Object-adaptive tracking for AR guidance system

Abstract

This paper proposes a model-based object-adaptive tracking method which uses both edges and feature points as vision cues and flexibly adjusts the contribution of each vision cue using a single parameter based on the characteristics of tracking object and the initial conditions. It will be shown that, in many situations where conventional object tracking methods do not work, the proposed method provides reasonably good results. The proposed object-adaptive tracking method worked at 20 fps on UMPC with an average tracking error within 3 pixels when the camera image resolution is 640 by 480 pixels and this real-time capability enabled the proposed method to be successfully applied to an augmented reality (AR) guidance system for the National Science Museum.