Real-time Marker Pose Tracking Robust to Occlusion

Abstract

Augmented reality (AR) systems require that a camera pose should be estimated in order to render 3D virtual objects on an input real scene, accordingly many camera pose estimation methods have been proposed. Among others, marker-based ones are comparatively easy to implement and able to quickly calculate an accurate camera pose. They have been applied to various AR applications for this reason. Their weakness to occlusion, however, has been a well-known problem. This paper proposes an effective solution which uses contour tracking and line fitting. The proposed method tracks markers by detecting lines closest to sides of them detected from the previous frame. Therefore it can work robustly even when sides of markers are partially occluded.