Trajectory-based anomalous event detection

(full paper available here).

The data sets for the main experiment are avaliable here. File names are in the format c<number of clusters>o<number of outliers>i<iteration>_{test|train}.mat. 10 iterations have been performed on each possible case, up to 10 clusters and 10 outliers max, for a total of 10x10x10 = 1000 train sets (plus the corresponding test sets). Each cluster contains 100 trajectories.
Datasets for comparison with discords are available here. Each dataset contains 260 trajectories: 250 normal ones, grouped in 5 clusters, and 10 anomalies.
All the data sets are in matlab ascii format, each row represents a 16-points trajectory in the form x_1, ..., x_16, y_1, ..., y_16. Last rows are anomalies.

The data sets have been created using our matlab trajectory generator.

Please acknowledge the source if you use these datasets in your works.