Hi - I'd like to introduce ground-truth labeled video sequences with interacting groups:

             http://groups.inf.ed.ac.uk/vision/BEHAVEDATA/INTERACTIONS/

The dataset has various scenario's of people acting out 

10 types of group interactions: InGroup, Approach, WalkTogether, Split,

Ignore, Following, Chase, Fight, RunTogether and Meet. The data 

includes individual interacting in groups and groups interacting. 

The data is captured at 25 frames per second. The resolution is 

640x480. The videos are available either as AVI's or as single JPEG files.

About 70K frames of the video sequences have ground truth bounding boxes 

of the people so far, in the VIPER XML format.

There is data for a ground plane homography.

Can I also mention other sites that might interest you or your new students:

             the CAVIAR ground-truthed video sequence data

                          http://groups.inf.ed.ac.uk/vision/CAVIAR/CAVIARDATA1/

             BEHAVE's crowd sequence data

                          http://groups.inf.ed.ac.uk/vision/BEHAVEDATA/CROWDS/

             CVonline (online vision "encyclopedia") now extended with

             subtrees on visual neurophysiology and psychophysics, and a

             list of imaging/vision related books:

                          http://homepages.inf.ed.ac.uk/rbf/CVonline/

             HIPR2 - JAVA-based interactive image processing tutorials

             including a pull-down plug-together workspace

                          http://homepages.inf.ed.ac.uk/rbf/HIPR2/

Enjoy, Bob

The development of the BEHAVE data was funded by the

UK's EPSRC in the BEHAVE project.

To the editor...

This letter was received from Bob Fisher (University of Edinburgh).  Note that Bob Fisher wrote an article about CVOnline (mentioned below) for the April 2005 issue of the IAPR Newsletter.

~ L. O’Gorman, ed.