Application of Wireless Sensor Networks for Road Monitoring Ondrej Karpis* Jozef Juricek** Juraj Micek*** *University of Zilina, Department of Technical Cybernetics Slovak Republic (e-mail:
[email protected]). ** University of Zilina, Department of Technical Cybernetics Slovak Republic (e-mail:
[email protected]) *** University of Zilina, Department of Technical Cybernetics Slovak Republic (e-mail:
[email protected])} Abstract: Rapid development of traffic in cities requires adaptive traffic flow control. Information on the number of vehicles in different lanes, their speed and their type is essential for adaptive management of intersections. Another important area is the vehicle parking. Detailed information about the number and location of parking spaces reduces the time vehicles spent in streets and contributes to increased throughput and reduction in emissions of Volatile Organic Compounds (VOC), Carbon Monoxide (CO), and Oxides of Nitrogen (NOX). This study aims to test the distance measuring sensor, which can be used for both problems mentioned above. Experiments with the sensor carried out in real world conditions revealed its both strengths and weaknesses. Keywords: vehicle detection, distance sensor, traffic monitor, wireless sensor network
1. INTRODUCTION Wireless sensor networks (WSN) represent new technology for effective solving of tasks that include distributed information sources. According to Huang (2009) contemporary applications of sensor networks can be divided into four basic groups: •
Environmental and habitat monitoring,
•
Medical diagnostics and health care,
•
Military surveillance and industry security,
•
Industry applications.
One of the most perspective areas from the first group of applications is traffic. Traffic systems are large-scale with distributed information sources by nature. Traffic system control requires different approaches than classical technology process control. Wireless sensor networks provide effective means for data collecting and transport. This makes from them an inseparable part of intelligent transport systems (ITS). Among the first successful applications of WSN in traffic belong systems for monitoring of transport flow parameters in order to determine transport intensity, to control crossroads with traffic lights, to forecast trends in traffic and so on (Wen, 2007; Corredor, 2008). Large part of intelligent transport systems is focused on urban transport infrastructure. Onward inhabitants cumulation in large cities together with growing living standard of common people, which is directly connected with growing number of vehicles, put highest demands on existing transport infrastructure. This trend is obvious not only in western
developed countries but also in fast developing countries of Asia. Significant change of transport infrastructure is usually not possible because of limited space. Therefore, maximizing of exploitation of existed infrastructure is of highest interest. Critical parts of road network are crossroads. Their improper management causes congestion and in addition to the direct economic impact it influences air quality and thus health. Adaptive management of crossroads allows for increased throughput of road network and significantly reduce air pollution caused by stationary cars in traffic jams. Of course, such management requires information on current traffic conditions in real time. At least it is necessary to know the number of cars entering the intersection from different directions. Ideally, we should know the type of vehicle approaching crossroad: be it car, truck or bus. This information is especially necessary with respect to much larger amount of emissions produced by trucks and buses. Another important area is the vehicle parking (Lee, 2008). It addresses problems: how to create sufficient number of parking spaces and how to make them available to the drivers. Creating the necessary number of parking spaces is possible by building parking houses. More complicated problem is how to ensure their optimal use. Without proper information system, it can be difficult to find free parking space. Looking for parking space can take too much time and has abovementioned undesirable effects on air quality. Parking houses should provide information on the number of free spaces and, ideally, on their specific location. In this case it is necessary to scan the occupancy of each parking space separately. For this type of application is particularly appropriate to use WSN, especially in terms of minimizing the cost of building the necessary infrastructure and also
because of its easy reconfiguration. Of course, WSN have its limitations, e.g. low transfer rate, that require using suitable methods of signal processing (Shuai, 2008). Systems monitoring traffic intensity can use several types of sensors: inductive, microwave, infrared, ultrasonic, magnetic etc. These systems are typically focused on vehicle detection only, possibly to distinguish it from non-motorized vehicles (bicycles). Some sensors also have the potential to distinguish the type of vehicle (passenger car, truck, bus). In this area the most frequently used are camera systems, with subsequent evaluation of the image. Such systems are relatively expensive, energy intensive and require high-speed communication channels. For these reasons they are not used in wireless networks. To distinguish between different types of vehicles it is possible to use distance measuring sensor. The sensor must be placed above the road so that vehicles pass beneath it. Identifying the type of vehicle can be based on the measured distance in time. Trucks and buses are higher than passenger cars and are also usually much longer.
which has an adverse impact on the quality of measured data. For example ultrasonic sensor SRF02 has response time of 70 ms within range up to 6 m. Sensors with larger range have even worse response time. For this reason we decided to use infrared distance sensor SHARP GP2Y0A700K0F. The sensor is suitable for measuring distances in the range of 1 to 5.5 m. It has a voltage output that is updated approximately every 20 ms, corresponding to a maximum sampling frequency of 50 Hz. On a passenger car with an average length of 5 m and maximum moving speed of 50 km/h (maximum permitted speed in Slovak cities) then falls at least 18 samples, i.e. one sample on every 28 cm of car length. This number should be sufficient to reliably distinguish the type of passing vehicle.
This paper focuses on the design and testing system based on infrared distance sensor. This sensor is able to provide data for both applications: adaptive crossroads control and parking space occupancy. The paper is divided into the following parts: the second part is devoted to a description of node hardware equipment. The third section describes the measurements made using the created system and their evaluation. The conclusion evaluates the potential of the proposed system and future work. 2. NODE HARDWARE EQUIPMENT Developing of WSN is a complex problem with many particular parts: network nodes design, developing of effective methods of data preprocessing and methods of reliable communication. Network node consists of sensor module, control and signal processing module and RF transceiver (Figure 1). Sensor and signal conditioning
MCU signal processing and device control
Wireless transciever
Power module
Figure 1. Node structure In Figure 2a and 2b are shown nodes developed at the Department of Technical Cybernetics of the University of Zilina. Figure 2a is a normal node (without batteries and the distance sensor), and figure 2b is a data collection node, which can be connected to the USB port of a computer. This node is used to collect information from other nodes for further processing. The principle of distance measuring sensor can be based on the reflection of either ultrasonic or infrared rays. Ultrasonic distance measuring sensors have a relatively low refresh rate,
a) Normal node Figure 2. Wireless node
b) Data collection node
The node is powered by batteries and DC/DC unit. Capturing and pre-processing of the measured signal is made by 32-bit microcontroller AT91SAM7S64. It has a build-in 8-channel, 10-bit AD converter. Microcontroller’s clock frequency is 48 MHz. Its core is built on the powerful ARM7 architecture, which allows engagement of the appropriate compression algorithms to increase wireless network throughput. Wireless communication is based on XBee-PRO module. This module complies with IEEE 802.15.4 specification and is designed for low-cost, low-power mesh networks. The module operates in the free 2.4 GHz ISM band. Transfer rate of the module is 250 kbps. The range is dependent on the type of environment, used antennas and the transmitter output power. Measurements of the range under real world conditions are presented in Micek, Karpis (2010). The modules used in nodes shown in figure 2 have integrated ceramic antennas. Data collection node is powered by USB. USB interface is compatible with USB 2.0 specifications. On the PC side, as a driver, freely available universal library libusb-win32 is used. Maximum data transfer rate between the computer and the node is 8 Mb/s, which far exceeds the actual speed of wireless network. Nodes have been designed so that they can be used in various applications. In addition to analog inputs they can also use
the serial interface RS232. In Micek, Karpis (2009) the nodes were used to transfer images from camera with serial output to PC. Other application involved transfer of audio signal in real time. 3. MEASUREMENTS
3,5
3
2,5
Voltage [V]
In order to verify the functionality of the developed system we realized two measurements under real world conditions. Both measurements were made at a road crossing the railway bridge. This place was chosen because we could place the sensor sufficiently low above the ground. The limited range of distance measuring sensor does not allow its arbitrary placement. The sensor was placed approximately 3.8 m above the ground. Tests were designed to verify whether the sensor could be used in real terms. Therefore we have not fully implemented a wireless network, but the measured data were stored on an SD card and evaluated off-line.
It is clear that in the normal state the sensor output is very noisy. Significant noise disallows for reliable passenger car detection. The only reliably detected vehicles are trucks and buses. Considerable noise is probably due to the fact that the sensor operates at its limits. Declared range of the sensor is only a 5.5 m, but it applies to the white reflecting surface with reflectance of 90 %. Another negative factor is the characteristics of the sensor that is relatively flat for greater distances (Figure 5). One of the possible ways of how to suppress unwanted noise is the averaging of samples. However, in this case the noise is too big and averaging would significantly affect the useful information.
2
1,5
1
0,5
0 0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
5
Distance [m]
Figure 5. Sensor characteristic During the measurement there were circumstances that further deteriorated the ability of the sensor to correctly measure the distance. At around 5:30 in the morning (after approximately 1 160 000 measured samples) it started raining. Rain caused the dispersion of the sensor’s infrared beam and the signal became unstable (Figure 6). After the rain the situation has not improved much because the road was still wet. Another negative factor was rough road surface.
Sensor
8
Figure 3. Sensor node placement
7
6
Distance [m]
The first measurement took 21 hours, from 23:00 to 20:00 of the following day. With the sampling frequency of 50 Hz we recorded 3 780 000 samples. Figure 4 shows the output from the sensor while bus passes below it.
5
4
3
8
2 7
1 6
Distance [m]
0 0
5
20000
40000
60000
80000
100000
120000
140000
160000
180000
200000
Sample 4
Figure 6. Sensor output during rain
3
2
1
0 0
100
200
300
400
500
Sample
Figure 4. Sensor output
600
700
800
900
1000
Measured values show that without additional modifications the sensor is not suitable for detection of passenger cars. To be able to verify the dynamic properties of the sensor, we decided to improve the parameters of the distance measuring sensor. There are only two possibilities: -
To improve the reflectance of the infrared signal,
To reduce the distance of the sensor from ground. This can be done only partially. It is possible to place the sensor too low because could cause damage of the sensor and/or put road users in risk.
the not we the
Among these options, we decided to try the first one: to improve the reflective properties of the road. This can be achieved in two ways: to colour the road with an appropriate paint or to install reflective material on the road. Using the paint is not desirable as it would permanently alter the appearance of the road. Therefore, we decided to place cardboard wrapped with aluminium foil on the road temporarily. Before the actual measurement, we tried more materials in the lab to found the one with best properties – aluminium foil. Cardboard had dimensions 60 x 80 cm and was placed under the sensor on the road. Since the used material was soft, road safety was not compromised. Figure 7 depicts sensor output after installing reflective material on the road.
8
7
6
Distance [m]
-
0 0
400
600
1
8 7 6 5 4 3 2 1 0
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
50000
Sample
1200
8 7 6 5 4 3 2 1 0
a
0
8 7
50
100
1400
1600
1800
2000
0
b
0 8 7 6 5 4 3 2 1 0
c
2 1 0
0
1000
Figure 9 shows four records of passing testing vehicle under the sensor. Figures a, and b, represent passing of vehicle in one direction and figures c, and d, passing in the opposite direction.
6 5 4 3
2
800
Figure 8. Shifting of reflective material
Distance [m]
Distance [m]
200
Sample
6
3
3
1
7
4
4
2
8
5
5
50
100
50
100
50
100
d
0
Sample
Figure 7. Improved signal reflection
Figure 9. Testing vehicle characteristics
It is obvious that the improvement of reflectance characteristics of the road have been successful and had a significant impact on reducing noise at the sensor output. Four major peaks displayed on the figure represent passage of a vehicle under the sensor.
All figures are similar to each other. This is caused by sensing the same vehicle, but also by the characteristics of the sensor itself. Note that at the beginning and end of the passage there are significant peaks in the opposite direction than we would anticipate. As if the distance of sensor from reflecting surface suddenly increased. This phenomenon is caused by the fact, that front and rear of almost all vehicles are oblique and infrared signal emitted by sensor will not return to the receiver. Sensor output is then close to zero, what corresponds according to sensor characteristics to a very large distance. Distances on all figures are limited to a maximum of 8 m. The central part of the signal (between the extreme peaks) is relatively flat and corresponds to the distance between sensor and top of the vehicle. This distance is of course less than the distance from the road. Bus passage record (Figure 4) is not bounded by extreme peaks. This is consistent with the mechanism of peak generation and shape of buses.
Notice the difference in distance measured with and without additional reflective material (Figures 7 and 4). If the signal is reflected by raw road surface, the mean of the measured distance is approximately 3.8 m, which corresponds to the actual distance of sensor from the road. After using the reflective material the average distance suddenly decreased to the value of only 3 m. Of course, we used the same formula to convert the measured voltage to distance. The fact that this discrepancy was not due to measurement error is documented by Figure 8, which captures the moment of the auxiliary reflecting material shifting. The shift was caused by bus passing over the material. This incident has showed us that the characteristic of the sensor is dependent on the reflecting material. This feature increases the uncertainty of the distance measuring. However, in some cases this feature could be useful. It could help to distinguish different color even among those vehicles of same type.
Pictures also show small evident peak near one of the main peaks. Location of the small peak corresponds to the transition of the vehicle in the same direction. Hence one can determine the direction of passing vehicle. It is assumed that
characteristics measured with sensor with improved static and dynamic parameters will make it possible to distinguish among individual vehicle types. Not only to recognize weather the passing vehicle is passenger or cargo vehicle, but also to distinguish e.g. Volvo from Volkswagen. Based on the evaluation of the test vehicle passage record, we were able to identify cars even in the first (noisy) measurement. Results of both measurements do allow basic classification of vehicles by type that is to distinguish passenger vehicles and trucks or buses. Each recorded sample has two main features: record length and mean height of the vehicle. Record length is referred to as the number of samples per vehicle. Vehicle height is calculated as the average level measured between two main peaks. It should be noted that the actual vehicle height and the measured distance is inversely dependent. A small measured distance corresponds to the high vehicle and vice versa. Figure 10 shows the characteristics of 105 vehicles. It is possible to recognize two, well separated groups of vehicles – passenger cars and trucks (buses). 3,5
The last thing we evaluated was the consumption of the sensor used. Operating supply voltage of the sensor is 5 V. The average supply current during measurements was 33 mA. These values are relatively high and restrict the use of the sensor in battery supplied applications. This shortcoming can be resolved when used in parking houses. 6. CONCLUSION This work was aimed to test the vehicle identification system based on infrared distance measuring sensor. Based on measured data and experience gained during the measurements, it is possible to evaluate the positive and negative features of the used sensor: +
Ability to distinguish passenger cars and trucks
+
High detection reliability (no false positives or false negatives during measurements)
−
Low range
−
Relatively low refresh rate
−
Dependence on weather conditions, especially the rain
−
Sensor characteristic is dependent on characteristics of reflecting material
−
Relatively high power consumption
−
Limited capabilities of sensor placement (must be placed above the road)
3
Distance [m]
2,5
2
1,5
1
0,5
0 0
10
20
30
40
50
60
70
80
Length [samples]
Figure 10. Vehicles characteristics It should be noted that we do not know the actual speed of vehicles passing under the sensor. Therefore we cannot accurately calculate the length of the vehicle. Slow passenger car may have a similar number of samples as a fast truck. Likewise, the exact height of the vehicle is not always possible to determine, given the properties of the used sensor. Improving the estimate of the length of the vehicle and its speed is possible by using two sensors, provided that we know the distance between them. Usually a pair of sensors used to evaluate the speed of vehicle is located relatively close to each other (at most a few meters) in order to clearly determine which two measurements correspond to the same vehicle. If we use distance measuring sensors of sufficiently high quality, we anticipate the ability to calculate the average speed of each vehicle even at a much greater distance between sensors. Each vehicle will have its own characteristics and the correlation analysis will make it possible to match individual records. Of course this will also require time synchronization of individual sensors, as described in Yuhe (2007).
Mentioned features of the sensor SHARP GP2Y0A700K0F significantly limit its use for traffic monitoring in real terms. However, it can be used in systems monitoring occupancy of parking spaces. In the future we want to focus on the development of distance measuring sensor with improved characteristics, especially with enhanced range, better maximum sampling frequency, weather resistance and independence on the reflecting material. With improved sensor we will attempt to identify individual vehicles based on their profile. Successful identification of the vehicle is a prerequisite to calculate the average speed between two sensors. REFERENCES Corredor, I., Garcia, A., Martinez, J., Lopez, P. (2008). Wireless Sensor Network-based system for measuring and monitoring road traffic. Proceedings of Collaborative Electronic Commerce Technology and Research. Lee, S., Yoon, D., Ghosh, A. (2008). Intelligent Parking Lot Application Using Wireless Sensor Networks. Proceedings of The 2008 International Symposium on Collaborative Technologies and Systems. 48-57 Micek, J., Karpis, O. (2009). Wireless automation and related problems. Proceedings of seminar New Trends in Cybernetics, Automation and Informatics. Gabcikovo
Micek, J., Karpis, O. (2010). Wireless Sensor Networks for Road Traffic Monitoring. Communications, Scientific Letters of the University of Zilina. in print Shuai, M., Xie, K., Ma, X., Song, G. (2008). An On-Road Wireless Sensor Network Approach for Urban Traffic State Monitoring. Proceedings of the 11th International IEEE Conference on Intelligent Transport Systems. 1195-1200 Wen, Y., Pan, J. L., Le, J. F. (2007). Survey on application of wireless sensor networks for traffic monitoring. Proceedings of the International conference on Transportation Engineering 2007. 2079-2084 Yuhe, Z., Xi, H., Li, C., Ze, Z. (2007). Design and evaluation of a wireless sensor network for monitoring traffic. Proceedings of the 14th world congress on intelligent transport systems, Beijing.