Abstract
A novel method for extracting “trip periods,” i.e., periods in which a person travels, from continuously collected sensor data, called a “trip-extraction method” hereafter, is proposed to make a sensor-based travel-behavior survey possible. There are mainly two drawbacks in previous studies that detect “stay periods,” i.e., periods in which a person stays within an area, by using the boundary of a “stay area,” i.e., an area in which a person stays and then regard the rest of the periods as trip periods: false positives caused by GPS-positioning errors and false negatives caused by short-distance trips within the boundary. This study solves these problems by using novel features that are effective even in the case where the GPS-positioning error is large and by classifying every single piece of GPS data into either trip periods or stay periods not on the basis of the stay-area boundary but on the newly proposed features. An experimental evaluation showed that the precision of the proposed method was 89.4%, which is much higher than that of conventional methods.
