Real-Time Stereo by Using Dynamic Programming

Abstract

A Dynamic Programming (DP) based algorithm that achieves real-time, high quality stereo-matching is presented. A special coarse to fine approach as well as the MMX based assembler implementation mainly contribute to the reached computation speed. A novel vertical smoothing approach by re-integrating paths inside the DP matching allows the avoidance of noisy horizontal strokes, so that high quality stereo-matching is achieved. The current implementation capable of running at about 30 FPS on an 2.2GHz PC, which is sufficient for the utilization in real-time applications.

Cite

Text

Forstmann et al. "Real-Time Stereo by Using Dynamic Programming." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2004. doi:10.1109/CVPR.2004.428

Markdown

[Forstmann et al. "Real-Time Stereo by Using Dynamic Programming." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2004.](https://mlanthology.org/cvprw/2004/forstmann2004cvprw-realtime/) doi:10.1109/CVPR.2004.428

BibTeX

@inproceedings{forstmann2004cvprw-realtime,
  title     = {{Real-Time Stereo by Using Dynamic Programming}},
  author    = {Forstmann, Sven and Kanou, Yutaka and Ohya, Jun and Thüring, Sven and Schmitt, Alfred},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year      = {2004},
  pages     = {29},
  doi       = {10.1109/CVPR.2004.428},
  url       = {https://mlanthology.org/cvprw/2004/forstmann2004cvprw-realtime/}
}