Extraction of pressure features for predicting driver posture

Zhao, Mingming<sup>1,2</sup>; Beurier, Georges<sup>3</sup>; Wang, Hongyan<sup>4</sup>; Wang, Xuguang<sup>5</sup>

Affiliations

¹Biomechanics at Université Claude Bernard Lyon 1 (UCBL1), France

²Tongji University, Shanghai, China

³Laboratory of Biomechanics and Impact Mechanics, a mixed research unit of UCBL1 and Université Gustave Eiffel (formerly known as IFSTTAR), France

⁴School of Automotive Studies at Tongji University in Shanghai

⁵Université Gustave Eiffel at the Laboratory of Biomechanics and Impact Mechanics, France

Abstract and keywords

Postural information of drivers is helpful for specifying take-over control as well as for developing intelligent protection systems in autonomous vehicles. Thanks to the ease of use and the robust performance under varying environments, a pressure sensor could be a good alternative or complementary to cameras for monitoring drivers’ postures, especially for the trunk and feet. However, association between pressure distribution and posture is still unclear, effective methods for

ostural classification from pressure measurement therefore need to be developed. The objective of this study was to propose a method for extracting relevant features from the original pressure distribution data to predict drivers’ posture. First, a large number of pressure parameters, including contact area proportions, centres of pressure and pressure ratios, were defined for generating pressure features and the relevancy was analysed by performing the Out-Of-Bag feature importance evaluation of a Random Forest classifier. Finally, the relative value changes of 15 important parameters were used as features for training the classifier, leading to an average accuracy of 77.4% across nine posture classes and 23 drivers in the Leave-One-Out cross-validation tests. This study will provide valuable insight for extracting features in order to develop robust postural monitoring systems using pressure measurement.

Keywords: Driver posture monitoring; driving safety; machine learning; pressure features

Links

URL: http://www.ircobi.org/wordpress/downloads/irc20/pdf-files/52.pdf

Cited Works (5)

Year	Entry
2017	Jorlöv S, Bohman K, Larsson A. Seating positions and activities in highly automated cars: a qualitative study of future automated driving scenarios. In: Proceedings of the 2017 International IRCOBI Conference on the Biomechanics of Injury. September 13-15, 2017; Antwerp, Belgium.13-22.
2006	Dingus TA, Klauer SG, Neale VL, Petersen A, Lee SE, Sudweeks J, Perez MA, Hankey J, Ramsey D, Gupta S, Bucher C, Doerzaph ZR, Jermeland J, Knipling RR. The 100-Car Naturalistic Driving Study: Phase II – Results of the 100-Car Field Experiment. Washington, DC: National Highway Traffic Safety Administration; April 2006. Report No. DOT HS 810 593.
2019	Zhao M, Georges B, Wang H, Wang X. Detection of driver posture change by seat pressure measurement. In: Proceedings of the 2019 International IRCOBI Conference on the Biomechanics of Injury. September 11-13, 2019; Florence, Italy.84-85.
2018	Jin X, Hou H, Shen M, Wu H, Yang KH. Occupant kinematics and biomechanics with rotatable seat in autonomous vehicle collision: a preliminary concept and strategy. In: Proceedings of the 2018 International IRCOBI Conference on the Biomechanics of Injury. September 12-14, 2018; Athens, Greece.106-113.
2020	Zhao M, Beurier G, Wang H, Wang X. Driver posture prediction using pressure measurement and deep learning. In: Proceedings of the 2020 International IRCOBI Asia Conference on the Biomechanics of Injury. 2020; Beijing, China.102-105.