Online Adaptive Gaussian Mixture Learning for Video Applications

Abstract

This paper presents an online EM learning algorithm for training adaptive Gaussian mixtures for non-stationary video data. Existing solutions are either slow in learning or computationally and storage inefficient. Our solution is derived based on sufficient statistics of the short-term distribution. To avoid unnecessary computation or storage, we show that the equivalent estimates can be accomplished by a set of recursive parameter update equations with one additional variable. The solution is evaluated against several existing algorithms on both synthetic data and surveillance videos. The results showed remarkable learning efficiency and robustness over current solutions.

Cite

Text

Lee. "Online Adaptive Gaussian Mixture Learning for Video Applications." European Conference on Computer Vision, 2004. doi:10.1007/978-3-540-30212-4_10

Markdown

[Lee. "Online Adaptive Gaussian Mixture Learning for Video Applications." European Conference on Computer Vision, 2004.](https://mlanthology.org/eccv/2004/lee2004eccv-online/) doi:10.1007/978-3-540-30212-4_10

BibTeX

@inproceedings{lee2004eccv-online,
  title     = {{Online Adaptive Gaussian Mixture Learning for Video Applications}},
  author    = {Lee, Dar-Shyang},
  booktitle = {European Conference on Computer Vision},
  year      = {2004},
  pages     = {105-116},
  doi       = {10.1007/978-3-540-30212-4_10},
  url       = {https://mlanthology.org/eccv/2004/lee2004eccv-online/}
}