deepGroup

Hierarchical deep neural network for activity analysis of surveillance videos at individual, group and overall level.

(THIS REPO IS NOT MAINTAINED. Please refer to https://github.com/albu5/deepGroupv2)

1. This is a research quality code, so don't expect it to work straightaway. However, I'll be delighted if someone uses my work. So you are encouraged to ask for help if you run into trouble raising an issue or contacting me via e-mail.

2. Please cite our work if you use codes or ideas from this repository. A bibtex style citation can be found in deepgroup-bibtex.bib

3. Both MATLAB and Python are used. Install appropriate python packages whenever needed (all required packages can be obtained from pip, conda etc).

4. Install keras https://keras.io/

5. Throughout this repository, collective activity dataset is referred to http://www-personal.umich.edu/~wgchoi/eccv12/wongun_eccv12.html Reference: A Unified Framework for Multi-Target Tracking and Collective Activity Recognition ECCV, 2012 W. Choi and S. Savarese

6. Group level annotations provided by Neha Bhargava are present in my_anno.zip

Overview

In total, we have 3 levels of hierarchy and 5 main stages of computation. We assume that individual tracklets are already known. The five stages are:

1. Orientation
2. Individual action
3. Group detection
4. Group activity
5. Scene activity

See individual folders for each stage.