Citation

Hu J, Liu W, Kang J, Yang W, Zhao H. Proc. Inst. Mech. Eng. Pt. D J. Automobile Eng. 2019; 233(9): 2323-2332.

Copyright

(Copyright © 2019, Institution of Mechanical Engineers, Publisher SAGE Publishing)

DOI

10.1177/0954407019857408

PMID

unavailable

Abstract

A novel method for the eight most common driver's distraction actions recognition is presented in this paper. To this end, a semi-cascade network (SCN) with very lightweight architecture is designed. The approach recognizes the morphology of the human face and hands to make judgments about the driver's actions rather than just judging facial information. In order to subdivide similar actions, a SCN structure which effectively reduces the network's scale is employed. A joint training approach is proposed for training the network and achieving 95.61% accuracy. In addition, to verify the validity of the method, a dataset containing 100,000 samples is created. Finally, a warning strategy is provided for our system and 93.9% warning rate for the driver's distraction behavior is achieved.

Language: en