Combining Local Search, Neural Networks and Particle Filters to Achieve Fast and Reliable Contour Tracking
- Peter Torma
- Csaba Szepesvari, Department of Computing Science; PI of AICML
LS-N-IPS is an extension of the standard N-IPS particle filter (also known as CONDENSATION in the image processing literature). The modified algorithm adds local search to the baseline algorithm: in each time step the predictions are refined in a local search procedure that utilizes the most recent observation. A critical choice in the design of LS-N-IPS is the way the local search is implemented. Here, we introduce a method based on training artificial neural networks for implementing the local search. In experiments with real-life data (visual tracking) the method is shown to improve robustness and performance significantly, surpassing the performance of previous state-of-the-art algorithms.
Citation
P. Torma, C. Szepesvari. "Combining Local Search, Neural Networks and Particle Filters to Achieve Fast and Reliable Contour Tracking". IEEE, January 2003.Keywords: | machine learning |
Category: | In Conference |
BibTeX
@incollection{Torma+Szepesvari:IEEE03, author = {Peter Torma and Csaba Szepesvari}, title = {Combining Local Search, Neural Networks and Particle Filters to Achieve Fast and Reliable Contour Tracking}, booktitle = {}, year = 2003, }Last Updated: January 04, 2007
Submitted by William Thorne