Title:	INTERACTION ERROR BASED VIEWPOINT ESTIMATION FOR CONTINUOUS PARALLAX ERROR CORRECTION ON INTERACTIVE SCREENS
Author(s):	Bastian Migge, Andreas Kunz, Tim Schmidt
ISBN:	978-972-8939-52-6
Editors:	Katherine Blashki
Year:	2011
Edition:	Single
Keywords:	User Modeling, Continuous Calibration, Parallax Distortion, Recursive Estimator
Type:	Full Paper
First Page:	93
Last Page:	104
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Many interactive screens suffer from the incoherence between image plane and interaction plane. The resulting gap causes parallax errors that hinder a precise interaction with the system. For many reasons, this gap cannot be physically reduced any further, while a software correction is still missing. Thus, this paper introduces an observation model for a continuous automatic recalibration controller of the touch sensitive surface. First, we show that the overall interaction error stems partly from the parallax error, which depends on the changing viewpoint of the user. Hence, a static calibration cannot overcome this error. Being not directly measurable, a continuously adapting correction controller sets the appropriate correction parameter based on updating the estimate of the user’s viewpoint from the history of his interaction errors. To estimate the user’s viewpoint in front of the screen based on the interaction error on the screen, we secondly investigate the correlation of the two domains from data of a user study, working on a large interactive screen with a significant gap between image plane and interaction plane. The correlation analysis shows significant differences in the interaction error stemming for different viewpoints, which allows the controller to infer the viewpoint. Finally, we model the results as a discrete observation model for the Partially Observable Markov Decision Processes (POMDP) correction controller.
	Go Back