Micc People Counting RGB-D imagery for visual surveillance

We acquired a dataset of RGB-D imagery under typical conditions for visual surveillance applications in crowded environmnents. We recorded three video sequences that encompass three common people grouping and motion scenarios:

• In the FLOW sequence we asked the participants to walk straight from one point to another of the room. This sequence is useful to evaluate people counting systems in scenarios such as retail access or underground/train station pedestrian walkways.
• In the QUEUE sequence we asked participants to act as if waiting in line. People move slowly forward as people ahead are served. This sequence is challenging since pedestrians can be absorbed by the background while waiting and is useful to test the robustness of the background modeling under real conditions.
• Finally, in the GROUPS sequence we asked participants to split into two groups and talk to each other without exiting the controlled area. This sequence represents scenarios related to public safety in open and closed spaces.

In all of the three sequences people are highly occluded. Table reports some statistics on the sequences in our dataset.

If you use this dataset please cite our paper: Real-time people counting from depth imagery of crowded environments, Enrico Bondi, Lorenzo Seidenari, Andrew D. Bagdanov, Alberto Del Bimbo, Proc. of IEEE International Conference on Advanced Video and Signal based Surveillance (AVSS), 2014, Seoul, Korea.

@InProceedings{avss14,
author = “Bondi, Enrico and Seidenari, Lorenzo and Bagdanov, Andrew D. and Del Bimbo, Alberto”,
title = “Real-time people counting from depth imagery of crowded environments”,
booktitle = “Proc. of IEEE International Conference on Advanced Video and
Signal based Surveillance (AVSS)”,
year = “2014”,
note = “(Oral)”,
address = “Seoul, Korea”,
url = “http://www.micc.unifi.it/publications/2014/BSBD14/PID3277847.pdf”
}