1 REAL-TIME IMAGE PROCESSING APPROACH TO MEASURE TRAFFIC QUEUE PARAMETERS. M. Fathy and M.Y. Siyal Conference 1995: Image Processing And Its Applications.

1

REAL-TIME IMAGE PROCESSING APPROACH TO MEASURE

TRAFFIC QUEUE PARAMETERS.

M. Fathy and M.Y. Siyal

Conference 1995: Image Processing And Its Applications

2

OUTLINE

Introduction

The queue detection algorithm

Motion detection algorithm

Vehicle detection algorithm

Results and discussion

Conclusion

Bibliography

3

INTRODUCTION

Measure of traffic queue is required in many situations

-Traffic jam

-Traffic accidents

-Adjusting time in traffic lights

Problems to measure the traffic in real-time

-Variations of light conditions

-Different shape or size of Vehicles

-Geometry of the scene

Objectives of the paper

-Measure in real time accurately queue

parameters like length or period of occurrence

4

INTRODUCTION

Previous works

-Rourke and Bell (1991): Method based in Fast Fournier Transformation

(FFT). This method do not measure the length. Very time-consuming.

-Hoose (1991): Do not measure length.

Introduction to the algorithm

Motion detection

Vehicle detection

Yes

No

This approach reduces the computational time

5

MOTION DETECTION ALGORITHM

The image is divided in sub-profiles.

Sub-profiles with different size to compensate:

−Effect of the transfer of the three-dimensional view of the camera to a two-dimensional image.

−Parameters to the camera like height of the camera, field of view and angle of the optical axes

By knowing the coordinates of 6 reference points of the real-world and the coordinates

of their corresponding images to make a geometric correction and measure length.

The size of the sub-profile depends on the resolution and the accuracy required, but

the size should be about the length of the vehicle.

A median filter is applied to the sub-profiles to remove the noise.

4th ave. New york (4-2-06)

6

MOTION DETECTION ALGORITHM

For each sub-profile are calculated the histogram for two consecutives frames

First frame Second frame Difference

Motion detected

Difference histogram with

high values

7

MOTION DETECTION ALGORITHM

First frame Second frame Difference

No Motion detected

Difference histogram with

Low values

8

VEHICLE DETECTION ALGORITHM

Most of the vehicle detection algorithms developed so far are based on a background differencing technique. However, this method is sensitive to the variations of ambient lighting and it is not suitable for real world applications.

The method used here is based on applying edge detector operators because edges are less sensitive to light variations

The edge detector, consisting of separable median filtering and morphological operators, SMED (separable morphological edge detector).

The Edge detector is applied to each sub-profile

Motion detection

Vehicle detection

Yes

No

9

VEHICLE DETECTION ALGORITHM

The histogram of each sub-profile is processed to select dynamic left-limit value and a

threshold value to detect Vehicles.

When the window contains an object, the left-limit of the histogram shifts towards the

maximum grey value. This process is repeated in 100 frames and the minimum of the left-

limit of these frames are selected as the left-limit for the next frames

The left-limit selection program selects a grey value from the histogram of the window,

where are approximately zero edge points above this grey value.

Histogram containing no object Histogram containing a small part of an object

Histogram containing a large part of an object

10

VEHICLE DETECTION ALGORITHM

For threshold selection, the number of edge points greater than the left-limit grey value

of each window is extracted for a large number of frames (200 frames) to get enough

parameters below and above a proper threshold value.

These numbers are used to create a histogram (horizontal: number of edge points greater

than left-limit: vertical: frequency of repetition of these numbers)

Peaks related to the frames passing a vehicle for that frame

Number of edge points greater than left-limit

Freq

uenc

y of

rep

etiti

on

Before median filter

Freq

uenc

y of

rep

etiti

on

Number of edge points greater than left-limit

After median filter

11

RESULTS AND DISCUSSION

Operations of the algorithms compared with manual observations of images confirm

that the queues are detected and its parameters are measured accurately in real-time.

The average processing speed is about 2 frames per second, enough for real-time.

The program works in such way that after 10s, the presence of the queue and its length

is reported

The algorithm is applied to each profile:

-If no vehicles are detected repeat the process for this sub-profile again

-If vehicles are detected, detection will be applied and the next sub-profile. I no vehicles are detected back to the previous sub-profile.

12

RESULTS AND DISCUSSION

Testing the method under different weather conditions

The results show that this queue measurement approach can determine the length of the queue to within 95% accuracy (5% error).

13

CONCLUSIONS

The algorithm uses a new technique by applying a combination of simple but effective operations and has been implemented in real-time.

In order to reduce the computation time, a motion detection operation is applied on all sub-profiles, while the vehicle detection operation is only applied when it is necessary.

The vehicle detection operation uses an edge-based technique which is less sensitive to noise.

The threshold selection for vehicle detection is done dynamically to compensate the effects of variations of lighting and it does not introduce any significant computational cost.

14

CONCLUSIONS

The results show that this queue measurement approach can determine

the length of the queue to within 95% accuracy.

This error is mainly due to the objects located very far from the camera

and can be reduced by adjusting the size of sub-profiles more

appropriately, by analysing camera parameters more accurately.

A practical implementation of this approach called ‘Variable Sign

System’, has been operational since early 1995. This system alarms the

drivers for heavy traffic, one kilometre before the intersection.

15

BIBLIOGRAPHY

HOOSE, N. (1991): ‘Computer Image Processing in Traffic Engineering’. Research Studies Press, Taunton.

INIGO, R.M. (1987): ‘Traffic monitoring and control using machine vision: a survey’, IEEE Trans. Indust. Elec., IE-32, (3), pp. 177-185.

SIYAL, M.Y., FATHY, M., and DARKIN, C.G. (1994): ‘Image processing algorithms for detecting moving objects’, Proc. of Third International Conference on Automation, Robotics and Computer Vision (ICARCV’94), Singapore.

IKRAM, W. (1990): ‘Traffic studies using imaging techniques’. PhD. thesis, UMIST.FATHY, M. (1991): ‘A RISC type programmable morphological image processor’. PhD.

thesis, UMIST.HOOSE, N. (1992): ‘Impact: an image analysis tool for motorway surveillance’, Trafic

Eng. & Control, pp. 140-147.ROURKE, A., and BELL, M.G.H. (1991): ‘Queue detection and congestion monitoring

using image processing’, Traffic Eng. & Control, pp. 412- 421.FATHY, M., SIYAL, M.Y., and DARKIN, C.G. (1994): ‘A low cost approach to real-

time morphological edge detection’, Proc. of IEEE TENCON Conference, Singapore.SCHALKOFF, R.J. (1989): ‘Digital Image Processing and Computer Vision’. John Wiley.

16

REAL-TIME IMAGE PROCESSING APPROACH TO MEASURE

TRAFFIC QUEUE PARAMETERS.

M. Fathy and M.Y. Siyal

Conference 1995: Image Processing And Its Applications

E N D