D-PET: A Direct 6 DoF Pose Estimation and Tracking System on Graphics Processing Units
ISCAS 2017
 
Hung-Yu Tseng           Po-Chen Wu           Yu-Sheng Lin           Shao Yi Chien
 
Abstract
 
Real-time recovering an accurate 6 DoF pose of a known planar target is essential for augmented reality and robotics applications. Despite several pose estimation tracking systems have been proposed over recent years, there is still the need for a more efficient and more accurate solution for general planar objects. In this work, we develop a innovative GPU implementation of a real-time pose estimation and tracking system. It consists of a pose estimation unit and a pose tracker unit. While the former computes an initial pose of a target using direct method, the latter realizes accurate pose tracking with a hierarchical search scheme. Experiments on both synthetic and real datasets demonstrate that the proposed algorithm performs favorably with various planar targets. By implementing our method on an embedded GPU, the system achieves to work at 11 FPS and is suitable for real-time applications.

Downloads
 
Paper [pdf]
Poster [pdf]

Bibtex
 
@inproceedings{tsengdpe,
    title={D-PET: A Direct 6 DoF Pose Estimation and Tracking System on Graphics Processing Units},
    author={Tseng, Hung-Yu and Wu, Po-Chen and Lin, Yu-Sheng and Chien, Shao-Yi},
    booktitle = {Proc. IEEE International Symposium on Circuits and Systems},
    year = {2017}
}