REAL TIME DROWSY DRIVER DETECTION USING HAARCASCADE SAMPLES

REAL TIME DROWSY DRIVER DETECTION
USING HAARCASCADE SAMPLES
Dr.Suryaprasad J, Sandesh D, Saraswathi V, Swathi D, Manjunath S
PES Institute of Technology-South Campus, Bangalore, India

ABSTRACT
With the growth in population, the occurrence of automobile accidents has also seen an
increase. A detailed analysis shows that, around half million accidents occur in a year , in India
alone. Further , around 60% of these accidents are caused due to driver fatigue. Driver fatigue
affects the driving ability in the following 3 areas, a) It impairs coordination, b) It causes longer
reaction times, and, c)It impairs judgment. Through this paper, we provide a real time
monitoring system using image processing, face/eye detection techniques. Further, to ensure
real-time computation, Haarcascade samples are used to differentiate between an eye blink and
drowsy/fatigue detection.

KEYWORDS
Face detection, Eye Detection, Real-time system, Haarcascade, Drowsy driver detection, Image
Processing, Image Acquisition .

1. INTRODUCTION
According to survey [1],driver fatigue results in over 50% of the road accidents each year. Using
technology to detect driver fatigue/drowsiness is an interesting challenge that would help in
preventing accidents. In the past various efforts have been reported in the literature on approaches
for drowsiness detection of automobile driver. In the last decade alone, many countries have
begun to pay great attention to the automobile driver safety problem. Researchers have been
working on the detection of automobile driver’s drowsiness using various techniques, such as
physiological detection and Road monitoring techniques. Physiological detection techniques take
advantage of the fact that sleep rhythm of a person is strongly correlated with brain and heart
activities. However, all the research till date in this approach need electrode contacts on the
automobile drivers’ head, face, or chest making it non-implementable in real world scenarios.
Road monitoring is one of the most commonly used technique, systems based on this approach,
include Attention assist by Mercedes, Fatigue Detection System by Volkswagon, Driver Alert by
Ford, Driver Alert Control by Volvo. All the mentioned techniques monitor the road and driver
behavior characteristics to detect the drowsiness of the automobile driver. Few parameters used
include whether the driver is following the lane rules , appropriate usage of the indicators etc.. ,
If there are aberrations in these parameters, above the tolerance level then the system concludes
that the driver is drowsy. This approach is inherently flawed as monitoring the road to detect
drowsiness is more of an indirect approach and also lacks accuracy.

Sundarapandian et al. (Eds) : ICAITA, SAI, SEAS, CDKP, CMCA-2013
pp. 45–54, 2013. © CS & IT-CSCP 2013

DOI : 10.5121/csit.2013.3805

Computer Science & Information Technology (CS & IT)

46

In this paper we propose a direct approach that makes use of vision based techniques to detect
drowsiness. The major challenges of the proposed technique include (a) developing a real time
system, (b) Face detection, (c) Iris detection under various conditions like driver position,
conditions
with/without spectacles, lighting, etc (d) Blink detection, and (e) Economy. The focus will be
placed on designing a real-time system that will accurately monitor the open or closed state of the
time
driver’s eyes. By monitoring the eyes, it is believed that the symptoms of driver fatigue can be
eyes,
detected early enough to avoid a car accident. Detection of fatigue involves the observation of eye
movements and blink patterns in a sequence of images of a face extracted from a live video. T
This
paper contains 4 sections, Section 2 discusses about the high level aspects of the proposed
high-level
system, Section 3 discusses about the detailed implementation, Section 4 discusses about the
testing and validation of the developed system and Section 5 presents the conclusion and future
presents
opportunities.

2. SYSTEM DESIGN
The system architecture of the proposed system is represented in Fig 1,

Fig 1. Proposed System Architecture

Fig 1. showcases the various important blocks in the proposed system and their high level
interaction. It can be seen that the system consists of 5 distinct modules namely,(a) Video
acquisition, (b) Dividing into frames, (c) Face detection, (d) Eye detection and (e) Drowsiness
detection. In addition to these there are two external typically hardware components namely,
Camera for video acquisition and an audio alarm. The functionality of each these modules in the
system can be described as follows:

47


Video acquisition: Video acquisition mainly involves obtaining the live video feed of the
automobile driver. Video acquisition is achieved, by making use of a camera.
Dividing into frames: This module is used to take live video as its input and convert it into a
series of frames/ images, which are then processed.
Face detection: The face detection function takes one frame at a time from the frames provided
by the frame grabber, and in each and every frame it tries to detect the face of the automobile
driver. This is achieved by making use of a set of pre-defined Haarcascade samples.
Eyes detection: Once the face detection function has detected the face of the automobile driver,
the eyes detection function tries to detect the automobile driver's eyes. This is achieved by
making use of a set of pre-defined Haarcascade samples.
Drowsiness detection: After detecting the eyes of the automobile driver , the drowsiness
detection function detects if the automobile driver is drowsy or not , by taking into consideration
the state of the eyes , that is , open or closed and the blink rate.
As the proposed system makes use of OpenCV libraries, there is no necessary minimum
resolution requirement on the camera. The schematic representation of the algorithm of the
proposed system is depicted in Fig 2. In the proposed algorithm, first video acquisition is
achieved by making use of an external camera placed in front of the automobile driver. The
acquired video is then converted into a series of frames/images. The next step is to detect the
automobile driver's face , in each and every frame extracted from the video.
As indicated in Figure 2, we start with discussing face detection which has 2 important functions
(a) Identifying the region of interest, and (b) Detection of face from the above region using
Haarcascade. To avoid processing the entire image, we mark the region of interest. By
considering the region of interest it is possible to reduce the amount of processing required and
also speeds up the processing, which is the primary goal of the proposed system.


48

Fig 2. High Level System Flow

For detecting the face, since the camera is focused on the automobile driver, we can avoid
processing the image at the corners thus reducing a significant amount of processing required.
Once the region of interest is defined face has been detected, the region of interest is now the
face, as the next step involves detecting eyes. To detect the eyes, instead of processing the entire
face region , we mark a region of interest within the face region which further helps in achieving
the primary goal of the proposed system. Next we make use of Haarcascade Xml file constructed
for eye detection, and detect the eyes by processing only the region of interest . Once the eyes
have been detected, the next step is to determine whether the eyes are in open/closed state, which
is achieved by extracting and examining the pixel values from the eye region. If the eyes are
detected to be open, no action is taken. But, if eyes are detected to be closed continuously for two
seconds according to[6] , that is a particular number of frames depending on the frame rate, then
it means that the automobile driver is feeling drowsy and a sound alarm is triggered. However, if
the closed states of the eyes are not continuous, then it is declared as a blink.

49


3. IMPLEMENTATION
The implementation details of each the modules can be explained as follows:

3.1 Video Acquisition
OpenCV provides extensive support for acquiring and processing live videos. It is also possible
to choose whether the video has to be captured from the in-built webcam or an external camera
by setting the right parameters. As mentioned earlier, OpenCV does not specify any minimum
requirement on the camera, however OpenCV by default expects a particular resolution of the
video that is being recorded, if the resolutions do not match, then an error is thrown. This error
can be countered, by over riding the default value, which can be achieved, by manually
specifying the resolution of the video being recorded.

3.2 Dividing into frames
Once the video has been acquired, the next step is to divide it into a series of frames/images. This
was initially done as a 2 step process. The first step is to grab a frame from the camera or a video
file, in our case since the video is not stored, the frame is grabbed from the camera and once this
is achieved, the next step is to retrieve the grabbed frame. While retrieving , the image/frame is
first decompressed and then retrieved . However, the two step process took a lot of processing
time as the grabbed frame had to be stored temporarily. To overcome this problem, we came up
with a single step process, where a single function grabs a frame and returns it by decompressing.

3.3 Face detection
Once the frames are successfully extracted the next step is to detect the face in each of these
frames. This is achieved by making use of the Haarcascade file for face detection. The
Haarcascade file contains a number of features of the face , such as height , width and thresholds
of face colors., it is constructed by using a number of positive and negative samples. For face
detection, we first load the cascade file. Then pass the acquired frame to an edge detection
function, which detects all the possible objects of different sizes in the frame. To reduce the
amount of processing, instead of detecting objects of all possible sizes, since the face of the
automobile driver occupies a large part of the image, we can specify the edge detector to detect
only objects of a particular size, this size is decided based on the Haarcascade file, wherein each
Haarcascade file will be designed for a particular size. Now, the output the edge detector is stored
in an array. Now, the output of the edge detector is then compared with the cascade file to
identify the face in the frame. Since the cascade consists of both positive and negative samples, it
is required to specify the number of failures on which an object detected should be classified as a
negative sample. In our system, we set this value to 3, which helped in achieving both accuracy as
well as less processing time. The output of this module is a frame with face detected in it.

3.4 Eye detection
After detecting the face, the next step is to detect the eyes, this can be achieved by making use of
the same technique used for face detection. However, to reduce the amount of processing, we
mark the region of interest before trying to detect eyes. The region of interest is set by taking into
account the following:


•
•

50

The eyes are present only in the upper part of the face detected.
The eyes are present a few pixels lower from the top edge of the face.

Once the region of interest is marked, the edge detection technique is applied only on the region
of interest, thus reducing the amount of processing significantly. Now, we make use of the same
technique as face detection for detecting the eyes by making use of Haarcascade Xml file for eyes
detection. But, the output obtained was not very efficient, there were more than two objects
classified as positive samples, indicating more than two eyes. To overcome this problem, the
following steps are taken:
•
•
•
•
•

Out of the detected objects, the object which has the highest surface area is obtained. This is
considered as the first positive sample.
Out of the remaining objects , the object with the highest surface area is determined. This is
considered as the second positive sample.
A check is made to make sure that the two positive samples are not the same.
Now, we check if the two positive samples have a minimum of 30 pixels from either of the
edges.
Next , we check if the two positive samples have a minimum of 20 pixels apart from each
other.

After passing the above tests, we conclude that the two objects i.e positive sample 1 and positive
sample 2 , are the eyes of the automobile driver.

3.5 Drowsiness detection
Once the eyes are detected, the next step is to determine if the eyes are in closed or open state.
This is achieved by extracting the pixel values from the eye region. After extracting , we check if
these pixel values are white, if they are white then it infers that the eyes are in the open state, if
the pixel values are not white then it infers that the eyes are in the closed state.
This is done for each and every frame extracted. If the eyes are detected to be closed for two
seconds or a certain number of consecutive frames depending on the frame rate, then the
automobile driver is detected to be drowsy. If the eyes are detected to be closed in non
consecutive frames, then We declare it as a blink.
If drowsiness is detected, a text message is displayed along with triggering an audio alarm. But, it
was observed that the system was not able to run for an extended period of time , because the
conversion of the acquired video from RGB to grayscale was occupying too much memory. To
overcome this problem, instead of converting the video to grayscale, the RGB video only was
used for processing. This conversion resulted in the following advantages,
•
•
•

Better differentiation between colors, as it uses multichannel colors.
Consumes very less memory.
Capable of achieving blink detection, even when the automobile driver is wearing
spectacles.

51


Hence there were two versions of the system that was implemented; the version 1.0 involves the
conversion of the image to grayscale. Currentlyversion 2.0 makes use of the RGB video for
processing.

4. TESTING
The tests were conducted in various conditions including:
1. Different lighting conditions.
2. Drivers posture and position of the automobile drivers face.
3. Drivers with spectacles.

4.1 Test case 1: When there is ambient light

Fig 3. Test Scenario #1 – Ambient lighting

Result: As shown in Fig 3, when there is ambient amount of light, the automobile driver's face
and eyes are successfully detected.

4.2 Test case 2: Position of the automobile drivers face
4.2.1 Center Positioned

Fig 4. Test Scenario Sample#2 driver face in center of frame

RESULT : As shown in Fig 4,When the automobile driver's face is positioned at the Centre, the
face, eyes , eye blinks, and drowsiness was successfully detected.


52

4.2.2 Right Positioned

Fig 5. Test Scenario Sample#3 – driver face to right of frame

RESULT: As shown in Fig 5,When the automobile driver's face is positioned at the Right, the
face, eyes , eye blinks, and drowsiness was successfully detected.
4.2.3 Left Positioned

Fig 6. Test Scenario Sample#4- driver face to left of frame

RESULT : As shown in screen snapshot in Fig 6,when the automobile driver's face is positioned
at the Left, the face, eyes , eye blinks, and drowsiness was successfully detected.

4.3 Test case 3: When the automobile driver is wearing spectacles

Fig 7. Test Scenario Sample#5 – Driver with spectacles

RESULT : As shown in screen snapshot in Fig 7,When the automobile driver is wearing
spectacles, the face, eyes , eye blinks, and drowsiness was successfully detected.

53


4.4 Test case 4: When the automobile driver’s head s tilted

Fig 8. Test Scenario Sample#6 – Various posture

RESULT : As shown in screen snapshot in Fig 8,when the automobile driver's face is tilted for
more than 30 degrees from vertical plane, it was observed that the detection of face and eyes
failed.
The system was extensively tested even in real world scenarios, this was achieved by placing the
camera on the visor of the car , focusing on the automobile driver. It was found that the system
gave positive output unless there was any direct light falling on the camera.

5. CONCLUSION
The primary goal of this project is to develop a real time drowsiness monitoring system in
automobiles. We developed a simple system consisting of 5 modules namely(a) video acquisition,
(b) dividing into frames, (c) face detection, (d) eye detection, and (e)drowsiness detection . Each
of these components can be implemented independently thus providing a way to structure them
based on the requirements.
Four features that make our system different from existing ones are:
(a)
(b)
(c)
(d)

Focus on the driver, which is a direct way of detecting the drowsiness
A real-time system that detects face,iris, blink, and driver drowsiness
A completely non-intrusive system, and
Cost effective

5.1 Limitations
The following are some of the limitations of the proposed system.
•
•

The system fails, if the automobile driver is wearing any kind of sunglasses.
The system does not function if there is light falling directly on the camera.


54

5.2 Future Enhancement
The system at this stage is a “Proof of Concept” for a much substantial endeavor. This will serve
as a first step towards a distinguished technology that can bring about an evolution aimed at ace
development. The developed system has special emphasis on real-time monitoring with
flexibility, adaptability and enhancements as the foremost requirements.
Future enhancements are always meant to be items that require more planning, budget and
staffing to have them implemented. There following are couple of recommended areas for future
enhancements:
•

Standalone product: It can be implemented as a standalone product, which can be
installed in an automobile for monitoring the automobile driver.

•

Smart phone application: It can be implemented as a smart phone application, which
can be installed on smart phones. And the automobile driver can start the application after
placing it at a position where the camera is focused on the driver.

REFERENCES
[1] Ruian Liu,et.a;, "Design of face detection and tracking system," Image and Signal Processing (CISP),
2010 3rd International Congress on , vol.4, no., pp.1840,1844, 16-18 Oct. 2010
[2] Xianghua Fan, et.al, "The system of face detection based on OpenCV," Control and Decision
Conference (CCDC), 2012 24th Chinese , vol., no., pp.648,651, 23-25 May 2012
[3] Goel, P, et.al., "Hybrid Approach of Haar Cascade Classifiers and Geometrical Properties of Facial
Features Applied to Illumination Invariant Gender Classification System," Computing Sciences
(ICCS), 2012 International Conference on , vol., no., pp.132,136, 14-15 Sept. 2012
[4] Parris, J., et.al, "Face and eye detection on hard datasets," Biometrics (IJCB), 2011 International Joint
Conference on , vol., no., pp.1,10, 11-13 Oct. 2011
[5] Peng Wang., et.a;., "Automatic Eye Detection and Its Validation," Computer Vision and Pattern
Recognition - Workshops, 2005. CVPR Workshops. IEEE Computer Society Conference on , vol.,
no., pp.164,164, 25-25 June 2005
[6] Picot, A. et.al., "On-Line Detection of Drowsiness Using Brain and Visual Information," Systems,
Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on , vol.42, no.3, pp.764,775,
May 2012
[7] Xia Liu; Fengliang Xu; Fujimura, K., "Real-time eye detection and tracking for driver observation
under various light conditions," Intelligent Vehicle Symposium, 2002. IEEE , vol.2, no., pp.344,351
vol.2, 17-21 June 2002.

REAL TIME DROWSY DRIVER DETECTION USING HAARCASCADE SAMPLES

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to REAL TIME DROWSY DRIVER DETECTION USING HAARCASCADE SAMPLES (20)

Recently uploaded (20)

REAL TIME DROWSY DRIVER DETECTION USING HAARCASCADE SAMPLES