Collaborative real-time traffic information generation and sharing
framework for the intelligent transportation system
Wei-Hsun Lee
a,b, Shian-Shyong Tseng
a,c,*, Wern-Yarng Shieh
da
Department of Computer Science, National Chiao Tung University, 1001 Ta Hsueh Road, Hsinchu 300, Taiwan, ROC b
Telecommunication Laboratories of Chung-hwa Telecom, 12, Lane 551, Min-Tsu Road, Sec. 5 Yang-Mei, Taoyuan 326, Taiwan, ROC c
Department of Information Science and Applications, Asia University, 500 Liufeng Road, Wufeng, Taichung 413, Taiwan, ROC d
Department of Computer and Communication Engineering, St. John’s University, 499, Sec. 4, Tam King Road, Tamsui, Taipei 251, Taiwan, ROC
a r t i c l e
i n f o
Article history: Received 3 October 2008
Received in revised form 2 September 2009 Accepted 4 September 2009
Keywords:
Collective intelligence Traffic status prediction Smart traffic agent
Intelligent transportation system (ITS) Knowledge-based system
a b s t r a c t
Real-time traffic information collection and data fusion is one of the most important tasks in the advanced traffic management system (ATMS), and sharing traffic information to users is an essential part of the advance traveler information system (ATIS) among the intelligent transportation systems (ITS). Traditionally, sensor-based schemes or probing-vehicle based schemes have been used for collecting traffic information, but the coverage, cost, and real-time issues have remained unsolved. In this paper, a wiki-like collaborative real-time traffic information collection, fusion and sharing framework is proposed, which includes user-centric traffic event reacting mechanism, and automatic agent-centric traffic information aggregating scheme. Smart traffic agents (STA) developed for various front-end devices have the location-aware two-way real-time traffic exchange capability, and built-in event-reporting mechanism to allow users to report the real-time traffic events around their locations. In addition to collecting traffic information, the framework also integrates heterogeneous external real-time traffic information data sources and internal historical traffic information database to predict real-time traffic status by knowledge base system technique.
Ó 2009 Published by Elsevier Inc.
1. Introduction
Real-time traffic information plays an important role in several applications of intelligent transportation system (ITS), such as advanced traffic management systems (ATMS), advanced traveler information system (ATIS), commercial vehicle operation (CVO) and emergency management system (EMS), etc. It is one of the most useful features especially for users on the road, because it can assist and support travelers and drivers to plan the trip before travelling and decide on the route to take in order to reduce travel time and improve travel safety. Moreover, the traffic information collected can also support traffic management administrators in making decisions, taking appropriate actions to alleviate congestions, and improving the global performance of the traffic network. However, with one-way broadcasting mechanism, most traffic information services lack collective intelligence of gathering real-time traffic information from the service terminal devices and user-reporting mechanism, and continuous real-time traffic information for the whole traffic network is also hardly available due to cost, sensor coverage, and several other issues such as data collection and transmission techniques.
0020-0255/$ - see front matter Ó 2009 Published by Elsevier Inc. doi:10.1016/j.ins.2009.09.004
* Corresponding author. Address: Department of Computer Science, National Chiao Tung University, 1001 Ta Hsueh Road, Hsinchu 300, Taiwan, ROC. Tel.: +886 3 5731966.
E-mail addresses:leews@cht.com.tw(W.-H. Lee),sstseng@asia.edu.tw(S.-S. Tseng),shiehwy@mail.sju.edu.tw(W.-Y. Shieh).
Contents lists available atScienceDirect
Information Sciences
On the other hand, mobile personal devices such as mobile phones, PDAs are equipped with location capability by con-necting to GPS module by wired or wireless scheme. More and more personal navigation device (PNDs) and ultra-mobile PC (UMPC), which are equipped with GPS are now capable of communication in order to retrieve real-time traffic information. All these devices have the features of mobility, communication, location-aware and user-interface capabilities. With the front-end devices having these features, many modern mobile applications such as location-based service (LBS), telematics, and ITS-related applications can be provided. For example, LBS, a kind of service providing appropriate location-aware infor-mation for users, accomplish the business processes by exchanging inforinfor-mation between the mobile front-end devices and the back-end systems. On the basis of such high penetration of mobile communication devices with location capability, the goal of this paper is to propose a collaborative traffic information generation and sharing framework for real-time traffic information collection, fusion and distribution.
The applications of traffic information are generally divided into three stages, which are data collection and cleansing, data fusion and integration, and data distribution. In the data collection and cleansing stage, raw data are collected from dif-ferent data sources, filtered and analyzed in the back-end system. The collected data are then integrated with the geograph-ical information system (GIS) and external traffic information data sources in the data fusion and integration phase. In the data distribution phase, several interfaces including web, radio, and mobile subscription are provided for users, where the mobile subscription scheme suitable for users with mobile communication devices can retrieve real-time traffic information provided by the traffic information center (TIC) according to the device location.
Traditionally, there are three schemes for collection of real-time traffic information[5]: site-based, sensor-based and probing vehicle-based data collection schemes. Site-based measurement collects vehicle license plate information and arri-val times at various checkpoints through automatic vehicle identification (AVI) technologies, matches the license plates be-tween consecutive checkpoints, and computes travel times from the difference bebe-tween arrival times. Vehicle-based methods analyze the raw data collected from fleet of probe vehicles by matching the vehicle tracks with geographical infor-mation system (GIS). Sensor-based scheme collects raw data from the stationary sensors like loop detectors, transponders or radio beacons installed at arterial roads. However, these traffic information collection methods have drawbacks and limita-tions. For example, site-based and sensor-based methods have the spatial coverage problem due to the fixed and limited number of sensors or AVI devices. Vehicle-based scheme[1,8,12]has cost, spatial and temporal coverage problems due to the very high cost for maintaining a dedicated fleet of urban network traffic probing vehicles. Moreover, the cost of real-time transmission for the whole traffic network in each data collection scheme is also very high[4].
As information exchanging and sharing through a public platform, e.g., the Wikipedia website, is the major trend of Web 2.0 technology, several wiki-like collective intelligence models or applications have been proposed[3,7,9,10]recently. Our idea is to create a wiki-like environment for exchanging and sharing real-time traffic information. Accordingly, a collabora-tive real-time traffic information collection, data fusion and distribution framework including a front-end smart traffic agent (STA), real-time traffic information exchange protocol (TIEP) and traffic data fusion and distribution back-end center (TIC) is proposed in this paper. In addition to two-way real-time location-aware traffic information exchange with TIC through TIEP, STA is also developed as a smart interface between the terminal and the driver for displaying real-time and predicted traffic information, and providing real-time traffic event-reporting interface to share local traffic information with other users. Sev-eral versions of STAs are developed for various location-aware front-end devices such as GPS smart phone, PDA, PND, or UMPC, which have two-way real-time traffic information exchange capability, and built-in event-reporting mechanism to allow users to report the real-time traffic events around their locations. TIEP is an XML-based traffic information exchange interface between STA and the back-end TIC. The back-end TIC includes data collection and cleansing, data fusion and inte-gration, and a knowledge-based system for traffic status prediction. Several external traffic information data sources includ-ing national traffic information center [17], national freeway traffic information[19], and police station real-time traffic information[18]are integrated into the TIC by adopting the optimal weight traffic information fusion scheme[13,14]to en-hance the traffic information prediction capability. A knowledge-based system with expert heuristic rules is also imple-mented in the TIC to predict short-term traffic status by combining real-time collected traffic information, historical traffic database, and multiple external traffic information data sources.
The rest of this paper is organized as follows. Section2introduces the design details of STA and TIEP according to the request-response model. In Section3, the proposed collaborative traffic information generation and distribution framework is discussed, which includes traffic information collection and cleansing, traffic information generation for urban network and freeway, and external traffic information data sources connection. The integration and information fusion of heteroge-neous multi-sources external traffic information data sources is discussed in Section4, where optimal weighting scheme developed from the Shannon entropy theory is applied. In Section5, a knowledge-based system for predicting the traffic sta-tus is discussed. Finally, concluding remarks are presented in Section6.
2. Smart traffic agent (STA) and traffic information exchange protocol (TIEP)
STA, a software agent for the front-end devices, provides an interface between the traveler and the TIC. It is installed on the front-end device with a two-way real-time data exchange channel involving TIEP for (1) automatically echoing the local traffic information detected by STA to the TIC, (2) automatically downloading the local real-time and predicted traffic infor-mation, and (3) manually reporting the traffic events or traffic status to the TIC by the traveler. The TIEP is defined in XML
format based upon the request-response model, where the request message is integrated with the echoing traffic informa-tion to be sent to the TIC, which includes GPS coordinates, moving direcinforma-tion, speed, traffic status informainforma-tion and scope of request. The frequency and scope of exchange concerning the local traffic information is determined by rules embedded in the STA, which is calculated using the traveling speed and the vehicle location. For example, the higher the moving speed of the STA, the larger the scope of request and the shorter the frequency in order to retrieve more up-to-date and more suitable traffic information.
The TIC response message consists of a list of traffic messages which are divided into two parts: text message information and average speed information. The response messages received by the STA can be employed to update the map-based user interface, which helps the travelers easily catch the local real-time and predicted traffic information. The text messages in-form the travelers of the local traffic messages consisting of hno., message type, data source, message, severity, coordinatei such as traffic events, CMS. For example, text message h3, ‘events’, ‘TIC’, ‘car accident’, 5, x1, y1i indicates that a car accident re-ported by STA users happened at the location ðx1; y1Þ which has the severity level 5, and text message h5, ‘CMS’, ‘EIOT’, ‘Link A1(North) strongly congested, please detour!’, 4, x2, y2i informs that a changeable message sign (CMS) system shows ‘‘Link A1(north) is strongly congested, please detour!” message at the location ðx2; y2Þ where the message comes from the data source ‘‘EIOT”. On the other hand, the average speed information contains the current average speed and predicted traffic condition consisting of hno., type, link/intersection id, avg. speed, current status, predicted status, coordinatesi, which informs the current average traveling speed and predicted traffic status of the links around the STA. Real-time traffic status and pre-dicted traffic status are combined in the TIC response message by considering the prepre-dicted STA location. As illustrated in Fig. 1, after an STA request is sent from location A, TIC responds the real-time (dotted small rectangle) and predicted (dotted large rectangle) traffic messages centralized by predicted location B by considering the moving direction, moving speed, and response packet size (response time). In the dotted small rectangle centralized by predicted STA location at point B, the TIC responding message contains only real-time traffic information (average speed, traffic events) since the predicted informa-tion makes less sense to the current status. In the large dotted rectangle, both real-time and predictive traffic informainforma-tion is contained in the response message to inform drivers of the current status and predicted traffic status around the STA.
There is a wiki-like user-centric traffic information sharing mechanism for users to manually report local traffic events or traffic status to the TIC by the user interface provided by the STA, as shown inFig. 2. The reported traffic event or status will not be distributed instantly until it is confirmed by another traveler (through the same mechanism) or confirmed by the external data sources. The intention of this double-check mechanism is to prevent the system from receiving fake messages.
3. Collaborative traffic information generation and distribution framework
As shown inFig. 3, the architecture of the collaborative traffic information generation and distribution framework in-cludes three parts: location-aware mobile device with STA, TIC for data collection and fusion, and external traffic information data sources, where various versions of STAs are developed for different front-end mobile devices depending on different hardware and operating system. The generated real-time traffic information can be distributed by STA (request and response model), radio broadcast network (RDS), and web site. The TIC, is connected to external real-time traffic information data sources, including E-IOT national traffic events database[17], TANFB freeway traffic information[19], PRS public traffic ser-vice[18], and parking slot databases.
3.1. Traffic information collection and cleansing
Based on the TIEP, the traffic information is exchanged between local STAs and the TIC. Each record of data exchange col-lected by the request-response model between STA and TIC is treated as a traffic information sample, which will be cleansed and analyzed to get the local traffic status at the STA location. The local traffic information uploaded by the STA includes moving speed, moving direction, GPS coordinates, STA state, and scope of request. The TIC then answers the request by
A
B
real-time and predictive real-time
querying the traffic information database for the traffic status around the GPS coordinates of the STA. The cleansing process filters out the useless raw data, e.g., incomplete data, incorrect data, outlier, or noise data. The location information of the STA (link, address) is transformed from the GPS data in the collected sample with the assistance of traffic network database and GIS. A traffic information sample Skð0ji;V; DÞ transformed from the STA uplink request Ukof a vehicle, as illustrated in Eq.
(1), consists of object id ð0j
iÞ, speed (V), direction ðDÞ of the vehicle when it communicates with the TIC at time t at GPS
coor-dinates ðx; yÞ, where 0jiis a spatiotemporal network object spatially indexed by network object id i (transformed from the location to address interpolation) and temporally indexed by time zone j (transformed from timestamp t).
UkðX; Y; t; V; DÞ ! Gis
SkðO j
i;V; DÞ ð1Þ
3.2. Freeway traffic information generation
The focus of traffic information for the freeway area is different from that for the urban area. Travelers are more con-cerned about the traffic event messages and ramp-to-ramp traveling time than the route path choice on the freeway. On the contrary, in the urban area, travelers are more concerned about the route path choices for avoiding the congestion. Since
Fig. 2. User-centric traffic status reporting user interface in STA.
the topology in the freeway is different from that in the urban network, the traffic information generation schemes use dif-ferent sampling strategies to generate the traffic information depending on the location of the traffic information in the free-way area or urban area. For example, for a vehicle equipped with STA traveling through a freefree-way, a series of samples uploaded from the same STA are collected by the TIC, and each sample can be transformed into the format of ðMx;TxÞ, where
Mxindicates the mileage and Txindicates the uploaded timestamp. By selecting the entry sample ðMs;TsÞ and exit sample
ðMe;TeÞ of an STA, the average traveling speed ðVs;mÞ between these two ramps can be calculated by ðMe MsÞ=ðTe TsÞ,
which is regarded as a case. For any two ramps k and j, the overall average traveling speed can be calculated by the arith-metic mean value ðVk;jÞ for all the cases of ðk; jÞ as shown in Eq.(2).
Vk;j¼
n ðjMk MjjÞ Pn
i¼1ðTk;i Tj;iÞ
ð2Þ
3.3. Urban network traffic information generation
Urban network consists of a set of network objects, each of which is either a link or an intersection where traffic conges-tions occur on some network objects in which the traffic demand cannot be fully serviced. The traffic status of the network can be represented by a set of network objects which constitute the network. Average speed of a link in a temporal period can be calculated by the arithmetic mean value shown in Eq.(2)of all the samples spatially falling into the link and tempo-rally falling into the temporal period. However, in urban area, it seems unreasonable to represent the link traffic status by the average speed for all the links because the service levels of road grades of different categories are different. We normalize the link traffic status by the categories of the links, as shown inTable 1, the service levels of the three road categories in urban area of Taiwan are defined from levels A–F[2]. The link traffic information can be transformed into the service level by map-ping the average speed to the road category and service level defined inTable 1. For example, average speed of 30 km/h indi-cates that the traffic status is good (level B) in street (category III), but slight congestion (level D) in the expressway or arterial road (category I).
On the other hand, the average intersection delay can represent traffic information for the intersection, and the delay be-tween two consecutive links, which is mostly caused by signal delay and queuing delay, can be classified by TD, LTD and RTD patterns according to the three possible directions from one link to another[4]. Eq.(3)shows the general format of inter-section delays (TD/LTD/RTD), where P is the pattern type, SOidand SIidare the two consecutive links where vehicle leaves
out the link SOidand comes into the link SIid, Tidis the temporal id, Davgis the average delay time of this intersection, and
Sup, Con are the support and confidence of the pattern, respectively.
½TD=LTD=RTD : ðP; SOid;SIid;Tid;Davg;Sup; ConÞ ð3Þ
For example, (‘RTD’, ‘L1’, ‘L2’, ‘W,P’, 40, 0.2%, 75%) represents that in the peak hours of workday, it takes 40 seconds to make a right-turn from link ‘L1’ to link ‘L2’, with the support being 0.2% and confidence being 75%. Intersection delay patterns can be discovered by sequential pattern mining or spatial and temporal sequence mining on all samples of intersection delay in a journey containing two consecutive samples with different links in the historical traffic information database.Fig. 4shows an example of RTD pattern: a probing vehicle driving north and then turning right to east, it reports TIS at location A of Link La
and reports consecutively TIS at location B of Link Lb. The symbols of the TIS format (T,L,X,Y,D,V) inFig. 4stand for timestamp
(T), link id (L), coordinates ðX; YÞ, direction (D) and speed (V). The distance da;dbinFig. 2stands for the distance from A or B to
the intersection of links Laor Lb, respectively. Assuming that in the short time interval between Tband Ta, the vehicle is
driv-ing at the speed of Vaat link Laand Vbat the link Lb. Then the right-turn delay (RTD) from Lato Lbcan be estimated by
sub-tracting travel time of daand dbfrom elapsed time between two TISs ðTb TaÞ.
The status of each intersection delay in an intersection is also classified into six service levels (A–F) by normalizing each intersection delay, i.e., by equally dividing the intersection delay samples into six levels according to the historical traffic information database. With the traffic status of all the links and intersections, the whole network status can be easily rep-resented on the map by coloring the objects in the user interface.
Table 1
Service level classification for the three road categories in Taiwan[2].
Road grade I II III
Free flow speed 55 (kph) 45 (kph) 40 (kph)
Service level Avg. speed (kph) Avg. speed (kph) Avg. speed (kph)
A (90%) >51 >43 >33 B (70%) 51–39 43–32 33–25 C (50%) 39–34 32–27 25–20 D (40%) 34–29 27–23 20–16 E (33%) 29–21 23–17 16–10 F (25%) <21 <17 <10
3.4. External traffic information data sources connection
In addition to the traffic information collected from LBS, external real-time traffic information data sources including na-tional traffic information center (E-IOT)[17], Taiwan Area National Freeway Bureau (TANFB) freeway traffic information sys-tem[19], public traffic service of police radio station (PRS)[18], and several parking slot databases are integrated in the framework to retrieve more driving assistance information such as real-time CMS (changeable message sign) information, and parking slot information. Accordingly, the traffic information in the framework is sufficient for not only reporting real-time traffic information but also predicting the traffic status and can be further fed back to these external data sources for sharing the information with travelers not equipped with STA.
3.5. Robustness of the system
In the proposed framework, huge number of STAs may request to a centralized TIC and overload it, so that the robustness of the TIC must be considered. For the robustness issue, several data servers can be set up in the TIC in order to handle large amount of STAs, and the load distribution can be done by distribution rule embedded in the STA or the corresponding data server assigning rule defined in the TIEP. The principle of these distribution methods is to averagely share the load to the data servers in the TIC.
4. Multi-sources heterogeneous traffic information fusion
The traffic information collected in the framework is classified by traffic events (message), and traffic status (link average speed, intersection status). The collected STA-centric echoing samples can be transformed into freeway traffic status, urban network link and intersection traffic status as discussed in Section3. The generated traffic status and user-reported traffic events are stored in the real-time traffic database, and the out-of-date data will be periodically moved into the historical traffic database.
In order to integrate the required external real-time traffic information data sources with the generated traffic informa-tion in the framework, an optimal weighting combinainforma-tion scheme developed from Shannon entropy theory[11]for traffic data fusion approach[13–15]is adopted. The internal traffic information in the framework and each of the external traffic data source is regarded as a data source ðSiÞ. The weighted combination fusion equation for a network object is shown in Eq.
(4), where Ojcan represent any attribute of a network object such as link average speed, intersection delays (TD/LTD/RTD),
and Wiis the weight for the data source Si.
Oj¼ X i Wi SiðOjÞ; X i Wi¼ 1 ð4Þ
The Shannon entropy is employed to measure the uncertainty and randomness of the collected data and external data sources, and the optimal weighting approach[14]tries to minimize the overall uncertainty by assigning the lower weight to higher entropy data sources. The entropy with conditional probability of data source Si, denoted as hið
c
Þ which is adoptedfrom[13], is shown in Eq.(5), where Pðhjj Þ is the probability of occurrence of the state h
c
jgiven the observed statec
. Theprobability can be summarized from the historical traffic database; for example, the probability of service level ‘C’ on road Ljcan be calculated by the speed samples which are located in Ljand falls in the range of level ‘C’ in the road category of Lj
divided by the total samples located in Lj. The weight assignment equation in optimal weighting approach[14]is shown in
Eq.(6), where S is the set of all data sources. It shows that the weight Wið
c
Þ has a negative relationship with h 2ið
c
Þ, so thatdata source i will be assigned a lower weight Wiif the entropy hi(uncertainty) is high.
A
B
d
ad
bA,B : two consecutive TISs with different link id A (Ta, La, Xa, Ya, Da, Va) B (Tb, Lb, Xb, Yb, Db, Vb) O
hið
c
Þ ¼ X Pðhjjc
Þ log Pðhjjc
Þ ð5Þ Wiðc
Þ ¼ 1 h2iðc
Þ P i2Sh2i1ðcÞ ð6Þ However, there are several traffic characteristics in temporal dimension such as peak hour, weekday, or midnight. Without considering temporal traffic characteristics, the probability calculation in the entropy hiðc
Þ equation defined in Eq.(5)maylose the precision for the ‘‘optimal weight” calculation. For example, traffic status characteristics on workdays are different from those on holidays, and traffic characteristics during AM peak hours are not the same as those during PM peak hours. In order to take the traffic status characteristics in temporal dimension into consideration, we divide the collected samples in the historical traffic information database into five temporal periods: {AM peak hour, PM peak hour, workday, holiday, mid-night}. The probability of occurrence of the state hjgiven the observed state
c
;Pðhjjc
Þ is modified into Pðhkjjc
Þ to extend thetemporal dimension consideration, where k is the index for temporal periods. The entropy defined in Eq.(5)is then modified to consider temporal dimension, as shown in Eq.(7), where hkið
c
Þ denotes the entropy of network object i at temporal index k.hkið
c
Þ ¼ X Pðhk jjc
Þ log Pðh k jjc
Þ ð7ÞIt is easy to map the events from the original entropy definition (Eq.(5)) to the new definition (Eq.(7)) by mapping the event timestamp to one of the five pre-defined temporal periods stated above. The advantage of adding this temporal dimension consideration is that the effects of the traffic events happened in different temporal periods can be considered differently. For example, a serious traffic accident event happened in the midnight may slightly affect the traffic flow, but the same event may seriously affect the traffic flow if it happened in the AM/PM peak hours.
5. Knowledge-based traffic status predicting system
The traffic event messages collected from the STA and external traffic data sources are used not only for informing current traffic status but also for predicting the traffic status. For each STA traffic information request, the prediction result is sent to the STA accompanied with real-time traffic information, as illustrated inFig. 1. The traffic status prediction is implemented by the knowledge-based system, in which expert heuristic rules are elicited to enhance the precision of traffic status predic-tion. There are many rules for traffic events to be considered for traffic status prediction on an STA request, such as event type, severity, event elapsed time, and spatial consideration related to the STA. For example, a traffic accident may have se-vere impact on the surrounding STAs, but the severity will be reduced as the distance increases.
Link service level estimation intersection delay estimation Spatiotemporal congestion patterns right turn delay left turn delay through delay is a is a includes Travel status prediction is a External event integration Optimal weight assignment related-to includes Intersection delay patterns includes includes Default values related-to related-to includes STA reaction events Internal event aggregation Dynamic linear combination include related-to include Historical Traffic info. DB mining mining aggregate STA Request Event
type & severity measurement Event elapsed time effect measurement STA spatial relationship measurement invoke includes includes includes
The design ontology of traffic status prediction knowledge base system is illustrated inFig. 5, which includes several de-sign concepts and relationships between these concepts, such as ‘‘is-a”, ‘‘includes”, ‘‘modify”, and ‘‘related-to”. These relation-ships connecting concepts in the ontology represent some interactions among them. For example, the ‘‘traffic status prediction” concept includes several concepts such as ‘‘dynamic linear combination”, ‘‘intersection delay estimation”, ‘‘link ser-vice level estimation”, and three traffic event factor measurement concepts. When the system receives an STA traffic status request, the ‘‘traffic status prediction” concept is then ‘‘invoked” to process the whole traffic status prediction process, which again invokes those concepts included in the ‘‘traffic status prediction” concept. The optimal weight assignment module inte-grates the internal STA reported events and the external traffic event information sources by optimal weighting linear com-bination scheme. On the other hand, the real-time link service level and intersection delay estimation modules retrieve the current link service level and intersection delay information from the real-time traffic information database. ‘‘Traffic status prediction” returns the current traffic status around the STA (real-time area inFig. 1) and predicts the traffic status (real-time and prediction area inFig. 1) by fine-tuning the current traffic status and considering the localized STA and traffic event im-pacts consideration modules including ‘‘event type and severity measurement”, ‘‘event elapsed time effect measurement”, and ‘‘event location to STA spatial relationship measurement”.
The ‘‘event type and severity measurement” module evaluates the effect of the event to the STA by its type and severity, and tunes the real-time evaluation result; for example, the service level prediction of the link and related intersections surround-ing the traffic accident event may be downgraded one or two levels compared with the current link service level and inter-section delay estimation. The ‘‘event elapsed time effect measurement” and ‘‘event location to STA spatial relationship measurement” modules tunes the link service level and intersection delay by considering the temporal distance (event elapsed time from the STA request time) and spatial relationship (distance from the event to the STA), respectively. The traf-fic status prediction knowledge base prototype system has been implemented by the expert system shell, DRAMA[16], a New Object-oriented Rule Model (NORM)[6]knowledge-modeled rule base system platform implemented using pure Java language, includes DRAMA Server, Console, Knowledge Extractor, and Rule Editor. Heuristic rules donated by the traffic do-main experts are categorized into several knowledge classes (KC), and stored at knowledge base of DRAMA server.
6. Conclusion
Real-time traffic information plays an important role in ITS, especially for travelers, fleet managers, and ATMS adminis-trators. However, the availability of real-time traffic information suffers from the cost, coverage, and real-time issues in the traditional data collection techniques, so compromise solution must be made among these considerations. A wiki-like col-laborative real-time traffic information generation and sharing framework involving a real-time traffic status prediction knowledge base system is proposed in this paper. By this collective traffic information generation scheme, more real-time traffic data will be collected cost-effectively and accurately, and the spatiotemporal coverage is better than the traditional traffic information collection scheme. Front-end STA, TIEP protocol, and back-end TIC constitute the traffic information shar-ing framework, where STA and TIEP are designed for the location-aware real-time traffic information exchange between the STA and TIC, which enables STA to automatically echo the local traffic information detected by STA to the TIC, automatically download the local real-time and predicted traffic information, and report the traffic events or traffic status to the TIC by the traveler.
In additional to real-time traffic information collection and generation from STA, the TIC back-end system adopts the Shannon entropy-based optimal weighting approach[14]to integrate external real-time traffic information data sources to minimize data uncertainty. A rule-based knowledge system is implemented to predict the short-term traffic status by fine-tuning the real-time traffic status by considering the event impact factor. Heuristic rules donated by domain experts improve the prediction accuracy. The methodology proposed in this paper turns the traditional government centralized data collection and distribution scheme into wiki-like distributed collective intelligence scheme, which may become the main stream for real-time traffic information collection and sharing scheme in the near future.
Acknowledgements
This work was partially supported by National Science Council of the Republic of China under Grant No. NSC95-2520-S009-008-MY3. The authors wish to express their appreciation to the Telecommunication Labs of Chunghwa Telecom Co. Ltd. for co-working the proposed framework.
References
[1] E. Chung, M. Sarvi, Y. Murakami, R. Horiguchi, M. Kuwahara, Cleansing of probe car data to determine trip OD, in: Proceedings of 21st ARRB and 11th REAAA Conference, Cairns, Australia, 2003.
[2] Tzu-Jung Cheng, Tyh-ming Lin, Yu-Ching Ou-Yang, 2001 Highway Capacity Manual in Taiwan, Government Publication No.: 1009001056, 2001, Available:<http://www.iot.gov.tw/>.
[3] V. Furtado, L. Ayres, M. de Oliveira, E. Vasconcelos, C. Caminha, J. D’Orleans, M. Belchior, Collective intelligence in law enforcement – the wikicrimes system, Elsevier Information Sciences 180 (1) (2010) 4–17.
[4] Wei-Hsun Lee, Shian-Shyong Tseng, Sheng-Han Tsai, A knowledge base real-time travel time prediction system for urban network, Elsevier Expert Systems with Applications 36 (3, Part 1) (2009) 4239–4247.
[5] H.E. Lin, R. Zito, A review of travel-time prediction in transport and logistics, in: Proceedings of the Eastern Asia Society for Transportation Studies, vol. 5, 2005, pp. 1433–1448.
[6] Y.T. Lin, S.S. Tseng, C.F. Tsai, Design and implement of new object-oriented rule base management system, Elsevier Expert Systems with Applications 25 (3) (2003) 369–385.
[7] I. Lykourentzou, K. Papadaki, D.J. Vergados, D. Polemi, V. Loumos, CorpWiki: a self-regulating wiki to promote corporate collective intelligence through expert peer matching, Elsevier Information Sciences 180 (1) (2010) 18–38.
[8] T. Nakata, J. Takeuchi, Mining traffic data form probe-car system for travel time prediction, in: Proceedings of ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Seattle, WA, USA, August 2004, pp. 817–822.
[9] Debbie Richards, A social software/Web 2.0 approach to collaborative knowledge engineering, Elsevier Information Sciences 179 (15) (2009) 2515– 2523.
[10] Martijn C. Schut, On model design for simulation of collective intelligence, Elsevier Information Sciences 180 (1) (2010) 132–155. [11] C. Shannon, W. Weaver, The Mathematical Theory of Communication Urbana, University of Illinois, 1964.
[12] J.S. Yang, Travel time prediction using the GPS test vehicle and Kalman filtering techniques, in: Proceedings of American Control Conference, Portland, OR, USA, 2005, pp. 2128–2133.
[13] Hsin-Chieh Wu, The Application of Entropy of Data Fusion of Traffic Information, Thesis of Master of NCTU, 2004, pp. 16–19. [14] Ruei-Feng Wu, The Optimal Weighting Approach for Traffic Data Fusion, Thesis of Master of NCTU, 2006, pp. 11–27.
[15] Yifeng Zhou, Henry Leung, Minimum entropy approach for multisensor data fusion, in: Proceedings of IEEE Signal Processing Workshop on Higher-Order Statistics (SPW-HOS ’97), 1997, p. 0336.
[16] DRAMA, Coretech Inc., 2003, Available:<http://www.ctknow.com.tw/rule.htm>.
[17] National traffic information center, Institute of Transportation MOTC, Available:<http://www.e-iot.iot.gov.tw/>. [18] Police station public traffic service, Police radio station, Available:<http://www.prs.gov.tw/>.
