A vehicle trajectory-based parking location recognition and inference method: Considering both travel action and intention

Abstract

Vehicle mobility impacts urban infrastructure planning, e.g., surging electric vehicle chargers in building parking lots. Existing methods for recognizing vehicle mobility patterns often ignore either drivers’ travel actions or intentions, thus misunderstanding their travel and parking behaviors. This study proposes a new Global Positioning System (GPS) trajectory-based machine learning method to reveal travel actions and intentions. It adopts DBSCAN clustering to recognize vehicles’ high-frequency access locations and then uses principal components analysis to infer building categories of their intended destinations. A dataset containing 10.37 million trips of 11,590 vehicles in Beijing was used to validate the method. Compared with conventional point-of-interest analysis, our method reveals a significant difference between drivers’ intended destinations and actual parking locations, resulting from the destinations with multiple functions or limited parking spaces. By travel intention analysis, 9,625 vehicles are identified as private vehicles, further distinguished into commuters (52.9 %) and non-commuters (47.1 %). The commuters have smaller travel ranges around their homes and workplaces ($\overline{R_g}$=11.0 km and $\overline{R_g^2}$=4.4 km), whereas the non-commuters have larger travel ranges and more chances of long trips ($\overline{R_g}$=12.7 km and $\overline{R_g^2}$=2.9 km). The proposed method extracts travel intention information from GPS trajectories, contributing to comprehensively understanding vehicle mobility and informing urban planning.