SAIVT-QUT Crowd Counting


This database contains three sequences annotated for crowd counting, captured on a university campus. Sequences are sparsely annotated (every 100 or 200 frames) over the approximately 5,000 or 10,000 frame sequence duration. Annotation is local, in that the location of each individual person in the frames is provided. Annotation is also provided for the PETS 2006 and 2009 datasets (although please note that you will need to download these databases separately). Contact Dr Simon Denman for more information.


The SAIVT-QUT Crowd Counting database is © 2012 QUT and is licensed under the Creative Commons Attribution-ShareAlike 3.0 Australia License.


To attribute this database, please include the following citation:

Ryan, David, Denman, Simon, Sridharan, Sridha, & Fookes, Clinton B. (2012) Scene invariant crowd counting and crowd occupancy analysis. In Video Analytics for Business Intelligence [Studies in Computational Intelligence, Volume 409]. Springer-Verlag, Germany, pp. 161-198. See the full paper on eprints.

Acknowledging the database in your publications

In addition to citing our paper, we kindly request that the following text be included in an acknowledgements section at the end of your publications:

We would like to thank the SAIVT Research Labs at Queensland University of Technology (QUT) for freely supplying us with the SAIVT-QUT Crowd Counting database for our research.

If you use the annotation data for either the PETS2006 or PETS2009 databases, you should also cite/acknowledge these databases in the appropriate method as outlined by the database creators, in addition to citing and acknowledging this database as outlined above and in the LICENCE.txt file.

Installing the database

Acknowledging the Database in your Publications Installing the SAIVT-QUT Crowd Counting database Download and unzip the following archive:  SAIVT_QUTCrowdCountingDatabase.tar.gz (933 MB). Upon downloading and unzipping the database, you should have the following directory structure:

+– LICENSE.txt
+– README.txt
+– Ryan2011 – Scene invariant crowd counting.pdf

+– Ryan2012 – Scene invariant crowd counting and crowd occupancy analysis.pdf

The database is located in the ‘Datasets/QUT’ subdirectory. Calibration and ground truth annotation is included within this directory, as well as a clean background image and region of interest for each of the three sequences. Ground truth annotation for the PETS2006 and PETS2009 databases are contained within the ‘Datasets/PETS2006’ and ‘Datasets/PETS2009’ subdirectories. A summary of this follows:

Video Calibration ROI Ground Truth (‘dot’ annotations) Initial Background Frame
PETS 2009 View 1 No (1) Yes Yes Yes Yes
 PETS 2009 View 2  No (1)  Yes  Yes  Yes  Yes
 PETS 2006 View 3  No (2)  Yes  Yes  Yes  No (3)
 PETS 2006 View 4  No (2)  Yes Yes  Yes  No (3)
 Camera A QUT  Yes Yes Yes  Yes  Yes
 Camera B QUT Yes Yes Yes Yes  Yes
 Camera C QUT Yes Yes Yes Yes  Yes