This page lists several sources of surveillance videos. The first set is associated with past or ongoing research efforts on the topic of video based surveillance, and usually involve specific content associated to some challenging aspect of video interpretation. When getting started, it is usually best to identify the simplest videos. These are usually videos with diffuse, consistent lighting (so, outdoors on an overcast day, or indoors but without glare on the floor), and possibly with few but frequently entering persons or objects. Later, more difficultones can include multiple persons, lots of occlusions, variable lighting conditions, etc.

3DPeS: Multiple videos from multiple views. Also has camera information and other additional meta-content if needed.

AVSS: Surveillance videos whose main characteristics include people leaving luggage or packages behind (left items was a hot topic back then), or cars parking.

CAVIAR: The classical surveillance video. Seen in many “early” publications on the topic. There are good starter videos in this batch.

Open ViSOR Videos: Collection of publicly available video to process for research purposes. Some are super short, some are long enough for purposes of this Learning Module. Requires spending a few minutes to find the ones that are right.

PETS: The Performance Evaluation of Tracking and Surveillance (PETS) videos are classics and well known. The research lab behind them has been running surveillance challenges for around a decade now, starting initially with good tracking, but moving on to action recognition and threat detection. The year that challenges occur, several video datasets are released. Sometimes video is re-used in later years for different challenge. Historical releases include PETS 2006, PETS 2009, PETS 2014, PETS 2015, PETS 2016. Older PETS data can be found here, but the documentation and reliability may not be as good. Some of the datasets are also good for starting, and also good for medium difficulty level.

Matlab has several videos available if the computer vision toolbox is installed (presumed to be the case here). They can be found in the directory:

MATLABROOT/toolbox/vision/visiondata

where MATLABROOT is the directory of your Matlab installation.

The files that are most pertinent to this module include:

1. atrium.avi
2. viptraffic.avi
3. visiontraffic.avi

Data sets, not so clear cut:

• i-LIDS (no longer available)
• TRECVID

Return