Low-Complexity Global Motion Estimation for Aerial Vehicles
Abstract
Global motion estimation (GME) algorithms are typically employed on aerial videos captured by on-board UAV cameras to compensate for the artificial motion induced in these video frames due to camera motion. However, existing methods for GME have high computational complexity and are therefore not suitable for on-board processing in UAVs with limited computing capabilities. In this paper, we propose a novel low complexity technique for GME that exploits the characteristics of aerial videos to only employ the minimum, yet, well-distributed features based on the scene complexity. Experiments performed on a mobile SoC platform, similar to the ones used in UAVs, confirm that the proposed technique achieves a speedup in execution time of over 40% without compromising the accuracy of the GME step when compared to a conventional method.
Cite
Text
Ramakrishnan et al. "Low-Complexity Global Motion Estimation for Aerial Vehicles." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2017. doi:10.1109/CVPRW.2017.55Markdown
[Ramakrishnan et al. "Low-Complexity Global Motion Estimation for Aerial Vehicles." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2017.](https://mlanthology.org/cvprw/2017/ramakrishnan2017cvprw-lowcomplexity/) doi:10.1109/CVPRW.2017.55BibTeX
@inproceedings{ramakrishnan2017cvprw-lowcomplexity,
title = {{Low-Complexity Global Motion Estimation for Aerial Vehicles}},
author = {Ramakrishnan, Nirmala and Prakash, Alok and Srikanthan, Thambipillai},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year = {2017},
pages = {402-410},
doi = {10.1109/CVPRW.2017.55},
url = {https://mlanthology.org/cvprw/2017/ramakrishnan2017cvprw-lowcomplexity/}
}