Online human action recognition framework

Authors:Řezníček Ivo, Hradiš Michal, Zemčík Pavel
Licence:no - free
Keywords:spacetime interest points, svm classifier, bag of words
The product is able to do detect human actions in video on the fly. It is based on online version of local spatio-temporal feature extractor for description of spatio-temporal interest regions. These local features are transformed by codeword transfomation into bag-of-words descriptors. The descriptors are constructed from the predefined number of input frames and are classified using arbitrary number of linear svm classifiers (one for each class to be detected). Classifier's output can be converted to decision, if the classified action is presented in an current part of a video or not. The product is modular. The feature extractor can be replaced by another type, the number of classifiers is not restricted, and even another type of classifiers can be used. All components, the codebook and all classifiers have to be prepared, trained and evaluated in another framework; here all of them are only used for online prediction.
Research groups:

Your IPv4 address:
Switch to IPv6 connection

DNSSEC [dnssec]