Citation

Cheng SY, Trivedi MM. IEEE Pervasive Comput. 2006; 5(4): 28-37.

Affiliation

Computer Vision and Robotics Research Lab., Science and Eng. Research Facility, Univ. of California, San Diego, San Diego, CA 92037-0434

Copyright

(Copyright © 2006, IEEE (Institute of Electrical and Electronics Engineers))

DOI

unavailable

PMID

unavailable

Abstract

An experimental vehicle with cameras and sensors was equipped to capture the vehicle dynamics, view of the road, and driver's body pose, to make a predictive turn-assistance safety system a reality. The aim of the research was to find the extent of using the body-pose information to detect and predict driver activities. The detection performance of a two-class pattern classifier was analyzed using receiver-operator characteristic curves, which describe the ability of the classifier to suppress missed detections and false alarms. Vision-based body-pose recovery is challenging because of the wide range of configurations and appearances a human body can assume and its tendency to occlude itself in images. The overall awareness of the environment, vehicle, and driver in the driving scenario allows for intelligent intervention in mitigating a dangerous situation. Intelligent driver support systems addresses some of the challenging multidisciplinary research problems in pervasive computing.