Head Dynamic Analysis: A Multi-view Framework Ashish Tawari and Moham M. Trivedi University of California, San Diego, USA {atawari,mtrivedi}@ucsd.edu Abstract. Analysis of driver’s head behavior is an integral part of driver monitoring system. In particular, head pose and dynamics are strong indicators of driver’s focus of attention. In this paper, we present a dis- tributed camera framework for head pose estimation with emphasis on the ability to operate reliably and continuously. To evaluate the proposed framework, we collected a novel head pose dataset of naturalistic on-road driving in urban streets and freeways. As oppose to utilizing all the data collected during the whole ride where for large portion of the time driver is front facing, we use data during particular maneuvers typically involv- ing large head deviation from frontal pose. While this makes the dataset challenging, it provides an opportunity to evaluate algorithms during non-frontal glances which are of special interest to driver safety. We con- duct a comparative study between proposed multi-view based approach and single-view based approach. Our analyses show promising results. 1 Introduction Automatic analysis of driver behaviors is becoming an increasingly important aspect in the design of Driver Assistance System (DAS). With driver distraction and inattention being one of the prominent causes of automotive collision, we require new sensing approaches with ability to continuously infer driver’s focus of attention. Eye gaze and movement are considered good measures to iden- tify individual’s focus of attention. Vision based systems provide non-contact and non-invasive solution, and are commonly used for gaze tracking. However, such systems are highly susceptible to illumination changes, particularly, in real- world driving scenario. Eye-gaze tracking methods using corneal reflection with infrared illumination have been primarily used in indoor [5] but are vulnerable to sunlight. While precise gaze direction provides useful information, coarse gaze direction, approximated by head pose and its dynamics, are often sufficient [6,3]. Head pose is a strong indicator of a driver’s field-of-view and current focus of attention. It is intrinsically linked with visual gaze estimation, the ability to characterize the direction in which a person is looking. This paper presents an automatic head pose tracking system for uninterrupted driver monitoring using distributed cameras. The two main contributions of this paper are the design of the hardware setup and annotation strategy for ‘ground truth’ data collection, and the development A. Petrosino, L. Maddalena, P. Pala (Eds.): ICIAP 2013 Workshops, LNCS 8158, pp. 536–544, 2013. c  Springer-Verlag Berlin Heidelberg 2013