13 October 2017 Covariance analysis for evaluating head trackers
Author Affiliations +
Abstract
Existing methods for evaluating the performance of head trackers usually rely on publicly available face databases, which contain facial images and the ground truths of their corresponding head orientations. However, most of the existing publicly available face databases are constructed by assuming that a frontal head orientation can be determined by compelling the person under examination to look straight ahead at the camera on the first video frame. Since nobody can accurately direct one’s head toward the camera, this assumption may be unrealistic. Rather than obtaining estimation errors, we present a method for computing the covariance of estimation error rotations to evaluate the reliability of head trackers. As an uncertainty measure of estimators, the Schatten 2-norm of a square root of error covariance (or the algebraic average of relative error angles) can be used. The merit of the proposed method is that it does not disturb the person under examination by asking him to direct his head toward certain directions. Experimental results using real data validate the usefulness of our method.
© 2017 Society of Photo-Optical Instrumentation Engineers (SPIE) 0091-3286/2017/$25.00 © 2017 SPIE
Donghoon Kang "Covariance analysis for evaluating head trackers," Optical Engineering 56(10), 103105 (13 October 2017). https://doi.org/10.1117/1.OE.56.10.103105
Received: 20 July 2017; Accepted: 21 September 2017; Published: 13 October 2017
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Head

Cameras

Error analysis

Databases

Analytical research

Matrices

Optical engineering

RELATED CONTENT


Back to Top