For the compilation of this dataset, examples from abandoned objects have been selected from the publicly available PETS 2006 Dataset.
PETS2006 contains multiple instances of abandoned luggaged at a
public scenario, captured at different angles. Such events can be
considered as anomalous, and should be detected by alrgorithms that
perform anomaly detection in video. The abandoned objects present in
these sequences are of very small size relative to the average moving
object, which makes the correct detection more challenging. Training
data and test data have are provided for each selected sequence. For
training, the first frames of the video (~1000 frames) are used. The
training video contains only normal behavior in the scene. A small
portion of the video, containing an instance of the abandoned object
(anomaly) is used as test sequence. Table 1 contains a summary of frame
durations of each selected sequence.
|