Citation

Wahlstrom J, Skog I, Handel P, Bradley B, Madden S, Balakrishnan H. IEEE Trans. Intel. Transp. Syst. 2020; 21(2): 669-679.

Copyright

(Copyright © 2020, IEEE (Institute of Electrical and Electronics Engineers))

DOI

10.1109/TITS.2019.2896708

PMID

unavailable

Abstract

Smartphone-based driver monitoring is quickly gaining ground as a feasible alternative to competing in-vehicle and aftermarket solutions. Currently the main challenges for data analysts studying smartphone-based driving data stem from the mobility of the smartphone. In this paper, we use kernel-based k-means clustering to infer the placement of smartphones within vehicles. The trip segments are mapped into fifteen different placement clusters. As a part of the presented framework, we discuss practical considerations concerning e.g., trip segmentation, cluster initialization, and parameter selection. The proposed method is evaluated on more than 10 000 kilometers of driving data collected from approximately 200 drivers. To validate the interpretation of the clusters, we compare the data associated with different clusters and relate the results to real-world knowledge of driving behavior. The clusters associated with the label "Held by hand" are shown to display high gyroscope variances, low maximum speeds, low correlations between the measurements from smartphone-embedded and vehicle-fixed accelerometers, and short segment durations.

Language: en