Supporting large-scale travel surveys with smartphones – A practical approach

Abstract

Collection of travel data is a key task of transportation modeling. Data collection is currently based on costly and time-intensive questionnaires, and can thus only provide limited cross-sectional coverage and inadequate updates. There is an urgent need for technologically supported travel data acquisition tools. We present a novel approach for supporting travel surveys using data collected with smartphones. Individual trips of the person carrying the phone are automatically reconstructed and trip legs are classified into one of eight different modes of transport. This task is performed by an ensemble of probabilistic classifiers combined with a Discrete Hidden Markov Model (DHMM). Classification is based on features extracted from the motion trajectory recorded by the smartphone’s positioning system and signals of the embedded accelerometer. Our approach can cope with GPS signal losses by including positioning data obtained from the mobile phone cell network, and relies solely on accelerometer features when the trajectory cannot be reconstructed with sufficient accuracy. To train and evaluate the models, 355h of probe travel data were collected in the metropolitan area of Vienna, Austria by 15 volunteers over a period of 2months. Distinguishing eight different transportation modes, the classification results range from 65% (train, subway) to 95% (bicycle). The increasing popularity of smartphones gives the proposed method the potential to be used on a wide-spread basis and can complement existing travel survey methods.