Direct Pose Estimation for Planar Objects


Po-Chen Wu1*
Hung-Yu Tseng1*
Ming-Hsuan Yang2
Shao-Yi Chien1

1 Media IC & System Lab, National Taiwan University
2 Vision and Learning Lab, University of California, Merced
(* indicates equal contribution)

Abstract

Estimating six degrees of freedom poses of a planar object from images is an important problem with numerous applications ranging from robotics to augmented reality. While the state-of-the-art Perspective-n-Point algorithms perform well in pose estimation, the success hinges on whether feature points can be extracted and matched correctly on target objects with rich texture. In this work, we propose a two-step robust direct method for six-dimensional pose estimation that performs accurately on both textured and textureless planar target objects. First, the pose of a planar target object with respect to a calibrated camera is approximately estimated by posing it as a template matching problem. Second, each object pose is refined and disambiguated using a dense alignment scheme. Extensive experiments on both synthetic and real datasets demonstrate that the proposed direct pose estimation algorithm performs favorably against state-of-the-art feature-based approaches in terms of robustness and accuracy under varying conditions. Furthermore, we show that the proposed dense alignment scheme can also be used for accurate pose tracking in video sequences.

Paper

Paper (20.7 MB)

Citation

Po-Chen Wu, Hung-Yu Tseng, Ming-Hsuan Yang, and Shao-Yi Chien, "Direct Pose Estimation for Planar Objects." Computer Vision and Image Understanding, 2018.

Bibtex

@article{DPE2018,
    author  = {Wu, Po-Chen and Tseng, Hung-Yu and Yang, Ming-Hsuan and Chien, Shao-Yi}, 
    title   = {Direct Pose Estimation for Planar Objects}, 
    journal = {Computer Vision and Image Understanding},
    year    = {2018}
}

Download

Code

Program Demonstration Video

Datasets

Synthetic Image Dataset Resolution: 800 ✕ 600
Focal Length   fx 1000.0
Focal Length   fy 1000.0
Principle Point   cx 400.5*
Principle Point   cy 300.5*
Templates
Datasets
Poses
README
*The principle points are used for 1-indexed programming languages (e.g., MATLAB). They should be shifted by -1 for 0-indexed programming languages (e.g., C++).

Notes

  • You can check the file size by moving your mouse over the corresponding download icon.

Results

• Synthetic Image Dataset

Cumulative percentage of poses whose rotation or translation errors are under thresholds specified in the x-axis over experiments on the proposed synthetic image dataset. There is a total of 8,400 poses estimated by each pose estimation approach.

• Visual Tracking Dataset

Cumulative percentage of poses whose rotation or translation errors are under thresholds specified in the x-axis over experiments on the visual tracking dataset. There is a total of 6,889 poses estimated by each pose estimation approach.
Demonstration video of results under the Unconstrained case with the Paris template. Poses are computed frame by frame without using temporal information.

• Object Pose Tracking Dataset

Cumulative percentage of poses whose rotation or translation errors are under thresholds specified in the x-axis over experiments on the object pose tracking dataset. There is a total of 20,988 poses estimated by each pose estimation approach.
Demonstration video of results under the Free Motion case with the City template. Poses are computed frame by frame without using temporal information.

• Textureless Planar Object Pose Tracking

References