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

net/publication/320736588

Automatic Parking Space Detection System

Conference Paper · October 2017

DOI: 10.1109/ICMIP.2017.4

CITATIONS READS
12 4,648

4 authors, including:

Nazia Bibi Hassan Dawood

University of Engineering and Technology, Taxila University of Engineering and Technology, Taxila
1 PUBLICATION   12 CITATIONS    45 PUBLICATIONS   167 CITATIONS   

SEE PROFILE SEE PROFILE

Ping Guo
Beijing Normal University
326 PUBLICATIONS   1,799 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

StarCraft View project

NSFC-CAS (U1531242) View project

All content following this page was uploaded by Nazia Bibi on 04 April 2019.

The user has requested enhancement of the downloaded file.

 2017 2nd International Conference on Multimedia and Image Processing

Automatic Parking Space Detection System

Nazia Bibi 1, Muhammad Nadeem Majid2 , Hassan Dawood3, and Ping guo4, *
1,2,3
Software Engineering Department UET, Taxila, Pakistan
3,4
Image Processing and Pattern Recognition Laboratory, Beijing Normal University, Beijing, China
e-mail: [email protected], [email protected], [email protected], [email protected]

Abstract—Searching a suitable parking space in populated

metropolitan city is extremely difficult for drivers. Serious II. LITERATURE SURVEY
traffic congestion may occur due to unavailable parking space. Various methods and techniques have been proposed to
Automatic smart parking system is emerging field and overcome the problem of parking in the congested areas.
attracted computer vision researchers to contribute in this Ming-Yee Chiu et al. proposed a method for counting the
arena of technology. In this paper, we have presented a vision
vehicles at the checkpoint from which the number of
based smart parking framework to assist the drivers in
available parking spaces can be counted [1]. The counting is
efficiently finding suitable parking slot and reserve it. Initially,
we have segmented the parking area into blocks using performed by installation of the induction loop sensors under
calibration. Then, classify each block to identify car and the road surface. Although the usage of sensors was less
intimate the driver about the status of parking either reserved costly, not easily affected by environmental conditions and it
or free. Potentially, the performance accuracy of detects accurately however, it installation was difficult and
recommended system is higher than state of the art hardware cause damage to roads. It was also difficult to maintain it in
solutions, validating the supremacy of the proposed framework. case of malfunction [2]. Moreover, the exact locations of free
parking area cannot be determined because the counting
Keywords-smart parking management; automatic parking; method is not able to give the detail information, it just
slot recognation; parking space detection; machine learning records the number of vehicles passing the checkpoints [3].
The other detection methods were based on use of
I. INTRODUCTION sensors like ultrasonic, infrared and microwave for the
detection of vehicles [4]. These sensors are placed beneath
Now a day’s most of the parking areas are manually every parking space. Wan-Joo Park et al. proposed the use
managed by human manpower and there is no automatic of ultrasonic sensors mounted on the cars to search for a free
system to manage the parking area in an efficient way. There parking space. The disadvantage of this method was that the
is great analogy that when a driver enters any of the parking sensors are easily affected by weather conditions like rain,
lot he must look for some kind of information board that tells temperature, snow and fast air breeze. Another method was
him about the status of the parking lot that whether it is fully presented by Vamsee K. Boda et al. based on wireless sensor
occupied, partly occupied or vacant. Most of the times the nodes. This method was less costly and it uses the wireless
drivers have to circle around the parking area in search of the sensor nodes implemented at the critical places like the lane
free parking space. This kind of problem mostly occur in turns, entrance and exit positions of the parking lot. The total
cities near the shopping malls, hospitals etc., where the number of cars in the parking area can be determined by the
number of vehicles is greater as compared to the parking difference of incoming and outgoing cars [5].
spaces. The other kinds of detection methods are presented based
The process for searching the free parking space is time on vision based methods. Through vision based methods, the
consuming and also wastage of fuel. Most of the times the whole parking area available for parking can be examined
parking spaces remain unoccupied, however the total though the camera, the data is than processed and the result
occupancy is low because of bad management of parking lot. generated will determine the exact number and location of
This causes ineffective use of the parking area and also the free parking spaces. Zhang Bin et al. proposed that vision
results in traffic jams and congestion near the parking lots. based parking space detection methods are very easy to
To properly manage the parking lot and display each install, low in cost and the detector can be easily adjusted
parking division’s information to the drivers before entering according to requirements. Moreover, the data obtained from
the parking lot have become an important issue to be images is very rich. However, the defects in the vision
resolved. In this paper, a system is proposed that will detect method are that the accuracy is highly dependent upon the
the total number of available parking spaces and displays the position of the camera.
information to the drivers so that they can easily parked their Thomas Fabian proposed an unsupervised vision based
cars. A web camera is used to get the images of the parking system for parking space occupancy detection. The proposed
area and image processing techniques are used to detect the system has low complexity in computation and needs less
presence or absence of cars to count and locate the available image frames per minutes. He claims that the major problem
parking spaces .The status of the parking lot is updated in images detection is the occlusions and shadows [6]. For
whenever a car enters or leaves the parking lot. unsupervised learning more advanced clustering algorithms

978-1-5090-5954-6/17 $31.00 © 2017 IEEE 11

DOI 10.1109/ICMIP.2017.4
are used. H. Ichihashi et al. proposed that the vision based
parking space detection system are mostly affected by
Live stream No. of
weather and lighting condition like the falling of rain drops video

Output
on the lens of camera during heavy rainfall. Low and high CPU Free and

Input
lighting conditions. For this reason the cameras are mostly Reserved
Parking
used for the detection of vehicles in the indoor parking areas
Blocks
not for the outdoor parking lots [7].
R. Yusnita et al. presented a method in which a brown
color round patch was drawn in each parking space manually. Recorded
When the system is initialized it looks for the rounded shape Video Environmental LCD
in each space, if patch is detected that particular space is File parameters Display
considered as free and will be displayed the driver [8].
When the patches are blocked by objects(vehicles) then
Figure 1. Block diagram of proposed system.
the system assumes the particular spaces are filled by
vehicles. The system was good enough for managing the
Initially, the parking arena have no parking lines. User will
parking lot, however it does not work well in heavy rainfall
manually input the coordinates of parking area and vehicle
and snow. N. True proposed an efficient parking space
detection by using the combinations of color histogram and intended to be parked. The system automatically generates
vehicle features detection [9]. Najmi Hafizi proposed an virtual parking lines keeping in view the size of vehicle. The
image-based method for detection of free slots in the outdoor maximum capacity of parking slots in our training model is
parking area. A low resolution web camera is utilized for fourteen. A unique numeric label is assigned to each parking
acquiring images of the parking lot that reduces the cost lot from 1-14.
greatly. The images acquired are preprocessed and then a After the parking arena is divided in the virtual blocks,
pair of ROI is applied on every division of the parking lot, our system will check the existence of car in each block.
which increases the reliability of detecting vehicles [10]. Binary filter is applied on image and then inverse binary to
In [11] an image processing technique was presented extract car as region of interest ROI. Computing the value of
that captures the brown circle drawn on the parking area and connected region in ROI and setting the threshold value
process it to detect whether that parking division is free or greater then eighty as reserved parking slot. The number of
reserved. In [12] an image of car is saved as reference and the free blocks will be indicated to the divers in green and
the other images are matched with the reference image by the reserved blocks will be indicated in red color.
edge detection technique and information about free and
reserved slots are displayed. There are number of methods IV. ALGORITHM OF PROPOSED SYSTEM
has been proposed for the extraction of features from the The main steps of the proposed algorithm for parking space
images like [13] - [17]. detection are shown in Fig. 2
In this paper, we have designed and implemented a
framework for automatic parking system. The experimental 1. System will get Livestream video of the parking lot
results have shown the remarkable accuracy, achieved by from camera.
proposed system as compared to state-of-the-art 2. Images are captured when a car enters or leaves the
methodologies. parking lot.
The proposed parking lot model is discussed in detail in 3. RGB Images are converted to grayscale images.
Section 3. In Section 4 the proposed algorithm of the parking 4. Do calibration
system is discussed in detail .Section 5 deals with
 First select the coordinates of the parking lot.
experimental results. Conclusion is given in section 6.
This will crop the extra space other than
III. PROPOSED PARKING MODEL AND METHODOLOGY parking lot from the image.
 Secondly select the coordinates of the single
The main flow of framework is shown is Fig. 1. Videos
parking slot. This will divide the parking lot
were acquired from the top view of parking arena, from ten
into equal size slots.
feet heighted camera. To strengthen the recognition capacity
5. Each block is converted from grayscale to binary and
of system video data was captured at different
then inverse binary to get the car in white color and
environmental conditions and temporal shifts.
parking area into black color.
Video is segmented into frames. Then from each
6. Threshold value is calculated in every block to detect
segment a key frame is extracted and further processing is
whether that block contain car or not.
applied on this key frame, to reduce computational
7. If value is less than threshold value than that block is
complexity. When radio controlled toy car enter or leave the
free and available for parking car and if value is greater
parking lot from parking arena, motion of car is estimated
than block is occupied.
by key frame subtraction.

12
 TPS = Total Parking slots
Input ANC = Actual Number of Cars
Livestream PNC = Predicted Number of Cars
video
Performance= 1 - ((| ANC - PNC |)/TPS) *100 (1)
The percentage of error in the proposed system will be
find by using the equation (2)

Get image Percentage Error = ((| ANC - PNC |)/TPS) *100 (2)
frames RGB to Grayscale

Get
coordinates
of parking Do Calibration
area

Get
coordinates
of car Parking area divided
into Blocks

Convert Block to
binary & than inverse
binary Figure 3. Free slots detection.

The testing of the proposed methodology is performed

Get value of on the model parking lot having total parking slots TPS=
connected region to 14.The test are performed in different weather conditions as
detect car shown in Table 1.
TABLE I. ACCURACY OF PROPOSED METHOD.
Vehicle Number Correct False Accuracy
No. of Free
and Reserved Appearance of tests detection detection
Blocks

Sunny Clear 50 50 0 100%

Figure 2. Algorithm of proposed system.
day
Occluded 50 48 2 96%
V. RESULTS AND DISCUSSION
The proposed method is implemented in MATLAB. The
online system is getting images from the camera while Cloudy Clear 50 49 1 98%
offline system is getting images from a video file. The result
day
of the online system shows that the proposed algorithms has
Occluded 50 46 4 94%
efficiently detected the available parking slots and notify the
drivers. The proposed algorithm is implemented on the
model parking lot having space for 14 cars. The Slots
having no car are shown as free while the Slots having car The accuracy of the proposed algorithm is found to be
in it are shown as reserved to drivers as shown in Fig. 3. 100%, 98% 96% and 94%.The results shows that when the
To test the performance of our proposed algorithm, the captured images of the parking lot are not clear because of
accuracy of the system is measured with images taken at less lighting or occlusions, the efficiency decreases and the
different time intervals. The performance is calculated by accuracy for detection’s reduces. It is observed that the
comparing the results of occupancy to the ground truth after average performance is 99.5 % and is very high as
every 5 sec. The performance of the proposed system is compared with other parking lot detections applications.
measured by the using the equation (1) The performance of the proposed algorithm in some cases
drops down because of the strong shadows. The accuracy of

13
the proposed work also depends on the type of camera used the cars in these virtual parking blocks and notify the drivers
for monitoring the parking lot. The output accuracy is about the available parking blocks.
shown below in Fig. 4. The correct detection of the cars is
shown in chart in blue color and the false detection of cars is
shown in orange color. Accuracy of different parking space
detection methods
60 99
50 48 49 50
50 98
97
No of tests

40

Accuracy
30 96
20 95
10 4 94
0 2 1
93
0
Twin ROI Edge our
sunny day occluded cloudy day occluded
based based system
sunny day cloudy day
detection detection
Apearance of vehicle
Figure 5. Accuracy of different parking system.
correct FALSE 线性 (correct )

VI. CONCLUSION
Figure 4. Vehicle detection in different conditions.
The main contribution of this study is to optimize the
The results of the system proposed are compared with identification of available parking slots to possibly reduce
already implemented systems. The average detection rates the congestion in parking arena. Due to advancement in
of our system is compared with the two techniques used in machine learning and vision base technology cost effective
existing parking systems for detection of free parking slots automatic parking systems facilitate the drivers to locate
as shown in Table 2. available spaces at parking arena. Future researchers can
focus on allocation specific location to customers already
TABLE II. COMPARISON OF PROPOSED METHOD WITH TWIN ROI registered from online parking management system.
BASED DETECTION TECHNIQUE [10],EDGE BASED DETECTION
TECHNIQUE[12]. ACKNOWLEDGMENT
Twin ROI Edge based Proposed This work is fully supported by the grants from the
National Natural Science Foundation of China (61375045),
detection detection approach
Beijing Natural Science Foundation (4142030) and the Joint
method method Research Fund in Astronomy (U1531242) under
cooperative agreement between the National Natural
Science Foundation of China (NSFC) and Chinese
Average 95% 97% 98% Academy of Sciences (CAS). Dr. Ping Guo. is the author to
whom the correspondence will be done.
detection Rate
REFERENCES
(car)
[1] Ming-Yee Chiu; Depommier, R.; Spindler, T.; , "An embedded real-
time vision system for 24-hour indoor/outdoor car-counting
applications," Pattern Recognition, 2004.
The graph showing the accuracy of different parking [2] Zhang Bin; Jiang Dalin; Wang Fang; Wan Tingting; , "A design of
system is given below in Fig. 5. parking space detector based on video image," Electronic
The accuracy of our proposed system is better than the Measurement & Instruments, 2009.
Color based twin ROI detection technique [10], Edge based [3] T. Mar; N. Marcel;, “ Video-based parking space detection,” 2012
[Online]. Available: http://www.ini.rub.de/data/documents/
detection technique [12] used in existing parking systems.In tschentscherneuhausen_parking_space_fbi2012.pdf
color based detection system and Twin ROI based [4] Ichihashi, H.; Notsu, A.; Honda, K.; Katada, T.; Fujiyoshi, M.; ,
system ,the efficiency goes down when the car and parking "Vacant parking space detector for outdoor parking lot by using
area is of the same color.. Our proposed method will surveillance camera and FCM classifier," Fuzzy Systems, 2009.
FUZZ-IEEE 2009.
automatically divide the parking area in multiple blocks by
[5] Boda, V.K.; Nasipuri, A.; Howitt, I.; , "Design considerations for a
drawing virtual lines on the parking lot and efficiently detect wireless sensor network for locating parking spaces," SoutheastCon,
2007.

14
 [6] Fabian, T. , "An Algorithm for Parking Lot Occupation Detection," [12] Banerjee, Sayanti, Pallavi Choudekar, and M. K. Muju. "Real time
Computer Information Systems and Industrial Management car parking system using image processing." Electronics Computer
Applications, 2008. Technology (ICECT), 2011 3rd International Conference on. Vol. 2.
[7] Ichihashi, H.; Katada, T.; Fujiyoshi, M.; Notsu, A.; Honda, K.; , IEEE, 2011.
"Improvement in the performance of camera based vehicle detector [13] Shaaban, Khaled, and Houweida Tounsi. "Parking Space Detection
for parking lot," Fuzzy Systems (FUZZ), 2010. System Using Video Images." Transportation Research Record:
[8] Yusnita, R.; Fariza N. ; Norazwinawati B.; “Intelligent Parking Space Journal of the Transportation Research Board 2537 (2015): 137-147.
Detection System Based on Image Processing,” International Journal [14] Singh, Himal Pratap, Om Prakash Uniyal, and Kireet Joshi. "An
of Innovation, Management and Technology, Vol. 3, No. 3, June Approach to Implement Cost Efficient Space Detection Technology
2012. with Lower Complexity for Smart Parking System." Indonesian
[9] N. True;, “Vacant Parking Space Detection in Static Images,” Journal of Electrical Engineering and Computer Science 15.3 (2015):
Projects inVision & Learning, University of California, 2007 [Online]. 415-419.
Available: http://www.cs.ucsd.edu/classes/wi07/cse190-a/reports [15] S. Saleh Al-Amri, N. V. Kalyankar, and Khamitkar S, “Image
/ntrue.pdf. segmentation by using thershold techniques,” Journal of Computing.
[10] Najmi Hafizi Bin Zabawi, Sunardi, Kamarul Hawari Ghazali, vol 2, Issue 5. MAY 2010, ISSN 2151-9617.
“Parking lot detection using image processing method”, October 2013. [16] Li, Shiqiang, Hussain Dawood, and Ping Guo. "Comparison of linear
[11] Yusnita, R., Fariza Norbaya, and Norazwinawati Basharuddin. dimensionality reduction methods in image annotation." Advanced
"Intelligent Parking Space Detection System Based onImage Computational Intelligence (ICACI), 2015 Seventh International
Processing." International Journal of Innovation, Management and Conference on. IEEE, 2015.
Technology 3.3 (2012): 232.
[17] Mehmood, Rashid, Rongfang Bie, Hussain Dawood, and Haseeb Ahmad. [18]
"Fuzzy Clustering by Fast Search and Find of Density Peaks." In 2015
International Conference on Identification, Information, and Knowledge
in the Internet of Things (IIKI), pp. 258-261. IEEE, 2015.

15

View publication stats