Action Detection

What is this?

This repository if for classifying videos/sequences containing actions, and classifying them. This is written in Python 3.

Currently we utilize the KTH dataset, which can be found here: http://www.nada.kth.se/cvap/actions/along with that, for the interest points we utilize the Spatio-Temporal Interest Points as defined by Laptev et al., which can be found here: https://www.di.ens.fr/~laptev/download/stip-2.0-linux.zip

Note: Regarding STIP, to work correctly with the current OpenCV, a workaround has to be made to link the correct libs.

Installing

This module uses scikit-learn, which can be installed by either the

pip install -r requirements.txt

or

pip install sklearn

Executing

For now, to execute

python3 classify.py

Files

The main dataset working, and preprocessing of the descriptors to form a vocabulary is done by KTH.py. The classify.py calls it for accessing the processed dataset.

The KTH.pkl contains the processed descriptors along with the classifiers and the rest. The sequences.txt and sequences.pkl contains the sequences of the actions to be classified.