Top-Down Tracking and Estimating 3D Pose of a Die

Abstract

Real-time 3D pose estimation from monocular image sequences is a challenging research topic. Although current methods are able to recover 3D pose, they are severely challenged by the computational cost. To address this problem, we propose a tracking and 3D pose estimation method supported by three main pillars: a pyramidal structure, an aspect graph and the checkpoints. Once initialized the systems performs a top-down tracking. At a high level it detects the position of the object and segments its time-space trajectory. This stage increases the stability and the robustness for the tracking process. Our main objective is the 3D pose estimation, the pose is estimated only in relevant events of the segmented trajectory, which reduces the computational effort required. In order to obtain the 3D pose estimation in the complete trajectory, an interpolation method, based on the aspect graph describing the structure of the object´s surface, can be used to roughly estimate the poses between two relevant events. This early version of the method has been developed to work with a specific type of polyhedron with strong edges, texture and differentiated faces, a die.