Index of /~tboult/PETS05/secure

[ICO]NameLast modifiedSizeDescription

[PARENTDIR]Parent Directory  -  
[TXT]README.html2014-05-29 04:49 2.9K 
[DIR]zod1/2004-10-12 10:21 -  
[DIR]zod2/2004-10-31 15:55 -  
[DIR]zod3/2004-09-09 08:10 -  
[DIR]zod4/2004-09-09 08:10 -  
[DIR]zod5/2004-09-09 08:11 -  
[DIR]zod6/2004-11-05 08:24 -  
[DIR]zod7/2004-09-09 08:11 -  
[DIR]zod7a/2004-11-05 11:35 -  
[DIR]zod8/2004-09-09 08:12 -  
[DIR]zod9/2004-09-09 08:12 -  
[DIR]zod10/2004-09-09 08:08 -  

PETS2005 coastal surveillance datasets

PETS2005 coastal surveillance datasets
T. Boult

These directories contains jpegs of challenging detection/tracking scenes on water. A growing issue for surveillance is applications in areas with complex often moving backgrounds such as those here. For these scenes it is not sufficient to have simple "video motion detection", rather the vision system must analyzes the various motions and decide what is significant.

As they would be used in real settings most of these examples include panning (and one example includes zooming) of the camera. This means that background adaption-based systems need to detect the change and adapt quickly to the new positions. In large outdoor setting PT(Z) functionality is critical given the large region that must be monitored.

Most of the examples use thermal cameras, one is a low-light camera. Some of the examples (e.g. zod4) have relatively small targets. Each is its own README. The "easiest" scenes are probably zod9 (stationary camera and eventually a large vessel) and 2 (target starts large so initial detection is easier).

Each of The final "test" sequences will be very similar to one of these examples.

Each directory has its own readme, a tar-gziped (winzip readable) file called Full-XXX.tgz and each file available individually. (The .tgz can be read/expanded with tar or with winzip). The XML "ground truth" for those that were made public are XXXXXX.xml

Currently only zod2, zo6 and zod7 has ground truth others are still in progress. The tool for ground-truthing was provided by the EC Funded CAVIAR project/IST 2001 37540, found at URL: http://homepages.inf.ed.ac.uk/rbf/CAVIAR/ and should be acknowledged if you use the ground truth data.

You can see T. Boult's results at http://vast.uccs.edu/~tboult/MOVIES/zodaic-manatee-8-02.avi http://vast.uccs.edu/~tboult/MOVIES/zodaic-manatee-8-02.mpg (higher quality) In those vido black boxes are hypothesized potential locations, and yellow-red boxes show detections (red =high confidennce yellow low). You can see our detection/tracking results on our GUI at http://vast.uccs.edu/~tboult/MOVIES/zodiac-gui.avi Which was was reported in the paper @inproceedings{boult2003geo, title={Geo-spatial active visual surveillance on wireless networks}, author={Boult, Terrance E}, booktitle={Applied Imagery Pattern Recognition Workshop, 2003. Proceedings. 32nd}, pages={244--249}, year={2003}, organization={IEEE} })

These images are Copyright (2004) T. Boult and may be used for PETS2005 with acknowledgment. For other uses please contact tboult at vast.uccs.edu. For any publication using this please cite
T. Boult, Coastal Surveillance Datasets, Vision and Security Lab, U. Colorodo at Colorado Springs, www.vast.uccs.edu/~tboult/PETS2005, 2005.