Cleansing of Probe Car Data to Determine Trip Od

GPS is increasingly being used to collect travel data as the cost of the equipment is relatively low and it is capable of providing continuous and accurate spatial information and speed in real time. One such example is the Internet Protocol probe car (IPCar) project in Japan which equipped probe cars (consisting of taxis and buses) with GPS. The aim of this project is to explore feasible real time applications of IPCar data either as a stand alone data source or with other data sources such as detector counts and automatic vehicle identification (AVI) travel time. The initial focus of this project is to provide travel time information. This study focuses on a data cleansing consisting of 6 steps to determine the OD pattern of the probes. The cleansing process addresses data errors and searches for trip ends and the results show that the cleansed data was accurate in terms of trip length distribution, in particular for trips of more than 1km. The probe data also matched 76%-83% of the trips from an independent data from a taxi management system. The OD pattern gives a macroscopic view of the taxi movement and shows that the main generator and attractor of trips are areas around Sakuragi-cho, Yokohama station, Honmoku, Negishi station and Totsuka. The OD pattern and desired line of the probe car also demonstrate that using taxis as probe vehicles only generate intense data for the small areas heavily serviced by taxis and therefore travel time at a higher level of confidence can only be predicted for these small areas. Future full scale implementation of the IPCar project may need to consider private cars and other sources of information such as detectors and AVI data, to supplement area with sparse taxi coverage. INTRODUCTION GPS is increasingly being used to collect travel data as the cost of the equipment is relatively low and it is capable of providing continuous accurate spatial information and speed in real time. One such example is the Internet Protocol probe car (IPCar) project in Japan which equipped probe cars (consisting of taxis and buses) with GPS. The aim of this project is to explore feasible real time applications of IPCar data either as a stand alone data source or with other data sources such as detector counts and automatic vehicle identification (AVI) travel time. The initial focus of this project is to provide travel time information and would later extend it to provide other information such as congestion level and emission level. The quality of travel time information from probe vehicles depends on the frequency of probe vehicles traversing a road link. A large sample of probe vehicles per link per unit time would provide travel time with a higher level of confidence. However, the frequency of probe vehicle is a function of the number of probe vehicles and distribution of probe vehicle trips over the network. In order to address the questions about distribution and frequency of probe vehicles, a detailed understanding of the origin destination (OD) pattern of IPCar is essential for the development of a travel time prediction model so as to meet the specified level of accuracy and confidence (see Figure 1). Only OD pattern of taxis are analysed as buses runs on a fixed route and schedule. From the OD patterns, the coverage of the probe cars at different times of the day and areas where the probe cars coverage is sparse can be determined. With the travel pattern, it is then possible to see areas where the probe cars coverage is sparse. Even in areas that are well covered, certain routes may be more popular at certain time of day. The next stage is to analyse the link characteristics to determine the frequency of probe car traversing each link per unit time interval. Once accurate OD information of taxis can be determined, it can be decided whether taxis are suitable probe cars to determine accurate travel time. This study focuses on the data cleansing of probe data to determine the OD pattern of the probes, shown as shaded boxes in Figure 1. The OD pattern gives a macroscopic view of the taxi movement. Detail study of link characteristics such as travel time variance and the development of a travel time prediction model using probe data will be published in the near future. In the following section, an overview of the IPCar system is provided. The steps involved in the cleansing of the probe car data and the trip distribution (OD) of the probe car are presented in detail in the subsequent sections. IPCAR SYSTEM The Yokohama IPCar project uses 179 vehicles consisting of 140 taxis and 39 buses. The experiment ran for 11 days in December 2001. The IPCar system is equipped with a GPS and a data logger. GPS collects position data at regular intervals. However, the IPCar system does not store the vehicles position at regular intervals. Instead it logs the state of events as either short stop (SS) or short trip (ST) (see Figure 2). The definition of a short stop is when the vehicle speed drops below 3 km/h. When the vehicle speed increases above 3 km/h, the event is considered as a short trip. In other words, instead of a time based data logging, the equipment is an event based data logging. So every time the event changes from SS to ST and vice versa, the GPS position and time stamp are recorded plus the event flag (eg. SS or ST). This approach reduces the amount of data stored and therefore less data transmission, without sacrificing the quality of the data (Horiguchi, 2002). Note that there is a maximum time limit of 30 seconds for a ST event. If a probe vehicle is moving for 2 minutes, it will be recorded as 4 consecutive ST events. However, there is no time limit for SS event. These event records are temporary buffered in the on-board equipment and transmitted by the polling request from the IPCar data centre. The interval of transmission is normally 1-5 minutes. In addition, the IPCar system records other parallel events information such as the status of the left and right blinkers, the hazard light and the parking brake (see Figure 2). There are instances where data are not recorded ie. contains gap. This could be due to communication or GPS errors. GPS errors might occur when a probe vehicle passes under an infrastructure such as tunnel, or when in the vicinity of elevated structures, the so called urban canyon. Gap could also occur when the engine is switched off because no data will be recorded. Taxi management system Three taxi companies took part in the experiment and one of the taxi companies also have their own Taxi Management System installed on the taxi. There are 16 probe vehicles with taxi management system. The taxi management system collects the time, date and position when the fare meter is on or off. In other words, when the taxi goes from in-service to not in-service and vice versa. This data source is used to verify the results from the data cleansing discussed in later section. DATA CLEANSING Before the probe data can be used to determine trip OD, the data needs to be cleansed because probe data is a continuous trajectory (see Figure 3) and also there are gaps in the data. Therefore, the data cleansing process for the OD analysis is to cut the “continuous” trajectories into trip ends by detecting the following events (see Figure 4). • Gap with parking brake event, • Long gap, • Gap with unrealistic speed, • Long stop, • Short stop with hazard light, and