TS90-2201.pdf

download TS90-2201.pdf

of 5

Transcript of TS90-2201.pdf

  • 7/27/2019 TS90-2201.pdf

    1/5

    Measure travel time by using Bluetooth detectors on

    freeway

    Yubin Wang, Jos Vrancken and Paul Seidel

    Abstract The common traffic sensors, such as loops,

    cameras and radars, are very expensive and difficult to

    maintain. Recently, Bluetooth detectors are hot alterna-

    tive sensors with relative low cost to provide reliable traf-

    fic information. In this paper, we will present travel time

    measurements by using Bluetooth detectors on freeways

    and an algorithm to calculate the average travel time.

    The test has been done at the Dutch traffic management

    system company Trinite Automation B.V. together withtraffic control center in Belgium. The average travel time

    from the Bluetooth data has been roughly compared

    with the loop detectors data. For both short routes and

    long routes, results are rather good compared with the

    Bluetooth test on urban road.

    I. INTRODUCTION

    In many countries, the road traffic is the

    most important but also the most problematic

    form of transportation. It includes problems like

    frequent congestion, high pollution and extremely

    high fatality rates in comparison with other

    transportation forms. In most countries, it is

    common to apply traffic management to improve

    capacity of the road infrastructures and reduce

    the negative effects of road traffic. Traffic data

    collection from road side equipments is essential

    for traffic management, more accurate traffic

    information we know, better traffic control we

    can perform. The common traffic data detectors

    are loops, cameras with license plate recognition

    and radars, however they have either high costsor high maintenance costs. GSM and GPS signals

    are also good sources to collect traffic data,

    Yubin Wang is with Trinite Automation B.V., Postbus 189, 1420

    AD Uithoorn, The Netherlands and the Faculty of Technology,

    Policy and Management, Delft University of Technology, 2600 GA

    Delft, The Netherlands [email protected]

    Jos Vrancken is with the Faculty of Technology, Policy and

    Management, Delft University of Technology, 2600 GA Delft, The

    Netherlands [email protected]

    Paul Seidel is with Trinite Automation B.V., Postbus 189, 1420

    AD Uithoorn, The Netherlands [email protected]

    however the data is not sufficient to know traffic

    state. Thus, recently a new type of detector

    by using the bluetooth signal can contribute to

    solving the high costs and the high maintenance

    costs problems. Bluetooth detectors are attractive

    alternative equipments to collect traffic data with

    relatively low cost.

    Elsewhere Bluetooth is also becoming very

    popular in traffic research and in operational

    systems, both for in-vehicle driver support and for

    roadside traffic data collection [1], [2], [3]. The

    data collected are mostly travel times and travel

    time variance, but also dynamic OD matrices are

    estimated on the basis of Bluetooth data.

    There are many advantages in using bluetooth

    detectors to measure traffic data.

    High detect rate: Nowadays, almost everyone

    has a mobile phone with a Bluetooth de-

    vice. Each Bluetooth device has a worldwide

    unique address; this address is detectable.

    Approximately 20 to 30 percentage of road

    users can be detected by using the bluetooth

    detector, so that we can receive guaranteed,

    accurate travel information.

    Very cheap: A Bluetooth receiver only costs

    about 10 Euro. We can plug it in a mini-

    computer, then it can be used to collect traffic

    data. Complete unit cost is much cheaper than

    the loop detectors and cameras if deployed on

    a large scale.

    Relative low maintenance costs: The device is

    very simple to install and it is also very easy

    to replace.

    Sufficient travel information of traffic state:

    Travel time of individual road users are mea-

    sured, thus the received traffic information is

  • 7/27/2019 TS90-2201.pdf

    2/5

    not fragmented and it can represent the real

    traffic state.

    There is a privacy problem with Bluetooth, but it

    can be handled by storing data in an anonymous

    way.

    With so many advantages, Bluetooth detectors areused to measure travel time on freeway in stead

    of cameras or induction loops. In this paper, we

    will present travel time measurements by using

    Bluetooth detectors and an algorithm to calculate

    the average travel time based on the measurements.

    The test in this paper was done at the Dutch

    traffic management system company Trinite

    Automation B.V. together with traffic control

    center in Belgium.

    I I . TEST IN BELGIUM

    In order to verify the accuracy of Bluetooth

    detectors for travel time measurements on freeway,

    a test in Belgium was done from 26-08-2010 till

    07-10-2010 on the freeway E313 (A13) between

    Geel and Antwerp. Four Bluetooth detectors are

    placed near ramps (See Figure 1). Figure 2 shows

    how the Bluetooth detector looks like in this test.

    Fig. 2. Bluetooth detector

    Total distance between point A and point D

    is around 36 km with a free flow travel time of 23

    minutes. Total distance divided by the maximum

    velocity is the free flow travel time. Table I shows

    the distance and free flow travel time between

    four points.

    Detectors Distance (km) Free flow travel time (min)

    A-B 22 14

    B-C 5 3

    C-D 9 9

    TABLE I

    DISTANCE AND FREE FLOW TRAVEL TIME

    The work principle of the Bluetooth detectors has

    been explained in many papers [1], [2], [3]. In the

    paper, we just give a short summary. We receive

    a list of detections with Bluetooth IDs for each

    measurement point every minute. Then we try to

    match the Ids at two measurement points. Once a

    match is found, travel time for the road user can

    be calculated. Figure 3 and Figure 4 show the

    measured travel time for each road user by usingBluetooth detectors. For the long route, many

    road users with high travel time are measured.

    This is due to the fact that there are gas stations

    in the middle of the route. Road users might stop

    at the gas stations and continue their trips after

    some while. There are not too many data with

    high travel time for the short route, since there

    are no gas stations or entry and exit points in the

    middle of the short route.

    Another interesting fact is that 3 to 10 roadusers are measured every minute for the long

    route and 8 to 20 road users are measured every

    minute for the short route. The reason is that there

    is no way out for the short route, so the road

    users detected at the starting point of the route

    are also detected at the end of the route mostly.

    However, the road users might exit in the middle

    of the long route. Thus, the detect rate for the

    short route is higher than the long routes.

    III. ALGORITHM: AVERAGE TRAVEL TIMECALCULATION

    As we discussed in the previous section, there

    are many hits with high travel time which are

    outliers. Thus, in order to calculate the correct

    travel time, we need to remove the outliers. In

    this paper, a simple algorithm is used to remove

    outliers and a moving average algorithm is used

    to calculate the average travel time.

  • 7/27/2019 TS90-2201.pdf

    3/5

    Fig. 1. Four Bluetooth detectors

    00:00 06:00 12:00 18:00 00:000

    1000

    2000

    3000

    4000

    5000

    6000

    Time

    TravelTime(Second)

    Bluetooth tracing

    Measured travel time

    Average travel time

    Fig. 3. Travel time from point A to point B

    Input detections are stored in an input table

    until a configured max-time (for example 2 hours)

    is reached. The configured max-time can boundthe measured travel time and remove outliers.

    Output detections are looked up in the input table

    at the moment they are received. If a match is

    found, the difference in time is stored in a match

    table. If the time is negative, it is not stored in

    the match table. Every minute, the match table of

    recent 10 minutes is analyzed. First, outliers are

    further removed by a simple algorithm. All hits

    that are more than 50 percent above the average

    are removed form the table. As long as there

    are values above this threshold, the average is

    re-calculated and the high values will be removed.Then, the average travel time is calculated by

    a moving average algorithm. The average travel

    time for each minute is the mean value of the

    recent 10 minutes measured travel time.

    IV. RESULTS AND ANALYSIS

    The average travel time is shown on the above

    figures. From the figures, we can see that there

    are the two lines on the long routes. It might be

  • 7/27/2019 TS90-2201.pdf

    4/5

    00:00 06:00 12:00 18:00 00:000

    100

    200

    300

    400

    500

    600

    700

    800

    900

    1000

    Time

    TravelTime(Second)

    Bluetooth tracing

    Measured travel time

    Average travel time

    Fig. 4. Travel time from point B to point C

    Fig. 5. Loop detector data

    due to the fact that there are two lanes on theroutes, cars on the left lane are driving faster than

    cars on the right lane. For the long route, the

    travel time goes higher to 1.5 times of free flow

    travel time around 7:00 AM. For the short route,

    the travel time goes higher to 4 times of free flow

    travel time around 7:00 AM and around 11:30 AM.

    While travel time is measured by Bluetooth

    detectors, traffic information for the same period

    is also collected by loop detectors. Figure 5shows the data from point A to point D. X axis

    is the time, y axis is the space between point

    A and point D. The color represents velocity.

    However, it is difficult to calculate the realized

    travel time by using loop the data. Thus, travel

    time is calculated by using the length divided

    by the average velocity. From figure 5, we can

    see that the velocity drops dramatically around

    7:00 AM for the long route and both around

  • 7/27/2019 TS90-2201.pdf

    5/5

    7:00 AM and 11:30 AM for the short route. The

    traffic jam causes the long travel time around

    7:00 AM for the long route and around 7:00

    AM, 11:30 AM for the short route. The rest is

    free flow. Roughly, travel time which measured

    by Bluetooth detectors is close to the travel time

    which measured by loop detectors.

    There are some peaks for the average travel

    time of the long route in the beginning and

    the end of the day. The peaks are not correct

    average travel time by comparing the travel time

    which measured by loop detectors. The reason is

    that there are too little Bluetooth measurements,

    thus the outliers can not be removed properly.

    However, there is no peak for the average travel

    time of the short route in the beginning and

    the end of the day. It is due to the fact thatthere are relative more hits per minute for the

    short route, thus the outliers can be removed easily.

    Compared with the Bluetooth test on urban

    roads [4], travel time measured by Bluetooth

    detectors is much reliable for both short routes

    and long routes on a freeway. The reason is

    that the distance between on-ramp and off-ramp

    is relatively large. Thus, it is relatively easy

    to remove the outliers, since the chance of

    go-and-return cars within a short time is low.

    V. CONCLUSIONS

    In this paper, we present travel time measure-

    ments by using Bluetooth detectors on freeways

    and the algorithm to calculate the average travel

    time, developed at Trinite Automation B.V. The

    average travel time measured by Bluetooth detec-

    tors, has been roughly compared with traffic time

    which measured by the loop detectors. For both

    the short route (5 km) and the long route (>10km),results are quite good compared with the Bluetooth

    test on urban road. Thus, Bluetooth detectors are a

    cheap and reliable alternative sensors on freeways.

    VI. FUTURE WORKS

    Since it is difficult to compare realized travel

    time by using the loop data and Bluetooth

    data, we will use GPS-based Floating Car data

    to compare with Bluetooth data in the future.

    A better filtering method will be used to

    remove the outliers.

    In the case there are too little hits, a different

    approach will be used to calculate the average

    travel time.

    Reliability of the average travel time based on

    number of hits will be considered. Predicted travel time based on the realized

    travel time will be calcuated.

    VII. ACKNOWLEDGMENTS

    The research leading to these results has sup-

    ported by Trinite Automation B.V. and received

    funding from the European Communitys Seventh

    Framework Programme (FP7/2007-2013) under

    grant agreement no. INFSO-ICT-223844 (project

    Con4Coord, a.k.a. C4C).

    REFERENCES

    [1] J. Barcelo, L. Montero, L. Marques, and C. Carmona. Travel

    time forecasting and dynamic od estimation in freeways based

    on bluetooth traffic monitoring. In Proceedings of the 89th

    annual meeting of the Transport Research Board 2010, pages

    117, January 2010.

    [2] Ali Haghani, Masoud Hamedi, Kaveh Farokhi Sadabadi, Stan-

    ley Young, and Philip Tarnoff. Freeway travel time ground

    truth data collection using bluetooth sensors. In Proceedings

    of the 89th annual meeting of the Transport Research Board

    2010, pages 122, January 2010.

    [3] Yegor Malinovskiy, Yao-Jan Wu, Yinhai Wang, and Un-Kun

    Lee. Field experiments on bluetooth-based travel time data

    collection. In Proceedings of the 89th annual meeting of the

    Transport Research Board 2010, January 2010.

    [4] Henk J. van Zuylen, Li Jie, and Lu Shoufeng. The use of

    bleutooth scanners for travel time measurements. 2010.