Deriving Trip Information From GPS Trajectories

Urban and regional planning is mainly based on travel demand models which aim at estimating changes in transportation activity over time. Recently, activity-based models have attracted an increasing interest. This new paradigm is based on the bottom-up idea, and as a result, raises an urgent need for household/personal travel survey (HTS/PTS) data.Traditionally, HTS Data was collected using Face-to-Face home interviews or computer-assisted-telephone interviews (CATI), which need the respondents recall travel details such as trip origins, destinations, start time, end time, trip modes and trip purpose. Compared to trip-based model, activity-based model also needs trip routes. Respondents often forget or slip some trips, which leads to poor accuracy to travel models. At the same time, high burden is also responsible for a low overall response rate.GPS can record location and time, which give us an accurate and detailed trip trajectories. This new technology presents a revolutionary method of travel survey. However, it calls for automated identification trips from trajectories and derive trip modes and trip purposes.The main study contents of this research include:(1) An object-oriented trajectory segmentation method was proposed, which used a bottom-up procedure to merge each level of trajectory details to form a semantic trajectory with trip information.(2) Combined with machine learning methods, trip information was derived, which included stops, trip modes and trip purposes.(3) An experimental travel survey was conducted in East China Normal University, which use 13 vehicle respondents and 11 personal respondents. Their travel data was collected to do the trip information deriving research described above.(4) A software was also developed to assist the research, which can automated import raw GPS data to a simple database, identify stops and gave an interactive GUI for modify derived trips and construct sample for machine learning.The results indicated that passive GPS travel survey method has low burden and high accuracy. The procedure we proposed have done well in trip identification and trip mode classification, however the method to infer trip purpose need to be improved.