An efficient pose recovery algorithm for vision based trajectory tracking applications

An accurate vision based trajectory tracking application requires accurate pose estimation of the camera (or the carrier vehicle) during the motion. Aside from the algorithmic and computational difficulties of pose estimation, challenges arise from the environmental changes or instantaneous imaging errors. Usually, in such cases, the true pose cannot be estimated. This causes the determination of incorrect motion directives to the robotic vehicle. To compensate this miscalculation a Pose Recovery algorithm is necessary to lead mobile vehicle to desired trajectory. This paper presents a new pose recovery algorithm that employs previously generated virtual poses of a reference object to correct miscalculated pose. In our experiments, we use the Mirage pose estimation to demonstrate the performance of the proposed recovery algorithm. We evaluated our algorithm on 3 different experiments using one reference object. According to the results, our pose recovery algorithm can be effective for majority of pose failure cases with very low positional errors.