Zhejiang University
|
Yueran Liu
Zhejiang University
|
Zhejiang University
|
East China Normal University
|
|
|
|
Planar object tracking plays an important role in AI applications, such as robotics, visual servoing, and visual SLAM. Although the previous planar trackers work well in most scenarios, it is still a challenging task due to the rapid motion and large transformation between two consecutive frames. The essential reason behind this problem is that the condition number of such a non-linear system changes unstably when the searching range of the homography parameter space becomes larger. To this end, we propose a novel Homography Decomposition Networks (HDN) approach that drastically reduces and stabilizes the condition number by decomposing the homography transformation into two groups. Specifically, a similarity transformation estimator is designed to predict the first group robustly by a deep convolution equivariant network. By taking advantage of the scale and rotation estimation with high confidence, a residual transformation is estimated by a simple regression model. Furthermore, the proposed end-to-end network is trained in a semi-supervised fashion. Extensive experiments show that our proposed approach outperforms the state-of-the-art planar tracking methods at a large margin on the challenging POT, UCSB and POIC datasets. |
In this paper, we propose a novel Homography Decomposition Networks approach to planar object tracking in video sequences, which decomposes the homography transformation into two groups, including a similarity group and a residual group. By estimating the similarity group firstly, the condition number of the entire system reduces substantially. Inspired by warped convolution, we employ a rotation-scale invariant convolution operator to predict similarity robustly. Then, the second stage predicts the residual transformation through the semi-supervised regression, where the residual transformation is the residual group with the extra error from the first stage. |
|
You can download here in case the url is blocked. We present tracking results on several sequence in POT. To clearly observe the tracking performance, we replace the object with other image in a promotional video. |
@misc{zhan2021homography,
title={Homography Decomposition Networks for Planar Object Tracking}, author={Xinrui Zhan and Yueran Liu and Jianke Zhu and Yang Li}, year={2021}, eprint={2112.07909}, archivePrefix={arXiv}, primaryClass={cs.CV} } |
Acknowledgements |