ISSN ONLINE(2278-8875) PRINT (2320-3765)

All submissions of the EM system will be redirected to Online Manuscript Submission System. Authors are requested to submit articles directly to Online Manuscript Submission System of respective journal.

DEVELOPMENT OF A DROWSINESS WARNING SYSTEM USING NEURAL NETWORK

Itenderpal singh1, Prof. V.K.Banga 2
  1. M.Tech. Scholar, Dept. of ECE, Amritsar College of Engineering & Technology, Amritsar, Punjab, India
  2. H.O.D, Dept. of ECE/EEE, Amritsar College of Engineering & Technology, Amritsar, Punjab, India
Related article at Pubmed, Scholar Google

Visit for more related articles at International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering

Abstract

In this paper, a vehicle driver drowsiness warning system using image processing technique with neural network is proposed. The proposed system is based on facial images analysis for warning the driver of drowsiness or inattention to prevent traffic accidents. The facial images of driver are taken by a video camera which is installed on the dashboard in front of the driver. A Neural network based algorithm is proposed to determine the level of fatigue by measuring the eye opening and closing, and warns the driver accordingly. The results indicated that the proposed expert system is effective for increasing safety in driving.

Keywords

Drowsiness warning system; accidents; face detection; eye detection; neural network

INTRODUCTION

Due to the increase in the amount of automobile in recent years, problems created by accidents have become more complex as well. Traditional transportation system is no longer sufficient. In recent years, the intelligent vehicle system has emerged and became a popular topic among transportation researchers. However, the research of safety in vehicle is an important subset of intelligent vehicle system research. Meantime, active warning system is one of the designs on active safety system. The safety warning systems, mostly active warning systems for preventing traffic accidents have been attracting much public attention [9].Safe driving is a major concern of societies all over the world. Thousands of people are killed, or seriously injured due to drivers falling asleep at the wheels each year. Recent studies show those drivers’ drowsiness accounts for up to 20% of serious or fatal accidents on motorways and monotonous roads, which impair the drivers’ judgment and their ability of controlling vehicles. Therefore, it is essential to develop a real-time safety system for drowsiness-related road accident prevention. Many methods have been developed and some of them are currently being used for detecting the driver’s drowsiness, including the measurements of physiological features like EEG, heart rate and pulse rate, eyelid movement, gaze, head movement and behaviors of the vehicle, such as lane deviations and steering movements. Among those different technologies, ocular measures, such as eye-blinking and eyelid closure, are considered as promising ways for monitoring alertness.
The aim of this paper is to develop a prototype drowsiness detection system. The focus will be placed on designing a system that will accurately monitor the open or closed state of the driver’s eyes in real-time. By monitoring the eyes, it is believed that the symptoms of driver fatigue can be detected early enough to avoid a car accident.

RELATED WORK

To analyze driver’s drowsiness several systems have been built. They usually require simplifying the problem to work partially or under special environments, for example D. Taneral et al. [6] presents an automatic drowsy driver monitoring and accident prevention system that is based on monitoring the changes in the eye blink duration. He proposed the method that detects visual changes in eye locations using the proposed horizontal symmetry feature of the eyes. This new method detects eye blinks via a standard webcam in real-time at 110fps for a 320×240 resolution. Flores Javier macro et al. [11] has presented a new Advanced Driver Assistance System (ADAS) for automatic driver’s drowsiness detection based on visual information and Artificial Intelligent. This system works on several stages to be fully automatic. In addition, the aim of the algorithm is to locate and to track the face and the eyes to compute a drowsiness index. Garcia .i et al [3] has presented a non-intrusive approach for drowsiness detection, based on computer vision. It is installed in a car and it is able to work under real operation conditions. An IR camera is placed in front of the driver, in the dashboard, in order to detect his face and obtain drowsiness clues from their eyes closure. It works in a robust and automatic way, without prior calibration. The presented system is composed of 3 stages. The first one is pre-processing, which includes face and eye detection and normalization. The second stage performs pupil position detection and characterization, combining it with an adaptive lighting filtering to make the system capable of dealing with outdoor illumination conditions. The final stage computes PERCLOS from eyes closure information. In order to evaluate this system, an outdoor database was generated, consisting of several experiments carried out during more than 25 driving hours. Sharma nidhi et al. [5] presented a novel approach to alert a driver who tends to doze off while driving to avoid road crashes. In her system using a small camera that points directly towards the driver’s face, an image is obtained. From that image, skin region i.e. face is segmented out using YCbCr colour space. Finally localization of eyes is done with fuzzy logic application to determine the level of fatigueness and then warn the driver accordingly.

OUR PROPOSED ALGORITHM

image
The first step is the image acquisition which is done by using video camera which takes the video of the driver and convert into image frames.
image
The second step is the face detection, in this we use viola Jones algorithm to detect face Viola-Jones algorithm is based on exploring the input image by means of sub window capable of detecting features. This window is scaled to detect faces of different sizes in the image. Viola Jones developed a scale invariant detector which runs through the image many times, each time with different size. Being scale invariant, the detector requires same number of calculations regardless of the size of the image.
image
image
image image image
The third step is eyes detection. Similarly, Eyes are detected by using this algorithm. To detect eyes we first detect nose and then detect pair of eyes. However, the RGB model includes brightness in addition to the colours. When it comes to human’s eyes, different brightness for the same colour means different colour. When analysing a human eye, RGB model is very sensitive in image brightness. The next step is to extract the features of eyes i.e. to convert RGB image into YCBCR image: The Cb and Cr components give a good indication on whether a pixel is part of the skin or not. This can clearly be seen in Figure.8, which are the Cb and Cr values of all the pixels that are part of the eye. There is a strong correlation between the Cb and Cr values of skin pixels, to reveal the comparison between eyes and non-eyes in the YCbCr space.
image
Figure 8.shows the image in YCbCr. Since the skin-tone colour depends on luminance, we nonlinearly transform the YCbCr colour space to make the skin clear. The main advantage of converting the image to the YCbCr domain is that influence of luminosity can be removed during our image processing. In the RGB domain, each component of the picture (red, green and blue) has a different brightness. However, in the YCbCr domain all information about the brightness is given by the Y component, since the Cb (blue) and Cr (red) components are independent from the luminosity. The next step is to calculate the mean and standard deviation of eyes i.e. for open, drowsy and close image
image
image
image
The last step is to train the network by using back propagation algorithm. Our proposed neural network model is as shown in figure 12.512 at the input nodes represent features of eyes and labels at output nodes represents classes.
image
image
As in Figure 13 i.e. list of inputs <512*60 double> are the features of the images that we are taken to train our neural network. These features are of open, drowsy and close eyes. 512 is size of our image and 60 are input images i.e. 20 of open eyes, 20 of drowsy eyes and 20 of close eyes.
image
image image image
The above figure 14, 15 and 16 are the three target patterns for our proposed neural network. In our work the target vector is encoded using one hot encoding method. One-hot refers to a group of bits among which the legal combinations of values are only those with a single high (1) bit and all the others low (0). For example, the output of a decoder is usually a one-hot code. Table A. displays the arrhythmia classes and their corresponding target vectors in one hot encoded form.
The last step is the Image recognition for alarming signal: The trained neural network easily predicted whether the eyes are open, close or drowsy. In this case if the input image as close eye of the driver. Our neural network recognize the image as shown below and automatically an alarming signal is generated that alerts the driver.
image

RESULTS AND DISCUSSIONS

The accuracy of event detection is the utmost important factor in determining the accuracy of drowsiness detection system. In this work, MLP feed forward neural network model was trained by back propagation algorithm to distinguish between three types of eyes arrhythmias. This paper presents the evaluation of designed classifier by means of different performance indices. The performance indices used here are
1. Confusion Matrix (Classification matrix).
2. Mean Square Error.
3. Error histogram.

A. CONFUSION MATRIX

The performance of classifier is analyzed using confusion matrix which is also known as table of confusion. It displays the number of correct and incorrect predictions made by the model compared with the actual classifications in the test data. The confusion matrix lists the correct classification against the predicted classification for each class. The number of correct predictions for each class falls along the diagonal of the matrix. All other numbers are the number of errors for a particular type of misclassification error. In Table B, we can see that when we consider overall data set than the accuracy reaches to 96.7% which is a very good amount. The rate of misclassification is less when compared to other phases. In this as we can see class1 is 0 times misclassified class 2 and class 3, class 2 was 0 times misclassified as class 1 and class 3 by 0.0%, class 3 was 1.7% misclassified as class 1 and class 2, and correctly classified as class 3.In Table we can see that the green boxes represent the final accuracy for each class as each class was correctly trained.
image
Finally overall accuracy is shown in blue box which shows that each class classification was correctly learned by the arrhythmia classifier with zero mean square error and within stipulated parameters related to LM algorithm and gives 96.7 % results.

B.MEAN SQUARE ERROR

The process of training a neural network involves tuning the values of the weights and biases of the network to optimize network performance, as defined by the network performance function net.performFcn. Mean Squared Error is the average squared difference between outputs and targets. Lower values of (MSE) indicate better performance of the network and zero means no error.
image
where, t –target, a - actual output, e – Error, N–Number of exemplars
The performance graph is shown in Fig 17.
image
shows that NN classifier was highly effective in knowing the nature of data set and knowing the causal association between different features like P peak, R peak, RR interval etc.

C. ERROR HISTOGRAM

Histograms are used to plot density of data, and often for density estimation: estimating the probability density function of the underlying variable. The total area of a histogram used for probability density is always normalized to 1. If the length of the intervals on the x-axis is all 1, then a histogram is identical to a relative frequency plot. There is no "best" number of bins, and different bin sizes can reveal different features of the data.
image

CONCLUSION AND FUTURE SCOPE

A non-invasive system to localize the eyes and monitor fatigue is developed. Information about the degree of eye closure is obtained through various self-developed image processing algorithms. During the monitoring, the system is able to decide if the eyes are opened, drowsy or closed. When the eyes are drowsy or closed, a warning signal is issued. Neural network provides a completely different, unorthodox way to approach a control problem, this technology is not difficult to apply and the results are usually quite surprising and pleasing.
For future scope we suggest that that one can work on more features that can include the change is size and shape of iris when the person is drunk or when there is glossy appearance to eyes or must work on the concept of Horizontal Gaze Nystagmus for better accuracy using other machine algorithm like SVM.

References