Methods Of Extracting Trip Information Using Survey Data Collected Through A Smartphone App

Developing accurate travel demand forecasting models requires a large amount of data.These data have traditionally been collected through household surveys.However,these surveys are often time-consuming to complete as they are normally carried out using paper or web-based questionnaires.Furthermore,administering these surveys requires a large amount of manpower,materials and resources,and users are often reluctant to participate in this kind of survey.Recently,technological developments such as smartphone App and mobile computing,have created new opportunities to collect this data more efficiently.Most smartphones have GPS tracking function which can be used track the trajectory of users.These data also contain rich information regarding the trips accomplished by smartphone users,such as trip's starting/ending time,traveling speed,mode and duration of each activity,which cannot be collected through traditional surveys.However,such information cannot be used directly to develop travel demand models as they are hidden inside trajectory data and need to be extracted,as raw trajectory data only contains the time,location and spot speed and acceleration/deceleration rate of each GPS point.Extracting the above information can be challenging,as a large amount of point data has to be analysed systematically and many factors must be considered,such as if the points are part of a trip or if they are collected at an activity location.Current methods used to extract such information are often inaccurate.This thesis aims to address these shortcomings by developing and evaluating new algorithms to extract trip information accurately from user's trajectory and socio-economic data.This thesis uses user's trajectory data and personal socioeconomic data collected by a smartphone APP developed by our team to extract the above trip information.The APP collects GPS trajectory data as well as the information on the travel purpose and travel modes of users,which are required inputs from the people participating in the survey.This thesis also proposes new algorithms to analyze the inherent space and time attributes of trajectory data and considers the effect of personal socioeconomic attributes.This thesis covers the following:(1)Firstly,in order to apply and verify the algorithms for extracting tripinformation,a novel travel survey system,which mainly includes a server and a smartphone App targeting to household/personal travel survey,is developed and then used to collect travel behavior data from the Qingpu district,Shanghai.As the data collected by the APP is not sufficient for some of the analyses presented in this thesis,additional vehicle trajectory data and data from traditional paper-based surveys were also collected and used to supplement the data collected from the proposed “server +smartphone App” system.The data is then processed and cleaned before being used as input into the models developed in this thesis.(2)Much of the research regarding activity point data does not distinguish between indoor and outdoor data.This thesis proposes an improved classification method based the location accuracy,speed,and maximum distance from the activity location.These data are then used to differentiate indoor activity points from outdoor points.This method also makes full use of the GPS feature that its location accuracy goes down quite bit under the roof to identify indoor locations where activities occur.The overall accuracy of this method is 96%.The identification of activity points also allows the trajectory to be divided into travel segments and activity segments,which serves as a foundation for the identification of travel purpose and travel patterns.(3)Social and economic attributes are strong factors influencing the travel purpose.To identify travel purpose,this thesis applies a neural network that uses socio-economic attributes(gender,age,occupation,and educational level),trip-related data(holidays,arrival time,travel time,and residence time),land use data and other attributes for identifying trip purpose.Due to the fact that the current trip purpose is often correlated with that of the previous trips,the trip purpose of last trip is therefore used as one of the inputs to the neural network model developed.The overall recognition accuracy of this method is found to be 94%.(4)A fuzzy decision tree model is developed for identifying/recognizing traveling mode,based on vehicle trajectory data,which includes the average velocity and variance of the velocity data collected by the our smartphone APP.In order to improve the classification accuracy of buses and cars,car ownership of the residents being surveyed is also used as an input of the model.Study results show that the recognition accuracy of the fuzzy decision tree model is 84%.The extraction methods proposed in this thesis can be used to obtain tripinformation of residents,including activity locations,traveling mode,and trip purpose.Although traveling mode and trip purpose are extracted with a relatively high precision,the classification of traveling mode is not as accurate due to data limitations.Despite this,the fuzzy decision tree model proposed is extensible,and additional input data can be used to further train the model and to improve the accuracy of the study results.The work discussed in this thesis is a part of the broader work for developing a smartphone App-based travel survey platform,which provides a way to enhance the efficiency of collecting and processing travel behavior data,to reduce the workload of travel surveys,and improve the accuracy of traffic demand forecasting models.