Welcome to the BMTT-PETS 2017 Surveillance
Challenges!
During this workshop, PETS datasets will also bring the
opportunity to evaluate algorithms working on surveillance; on
the detection of events of interest. PETS datasets give the
opportunity to test detection of commonly researched events in
land case and maritime domains. We continue the evaluation theme
of on-board surveillance systems for protection of mobile
critical assets as set in PETS
2016. Two datasets are provided as in PETS 2016: (1) a
multi sensor dataset, as used for PETS2014 to PETS 2016, which
addresses protection of trucks (the ARENA Dataset); and (2) a
maritime dataset - the IPATCH Dataset - addressing the
application of multi sensor surveillance toprotect a vessel at
sea from piracy. When authors are not directly working on
tracking, the dataset includes tracking data that can be used
as input for algorithms working specifically on event
detection tasks of surveillance. You can find a
description of the PETS datasets here.
- The task is to detect any of the 'Atomic' or 'Complex' events in the Surveillance Challenges and report the frame at the start of the event and the frame at the end of the event.
- To download the video sequences for the surveillance challenges (and corresponding tracking results for authors not working on low-level video analysis), please fill out the following User Agreement. You will receive a link to download the dataset after the User Agreement has been registered.
- A paper detailing the employed methodology must be submitted on CMT following the paper format and length rules as the main CVPR conference. Please follow the paper preparation instructions.
- Deadline for submission on CMT is April 12th, 2017!
- All presented papers must submit the results of their
method in a zip file as supplementary material on CMT. The results must be formatted into
workshop-specific XML files: - Accepted papers will appear in the proceedings.
- Each accepted paper will have either an oral or a poster presentation.
- If you have problems or questions for these challenges you can send a message to our surveillance challenge chairs.
Welcome to the CVPR 2017 Atomic event detection Challenge!
PETS datasets include a series of abnormal events that
when detected should
allow for behaviour understanding. Recorded behaviours that can
be clear indications of abnormalities
are:
- person falling or pushed to ground: Person loosing balance and falling to ground. Can be caused by themselves or by a third person.
- person or skiff speeding up: Sudden acceleration of the mobile object.
- person or skiff (boat) loitering: Detected object stands/moves slowly in the same area.
- person or skiff group formation: A mobile comes close to another and hold an interaction.
- person or skiff group separation: A mobile departs from a group.
- person or skiff moving around the mobile asset: Mobile object moving and covering at least two sides of the vehicle or vessel.
- person or skiff suddenly changing direction: Mobile object sudden change of trajectory.
The task in this challenge is to detect any of the events in this category and report the frame at the start of the event and the frame at the end of the event.
Sequences that can be processed in this category are
those belonging to the following Table:
|
Sequences that can be processed for the 'Atomic event' detection challenge |
|
|
ARENA Dataset |
IPATCH Dataset |
|
|
|
|
|
|
(clicking on the sequence ID gives a summary video with all recording cameras of the acted scenario.)
Welcome to the CVPR 2017 Complex threat event detection Challenge!
PETS datasets include a series of ‘complex’ threat/criminal behaviours to be detected by a surveillance system.
Recorded ‘complex’ threat/criminal behaviours are:
- attack to person: Physical and intentional aggression to
driver where they are hit or menaced with an arm,
and possibly brought to the ground. - stealing from vehicle: Someone penetrates the vehicle completely or partially and departs with an object removed from the vehicle.
- attack to vessel: Abnormal approach to the vessel of a boat (possibly from a sudden change of direction and/or the boat speeding up) that concludes with the skiff staying at the vessel starboard.
The task in this challenge is to detect any of the events in this category and report the frame at the start of the event and the frame at the end of the event.
Sequences that can be processed in this category are those belonging to the following Table:
| Sequences that can be processed for the
'Complex threat event' detection challenge |
|
|
ARENA Dataset |
IPATCH Dataset |
|
|
|
(clicking on the sequence ID gives a summary video with all recording cameras of the acted scenario.)
