes. The first technique,while most accurate, is not realistic, since sensing electrodeswould have to be attached directly on to the driver’s body, andhence be annoying and distracting to the driver. In addition,long time driving would result in perspiration on the sensors,diminishing their ability to monitor accurately. The secondtechnique is well-suited for real world driving conditions sinceit can be non-intrusive by using video cameras to detectchanges. Driver operation and vehicle behavior can beimplemented by monitoring the steering wheel movement,accelerator or brake patterns, vehicle speed, lateralacceleration, and lateral displacement. These too are nonintrusiveways of detecting drowsiness, but are limited tovehicle type and driver condition. The final technique fordetecting drowsiness is by monitoring the response of thedriver. This involves periodically requesting the driver to senda response to the system to indicate alertness. The problemwith this technique is that it will eventually become tiresomeand annoying to the driver.The propose system based on eyes closer count & yawningcount of the driver. By monitoring the eyes and mouth, it isbelieved that the symptoms of driver fatigue can be detectedearly enough to avoid a car accident. The eye blink frequencyincreases beyond the normal rate in the fatigued state. Inaddition, micro sleeps that are the short periods of sleeplasting 3 to 4 seconds are the good indicator of the fatiguedstate, but it is difficult to predict the driver fatigue accuratelyor reliably based only on a single driver behavior.Additionally, the changes in a driver’s performance are morecomplicated and not reliable so in this system secondparameter is also considered which is a yawning count. Inorder to detect fatigue probability the facial expressionparameters must be extracted first. As fatigue level can be properly characterized by eyes and mouth movements, avision sensor is needed to recognize and track the eyes and themouth, a normal video-camera as a vision sensor can be useon the premise with the environment is bright enough. Here,eye closing count, and yawn count in successive frames can bedetected using a web camera. In real time implementation if30 fps considered then it is observed that successive frameshave same information ,so instead of considering all frames ofvideo files, select frames such that which gives moreinformation but less computational requirements. In thisproposed system instead of analyzing complete frame of videofile, eyes and mouth portion are separated after detecting facearea , the facial features in these regions are considered indetail and corresponding eyes closing count & yawning countcan be obtained using correlation method for this type ofDigital Image Processing filtering and segmentation processhas to be carried out. In this system a feed forwardbackprapogation neural network is used along with Geneticalgorithm for giving more intelligence to the system since theinput parameters are highly unreliable and non linear innature ,it provides the optimized structure of the NN whichcan give result close to that of Fixed structured NN.2. System Overview
The complete block diagram representation of the proposedsystem is as shown in figure 1.while flowchart of the majorfunctions of The Drowsy Driver Detection System is shown inFigure.2 After inputting a facial image, the skin color basedalgorithm is applied to detect the face in the image.. Using thesides of the face, eyes portion and mouth portion of the imageis separated from which the open or closed state of the eyesare detected along with yawning count. This two parameterswhich are the output of the DIP Module are given as the inputto the Hybrid Intelligent System which is a combination ofNeural network and Genetic Algorithm, which can givecorresponding fatigue probability, Depending on the fatiguedprobability ,system draws the conclusion that the driver isfalling in fatigued state and issues a warning signal. In this system AVI file is used which is converted toframes and random frames are selected from it for furtherprocessing. The next step will be of eye and lip detection.Eyes are located by performing some morphologicaloperations on the face. This is done by converting the image toa binary image, based on threshold. In the binary image thereare two significant intensity changes that can be seen. The firstintensity change is the eyebrow, and the next change is the upper edge of the eye. The state of the eyes (whether it is openor closed) is determined by distance between the two intensitychanges. When the eyes are closed, the distance i.e. the no. ofwhite pixels between the two intensity changes is larger ascompared to when the eyes are open. The number of whitepixels between the two intensity changes is recorded. For Lipdetection R/G ratio process is used.The parameter extraction isperformed by DIP Module which is then provided to Geneticprocess based Optimized Neural Network, which givescomparative result of fatigues.
2.1. Face Detection
Human face localization and detection is often the first stepin applications such as video surveillance, human computerinterface, face recognition and /or facial expressions analysis,and image database management. A lot of research has beendone in the area of human face detection .In prior studies, different human skin colors from differentraces have been found to fall in a compact region in colorspaces. Therefore skin can be detected by making use of thiscompactness. The face detection is performed in three steps .The first step is to classify each pixel in the given image as askin pixel or a non-skin pixel. The second step is to identifydifferent skin regions in the skin-detected image by usingconnectivity analysis. The last step is to decide whether eachof the skin regions identified is a face or not. After theprobable location of the face is found the left and the rightedges of the face is determined
download full report
http://paper.ijcsns07_book/200903/20090311.pdf
