Robust 3D object tracking using an elaborate motion model

This paper proposes a new method for robust 3D object tracking from a single RGB image when an object model is available. The proposed method is based on image alignment between consecutive frames over a 3D target object. Different from conventional methods that only rely on image intensity for the alignment, we model intensity variations using the surface normal of the object. From this model, we also define a new constraint for the pose estimation, leading to significant improvement in the tracking robustness. In experiments, we demonstrate the benefits of our method by evaluating it under challenging tracking conditions.