Travel survey based on cellular phone data has been proved to be effective in extraction of travel characteristics of residents. Previous studies mostly use rule-based methods that have poor effect on outliers. Many of them also have not paid enough attention to the location oscillation phenomenon. Besides, due to the general lack of ground truth, the accuracy of these methods is difficult to assess. This paper proposes a three-step method for recognizing activity locations. Firstly, an equal-time interval interpolation method is used to balance the time weight of each trace. Secondly, an agglomerative hierarchical clustering algorithm is applied to merge the traces into different clusters. Finally, a new method of correcting location oscillation is proposed, to solve the problem that clusters in the same activity location oscillate. Meanwhile, this paper develops a new procedure to collect cellular phone dataset containing the ground truth information, so that the reliability of the proposed method can be evaluated. Results show that the method is good at identifying activity locations. The average recognition accuracy and distance error are over 85% and within 146 m, respectively. The average errors of departure and arrival time are 7.7 min and 5.3 min, respectively. We hope this study is useful to other scholars who are interested in algorithm evaluation and practical application of travel survey based on cellular phone data.

CICTP