2D-3D association for model-based 3D tracking of rigid objects via association space restriction

Abstract

The three dimensional (3D) tracking of rigid objects is required in many applications such as 3D television (3DTV) and augmented reality. The availability of consecutive camera positions enables 3D scene reconstruction in 3DTV applications. On the other hand, in augmented reality applications, the knowledge of the pose between camera and scene (or object) reference frames enables the addition of virtual objects to the scene. The necessity of accurate and robust pose estimates for such applications, makes the 3D tracking topic a deeply research field in computer vision. Hence, in this paper, in order to enable 2D-3D association, which is one of the most crucial requirements for high quality model-based 3D tracking, a method which is based on restriction of the association space by utilization of projective invariant properties,is proposed. The reliability of the proposed method is proved by comparisons with RANSAC, perspective factorization and SOFTPOSIT based methods by using real and artificial data.