Automatic segmentation and classification of movement trajectories for transportation modes

Abstract

The knowledge of the transportation mode used in a movement trajectory (derived in form of timestamped positions) is critical for applications such as travel behaviour studies. This thesis presents a method for segmenting movement data into single-mode segments and their classification with respect to the used transportation mode. The method relies on concepts found in expert systems, most notably membership functions, fuzzy logic, and certainty factors. A prototype, which may serve as a framework for managing travel behaviour surveys has been built in order to validate the presented theories and to classify the available test dataset. The transportation modes that this system classifies are walking, bicycle, tram, car, bus, train, underground, sailing boat, ferry, and aircraft. This research also investigates the performance of OpenStreetMap data in solving this problem. This free source of geodata proved to be crucial for the classification, where the ten transportation modes are discerned with various indicators mostly derived from the geodata, for instance, the proximity of the trajectory to the tram network and the information whether the movement has been made on a water surface or not. The classification relies on eliminating unlikely transportation modes by values set with a number of empirically derived fuzzy membership functions, and by using the selected combination of indicators it is possible to distinguish in between transportation modes which exhibit a similar behaviour (e.g. a car and bus in urban areas). Finally, the classification results are attached with a certainty value. The results are supplemented with additional mode-related information, e.g. the name of the departure train station. The segmentation has been done by detecting potential transition points between two transportation modes as brief stops. After each segment between consecutive potential transition points is classified, adjacent segments with the same classification outcome are merged (and removing the transition point in between), and keeping only the actual transition points where the transportation modes had been changed. The method solves the problem with noisy data, and traffic congestions which bias the indicators by using additional statistic values. The classification of gaps in the data (e.g. caused by a signal shortage during the logging of a trajectory) derived satisfying results, and segments with only their starting and ending point have been successfully classified. Thanks to the coverage of the OpenStreetMap data, in this research trajectories located outside the Netherlands (e.g. Norway and Denmark) have been segmented and classified as well. The accuracy of the classification with the developed prototype, determined with the comparison of the classified results with the reference data derived from manual classification, is 91.6 percent.