A WINDOW-BASED IMAGE PROCESSING SYSTEM FOR ITELLIGENT
TRANSPORTATION SYSTEM
M Y SIYAL
School of EEE, Information Engineering Division
Nanyang Technological University
SINGAPORE 639798
Abstract: -Various researchers for a number of years have investigated traffic data collection and analysis. Many
techniques have been proposed to speed up the operations and some intelligent approaches have been developed to
compensate the effects of lighting, shadows and occlusions. To achieve real-time processing, it is necessary to
reduce the amount of data to be processed, so the first question is to determine the suitable location for the key
regions or windows. Previously, researchers have either used full frame processing approach, which requires more
computing power and thus is not practical for real-time applications, or have used window based techniques, but
defining of windows was done manually, which is not practical in real-world traffic applications. In this paper, an
automatic approach is described to measure the size and location of windows required for this purpose. The results
demonstrate that this method provides better results than the fixed window size.
Keywords: - Image Processing, Transportation, Pattern Recognition, and Segmentation.
1. Introduction
The real-time measurement and analysis of
various traffic parameters are increasingly required
for traffic control and management. Image processing
technique is now considered as an attractive and
flexible method for automatic analysis and data
collections in traffic engineering [1-4].
The first image processing algorithm required in
traffic application is filtering and segmentation.
Background differencing is the simplest technique to
segment the traffic images [5]. However, this
technique has problems of accurately updating the
background frame and automatically selecting a
suitable threshold value. Edge detection based
segmentation of traffic scene has the advantage of
being less sensitive to variation of ambient lighting
and shadows [6]. However, combined background
differencing and edge detection has the advantage of
eliminating stationary vehicles, shadows and the road
markings and is less sensitive to variations of lighting
[2]. We use this vehicle detection technique to detect
moving vehicles and measure traffic parameters such
as vehicle count and the queue parameters.
To achieve, real-time processing it is necessary
to reduce the amount of data to be processed, so the
first question is to determine the suitable location for
the key regions or windows. Previously, researchers
have either used full frame processing approach
[1,5,7], which requires more computing power and
thus is not practical for real-time applications, or have
used window based techniques [3], but defining of
windows was done manually, which is not practical in
real-world traffic applications. In this paper, an
automatic approach is described to measure the size
and location of windows required for this purpose.
We have conducted extensive tests and experiments
using automatic window approach and have compared
results with the fixed window size methods. The
results show that our approach provides better results.
2. The Vehicle Detection Process
The algorithm used for vehicle detection is based
on applying low pass filtering and combined
background differencing and edge detection
operations on windows located across the roads. This
vehicle detection technique has better performance
than the background differencing operations as
colours of vehicles and the ambient lighting changes
in traffic scenes are less sensitive to edge detection.
Following the application of vehicle detection
operation, the number of pixels having greater value
than the threshold is used to recognise a vehicle [6].
2.1 The size of Window
The size of the window has a strong effect on the
performance of the whole system. The window size
should be located vertical to the road direction for
vehicle counting. As the width of a window is
increased up to the length of vehicles, the accuracy
increases but the computation time increases, which
slows down the whole system. On the other hand, the
width of the window is dependent on the road type, as
in highways (expressways, motorways etc.) the
vehicles move more rapidly than other roads. On
highways the width of the windows should be more,
as there is a risk of missing vehicles on some frames.
We have conducted extensive experiments in
order to determine the best window size for various
situations. The percentage of errors of vehicle
detection algorithm versus window width for a road is
shown in figure 1. The graph of the computation
times against window width for the same road is
shown in figure 2. As it can bee seen, a compromise
can be made between the accuracy needed and a
processing speed required by selecting the width of
the window.
In our system the area to locate windows and the
number of lanes on the road and the road type is
determined by the operator manually. Then the size of
windows is computed and is located automatically
across the road. The width of the windows is adjusted
every half an hour time periods regularly as the
relative speed of the vehicle changes during daytime.
The width of a window is computed as follow:
W=L*V
(i)
Where L is the length of each window and is
defined on the location of window on image and V is
a speed factor determined by the operator and is
adjusted automatically by comparing relative speed of
vehicles continuously.
Following computing the windows and their
sizes, the vehicle detection program uses the coordinates of each window to detect vehicles.
3. Measurement Of Traffic Parameters
As described earlier, to detect vehicles, windows
with suitable sizes are placed across each lane of the
road. In this vehicle detection approach, the edge
detector is applied to each window, and the number of
edge points is compared with a threshold value to
decide, whether the window contains the object or
not. The threshold value is automatically calculated
by processing the histogram of each window.
Following the application of edge detection
operation to the windows a status vector is created.
To count the number of vehicles, the status
vector is analysed. In this manner, for each frame, if a
vehicle is detected in the window, a ‘1’ is stored at
the status vector of the window; otherwise, a ‘0’ is
stored at the status vector of the window. The group
of 1s (ones) corresponds to a vehicle and a group of
0s (zeros) corresponds to the distance between two
vehicles.
The vehicle detection operation along with a
motion detection operation is used for queue length
measurement. In this case the window is located
along the road direction. The algorithm used to detect
and measure queue parameters consists of two
operations, one involving motion detection and the
other vehicle detection. These operations are applied
to a window to detect the size of the queue. As the
microcomputer systems operate sequentially, a
motion detection operation is firstly applied and then
if the algorithm detects no motion, the vehicle
detection operation is used to decide whether there is
a queue or not. The reason for applying motion
detection operation first is that the traffic scenes we
analysed for queue detection are expected to contain
vehicles and in this case vehicle detection mostly
gives the positive result, while in reality there may
not be any queue at all.
The method for motion detection is based on
differencing two consecutive frames and applying
noise removal operators. The vehicle detection
algorithm is only applied when the motion detection
algorithm detects 'no motion'. A graph for errors and
processing speed against window width for queue
operation is shown in figure 3. In this approach we
use linear transformation to detect and measure queue
from the front to the rear side of the image.
4. Results
The algorithms were tested under various traffic
and lighting conditions using an Intel Pentium-based
computer system operating at a clock speed of 200
MHz together with a frame grabber. The algorithm
has been implemented in a user-friendly package in
MS Windows environment. We have conducted
extensive experiments for a continuos period of 6 - 8
hours.
Figure 4 shows the results of counting vehicles
with fixed window size, while figure 5 shows the
result when the program automatically adjusts the
window size. As it can be seen, with this approach,
the accuracy of vehicle detection has been improved.
The result of queue parameter is shown in figure 6.
As it can be seen, with this approach, the status of
traffic can be established. Depending on the length
and the speed of the queue, motorists can be informed
about the traffic status and even could be redirected to
alternative routes.
[3]
5. Conclusion
In this paper a novel image processing approach
was described to analyse and measure road traffic
parameters. An automatic window size computation
was introduced to enhance the performance of edgebased vehicle detection operation. Results
demonstrated that this approach is more accurate than
the fixed window size approach.
To further increase the accuracy in the case of
vehicle not moving on their lanes, half size windows
could be used. This approach could detect
motorcycles or other small objects or people. If the
window detects an object and its adjacent window
doesn’t detect it, the object is considered to be
motorcycle, people or any other small object. Also by
taking the queue parameters into account, the traffic
status could be determined and the traffic lights could
be adjusted automatically.
[4]
[5]
[6]
[7]
References
[1] Dickinson, K. W. & Waterfall R. C., 1984,
"Image processing applied to traffic: Practical
experiences”, Traffic Engineering and Control,
pp. 60-67.
[2] M Fathy and M Y Siyal, 1995, “An image
detection technique based on morphological edge
[8]
detection and background differencing for realtime traffic analysis”, Pattern Recognition
Letters, Vol. 16, pp. 1321-1330.
Hoose, N., 1991, " Computer image processing in
traffic Engineering", Taunton, Research Studies
Press.
Ali, A T and Dagles, E L, 1992, “Recent
progress in real-time Image analysis for realworld traffic analysis”, ICARVC’92, second
international conference on Automation, Robotics
and Computer Vision proceedings, vol. 1, pp. 10 15.
M Fathy and M Y Siyal, 1995, ”A window-based
edge detection technique for measuring road
traffic parameters in real-time”, real-time Imaging
Academic Press, Vol. 1, pp. 297-305.
Fathy M and Siyal M Y, 1995, “A real-time
Image Processing Approach To Measure Traffic
Queue Parameters”, IEE Proceedings on Image,
Vision and Signal processing, Vol. 142, No. 5,
pp. 297-303.
M Y Siyal and M Fathy, 1998, "A window-based
image processing technique for quantitative and
qualitative analysis of road traffic parameters”,
IEEE Transactions on Vehicular Technology,
Vol. 4, pp. 1342-1349.
Siyal M Y and Atiquzaman M, “A Parallel
Pipeline Based Multi-Processor System For Real
Time Measurement of Road Traffic Parameters”,
Real-Time Imaging, Academic Press, USA, Vol.
6, No.3, pp. 241-249, June 2000
Window Size
25
Error %
20
15
10
5
0
0
5
10
15
20
25
30
window Width (Pixels)
35
40
45
50
Figure 1. Graph of window width versus percentage of errors for detecting vehicles
Time (ms)
Window Size
160
140
120
100
80
60
40
20
0
0
5
10
15
window Width (Pixels)
20
25
30
Figure 2. Graph of window width versus processing speed for detecting vehicles
Error %
Queue
16
14
12
10
8
6
4
2
0
0
1
2
3
4
5
6
7
Window Width (pixe ls )
8
9
10
11
12
8
9
10
11
12
Queue Speed
500
Time
400
300
200
100
0
0
1
2
3
4
5
6
7
Window Width (pixe ls )
Figure 3. Graph of window width versus processing errors and speed for queue parameters
Computer
No of Vehicles
1200
Manual
1000
800
600
400
200
0
Sunny
Cloudy
Rainy
Dusk
Weather Conditions
Figure 4. Results of Counting Vehicles (fixed Window Size)
Computer
1200
Manual
No of Vehicles
1000
800
600
400
200
0
Sunny
Cloudy
Rainy
Dusk
Weather Conditions
Figure 5. Results of Counting Vehicles (automatic Window size)
Manual
Computer
V ery heav y Traf f ic
Heavy Traf f ic
Normal Traffic
Light Traf f ic
10
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
1610
Time
Figure 6. Results of Queue Parameters
1700
1800
1900
2000
2100
2200
2300