Generating Probabilistic Path Observation from GPS Data for Route Choice Modelling
J Chen, J Newman, M Bierlaire, Ecole Polytechnique Federale de Lausanne, CH
Modeling mobility patterns with discrete choice models and data from Nokia smart phones
Jingmin Chen, Jeffrey Newman and Michel Bierlaire
Ecole Polytechnique Federale de Lausanne, Transport and Mobility Laboratory, Station 18, CH-1015 Lausanne, Switzerland
In order to get high quality mobility data, recent surveys apply devices with GPS functionality as mobility tracking tools. Recently developed cell phone technologies offer us the possibility to use it as an even further advanced survey tool, e.g., by recording the user's choice decisions as well as his/her situation when the choice is made. Motivated by this possibility, we launch a project in collaboration with the Nokia Research Center in Lausanne on using GPS-capable smart phones to collect data and estimate state-of-art discrete choice model from the data.
In this project, we give out 100 Nokia smart phones with pre-installed data collection software to the respondents. Each respondent will use it as his/her regular cell phone, carrying it along with him/her throughout the day, while the software constantly records data and sends this data regularly to a server. This is done automatically and does not need to be triggered by the respondent. Technically, the data made available through the software on the smart phone, N95 for instance, is not only from the cell phone's regular functions such as calendar, and sound recorder, but also from internal additional chips such as GPS receiver, wifi receiver, accelerometer, light sensor, and camera. However, due to privacy issues, an agreement is made with each respondent that indicates which types of data the respondent would like to provide. Given this wealth of available information, we are interested in revealing the respondents' mobility patterns.
In addition to the automatically collected data from the smart phones, online surveys are conducted for different research purposes: for example, for the purpose of route choice modeling, a prompted recall questioning survey asks respondents for their trip purposes and transportation modes. During this survey, data recorded from the smart phone is harnessed to help the respondents to remember their choice situations. For example, GPS tracks are plotted on a map as auxiliary information. This method also reduces the respondents' burden during the survey.
In order to avoid biases when matching the GPS data to the real network, a recently developed modeling approach with network-free data is used to identify travelers' route choice decisions. Instead of identifying exactly one path through the network based on the recorded GPS track, which is the method used in standard map-matching algorithms, this modeling approach produces probabilities for several paths which the GPS track may belong to. An initial test on the N95 phone shows that the recorded GPS data quality is not as good as the data recorded from the dedicated hand-held GPS tracking device "MobilityMeter". This imprecision in the GPS data further motivates the application of the approach with network-free data. It is also noteworthy that the smart phone provides functions such as Assisted-GPS that enhance the GPS performance during startup or in poor signal conditions.
This paper reports on (i) the design of a data collection campaign using GPS-capable smart phones and online surveys, (ii) associated findings such as advantages and disadvantages of using smart phones as tracking devices, and (iii) preliminary model estimation results from the collected data.
Association for European Transport