2008 IEEE Intelligent Vehicles Symposium

Eindhoven University of Technology

Eindhoven, The Netherlands, June 4-6, 2008

Parking Space Detection Using Ultrasonic Sensor

in Parking Assistance System
Wan-Joo Park, Byung-Sung Kim, Dong-Eun Seo, Dong-Suk Kim and Kwae-Hi Lee

Abstract— This paper deals with parking space detection by of Flight), which is travel time from a sensor to an object with
using ultrasonic sensor. Using the multiple echo function, the acoustic wave. A sensor circuit makes amends the amplitude
accuracy of edge detection was increased. After inspecting effect of the detected echo that has spreading loss and air loss
on the multiple echo function in indoor experiment, we applied occurred in proportion to TOF, to during travelling [4].
to 11 types of vehicles in real parking environment and made
experiments on edge detection with various values of resolution. Equation (1) shows, for compensated echo signal, the
We can scan parking space more accurately in real parking distance, R is obtained by TOF measurement.
environment. We propose the diagonal sensor to get information cto
about the side of parking space. Our proposed method has R= (1)
benefit calculation and implementation is very simple. 2
where c = 340 m/s at 20 °C

I. INTRODUCTION When making an environment map by ultrasonic sensor,

Parking is considered for a driver to be the hardest when the most critical problem is angle information as uncertain as
driving a car because parking is a high-stress maneuver. This beam width ( ± θ0 ). In Fig. 4, to illustrate, the error of the
paper deals with parking space detection by using ultrasonic single echo function shows that error of edge detection,
sensor. As safety and convenience for driving are paid which estimates larger than an original object’s width. The
attention, many researches are progressing vigorously for beam width is determined by the center frequency ( f R ) and
parking assistance system. Research of making Parking diameter of the transmitter ( a ) in equation (2) [5].
Environmental Map using ultrasonic is developed with 0.61λ
various methods. θ 0 = sin −1 ( ) (2)
The most fundamental method is to determine the direction a
of measurement distance with right angle under the where λ = c / f R
assumption that the surface is rough [1], [2]. But this method
has the error increment in proportion to the distance because Consequently parking space is measured more narrowly
of uncertain direction within beam width range. To identify than real parking space because of uncertain angle
corner, plane and edge of the surface of an object, Bozman information of ultrasonic sensor. To solve edge detection
modeled in ultrasonic and reflection surface [3]. Additionally, error, we can reduce the beam width by increasing the center
another method is detecting edge using ultrasonic sensors that frequency, f R or increasing the diameter of the transmitter,
have a wide beam width [6]. Moreover Satonaka got the a . However, this method has a serious problem that multi
parking space with modeling in the rounded edge side of path occurs because of reduced incidence angle when
vehicles [7]. It needs to make lots of calculations. For ultrasonic beam width is decreased [5].
example, if there is right angle structure in an underground
parking lot, Satonaka’s method can not be applied. There are various sensors for parking assistance system
The ultrasonic sensor measures the distance by TOF (Time including LRF (Laser Range Finder), Vision sensor and SRR
(Short Range Radar). LRF is able to get accurate information
Manuscript received January 10, 2008. of angular distance within 180°. However it can not detect a
Parking Space Detection Using Ultrasonic Sensor in Parking Assistance projected structure except scanning surface because LRF
System detect not the volume but the point. To make up for this
Wan-Joo Park is with the Electrical Engineering Department, Sogang
University, Seoul, Korea, phone: 82-10-3264-3308; fax: 82-2-704-4088;
problem, LRF is installed with tilt toward bottom, which
e-mail: wanjoo@sogang.ac.kr). brings more accurate 3D information [8]. Despite such
Byung-Sung Kim is with the Electrical Engineering Department, Sogang advantages, Laser can not detect black color surface of
University, Seoul, Korea, (e-mail: happykbs@sogang.ac.kr).
vehicles for natural property of light. It is really founded in
Dong-Eun Seo is with the Electrical Engineering Department, Sogang
University, Seoul, Korea, (e-mail: kiciflash@gmail.com). experiment status. Therefore LRF can not be applied to
Dong-Suk Kim is with the Mando Cenral R&D Center, Gyeonggi, Korea, parking assistance system because of such a fatal problem. In
(e-mail: greenhupa@mando.com). case using vision sensor, lane detection at parking lot is being
Kwae-Hi Lee is with the Electrical Engineering Department, Sogang
University, Seoul, Korea, (e-mail: khlee@sogang.ac.kr). developed actively. Vision sensor has good points that it has

978-1-4244-2569-3/08/$20.00 ©2008 IEEE. 1039

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.
 Fig. 2. Classification of surface shape for right angle edge side and rounded
edge sides.

each Tres (Multiple echo Resolution Time). Then the distance

is calculated by TOF of the first echo. The object is identified
by the multiple echo function.

cTres
Rres = (3)
2
Fig. 1. The single echo function and the multiple echo function
As an equation (3), we can calculate the radial resolution,
much information more than range sensor has. Thus, it can Rres of multiple echoes, the minimum distant difference
decide various situations and cope with status. However, if a among objects or the surface of an object, by checking
lane is damaged or is hid by another vehicle then vision interval time of multiple echo, Tres .
sensor can not detect the parking lot in spite of enough
parking space. In case using SRR (Short Range Radar), SRR
is able to get information about angles, distances and relative III. CLASSIFICATION OF SURFACE SHAPE
velocity and etc [9]. However, SRR has some problems that AND ERROR INCREMENT AT ROUNDED SHAPE
have to solve the scattering phenomenon on edge and its price
is far higher than other sensors. Fig. 2 shows that classification of surface shape by the
multiple echo function using equation (4).
The conventional research is building an environment map
by transmitting ultrasound and receiving just one echo. We R2nd -R1 st > Threshold : Edge
propose, however, once transmitting ultrasound and (4)
receiving multi-echoes more than 2 times. In accordance with R2nd -R1 st ≤ Threshold : Plane
this method, we can build more accurate environment map. where : R1st is the calculated distance by TOF of 1st echo
Besides we propose the diagonal sensor. It offers information
R2nd is the calculated distance by TOF of 2nd echo
about the side of parking space.

In Fig. 2 (a), since R2 nd -R1st is about Rres and smaller than

II. MULTIPLE ECHO threshold, the surface shape is estimated to plane. In Fig. 2 (b),
on the other hand, R2 nd -R1st is about the depth of object and
As you can see in Fig. 1, the single echo function is that larger than threshold. Therefore the surface shape is estimate
once transmitting ultrasound pulse set is sent and just one to edge when we assume the depth of the object is even
echo s received. Then the distance is calculated by TOF. The deeper than threshold. In other words, the 1st echo is reflected
ultrasonic sensor detects the nearest object placed in effective from the forward part of an object and the 2nd echo is reflected
beam width by the single echo function. And it estimates the from the back side. When ultrasonic sensor passes by the
direction of the object or assumes the direction is orthogonal object, the back side is detected as you can see in Fig. 2 (c). In
angle. Contrarily, the multiple echo function is after this case, R2nd -R1st is about Rres and smaller than threshold.
transmitting ultrasound pulse set once and the first echo is Then the surface appearance is estimated to plane. In reality
gotten. Also it has to check the more multiple echo signal at parking status, however, the edge of vehicles is too

1040

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.
 Fig. 4. Mobile robot, PIONNER 2-AT2 and it is possible to rotate sensor.

In flat board case, when Rres is 10cm, the smallest value,

has the best performance because flat board (an ideal case)
Fig. 3. In indoor experiment, comparison with the single echo function and
the multiple echo function. has not concave and bulge areas. Ultrasonic sensor can
receive regular reflection wave.
rounded-shaped to get accurate information. Fig. 2 (d)-(g)
shows errors of the edge detection. Whereas in right angle As another method to reduce the error of edge detection,
edge (fig. 2(b)) R2nd -R1st is larger than threshold, Fig. 2 (e) we need to adjust the amplitude gain of receiving part. The
reflection power of round-shaped surface is smaller than that
shows that R2 nd -R1st is smaller than threshold. Hence despite
of the plane-shaped surface. Reducing the amplitude gain of
edge point, the surface shape is estimated to plane. Also, in receiving part has an effect on making effective beam width
Fig. 2 (f), the surface shape is miscalculated because narrow. As the result of adjusting the amplitude gain of
R2nd -R1st is smaller than threshold. In status Fig 2 (g), since receiving part, we can reduce the error in edge detection.
R2nd -R1st is larger than threshold, the surface shape is
The sensor in the orthogonal direction of moving direction
estimated to edge. Therefore Fig 2 (g) shows enormous error.
gets information related to only parallel side. As we can see in
Consequently to reduce such above shown errors, we need
Fig. 3, by above referred method, we couldn’t get information
to adjust the radial resolution of the multiple echo function,
related to the side of parking space. However we
the amplitude gain of receiving part and rotation of ultrasonic
experimented to get the data of the side by rotating diagonal
sensor appropriately.
sensor with various angles. By data fusion both from
orthogonal and from diagonal sensors, we can get more
IV. ADJUST OF RESOLUTION, GAIN AND ANGLE accurate parking environment map. Fig. 4 shows it is possible
to rotate the sensor on auto-mobile robot.
The radial resolution of multiple echoes ( Rres ) means the
resolution of detection distance by calculating minimum V. EXPERIMENTAL RESULTS
interval time of between the 1st echo and the 2nd echo ( Tres ).
For instance, when Rres is 30cm, ultrasonic sensor can detect A. Experimental environments
the object or the surface shape of the object placed in more In this experiment, we used the mobile robot, PIONEER
than 30cm after receiving the 1st echo. In conformity with 2-AT2 (P2-AT2) of Active Robot Company. P2-AT2 is
adjusting Rres , accuracy of edge detection is variable. driven by the selection of direction, speed and distance in
VC++ software created by our team. The AVR processer
Our goal is to detect the parking space of vehicles. When
measures the movement distance of the robot and receives the
we assume the vehicle is 150cm wide, we want to get value of
TOF of ultrasound wave with real time processing. The AVR
R2nd -R1st with near 150cm. Because vehicles are designed
processer sends ultrasonic data including robot’s location to
variously and have many concave and bulge areas (especially VC++ software of notebook via UART port. This system can
wheel parts), the surface of vehicle generates irregularly conveniently scan parking space in any parking environment.
diffused reflection wave. Too small Rres brings bad effect The speed of the mobile robot is fixed on 5km/h and the
since sensor receives the continuous echoes. On the other distance from the mobile robot to vehicles is fixed on 1m.
hand, sensor can not receive the echo within Tres after The Ultrasonic sensor is Polaroid 600 series smart sensor
receiving the 1st echo. It is say that we can not get the of SensComp company. It’s f R is 50 kHz and a is 3.87cm,
information. Too large Rres makes errors of edge detection. therefore, beam width, θ0 is 12.4779°. As the result of

1041

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.
 TABLE 1
ERROR OF THE EDGE DETECTION ABOUT THE SINGLE ECHO FUNCTION AND
THE MULTIPLE ECHO FUNCTION AT 10, 50, 70CM RESOLUTION.

Resolution Single echo Multiple echo Unit

Mean 49.82 26.05 Cm
Variance 254.88 168.82 Cm
10cm
RMS 52.26 29.05 Cm
MAX 83 42 Cm
Mean 45.58 18.80 Cm
Variance 243.10 155.21 Cm
50cm
RMS 50.97 22.48 Cm
MAX 83 40 Cm
Fig. 5. In real parking environment, comparison with the single echo function
and the multiple echo function for various vehicle types. 70cm_front_single
Mean 49.01 15.42 Cm
means that error of edge detection of front side at 70cm resolution using the Variance 266.98 172.52 Cm
70cm
single echo function. ( 1 : AVANTE XD, 2 : BMW330i, 3 : CARSTAR, 4 : RMS 51.6 20.16 Cm
SANTAFE, 5 : SEPHIA, 6 : SPOTAGE, 7 : OPIRUS, 8 : KORANDO, 9 : MAX 91 32 Cm
SM3, 10 : GRANSER XG, 11 : EF SONATA )
TABLE 2
experiment, θ0 is measured to 11.8 ~ 12.4°. The detective ERROR OF THE EDGE DETECTION
TO ADJUST THE AMPLITUDE GAIN OF RECEIVING PART.
range of sensor is 0.25 ~ 8m and installment height of one is
the bumper height of vehicles. The sampling period is 69ms Types of Front Rear
(14.5Hz) and Rres is 10cm in indoor experiment. Unit
Vehicle Full Gain Low Gain Full Gain Low Gain
The reference map is created by Laser Range Finder (LRF) GRANDEUR
+25.7 +19.9 +35.7 +16.6 Cm
of SICK to get the accurate parking environment map. LRF XG
can get distance data within 0 ~ 180° direction with 1° +21.0 +20.3 -13.0 0.5 Cm
interval, and the reference map is created. AVANTE +16.4 +16.9 -8.0 -10.0 Cm
+21.3 -10.0 0.3 0.2 Cm
+26.8 +23.5 +16.3 0.3 Cm
B. Experimental results
CARENS +24.7 +23.5 +18.1 +11.6 Cm
Fig. 3 shows the difference of error in between the single +30.9 +23.5 +13.7 +11.6 Cm
echo function and the multiple one when using ideal flat +24.3 -6.8 -3.9 -2.7 Cm
board in indoor experiment. The reason of the existence of LAVITA -6.3 -8.5 +3.4 0.3 Cm
such errors is uncertainty direction angle of ultrasonic. +22.8 -5.0 0.4 0.5 Cm
At the 1st edge, whereas the error of the single echo
function was 10.76cm, the error of the multiple one was 91cm, the maximum error of the single echo function, became
3.71cm. Hence the multiple echo function had an effect to 32cm with the multiple echo function. Its value was about one
reduce the error of 7.05cm. At the 2nd edge, 14.9cm, the error third times. Also variance 266.98 of the single echo function
of the single echo function, was reduced to 3.9cm by the became 172.52 with the multiple echo function. It means the
multiple echo function. The gap was 11cm and it was more stable edge detection. Fig 5 shows the error of 70cm
immense gains. In indoor experiment, the error of the single resolution for the front edge and the rear edge of vehicles. 11
echo function was less than 15cm but that of multiple one was types of vehicles include sedan and SUV of various
less than 5cm. companies (HUNDAI, KIA, SSANG YOUNG, RENAULT
SAMSUNG, and BMW) We can see the reduction of edge
Fig. 5 and Table 1 are the experimental results for other 11 detection error in proportion to larger Rres . If we assume a
type’s vehicles in real parking environment. We made
experiments on edge detection with various value of Rres vehicle is 150cm wide, when Rres is 70cm, about half of
from 10cm to 100cm. Table 1 shows the error of edge vehicle width, we would get good results from experiments.
detection with the single echo function and the multiple echo As another method to reduce the error of edge detection,
function for Rres of 10, 50 and 70cm. According to the result, we need to adjust the amplitude gain of receiving part.
when Rres was 70cm, the best performance was at the mean Table 2 shows the reduction of edge detection error in
value with 15.42cm and the maximum value with 32cm. The proportion to less amplitude gain. If the edge is right angle
multiple echo function performed better than the single echo shaped, we would achieve the desired result. In case of
function. And the maximum error of the multiple echo GRANDEUR XG, the rear edge side has larger curvature
function was less than half of the error of the single echo than the front edge side. At full gain, 35.7cm error became
function for each resolution. Especially at 70cm resolution, 16.6cm error by adjusting amplitude gain of receiving part. In

1042

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.
 [CERATO]Tilt a Sonar at 60degree of the single echo function became 32cm with the multiple
4000
echo function. Its value is about one third times. Also
3500 variance 266.98 of the single echo function becomes 172.52
3000 with the multiple echo function. It means the more stable edge
detection.
2500
Distance [mm]

2000
To reduce the error in edge detection, we inspected closely
1500 edge which had errors more than 20cm. These edges have
large curvature. Therefore we reduced the amplitude gain of
1000
receiving part because the effective beam width of ultrasonic
500 0 degree sensor reduces proportionately to the amplitude gain of
60 degree
0 loci receiving part. In case of GRANDEUR XG, 35.7cm error of
-500
Object the rear side edge became 16.6cm error.
0 2000 4000 6000 8000 10000 12000 14000
Distance [mm]
We propose the diagonal sensor to get information about
Fig. 6. The parking environment map using the orthogonal sensor and the
diagonal sensor in real parking status. the side of parking space. We experimented to get the data of
the side by rotating diagonal sensor with various angles
case AVANTE, CARENS and LAVITA, the front edge side within 0 ~ 90°. As above shown result, it is appropriate to use
has more large curvature than the rear edge side. By adjusting 60° diagonal sensor for recognition of the side of parking
the amplitude gain of receiving part, we obtained good results. space.
The reason is that the effective beam width of ultrasonic
sensor reduces proportionately to the amplitude gain of We compared with the performance by other methods.
receiving part. However we should be cautious that too low Satonaka’s method using modeling of rounded shape of
amplitude gain makes the sensitivity of ultrasonic debased. vehicle edge side had edge detection error less ±30cm for 5
types of vehicles [7]. Our proposed method is verified by
Fig. 6 shows the result from the sensor in orthogonal 11types of vehicles such as sedan and SUV of various
direction of the robot’s moving direction and 60° diagonal companies. We obtained the final result of this work that edge
sensor. Using the sensor in orthogonal direction, we got detection error was from -10cm to 30cm. Furthermore,
information related to just parallel side of the robot’s moving proposed method has benefit calculation and implementation
direction. To get data of the side of parking space, we need is very simple. Whereas Satonaka’s method has enormous
diagonal sensor. 60° data in Fig. 6 shows the right side of error for right-angled post at parking lot or right-angled edge
parking space. The data of left side of parking space could be such as rear side of a truck, our proposed method can detect
obtained by -60° diagonal sensor. In Fig. 6, there is not left more accurate edge detection in this case.
side because of absence of -60° diagonal sensor. This result
shows the accurate information about parking space. The
wrong data from the distance over about 3m are able to be ACKNOWLEDGMENT
removed by orthogonal direction data. Wan-Joo Park thanks Ji-Yeon Song for help me write this
paper.

VI. CONCLUSION
REFERENCES
Using the multiple echo function, we could detect parking
space more accurately in real parking environment. First we [1] J. L. Crowley, "Dynamic world modeling for an intelligent
inspected effect on the profit of the multiple echo function in mobile robot using rotating ultrasonic ranging device", in Proc.
indoor experiment. 10 ~ 15cm, error of the edge detection IEEE Int. Conf. Robotics Automat., 1985, pp. 128-135.
with the single echo function becomes less than 5cm with the [2] A. Elfes, "Sonar based real-world mapping and navigation",
IEEE J. Robotics Automat., vol. 3, pp. 249-265, 1987.
multiple echo function in indoor experiment. Therefore we [3] Ömür Bozma and Roman Kuc, "Building a Sonar Map in a
confirmed utility of the multiple echo function. Specular Environment Using a Single Mobile Sensor", IEEE
Trans. on Pattern Analysis And Machine Intelligence, vol. 13,
Next step, we applied to 11 types of vehicles in real No. 12. 1991.
parking environment. Using the multiple echo function we [4] Polaroid Corp., "Ulrasonic Raging System Handbook,
got more accurate edge detection than the single echo Cambridge", MA, 1984.
[5] S. K. Nayar, K. Ikeuchi and T. Kanade, "Surface Reflection:
function. We made experiments on edge detection with Physical and Geometrical Perspectives", Int. Proc. of Image
various value of Rres from 10cm to 100cm. We could get Understanding Workshop 1990, pp. 185-212, 1990.
good performance at 70cm Rres . The maximum error 91cm

1043

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.
 [6] Ariel Di Miro, Bernhard Wirnizer etc.al,"Ultrasonic Image
Formation Using Wide Angle Sensor Arrays and Cross-Echo
Evaluation", in Proc. of the 24th DAGM Symp. on Pattern
Recognition, pp.421-428, 2002.
[7] Hisachi Satonaka, Masato Okuda, "Development of Parking
Space Detection Using Ultrasonic Sensor", 2006 ITS World
Congress, 2006.
[8] Ho Gi Jung, Dong Suk Kim, Pal Joo Yoon, and Jai Hie Kim,
"3D Vision System for the Recognition of Free Parking Site
Location", The 13th International Pacific Conference on
Automotive Engineering, 2005.
[9] Stenfan Görner, Herrmann Rohling, "Parking lot Detection
with 24GHz Radar Sensor", 3rd international Workshop on
Intelligent Transportation WIT 2006.

1044

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 12, 2009 at 02:10 from IEEE Xplore. Restrictions apply.