Sleep Detection System Using Matlab Image Processing

SLEEP DETECTION SYSTEM USING MATLAB IMAGE

PROCESSING
1
DIGVIJAY R, 2VIGNESH SRIDHARAN
1,2
BE 3rd Year EEE, Anna University

Abstract: The objective was to develop a non-invasive system for detecting the closing of eyes of a person driving an
automobile and provide an alarm indication thus preventing road accidents from occurring. Live video relay of the driver's
eyes is processed using Image Processing in MATLAB to detect whether the eye is closed for more than a fixed duration
thus indicating conditions of fatigue, alcohol consumption etc. The system proves to be more accurate and safe compared to
the existing sleep detection system developed using Infrared Sensors and Micro-processors.

I. CLAIMS an alarm indication if the eyes are detected to be

closed for more than a specified amount of time.
What is claimed is, MATLAB 2013a Image processing techniques are
adopted to detect the closure of the eye by sectioning
1. A non-invasive system for detecting the closing of only that portion of the driver's face from a live video
eyes of a person driving an automobile thereby relay obtained using a front camera.
identifying the condition of the person (Fatigue,
Alcohol induced, sleep derived) using live video relay III. DESCRIPTION
obtained from a webcam and processed using Image
Acquisition Toolbox and Image Processing Toolbox 1. Processing a Static Image
in MATLAB/2013a. The method is based on the Viola-Jones algorithm
[2]. The project started off with detecting the eyes of
2. The method of Claim 1, wherein a further alarm a static image stored in the computer. The first step
system is incorporated for warning the driver or the involved storing the image in a variable mentioning
surrounding personnel. the location and the type of image. From the given
3. The method of Claim 1, Advantageous over the image, only the eyes are sectioned out and processed
current system which uses Infrared Sensors and to detect for closure or fatigue. The image is
micro-processors, processed only to detect the eye region of the image
Greater accuracy over the current system which is by giving the position, width and height of the region
susceptible to various other light bands like sunlight, as inputs to the rectangle () function. The position,
headlights of other vehicles. width and height are obtained by using the Vision
Higher degree of safety over the current system class in MATLAB. The built in object detector
which causes Photo-keratitis (Inflammation of the function Cascade Object Detector is used to detect the
Cornea) and retinal damage because of long-term eyes. The Eye Detect object is given as input to the
exposure to infrared rays. step function along with the image and the values
4. The method of Claim 1 offers more flexibility in returned correspond to the X-Coordinate, Y-
terms of changing the module design for different Coordinate, Width and Height of the eye region. The
driving environments. image is then cropped using the imcrop () function
5. The method of Claim 1, is more efficient and with one input as the n*4 matrix and the other being
economical than the existing method designed using the image itself. The RGB image thus obtained is first
Hardware components. converted to its equivalent grayscale form using the
rgb2gray() function. This is followed by converting
II. INTRODUCTION the thus obtained gray scale image to its black and
white form using the im2bw () function. The BW
With the ever increasing population and usage of image thus obtained is then dilated to get only the
automobiles, there is an increase in the number of eyes. The purpose of performing the dilation function
fatalities as well. India, unfortunately, boasts of a is to enhance the foreground features.
very high number of 142,485 traffic-related fatalities IM2 = imdilate (IM,SE) dilates the grayscale, binary,
[1]. There are a number of reasons that can be or packed binary image IM, returning the dilated
attributed to this astonishing statistic, a few of image, IM2. SE is a structuring element object, or
primary concern being Fatigue, Alcohol Consumption array of structuring element objects, returned by the
and Sleep Deprivation. STREL function. The basic effect of the operator on a
Hence, we developed a method to test for the closing binary image is to gradually enlarge the boundaries of
of eyes of a person driving an automobile and provide regions of foreground pixels (white pixels). Thus

Proceedings of 2nd IRF International Conference, 9th February 2014, Chennai India. ISBN: 978-93-82702-57-3

33
 Sleep Detection System Using Matlab Image Processing

areas of foreground pixels grow in size while holes

within those regions become smaller.

Fig. 3 Dilated Image

2. Processing a live feed

The next step of the project was to perform the same
on a live video feed obtained by either using an
external USB operated camera or by using the built-in
webcam. The accuracy of this method of eye
detection is based on the sensitivity of the camera. It
is found to have a direct relationship with the
accuracy. The greater the accuracy needed, the better
quality of webcam has to be used. The first step
towards implementing this, is to first identify the
webcam drivers installed and then configure the
webcam to obtain the necessary video feed. The
associated webcams were identified by using the
imaqhwinfo() function. The next step was to
configure the webcam and assign the video
properties.
This involved setting the FramesPer Trigger and
Returned Color Space properties of the video object.
The live feed was then obtained using the start(video-
object) function.
The vision. Cascade Object Detector statement for
detecting the face was used to initialize an object
FaceDetect. The next step was to crop the image such
that only the face is retained static for further eye
detection. This is achieved by visualizing the live
video feed as individual frames and processing each
frame distinctly. The vision. Cascade Object Detector
for detecting the eye region was used to initialize an
object EyeDetect. The video capturing was initially
performed for the first 50 frames. The video was
converted to individual frames using the
getsnapshot() function which returns a matrix
corresponding to an RGB image. The next step
involved was similar to identifying the eye region in a
static image, the difference being instead of the image
being stored in the computer memory, it is stored
virtually in a MATLAB script.
Since the getsnapshot() function works by contacting
the webcam every time it is called, the processing
time is increased. In order to minimize the time taken
by the getsnapshot(), the triggerconfig() property of
the video object was set to manual mode. The
Fig. 1 Static Image File EyeDetect object is given as input to the step function
along with the image and the values returned
correspond to the X-Coordinate, Y-Coordinate,
Width and Height of the eye region.
The image is then cropped using the imcrop()
function with one input as the n cross 4 matrix and
the other being the image itself. The RGB image thus
obtained is first converted to its equivalent grayscale
form using the rgb2gray() function. This is followed
by converting the thus obtained gray scale image to
Fig. 2 BW Image after processing its black and white form using the im2bw() function.
Proceedings of 2nd IRF International Conference, 9th February 2014, Chennai India. ISBN: 978-93-82702-57-3

34
 Sleep Detection System Using Matlab Image Processing

The BW image thus obtained is then dilated to get

only the eyes.
The dilated image thus obtained is a matrix.
The sum of all the elements of the matrix are obtained
using the sum() function. The value of this sum is
compared with a temporary value(20) , which is used
to decide whether the eyes are closed or not. This
temporary value is the specified time frame to check
for closure of the eyes. This is done by calculating the
sum of the elements of the matrix for a particular set
of frames(50).
If the value does not exceed the threshold and stays
constant for a short time it corresponds to a normal
blink. If the value stays for more than 5 frames it
means that the eye has been closed for more than 5
frames indicating a drowsy state. The MATLAB code
immediately sends a warning message that the driver
is drowsy. This can be improved by interfacing an
audio amplifier and speaker to produce a warning
sound or can be interfaced with a vibrator to produce
a vibration.

Fig. 4 Live Video Relay from a web camera

Fig. 5 Face Detect from live feed

Fig. 6 Eye Detect and Cropped

CONCLUSION

The sleep detection system proposed has a specific

set of advantages over the existing method and proves
to be more efficient and economical in comparison.

Proceedings of 2nd IRF International Conference, 9th February 2014, Chennai India. ISBN: 978-93-82702-57-3

35
 Sleep Detection System Using Matlab Image Processing

Although there is a constraint on the quality of the REFERENCES

camera required for processing the live video feed,
this method offers a non-invasive system for sleep [1] http://en.wikipedia.org/wiki/List_of_countries_by_traffic-
related_death_rate
detection. An improvement over the existing method, [2]http://en.wikipedia.org/wiki/Viola%E2%80%93Jones_object_
this system is indeed a feasible and easily detection_framework
implementable alternative. [3] http://www.mathworks.in/help/vision/ref/vision.cascadeo
bjectdetector.step.html
[4] http://stackoverflow.com/questions/15748903/matlab-exa
ctly-timed-getsnapshot-for-real-time-event-analyzing

Proceedings of 2nd IRF International Conference, 9th February 2014, Chennai India. ISBN: 978-93-82702-57-3

36