Taxi Queue, Passenger Queue or No Queue? - ?· Taxi Queue, Passenger Queue or No Queue? ... relatively…

Download Taxi Queue, Passenger Queue or No Queue? - ?· Taxi Queue, Passenger Queue or No Queue? ... relatively…

Post on 12-Jun-2018

220 views

Category:

Documents

2 download

Embed Size (px)

TRANSCRIPT

  • Taxi Queue, Passenger Queue or No Queue?

    A Queue Detection and Analysis System using Taxi State Transition

    Yu Lu, Shili Xiang, Wei Wu

    Institute for Infocomm Research, A*STAR, Singapore

    {luyu, sxiang, wwu}@i2r.a-star.edu.sg

    ABSTRACTTaxi waiting queues or passenger waiting queues usually re-flect the imbalance between taxi supply and demand, whichconsequently decrease a citys trac system productivityand commuters satisfaction. In this paper, we present aqueue detection and analysis system to conduct analytic-s on both taxi and passenger queues. The system utilizesthe event-driven taxi traces and the taxi state transitionknowledge to detect queue locations at a coordinate leveland subsequently identify 4 dierent types of queue context(e.g., only passengers queuing or only taxis queuing). Morespecifically, it adopts the novel and easy-to-implement al-gorithms to selectively extract taxi pickup events and theircritical features. The extracted taxi pickup locations arethen used to detect queue locations, and the extracted crit-ical features are used to infer queue context. The extensiveempirical evaluations, which run on daily 12.4 million taxitrace records from nearly 15000 taxis in Singapore, demon-strate the high accuracy and stability of the queue analyticsresults. Finally, we discuss the real world deployment issuesand the gained insights from the queue analysis results.

    1. INTRODUCTIONIn the densely populated Asian cities (e.g., Singapore, Bei-jing and Taipei), relatively cheap taxi fares and large num-ber of taxis greatly facilitate the pervasive usage of taxisby urban citizens for various purposes, such as traveling be-tween oce and home, purchasing groceries at supermarketsand visiting friends. It is relatively dierent from the tax-i usage at many cities in US or Europe, where taxis morefrequently serve airport routes and do not cover all urbandistricts. The taxi usage characteristics in the Asian citieseasily cause that the temporal and spatial imbalance of taxisupply and demand occurs frequently: taxis would queue upfor passengers due to temporarily low taxi demand but highsupply nearby; passengers would queue up for taxis due totemporarily high taxi demand but low supply; in many timeperiods, taxis and passengers would concurrently queue upas both taxi demand and supply are high. Such queuing

    c 2015, Copyright is with the authors. Published in Proc. 18th Inter-national Conference on Extending Database Technology (EDBT), March23-27, 2015, Brussels, Belgium: ISBN 978-3-89318-067-7, on OpenPro-ceedings.org. Distribution of this paper is permitted under the terms of theCreative Commons license CC-by-nc-nd 4.0

    (a) Taxi Queue (b) Passenger Queue

    Figure 1: Dierent Types of Queue in Singapore

    events usually not only reduce the productivity of an urbantrac system, but also greatly decrease the satisfaction ofpublic commuters as well as taxi drivers. Fig. 1 illustrates ataxi queue and a passenger queue that both frequently occurin Singapore.

    Properly and accurately detection of queue locations andqueue context would benefit many parties and stakeholders.The real time queuing events information and their long-term patterns can be used in the recommendation systemsfor taxi drivers and commuters (e.g., suggest commuters tothe nearby taxi queue locations). The information can al-so be used in the taxi operators booking and dispatchingsystems (e.g., guide available taxis to passenger queue lo-cations). Moreover, the government agencies need such in-formation to understand the imbalance between taxi supplyand demand, and accordingly take necessary actions (e.g.,increase operating taxis or adjust taxi fares).

    Motivated by the availability of abundant information intaxi traces, e.g., GPS locations and taxi states, using taxitraces to design and build a city scale queue detection andanalysis system is a promising solution. However, it is anopen and non-trivial problem. Firstly, taxi queuing for pas-sengers is not simply a passenger pickup, dropo or vehicleparking event, only the GPS coordinates and the binary taxistates (occupied or non-occupied) are not enough to captureit. Secondly, passenger queuing for taxis is even more di-cult to detect, as no any direct information from the passen-ger side and no apparent clue in taxi traces. Thirdly, bothtaxi queuing and passenger queuing are highly dynamic interms of time and locations, which not only repeatedly occurat fixed taxi stands or during peak hours.

    In this paper, we present a practical system that captures

    593 10.5441/002/edbt.2015.60

    http://OpenProceedings.org/http://dx.doi.org/10.5441/002/edbt.2015.60

  • taxis and passengers queuing activities at a fine-grained s-cale, i.e., at the individual coordinate level rather than aregion or zone, and subsequently analyzes dierent types ofqueue context. We summarize the key contributions of thiswork as follows:

    We propose a novel approach, using multiple taxi s-tates and their transition information, to conduct thecity scale queue analytics for both taxis and passen-gers.

    We design and implement a two-tier queue analyticsengine, where the lower tier module detects queuinglocations and the upper tier module identifies queuecontext based on the selected taxi pickup events andfeatures.

    We conduct the extensive empirical evaluation of ourqueue analytics results before we deploy the systemin the real world. We demonstrate its stability usinglarge scale taxi traces, and its accuracy using variousother data sources (e.g., landmark information, failedtaxi booking data).

    The rest of the paper is organized as follows: section 2 in-troduces the background of the dataset, and then section 3depicts our overall system architecture and define the queuetypes. In sections 4 and 5, we describe our queue spot de-tection and queue context disambiguation modules in detailrespectively. Extensive empirical evaluations are conductedin section 6, which is followed by a discussion on deploymentand other issues in section 6. The related work is present-ed in section 8. At last, we conclude with future work insection 9.

    2. TAXI STATE AND EVENT-DRIVEN LOG2.1 Mobile Data TerminalAs part of the taxi operators eorts on improving theirquality of service, each taxi in Singapore is equipped witha specifically designed device, called mobile data terminal(MDT), which is mainly used to handle taxi bookings andmonitor a taxis real time status. More specifically, it re-ceives taxi booking tasks from the backend service (taxi cal-l center), and sends back taxi drivers decision (accept orreject the task) via general packet radio service (GPRS).Moreover, MDT keeps logging and updating a taxis realtime state by collecting the information from taxi meter,roof-top signs and its frontend touch screen. Fig. 2 simplydepicts an MDT system on a Singapore taxi, where MDTis hardwired directly to dierent on-vehicle devices and pro-vides taxi drivers a multifunctional touch screen.

    2.2 Taxi StateBased on the collected real time information, the MDT de-vice is able to precisely identify 11 dierent taxi states. Ta-ble 1 lists all the taxi states with their descriptions. The taxistate transitions mainly depend on the type of a taxi job. Inprinciple, all taxi jobs can be classified into two categories:street job and booking job.

    A street job means a taxi picks up new passengers by streethail, and the following is the typical taxi state transitionson a street job:

    Figure 2: A simplified telematics system on a Sin-gapore Taxi

    a) a passenger hails down a taxi with FREE state along aroad or a taxi stand.

    b) the taxi driver starts the taximeter for a new trip, andmeanwhile the MDT updates the taxi state to POB.

    c) during the trip, the taxi state keeps POB while the MDTperiodically updates the taxi GPS location.

    d) the taxi is approaching the destination and the driverpresses the STC button on the MDT touch screen toupdate the taxi state to STC.

    e) upon arrival of the destination, the driver presses the but-ton on the taximeter for printing the receipt, and mean-while the MDT updates the taxi state to PAYMENT.

    f) once the driver resets the taximeter after the passengeralights, the MDT automatically updates the taxi state toFREE again.

    A booking job means a taxi picks up new passengers, whohave made a booking via telephone, short message service(SMS) or mobile phone applications (apps). The typical taxistate transitions on a booking job can be described as below:

    a) a passenger makes a taxi booking, and the backend ser-vice dispatches the booking information to the nearbytaxis with FREE or STC state.

    b) a taxi driver successfully bids the booking job by pressingthe button on the MDT touch screen, and meanwhile theMDT updates the taxi state to ONCALL.

    c) upon arrival of the booking pickup location, the MDTupdates the taxi state to ARRIVED.

    d) if the passengers do not show up within a specific timeperiod (e.g., 15 minutes), the MDT updates the taxi stateto NOSHOW first and then to FREE within 10 seconds.

    e) if the passenger gets on the taxi in time, the MDT up-dates the taxi state to POB once the driver starts thetaximeter.

    f) the subsequent taxi state transitions are the same asstreet jobs procedure, i.e., from street jobs step c) tostep f).

    Fig. 3 illustrates a complete taxi state transition diagram,which includes the procedures of both street jobs and book-ing jobs.

    594

  • Table 1: Taxi State and Description

    Taxi State Description FREE Taxi unoccupied and ready for taking new passengers or bookings POB Passenger on board and taximeter running STC Taxi soon to clear the current job and ready for new bookings

    PAYMENT Passenger making payment and taximeter paused ONCALL Taxi unoccupied, but accepted a new booking job AR

Recommended

View more >