The Smartphone-Based Person Travel Survey System: Data Collection, Trip Extraction, and Travel Mode Detection

Travel data is vital for understanding individual travel behaviors and estimating travel demand. Compared with traditional travel surveys in which respondents were asked to recall their trips, smartphones can record GPS trajectories actively at a high level of accuracy in both space and time. However, how to identify trips and detect travel modes from raw GPS data in an efficient way is still challenging. To address these issues, we firstly target high-accuracy data collection through developing a smartphone-based person travel survey system which comprises of three components, i.e., data collector, data processor, and data validator. The data collector is a smartphone app for collecting GPS trajectories; the data processor is a server embedded with rule-based algorithms to extract travel-activity information; and the data validator is a webpage to present the self-extracted travel information for respondents’ validation. Secondly, we propose a hybrid random forest and merging algorithm to detect multiple travel modes in complex trips. The algorithm reaches an accuracy of 100% for simple trips and 95.7% for multimodal trips. These results indicate that the proposed hybrid random forest and merging algorithm can help identify travel modes of complex trips more accurately and efficiently.

I. Introduction

Travel data is crucial for understanding travel behaviors and has been broadly applied in the fields of urban transportation planning and management [1]. Traditional travel surveys collected travel data employing face-to-face interviews, mail-or-phone interviews, and computer-assisted interviews, with much labor and time consuming but low-quality data obtained [2], [3]. Thanks to the rapid development of Information and Communication Techniques (ICTs), massive sources of data (e.g., cellular data and vehicle sensor data) have been passively generated at low cost and commonly used in the research of travel behaviors [4]. However, these data have certain drawbacks such as sampling bias, coarse spatial resolution, and validation issues [5]. Currently, smartphones embedded with GPS modules and accelerometers have been applied in travel data collection [2], [6]. Despite recording personal daily trajectories at a high level of accuracy in temporal and spatial aspects, smartphones can not provide travel and activity information. Consequently, how to efficiently extract trips and detect travel modes from GPS trajectories becomes a great challenge in the applications of smartphone-based survey systems.