International Journal of Computer Applications (0975 - 8887)

Volume 182 - No.5, July 2018

A Smart Image Processing-based System for Parking

Space Vacancy Management

Benjamin Kommey Ernest O. Addo Andrew S. Agbemenu

Department of Computer Engineering Department of Computer Engineering Department of Computer Engineering
KNUST KNUST KNUST
Kumasi, Ghana Kumasi, Ghana Kumasi, Ghana

ABSTRACT In order to assist drivers in this regard, some parking guidance sys-
tems that attempt to notify the drivers of vacant spots have been
Drivers often encounter problems associated with locating empty designed and installed in various parking lots. A very common sys-
parking slots in parking areas. This paper presents a smart tem features sensors and light indicators at each parking space. The
parking lot management system which operates using image indicator which is lit by default is positioned such that it can be
processing. An image processing algorithm is used to detect viewed from an appreciable distance and turns off only when a
empty parking areas from aerial images of the parking space. vehicle occupies the spot. Drivers are able to spot vacant spaces
The algorithm processes the image, extracts occupancy infor- from a distant without having to move around the whole parking
mation concerning spots, and their positions thereof. The sys- lot. There are numerous variants of this basic system proposed in
tem also reports if individual parking spots are occupied or oth- recent years. These systems mainly vary based on the vehicular
erwise. Occupancy information is made available to newly ar- detection mechanism employed. The methods used ranged from
riving drivers by projecting it unto large displays positioned at the use of radio frequency identification (RFID) technology [4, 2]
vantage points near the vicinity. The smart parking lot manage- or infrared sensors [11] to ultrasonic sensors [7]. Among these is
ment system reduces the stress and time wastage associated with an RFID based smart parking system prototype developed by [8]
car parking and makes management of such areas less costly. which implements an automated process of vehicular checking in
and checking out for parking lots using RFID reader devices. In
General Terms another system proposed by [13], the effect of the Earth’s magnetic
field on magnetic sensors is harnessed to determine the vacancy
Parking Space Vacancy Management, Image Processing status of parking spaces. Although these mechanisms achieve the
basic intended function of vehicular detection, their operations are
Keywords often marred by environmental factors such as changes in weather
conditions and/or the presence of electromagnetic interferences.
Edge Detection, Morphological Dilation,WiFi, ROI, FOV
Upon detection of a vehicle in the allotted space, the system re-
quire a way to inform drivers of the occupancy status of that space.
1. INTRODUCTION In an attempt to do this, some systems implement vehicular ad hoc
In recent times, household vehicles have increased rapidly with networks (VANETs) with the assumption that the vehicles are ca-
population growth and economic development. Meanwhile, spaces pable of transmitting and receiving information to and fro sensors
required to park these cars in public places has rather been in short within a certain range [15, 3, 9]. The drawback of VANET systems
supply and are becoming very costly too [14, 6, 12]. This rising im- is that they require that vehicles have a dedicated tamper-resistant
balance is very evident in major attraction areas that witness the in- communication device: a feature that older vehicles lack. Without
flux of large numbers of people visiting in motor vehicles. Finding modification to these vehicles, the operation of VANET system is
parking spaces in such places during events is challenging. Vehi- flawed. There is therefore the need for a parking guidance system
cle owners often undergo the extremely frustrating process of driv- that works well with all types of vehicle without modication irre-
ing through the entire area in an unguaranteed search of available spective of their types and natures. An additional factor of concern
parking spots. Productive working hours are lost, fuel is expended, for VANET systems is the network infrastructure employed for the
and carbon emissions from conventional engine vehicles increases. vehicle detection sensors. Traditional systems such as [13] have
Management of these areas often resort to simply providing more often used wired infrastructure however recent developments have
parking spaces in order to check this problem. This effort how- suggested the use of wireless sensor networks. In these networks,
ever, compounds the existing problem as locating vacant spots be- all detection sensors are able to wirelessly communicate with a cen-
come increasingly difficult when the lot is more than half filled. It is tral controller that has information about all nodes. Sensors can be
therefore quite clear that proper management of available spaces is repositioned effortlessly without the need for costly reinstallations.
the most suitable solution. Contrary to the conventional use of war- [5, 10] used ZigBee networks in their system implementations. In
dens, smart systems that provide readily available spot vacancy in- the work done by [3], WiFi was employed in the broadcast vacancy
formation are needed to effectively satisfy these parking demands. reports in the infrastructure. This paper presents the design and im-

1
 International Journal of Computer Applications (0975 - 8887)
Volume 182 - No.5, July 2018

plementation of a smart system which uses image processing meth-

ods for vehicular detection coupled with wireless network technol-
ogy for monitoring parking areas.

2. PROPOSED SYSTEM MODEL

9
8
9
8
2.1 Architecture of System

10
10

7
7

11 12
11 12

6
6
The smart parking system proposed in this paper consists of three

ROI 2

5
5
ROI 1

ROI 4
ROI 3
main components. These are the parking detection nodes with in-

13
13

4
4
corporated WiFi access points (APs) deployed within each ma-

14
14

3
3

15
15

2
jor section of the parking facility, a wireless local area network

16
16

1
1
(WLAN) integrated local base station, and a notification system
for information delivery. The architecture of the proposed system
is shown in Figure 1.

Fig. 2. Camera positions with their respective FOVs and assigned ROIs.

MCU s AP 2.2 System Operation

Base Station
Node
The operation of the system involved four major sub-operations as
Aerial Cameras shown in Figure 3
9
8
9
8

10
10

7
7

Image
11
11

6
6

Acquisition
12
12

5
5

and Processing
13
13

4
4

14
14

3
3

15
15

2
2

16
16

1
1

Display Device

Parking Area
System Data
Initialization Interpretation
Fig. 1. Architecture of proposed parking management system.

The parking area is logically demarcated into major sections for Display
easy deployment of parking detection nodes. Each major section of Update
the area is equipped with a detection node. A node consists of wide
field view cameras for taking snapshots of the assigned parking mi-
nor section at regular intervals and IEEE 802.11 b/g/n compatible
microcontroller unit (MCU) for communication between the node Fig. 3. Operational modules of proposed system.
and the base station. As shown in Figure 2, a node is affixed at a
higher altitude close to the perimeter of its assigned parking section These modules are system initialization, image acquisition and pro-
such that the node overlooks the area and the cameras are able to cessing, data interpretation, and display update. The last three pro-
capture their individual regions of interest (ROIs) or minor sections cesses were then repeated as long as the system was active. The ini-
in their field of view (FOV). tialization of the parking guidance system takes place once, when
Each node’s control unit is built around the TI CC3200 microcon- the system is being set up for the first time or after a replacement
troller which operates on an industry-standard ARM Cortex-M4 of any of the systems module (shown by dashed arrow). During
core running at 80 MHz. The cameras are connected to the MCU initialization, a refresh signal is sent from the node’s controller to
via the chip’s fast parallel camera peripheral interface. The image the image sensors in order to activate the process. The aerial cam-
sensor interface receives captured images from the camera imme- eras capture images of their assigned parking lot section whiles the
diately they are taken. The controller forwards the collected data to area is empty. The images are transferred from the cameras to the
the base station over the setup local area network. For larger park- MCU via the fast parallel image sensor interface. The MCU pack-
ing area, repeaters could be employed to enhance wireless com- ages the received images into TCP packets and transmit the data
munication between the nodes and the central station by regen- from its configured on-chip Websocket (RFC 6455) HTTP server
erating signals transmitted over the network before they becomes over WiFi to the base stations Websocket client application. The
too weak or corrupted. The base station receives the data from the images are extracted from the received packets and are then pro-
nodes, processes them, retrieves vacancy information from the im- cessed using the image processing algorithm described in a later
ages, and updates the occupancy status of the parking lot and its section of this paper. Coordinates of the marked individual park-
vacancy thereof on the display positioned at vantage points. ing slots and their boundaries are obtained and stored for reference

2
 International Journal of Computer Applications (0975 - 8887)
Volume 182 - No.5, July 2018

during subsequent image processing of images. These coordinates Canny edge detection is performed to identify the edges of elements
could be used as references as long as positions and orientations in the binarized image. The edge detection process uses the first
of the nodes are fixed. Replacement of faulty cameras, change of derivative of the Gaussian to approximately optimize the product of
positions of existing devices or installation of new ones would re- signal-to-noise ratio and localization. The Canny edge detector first
quire re-initialization of the system thereafter. Figures 4, 5, 6, 7, smoothens the binarized image with a Gaussian filter with kernel
and 8 illustrate the processing of the initial frame. At the base size (σ) of 5 to reduce noise and unwanted details and textures. The
station, the received images undergo a preprocessing stage which gradient is then computed using the Sobel-Feldman gradient oper-
involves the reduction of the cost of computing the image informa- ator. Non-maximum suppression is performed on the combined re-
tion [1]. The coloured image is grayscaled using the Luminosity sult of the Sobel-Feldman convolutions to thin out edges. Double
method which calculates each resultant gray image pixel, Igray as thresholding using thresholds, T1 = 0.6 and T2 = 0.85 was done to
a weighted sum of the three linear-intensity RGB values. Igray is ensure less noise and fewer false edges. Finally edge tracking by
related to its respective colour image RGB values: Ired , Igreen , and hysteresis was performed to determine which weak edges were ac-
Iblue by equation 1. tual edges. Figure 6 shows the image obtained from applying the
canny edge detection algorithm to the image in Figure 5.
Igray = 0.213(Ired ) + 0.715(Igreen ) + 0.072(Iblue ) (1)
Figure 4 shows the image of an empty minor section of a test park-
ing area.

Fig. 6. Strong edges identified in binarized test image using the Canny
edge detection algorithm.

Fig. 4. An empty minor section of a test parking space.

A four stage morphological dilation was performed on the resulting
image for better and clearer outer borders. Four different flat linear
The grayscaled image is then linearly normalized by contrast structuring elements were applied in the morphological operations.
stretching to adjust image intensity values and compensate for poor The shapes and sizes of the structuring elements (SEn ) were:
image contrast which may occur due to glare. The stretched image  
is smoothed to remove noise and enhance accuracy of object de- SE1 = 1 1 1
tection and then binarized using Otsu’s global thresholding method
which chooses the threshold to minimize the intraclass variance of " #
1
the thresholded black and white pixels. The resulting blurred and SE2 = 1
binarized image is shown in Figure 5. 1
" #
1 0 0
SE3 = 0 1 0
0 0 1
" #
0 0 1
SE4 = 0 1 0
1 0 0
The outputs of the complete dilation procedure are shown in Fig-
ure 7. A flood-fill operation is performed on background pixels of
the binary image to fill holes in detected objects as shown in Fig-
ure 8. Each hole is a set of background pixels that cannot be reached
by filling in the background from the edge of the image.
With the initialization of the system, the ROIs in Figures 7 and 8
are identified, marked out as shown in Figures 9 and 10 respec-
Fig. 5. Grayed, linearly normalized, and binarized image of empty minor tively. Properties of the marked regions such as coordinates of ele-
section. ment edges and mean pixel count are stored. Figure 11 shows the

3
 International Journal of Computer Applications (0975 - 8887)
Volume 182 - No.5, July 2018

Fig. 7. Resulting edge image after fourth dilation Fig. 10. Identified and marked park spaces in ROI of Figure 8

Fig. 11. A graph showing the mean intensity per empty parking region
from Figures 9 and 10

count infers occupancy of the corresponding spot. This occupancy

information is then sent from the base station to the display devices
for use by incoming drivers.

3. TESTING, RESULTS AND DISCUSSION

The proposed parking space vacancy management system was
tested to determine its ability to accurately extract vacancy infor-
mation from captured images of the parking space. A test image
was obtained using an 8MP aerially positioned camera. The im-
age used is shown in Figure 12. The image was grayed, stretched,
Fig. 8. Dilated edge image with holes filled
smoothed, and binarized by using the image processing techniques
described earlier. Figure 13 shows the grayed, smoothed and bina-
measured mean pixel value per identified spot in Figures 9 and 10 rized test image.
for the empty parking space during initialization.

Fig. 12. Test image showing parking space minor section with cars in
Fig. 9. Identified and marked park spaces in ROI of Figure 7 some spots.

During subsequent operations of the system, parking areas in the The Canny edge detection algorithm was applied to identify strong
filled binarized images are marked with bounding-boxed regions edge elements in the binarized image. The obtained edge image is
based on the coordinates obtained from the initialization process. shown in Figure 14. The edge image was then dilated with the four
Pixel count in each region is determined and compared against stage morphological procedure using the four structural element
the stored initial values. Parking spots with cars would have extra objects stated in the previous section (Figure 15). Holes in the di-
edges within the marked regions which increases the mean inten- lated test image were then filled. Figure 16 shows the result of the
sity within the regions. A substantial increase in a region’s pixel background pixel flood-filling procedure.

4
 International Journal of Computer Applications (0975 - 8887)
Volume 182 - No.5, July 2018

Fig. 13. Grayed, contrast stretched, and binarized test image Fig. 16. Dilated edge test image with holes in detected objects filled

Fig. 14. Edge test image obtained using the Canny algorithm. Fig. 17. Marked park spaces in dilated test edge image

Fig. 15. Completely dilated test edge image

Fig. 18. Marked park slots in filled edge image
The parking spots in Figures 15 and 16 were all marked as shown in
Figures 17 and 18 and their individual bounded pixel measurements The dilated image approach recorded a mean percentage increase
were taken. The graph in Figure 19 shows the pixel counts for each in pixel count was 133.7% with an MAD of 86.5. It was observed
park spots in both images. that the use of filled edge images provided greater intensity in occu-
Graphical data in Figure 19 shows a substantial increase in the pied parking regions and hence provided a much more solid basis
mean pixel count in regions which had cars occupying park spots for declaring occupancy especially in areas which lie in the mid-
when compared to the data in Figure 11’s graph. In spot vacancy de- dle of the camera’s FOV as compared to the use of only the dilated
termination, the average percentage increase in intensity recorded edge images. However, due to the position and height of the camera
for occupied spots when using the filled images (Figures 10 and used, the filled image approach faired averagely in detecting vacant
18) was 166.9% with a mean absolute deviation (MAD) of 96.4. spots (spots 6 and 8) located between occupied ones at the extreme

5
 International Journal of Computer Applications (0975 - 8887)
Volume 182 - No.5, July 2018

5. REFERENCES
Fig. 19. A graph showing the mean intensity per parking region from Fig-
ures 17 and 18
[1] A. S. Agbemenu, J. Yankey, and E. O. Addo. An automatic
number plate recognition system using opencv and tesseract
right side of the sensor’s FOV. In these instances, the dilated edge ocr engine. International Journal of Computer Applications,
image approach exhibited good success. 180:1–5, 2018.
[2] J. P. Benson, T. O’Donovan, P. O’Sullivan, U. Roedig, and
4. CONCLUSION C. Sreenan. Car-park management using wireless sensor net-
works. In 31st IEEE Conf. Local Computer Networks, pages
The proposed system accurately detected the presence of cars in 588–595, 2006.
parking slots. The filled image approach performed better than the [3] M. Caliskan, D. Graupner, and M. Mauve. Decentralized dis-
use of dilated edge images to determine vacancy of parking spots. covery of free parking places. In VANET 06: 3rd international
The two approaches could be fused into one system. The camera workshop on Vehicular ad hoc networks, pages 30–39, 2006.
position could be adjusted to improve the performance. From the
[4] S. C. Hanche, P. Munot, P. Bagal, K. Sonawane, and P. Pooja.
above set of test results, it has been shown that the proposed image-
Automated vehicle parking system using rfid. International
processing-based system is a viable option for parking space va-
Journal on Research and Development, 1:89–92, 2013.
cancy management. Other technologies such as automatic number
plate recognition and traffic light control could be conjoined with [5] Y. Hirakata, A. Nakamura, K. Ohno, and M. Itami. Naviga-
this system to form holistic intelligent transport systems. tion system using zigbee wireless sensor network for parking.
In 12th International Conference on ITS Telecommunications,
pages 605–609, 2012.
[6] T. Huang, Z. Zeng, and C. Li. Design and implementation of a
prototype smart parking (spark) system using wireless sensor
networks. In International Conference on Neural Information
Processing, pages 506–515, 2012.
[7] A. Kianpisheh, N. Mustaffa, P. Limtrairut, and P. Keikhos-
rokiani. Smart parking system (sps) architecture using ultra-
sonic detector. International Journal of Software Engineering
and Its Applications, 6:55–58, 2012.
[8] Z. Pala and N. Inanc. Implementation of rfid technology in
parking lot access control system. In 12th International Con-
ference on ITS Telecommunications, pages 1–3, 2007.
[9] R. Panayappan, J. M. Trivedi, A. Studer, and A. Perrig. Vanet-
based approach for parking space availability. In VANET 07:
Fourth ACM International Workshop on Vehicular Ad-Hoc
Network, pages 75–76, 2007.
[10] S. Srikanth, P. Pramod, K. Dileep, S. Tapas, M. U. Patil, and
C. B. N. Sarat. Design and implementation of a prototype
smart parking (spark) system using wireless sensor networks.
In International Conference on Advanced Information Net-
working and Applications Workshops, pages 401–406, 2009.
[11] S. Srikurinji, U. Prema, S. Sathya, and P. Manivannan. Smart
parking system architecture using infrared detector. Interna-
tional Journal of Advanced Information and Communication
Technology, 2:1113–1116, 2016.
[12] D. Teodorovic and P. Lucic. Intelligent parking systems.
European Journal of Operational Research, 175:16661681,
2006.
[13] J. Wolff, T. Heuer, H. Gao, M. Weinmann, S. Voit, and
U. Hartmann. Parking monitor system based on magnetic eld
sensors. In IEEE Conference on Intelligent Transportation
Systems, pages 1275–1279, 2006.
[14] G. Yan, S. Olariu, M. C. Weigle, and M. Abuelela. Smart-
parking: A secure and intelligent parking system using notice.
In 11th International IEEE Conference on Intelligent Trans-
portation Systems, pages 569–574, 2008.
[15] H. Zhao, L. Lu, C. Song, and Y. Wui. Ipark: Location-
aware-based intelligent parking guidance over infrastructure-
less vanets. International Journal of Distributed Sensor Net-
works, 2012:1–2, 2012.