See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341317210

Automatic and Secure Electronic Gate System Using Fusion of License Plate,
Car Make Recognition and Face Detection

Conference Paper · February 2020

DOI: 10.1109/ICIoT48696.2020.9089615

CITATIONS READS

14 2,471

5 authors, including:

Razan Mohammed Suchithra Kunhoth

Taif University Qatar University
2 PUBLICATIONS 14 CITATIONS 18 PUBLICATIONS 151 CITATIONS

SEE PROFILE SEE PROFILE

Somaya Al-ma'adeed
Qatar University
297 PUBLICATIONS 7,485 CITATIONS

SEE PROFILE

All content following this page was uploaded by Suchithra Kunhoth on 06 June 2020.

The user has requested enhancement of the downloaded file.

 Automatic and Secure Electronic Gate System
Using Fusion of License Plate, Car Make
Recognition and Face Detection
Ghaida Saadouli, Maha Ibrahim Elburdani, Razan Mohammed Al-Qatouni, Suchithra Kunhoth, Somaya Al- Maadeed
Department of Computer Science and Engineering
Qatar University
Doha, Qatar
[email protected], [email protected], [email protected], [email protected], [email protected]

Abstract— In this paper, we propose an automatic electronic not be enough to recognize a car in certain instances, because
gate system with authenticity confirmation from 3 individual it can be easily forged [2]. For example, if the thief wants to
modules, viz., car make and model, license plate and face enter a secured area and his car is pawned, he can simply
detection. The ultrasonic sensor detects the stopped cars and replace the license plate of his car by another plate number.
initiate the camera to capture the car image. Connected Hence we cannot solely depend on the ANPR systems in more
Components identification and Optical Character Recognition secured areas such as military compounds, governmental
(OCR) algorithms are performed to recognize the characters organizations such as ministry offices. The recognition of car
and numbers in the car plate. The car make and model detection make and model complements the ANPR to build a bit more
algorithm uses feature extraction algorithms based on
secure system. It serve as an effective aid for police forces in
Difference of Gaussians (DoG) detector and Scale Invariant
Feature Transform (SIFT) descriptor. The Euclidian distance
the investigation of criminal offenses involving vehicles.
measure identifies the suitable match for a query image with the Moreover, this can also be used to track employees in some
one in database to determine the car make and model. The face secured areas like airports and military buildings. The
detection of driver is carried out to ensure the security of identification of face of the driver can add further
intended premise of the system deployment. It is based on the confirmation to indicate the real identity of the vehicle. An
Viola Jones algorithm. In the final stage, matching algorithms automatic gate control system with a multi-level
are applied to decide whether the image of the car plate, make authentication schema built on combining multiple cues from
and model and the face does not conflict with the details stored the vehicle can thus control unauthorized entry.
in the database. The smart electronic gate will open and let the
car enter when the authenticity is confirmed. The system is In this work, we propose a smart secure electronic gate
evaluated by building a prototype electronic smart gate using based on the functionalities of license plate, car make and
Logitech C920 HD pro webcam, Toy car and LEGO NXT Motor model, and face detection. An ultrasonic sensor is utilized to
Controlled Gate. Moreover we built a preliminary database detect the event of the cars being stopped at the gate. The three
from the images captured by surveillance cameras for make and subsequent modules takes the image of captured car and
model recognition of Qatar cars. Experimental evaluation on perform the respective analysis. Once we have the information
this dataset delivered an accuracy of 75%. of number plate, car make and model and the driver, we can
confirm the entry based on a matching with the details stored
Keywords—car make and model, face detection, license plate in a database. In the situations of data mismatch, a security
recognition, SIFT, LEGO NXT guard can turn up to provide the required assistance.
I. INTRODUCTION
Security is a major concern that should be managed
diligently when it comes to the matter of accessing certain
restricted premises. Surveillance cameras are widely
implemented which allows maintaining a digital record of
persons and cars entering the concerned location. Computer
vision and image processing approaches provide an automated
system for analyzing these images and videos. Since the
number of vehicles are increasing day by day, these Figure 1. Smart gate equipped with computer vision based
technologies can provide easier and faster solutions in such authentication techinques
scenarios.
This paper is organized as follows. Section II provides a
Automatic license plate recognition (ANPR) systems [1] brief review of the related researches carried out in the past
have become a very useful technology to identify traffic years. Section III looks into the significant benefits of such a
offenses and thereby control it to a limited extent. Additional system and Section IV describes the methodology of our
benefits such as theft control or prevention, access control system. The description and testing of our prototype is
management etc. can ensure the safety of persons as well as provided in section V and VI. The details of our car dataset
cars. Nowadays ANPR systems are common in many public
areas such as parks, shopping malls etc. However, ANPR may
and the corresponding experimental results are added in work in [12] reviews several researches that have
section VII. Finally concludes the paper with section VIII. implemented ALPR in the mobile-based platform.
II. RELATED WORK B. AVMMR techniques
There are several research works available in the Zafar et. al [13] proposed a 2D Linear Discriminant
Intelligent transportation systems (ITS) field involving the analysis(LDA) based approach for car make and model
ANPR and Automatic vehicle Make and model recognition recognition. Experiments demonstrated that 2D LDA
(AVMMR) techniques. They are summarized here under performed better than the Principal Component Analysis
specific subsections, along with the researches related to face (PCA). A hierarchical classifier was designed in [14] that
detection. determine the class of vehicle at first and then proceed to
recognize vehicle make and model in a smaller group. The
A. ANPR techniques algorithm produced 96% accuracy on a dataset of over 280
Scale invariant feature transform (SIFT) describing local back view images of vehicles. Moreover, robustness to
image features are widely used in pattern recognition tasks [3]. illumination and weather conditions were guaranteed. Vehicle
The work in [4] is based on SIFT features. However model recognition in [15] used the geometry and appearance
background objects, poor illumination, blurring are found to of car emblems in the captured rear view images. Using linear
affect the recognition rate. A different approach based on the SVM binary classifier with HOG features, an accuracy of
SIFT algorithm was proposed in [5] for the localization of the 93.75% was obtained on a dataset of 1342 images of 8 car
license plates. It finds keypoint features in the images, which make and 28 different car models. A smart camera equipped
were used in the matching process between input images and with ANPR, AVMMR as well as color recognition
characters templates. The recognition was based on the capabilities are described in [16].
peculiarities in transition among pixels of characters. The The make and model recognition uses scalable vocabulary
behavior of these transitions which is unique to each character tree (SVT) which is built on the Speeded Up Robust Feature
helps in their classification. An experimentation of two sets of (SURF) descriptors [16]. The ANPR methodology follows a
60 brazilian License Plate images each yielded an accuracy of series of steps such as otsu binarization, canny edge detection,
88.33% and 70%. Another work [6] used SIFT based template contour extraction, and the usage of an OCR Reader. The
matching technique for localization of Jordanian number developed iCamera system had processing speed of 10 fps.
plates. The characters were further recognized on application Two specific methodologies for MMR are proposed in [17].
of OCR in the segmented license plates. The real time system relies on SURF and SVM for the make
Localisation of blue color areas in HSV color space was and model recognition. The visual content classification
performed as an intial step in the license plate recognition approach for a non- real time scenario make use of the Edge
method proposed by Azad et al. [7]. The method focused on Histogram, SURF and SIFT descriptors. The latter method
the detection of standard Iranian plates. Color and the yielded higher classification accuracy. The work in [18]
geometric aspects served as the cues to localize the license studied the effectiveness of combining several local feature
plates in the HSV image which is subdivided into numerous point detectors with the SIFT descriptor. The experimental
blocks. Any tilts in the located license plate was managed results on four databases was analyzed to report the
utilizing the features of pixel arrangements and yielded a good comparatively higher recognition rates resulting from the
result of about 45 degrees. Normal factoring with a similarity SIFT- DoG, SIFT- Multiscale Hessian, SIFT- Multiscale
measure was used in the character recognition stage and the Harris combinations.
resultant accuracy was 97%. An ANPR system focused on The recent work in [19] used frontal view of the vehicle
Indian license plates in [8] uses the template matching images with a SqueezeNet deep learning architecture for the
technique for character recognition. Segmentation stage realization of an AVMMR system. Experiments on over
suggests the use of rectangular projection of number plate or 291,602 images comprising 766 classes yielded an accuracy
an alternative method of template matching. However, testing of 96.33 %.
was conducted only in a very small set of 8 images with
various shape, illumination, view-angle, distance etc. C. Face Detection techniques
ANPR system in [9] makes use of the Fourier transform to Face detection and Face recognition methodologies are
detect inherent spatial frequency of the characters in a license commonly used for biometric authentication purposes [20].
plate. Once the plates are localized using fourier analysis, Viola- Jones proposed a major approach for face detection
characters are segmented based on connected components and which ensures faster detection along with high detection rate
recognized using Support Vector Machines (SVM). The [21]. The classifier was built on the AdaBoost algorithm,
character recognition method in [10] also uses SVM on which enables to select a small number of critical visual
providing the extracted normalized character subimages as the features from enormous potential features. The cascading of
feature vectors to perform the recognition. classifiers served to reduce the computation time as well. Face
The license plate recognition task is addressed using a You detection based on skin color likelihood utilizes a stochastic
only look once (YOLO) deep learning framework in [11]. model to compute the similarity between skin and non-skin
Using 7 convolutional layers, the system exhibited like regions [22]. Local Binary Patterns (LBP), Haar like
approximately 98.22% and 78% accuracy for license plate features was used to build the cascaded classifier. The concept
detection and recognition respectively. ANPR was commonly of snakes, which is contour or deformable curve was also used
implemented in a PC based platform to enable the processing for face detection. The application of color information in the
of high quality images in a shorter period of time. But snake models led to the facial feature extraction algorithm in
nowadays mobile based platforms are also being considered [23].
for performing the automatic number plate recognition. The
 Apart from the feature based approaches, statistical as well (3) The categorized information on the incoming faculty,
as neural network based methods were also found for face staff, students and guests can be utilized in managing
detection. Reference [24] proposed a hierarchical face the parking spaces to be allotted for each of them
detection system. The eigenface algorithm was used to select (4) A statistical data can be collected with respect to the
the possible face candidates which are further examined by a type of visitors as well as the model of car used
neural network. The template based face verification schema
was used to confirm each face region from the output of the IV. PROPOSED METHOD
neural network. Multi-order statistical descriptors capable of The Proposed methodology for the secure automatic gate
object classification and face recognition was introduced in system is depicted in the Figure 2.
[25]. Dimensionality reduction of descriptors using multiple
linear discriminant analysis and kernel linear discriminant
analysis makes it a faster algorithm suitable for potential Car stops at the
application in real time object classification including faces. gate
The method proposed in [26] was able to detect faces from
different angles and can handle occlusion to some extent. It is
a deep dense face detector based on deep convolutional neural
Sensor Sends Signal to
network. Another deep learning framework for face detection Camera
can be found in [27]. It is an extension of the faster Region
based CNN framework for generic object detection. Several
strategies for adapting the technique for resolving face Image Acquisition
detection tasks were proposed, including feature
concatenation, multi-scale training, hard negative mining etc.
III. SIGNIFICANCE OF THE SECURE AUTOMATIC GATE Preprocessing
SYSTEM
An automatic gate was proposed in [28] for controlling the
entry to a building gate. The system relies on a patch based
matching process wherein a patch of the entering vehicle Make and Model License Plate Detection of
image is matched with the vehicle images of database. The Recognition Recognition Driver Face
gate control system in [29] utilizes the license plate
information for the authorization purpose. The automatic
parking management and fee collection system [30] depends
solely on the license plate recognition for the access control.
Similar work is reported in [31] by the implementation of an Database
intelligent secure garage system based on the license plate
recognition technique. A recent work [32] was also found to
rely merely on the license plate recognition for realizing an
automated gate. The automated security gate attendant [33] Yes All Details
No
Match
capture both the visitor face and the license plates. The
authentication is based on the name entered by the visitor in
the interface as well as the license plate number. The security Report to Security
Gate Opens Guard
cannot be assured completely since the name entered by a
visitor cannot be considered as a valid information. The
system proposed in [34] includes automatic license plate
Figure 2. Proposed framework for the secure automatic gate
recognition, vehicle make and model detection and under-
vehicle inspection. The combination of these methods was
A. Pre-processing and extraction of region of interest
suggested to minimize the limitations of each individual
(ROI)
technique and increase security in vehicle inspection.
Database of toys cars is used in the implementation of our
The fusion of three techniques; License plate recognition, prototype. The database consists of 10 cars of different make
vehicle make and model recognition as well as face detection and model. The pre-processing of the captured real time image
can obviously strengthen the security level and can build a involves conversion to grayscale and noise removal by median
highly secure automatic gate at the entrance of restricted areas. filtering. Binarization and the connected component labelling
We aims to perform the recognition of Qatari car plates with serves to identify the target license plates based on their
a design suitable for replacing the traditional security gates in specific structural properties (Figure 3). The corresponding
Qatar University. The potential benefits of the system include: regions of interest (ROI) extraction identifies the area around
(1) We will have an opportunity to acquire more the headlights, upper and bottom grills. This region contains
information about the persons entering the campus the discriminating information capable of recognizing the car
which in turn enhances the security in the premises makes and models. The ROI selected from a sample image is
shown in the Figure 4.
(2) The waiting times due to the screening procedures
such as manual checking of the identity card can be
reduced
 Toy cars of apparently big size are used in this work since they
have a convenient size, which help to detect and recognize the
plate number.A street prototype is also prepared to setup the
demo.
Figure 3. Extracted License Plate
B. Software Design
The software requirements include MATLAB R2013a or
above with Image Processing and Computer Vision Toolbox,
and VLFeat Library. NXT Lego Software needs to be installed
and integrated with MATLAB. The LEGO Mindstorms NXT
is used as the electronic gate system in this design.
VI. TESTING THE PROTOTYPE
The system testing was done by simulating different
scenarios wherein different toy cars was prompted to enter
the gate. Figure 5 shows the interface of our application with
the corresponding results. The sensor detected the stopped
Figure 4. ROI selection for make and model recognition car, which is a Ferrari toy car. The program was able to
recognize its make and model successfully. Then, it
B. Make and Model Recognition
recognizes the plate number which is 12889. All the results
The AVMMR system follows the algorithm in [18], including the detected driver’s face are displayed on the
where the interest points are detected using DoG and feature screen. Finally, the car is granted permission by enabling the
extraction from those keypoints are done using SIFT opening of gate.
descriptors. The training part computes the SIFT features
from various images of different car make-models. A
euclidean distance based matching between the query
descriptor and all the prestored descriptors from the training
image points out the closest match for the test image. The
best match identified by the smallest distance determines the
make and model of the car.
C. Face Detection
The Face Detection algorithm was implemented using
Viola Jone’s method [35].
D. Number Plate Recognition
This module accepts the cropped license plates from the
preprocessing stage. Following the segmentation and
Figure 5. Interface of our system with the test results
character recognition, it provides the plate numbers as text
data. First, a dilation operation is applied to the binarized
The test results on the dataset demonstrated 100%
image to split the characters from each other if the characters
accuracy in recognizing the make and model of the car.
are close. Second, a horizontal and vertical line scanning are
However there was one wrong detection for license plate and
performed to find each character region. Then, each character
two wrong detections for face which can be due to light
is cropped out based on the start and end point information in
reflections on the windshield.
the horizontal direction.
More accurate algorithms can be experimented for the
The next step is to normalize the cropped characters by
recognition phases which helps to implement the system in a
making sure that there are no extra white spaces and they are
real scenario apart from this prototype. As far as the prototype
equal in size and matches the size of character images stored
is concerned, the recognition and detection takes only 2
in the database. The character image is compared with the
seconds. Optimized algorithms serve to deliver comparable
ones in the database using template matching and the best
processing times, which turns out to be a promising concept.
similar character is sorted out to recognize the target
The assured security is really helpful in areas such as
character.
international airport where we are safe to leave our cars
V. BUILDING THE PROTOTYPE especially for long term parking. The instances of car thefts
can be ruled out. Military premises can guarantee the
The hardware and software aspects in the design of authorized vehicle entries without a security personnel. These
prototype are described in the subsequent sections. systems tend to be more secure and unbiased.
A. Hardware Design
VII. CAR MAKE AND MODEL RECOGNITION ON QATAR
Logitech C920 HD pro webcam was used to capture UNIVERSITY SURVEILLANCE DATA
the frontal view images of the stopped car. An ultrasonic
sensor senses the event of stopping the car. LEGO There is no publicly available dataset for make and model
Mindstorms NXT, the programmable robotics kit with sensor, recognition of Qatar vehicles. We built a dataset for the same
motors was used to build the robot to control gate movement. utilizing the images captured by surveillance cameras at the
university gates. The dataset is comprised of 225 images in manual screening or the installed card reader. The system tests
total constituting 24 different make-models (Table 1). conducted with the help of our prototype demonstrated the
feasibility of real time implementation. Another contribution
TABLE 1. DETAILS OF QU CAR DATASET of the work is the dataset for Qatar car make and model
recognition which is built on real surveillance images. The
SL. No Car Make Types Number of preliminary results are promising and we intend to extend and
Models used build a complete dataset to include the different types of
1 Chevrolet 1 vehicles currently in use in the state of Qatar.
2 GMC 2
3 Honda 3 ACKNOWLEDGMENT
4 Lexus 2 This publication was supported by Qatar University
5 Mitsubishi 3 Collaborative High Impact Grant QUHI-CENG-18/19-1. The
6 Nissan 4 findings achieved herein are solely the responsibility of the
7 Toyota 9 authors. The contents of this publication are solely the
responsibility of the authors and do not necessarily represent
the official views of the Qatar University.

REFERENCES
[1] National Police Chief's Council. "Automatic Number Plate Recognition
(ANPR)". [Online]. Available:
https://www.npcc.police.uk/FreedomofInformation/ANPR.aspx.[Acce
ssed: Nov 28, 2019].
[2] C. Parker. (2 December 2018). "Warning to Brits over fake number
plates as simple tips could stop YOU becoming a victim". [Online].
Available: https://www.thesun.co.uk/uncategorized/7879382/warning-
brits-fake-number-plates/.[Accessed: Nov 28, 2019].
Figure 6. Sample images from our database
[3] S. Almaadeed, A. Bouridane, D. Crookes, O. Nibouche, "Partial
shoeprint retrieval using multiple point-of-interest detectors and SIFT
The variability of dataset is attributed to the different descriptors," Integrated Computer-Aided Engineering, vol. 22, (1), pp.
natural lighting conditions and the security gates from which 41-58, 2015.
the images are taken. Compared to the existing datasets, the [4] M. Zahedi and S. M. Salehi, "License plate recognition system based
images are taken while the vehicles are in motion which may on SIFT features," Procedia Computer Science, vol. 3, pp. 998-1002,
2011.
result in some blurry images as well.
[5] F. A. da Silva, A. O. Artero, M. S. V. de Paiva, and R. L. Barbosa.
Experiments were carried out on our raw dataset for the (2013), "ALPRs-A new approach for license plate recognition using
car make and model recognition. The Region of interest is the SIFT algorithm". [Online]. Available:
identified as the frontal region containing headlights and https://arxiv.org/abs/1303.1667. [Accessed: Nov 28, 2019].
grills. It is cropped out manually and used for recognizing the [6] K. M. A. Yousef, M. Al-Tabanjah, E. Hudaib, and M. Ikrai, "SIFT based
make and model of car. All the images are further resized to automatic number plate recognition," in 2015 6th International
Conference on Information and Communication Systems (ICICS),
1024x512 to have uniform dimensions. The experimental 2015, pp. 124-129.
results for car make and model recognition using the [7] R. Azad, F. Davami and B. Azad, "A novel and robust method for
approach in [18] is indicated in the Table 2. The SIFT automatic license plate recognition system based on pattern
descriptor is combined with the various detectors and the recognition," Advances in Computer Science: An International Journal,
comparitive analysis indicate that SIFT- DoG and SIFT- vol. 2, (3), pp. 64-70, 2013.
Hessian combination delivers the maximum accuracy of [8] K. Bhosale, J. Jadav, S. Kalyankar, R. Bhambare, "Number Plate
Recognition System for Toll Collection," International Journal of
74.63%. Emerging Technology and Advanced Engineering, vol. 4, (4), pp. 729-
732, 2014.
TABLE 2. MAKE- MODEL RECOGNITION ACCURACIES [9] G. Lekhana and R. Srikantaswamy, "Real time license plate recognition
system," International Journal of Advanced Technology & Engineering
Method Accuracy (%) Research, vol. 2, (4), pp. 5-9, 2012.
DoG 74.63 [10] A. Kaur, S. Jindal and R. Jindal, "License plate recognition using
Hessian 74.63 support vector machine (SVM)," International Journal of Advanced
Harris Laplace 68.66 Research in Computer Science and Software Engineering, vol. 2, (7),
2012.
Hessian Laplace 65.67 [11] R. Chen, "Automatic License Plate Recognition via sliding-window
Multiscale Harris 71.64 darknet-YOLO deep learning," Image Vision Comput., vol. 87, pp. 47-
Multiscale Hessian 73.13 56, 2019.
[12] L. Connie, C. K. On and A. Patricia, "A Review of Automatic License
Plate Recognition System in Mobile based Platform," Journal of
VIII. CONCLUSION AND FUTURE WORK Telecommunication, Electronic and Computer Engineering (JTEC),
vol. 10, (3-2), pp. 77-82, 2018.
We proposed a real time automatic gate with the fusion of [13] I. Zafar, E. A. Edirisinghe, S. Acar, H. E. Bez, "Two-dimensional
car make and model recognition, license plate recognition as statistical linear discriminant analysis for real-time robust vehicle-type
well as face detection. The deployment of such a gate can recognition," in Real-Time Image Processing, 2007, pp. 649602.
minimize the human effort and increase the level of security [14] H. Emami, M. Fathi and K. Raahemifar, "Real time vehicle make and
in the intended premise. The faster authentication approach model recognition based on hierarchical classification," International
Journal of Machine Learning and Computing, vol. 4, (2), pp. 142, 2014.
can reduce the traffic blocks that may arise as a result of
[15] D. F. Llorca, D. Colás, I. Daza, I. Parra, M. A. Sotelo, "Vehicle model on International Conference on Multimedia Retrieval, 2015, pp. 643-
recognition using geometry and appearance of car emblems from rear 650.
view images," in 17th International IEEE Conference on Intelligent [27] X. Sun, P. Wu and S. C. Hoi, "Face detection using deep learning: An
Transportation Systems (ITSC), 2014, pp. 3094-3099. improved faster RCNN approach," Neurocomputing, vol. 299, pp. 42-
[16] R. Baran, T. Rusc and P. Fornalski, "A smart camera for the surveillance 50, 2018.
of vehicles in intelligent transportation systems," Multimedia Tools [28] K. Sundus and S. Mamare, "Using Digital Image Processing to Make an
Appl, vol. 75, (17), pp. 10471-10493, 2016. Intelligent Gate," (IJACSA) International Journal of Advanced
[17] R. Baran, A. Glowacz and A. Matiolanski, "The efficient real-and non- Computer Science and Applications, vol. 5, (5), 2014.
real-time make and model recognition of cars," Multimedia Tools [29] M. M. Aziz, "Ishik University Gate Control based on Kurdistan License
Appl, vol. 74, (12), pp. 4269-4288, 2015. Plat," International Journal of Enhanced Research in Science,
[18] S. Al-Maadeed, R. Boubezari, S. Kunhoth, A. Bouridane, "Robust Technology & Engineering (2319-7463), vol. 5, (1), pp. 34-38, 2016.
feature point detectors for car make recognition," Comput. Ind., vol. [30] M. M. Rashid, A. Musa, M. A. Rahman, N. Farahana, and A. Farhana,
100, pp. 129-136, 2018. "Automatic parking management system and parking fee collection
[19] H. J. Lee, I. Ullah, W. Wan, Y. Gao, Z. Fang , "Real-time vehicle make based on number plate recognition," International Journal of Machine
and model recognition with the residual SqueezeNet architecture," Learning and Computing, vol. 2, (2), pp. 94, 2012.
Sensors, vol. 19, (5), pp. 982, 2019. [31] A. U. Ahmed, T. M. Masum and M. M. Rahman, "Design of an
[20] S. Al-Maadeed, M. Bourif, A. Bouridane, R. Jiang, "Low-quality facial automated secure garage system using license plate recognition
biometric verification via dictionary-based random pooling," Pattern technique," International Journal of Intelligent Systems and
Recognit, vol. 52, pp. 238-248, 2016. Applications, vol. 6, (2), pp. 22, 2014.
[21] P. Viola and M. J. Jones, "Robust real-time face detection," International [32] L. T. A. Al-Mahbashi, N.A. Yusof, S. Shaharum, M. S. Karim, A. A.
Journal of Computer Vision, vol. 57, (2), pp. 137-154, 2004. Faudzi, "Development of automated gate using automatic license plate
[22] Y. Ban, S. K. Kim, S. Kim, K. A. Toh, S. Lee, "Face detection based recognition system," in Proceedings of the 10th National Technical
on skin color likelihood," Pattern Recognit, vol. 47, (4), pp. 1573-1585, Seminar on Underwater System Technology, 2019, pp. 459-466.
2014. [33] P. Bongard, Automated Security Gate Attendant, United States patent
[23] K. Seo, W. Kim, C. Oh, J.-J. Lee, "Face detection and facial feature US 8,254,631. Aug 28 2012.
extraction using color snake," in Proceedings of the 2002 IEEE [34] C. Anagnostopoulos, I. Giannoukos, T. Alexandropoulos, A Psyllos, V.
International Symposium on Industrial Electronics, L'Aquila, Italy, Loumos, and E. Kayafas , "Integrated vehicle recognition and
2002, pp. 457-462. inspection system to improve security in restricted access areas," in
[24] C. Tsai, W.C. Cheng, J.S. Taur and C.W. Tao, "Face detection using 13th International IEEE Conference on Intelligent Transportation
eigenface and neural network," in 2006 IEEE International Conference Systems, 2010, pp. 1893-1898.
on Systems, Man and Cybernetics, 2006, pp. 4343-4347. [35] Mathworks, Vision Cascade Object Detector. [Online]. Available:
[25] A. Mahmood, M. Uzair and S. Al-Maadeed, "Multi-order statistical https://www.mathworks.com/help/vision/ref/vision.cascadeobjectdete
descriptors for real-time face recognition and object classification," ctor-system-object.html [Accessed: Nov 28, 2019]
IEEE Access, vol. 6, pp. 12993-13004, 2018.
[26] S. S. Farfade, M. J. Saberian and L. Li, "Multi-view face detection using
deep convolutional neural networks," in Proceedings of the 5th ACM

View publication stats