A one-day GPS trajectory is first split into trips and each of these
resulting trips is then divided into single-mode segments (hereafter
referred to as segments) for this detection process. The
method used for trip detection is described in one of our previous
studies (Xiao, Juan, & Gao, 2015). The basic ideas underlying this
study are the assumption that the trip ends take the form of either
point clustering or an abrupt direction change in the case of GPS
normal recording, while they present a dwell exceeding a given
critical duration during signal loss. The segments are extracted
from trips according to the method proposed by Gong et al.
(2012). By means of a prompted recall survey, the information is
corrected if necessary. There is no heavy burden on respondents
as expected, since most travel information can be correctly
detected. Such validated travel information is regarded as the
‘ground truth’ in the subsequent study. Each respondent is
required to participate in the survey for at least five continuous
days. Ultimately, 1636 person-day positioning data from 202
respondents was collected. After data screening to ensure completeness
and logical behavior, 1248 person-day data streams were
retained for subsequent analysis.