Video dataset.
We have grouped all the test sequences into different complexity categories depending on two aspects:
People classification complexity, defined as the difficulty to classify moving and temporally stationary people in a scenario. It is related with the number of pedestrians, their velocity, partial occlusions and pose variations.
Background complexity, defined as the difficulty to extract the foreground due to the presence of edges, multiple textures, lighting changes, reflections, shadows and objects belonging to the background.
A description of complexity levels of the associated content is shown in Table:
Category |
Complexity |
Classification |
Background |
C1 |
Low |
Low |
C2 |
Medium |
Low |
C3 |
Medium |
Medium |
C4 |
High |
Low |
C5 |
High |
High |
Sequences have been extracted from public datasets related with the people detection/object classification task. They are:
1) PETS2006: 2 sequences. PETS_S1-S2.
2) WCAM: 2 sequences. WCAM_S1-S2.
3) VISOR: 5 sequences. VISOR_S1-S5.
4) CVSG: 18 sequences. CVSG_S1-S15.
5) The well known “hall monitor” sequence: 1 sequence. hall_monitor.
6) AVSS2007: 1 sequence. AVSS_S1.
7) TRECVID2008: 61 sequences. TRECVID_DEV08_S1-S61. (New content, March 2011)
Finally our video dataset stores 90 (29+61) video sequences manually annotated. (New content, March 2011)
|