Real-time 3D pose estimation from monocular imagesequences is a challenging research topic. Althoughcurrent methods are able to recover 3D pose,they require a high computational cost to processhigh-resolution images in a video sequence at highframe-rates. To address that problem, we introducethe new concept of check-points. They are the minimumnumber of points needed to detect a 3D objectmotion. Our method tracks the 2D projections of thecheck-points over a 2D maximum pyramid. To handlelarge displacements of the object, our approachevaluates the projection of the check-points at highestlevels of the pyramid. Moreover, it refines thepose localization by utilizing the check-points at lowestlevels of the hierarchy. We show that just checkinga few cells per frame, our method estimates the3D pose of the tracked object. This early versionof the method works with a specific type of object a3D cube, with six well distinguished faces and whichsalient features in all the faces are dots, a die.


Torres Garcia, F., & Kropatsch, W. (2013). Top-down 3D Tracking and Pose Estimation of a Die Using Check-points. In W. Kropatsch, F. Torres Garcia, & G. Ramachandran (Eds.), Proceedings of the 18th Computer Vision Winter Workshop (pp. 102–109). Prip 186/3. http://hdl.handle.net/20.500.12708/54499