19 September 2016 Driver head pose tracking with thermal camera
Author Affiliations +
Abstract
Head pose can be seen as a coarse estimation of gaze direction. In automotive industry, knowledge about gaze direction could optimize Human-Machine Interface (HMI) and Advanced Driver Assistance Systems (ADAS). Pose estimation systems are often based on camera when applications have to be contactless. In this paper, we explore uncooled thermal imagery (8-14μm) for its intrinsic night vision capabilities and for its invariance versus lighting variations. Two methods are implemented and compared, both are aided by a 3D model of the head. The 3D model, mapped with thermal texture, allows to synthesize a base of 2D projected models, differently oriented and labeled in yaw and pitch. The first method is based on keypoints. Keypoints of models are matched with those of the query image. These sets of matchings, aided with the 3D shape of the model, allow to estimate 3D pose. The second method is a global appearance approach. Among all 2D models of the base, algorithm searches the one which is the closest to the query image thanks to a weighted least squares difference.
Conference Presentation
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
S. Bole, S. Bole, C. Fournier, C. Fournier, C. Lavergne, C. Lavergne, G. Druart, G. Druart, T. Lépine, T. Lépine, } "Driver head pose tracking with thermal camera", Proc. SPIE 9974, Infrared Sensors, Devices, and Applications VI, 99740P (19 September 2016); doi: 10.1117/12.2237723; https://doi.org/10.1117/12.2237723
PROCEEDINGS
10 PAGES + PRESENTATION

SHARE
Back to Top