|
Do you have an interest in automatic event detection systems? This is a challenge on a well-renowned dataset, with a twist: variable levels of compression have been applied, to understand how that degradation affects the performance of algorithms, and what measures can mitigate the problems. Researchers can share results in a special session of the ICDP conference. For this special session, an abstract will be due by 3rd September, with full papers submitted by 1st December.
|
|
Motivation
The video quality of surveillance systems can be compromised to facilitate e.g. long distance wireless transmission or low bit-rate storage. The extent to which this affects the effectiveness of video analytics systems is not fully known. You are invited to use the challenge datasets to investigate techniques to minimise the adverse effect of video compression on surveillance event detection systems. Papers are solicited that describe the techniques and experiments, to be presented in a special session of the ICDP conference, and included in the proceedings (indexed by INSPEC). Clich here to register
Examples of frames:
Methodology
The challenge dataset consists of approximately seven hours’ original recordings, split between two scenarios, and recoded at six operating points of bit-rate and frame-rate. If you have not already signed the i-LIDS end-user agreement, then you will need to agree to the conditions of use. You will then receive username and password so that you can download the files. Clich here to register
A simple program will be made available for you to derive the performance of methods (ICDP_stats_tool.zip), using the following evaluation metrics: precision, recall and subsequent f-measure. You can submit your (raw) results to the organizers, in the prescribed xml format (example.xml). The special session will include an announcement of overall results (with prizes) and time for comparative analysis.
Datasets
Click here to register and download datasets
It is proposed to use eight sequences, each encoded six times – a total of 48 files:
| Sterile Zone Scenario |
Parked Vehicle Scenario |
| filename |
duration |
# events |
filename |
duration |
# events |
| SZTEA101b |
49:46 |
15 (dusk) |
PVTEA102b |
42:07 |
17 (day) |
| SZTEA102a |
36:39 |
13 (day) |
PVTEA103a |
1:03:54 |
28 (day) |
| SZTEA104a |
1:31:18 |
31 (night) |
PVTEA104b |
31:04 |
10 (night) |
| SZTEA103a |
47:14 |
17 (day) |
PVTEA101a |
1:21:29 |
33 (day,rain) |
| totals |
3:44:57 |
76 |
totals |
3:38:34 |
88 |
Table 1: i-LIDS recordings for challenge
With the following encoding regimes:
| Name |
Descriptiono |
| 25fps low |
MPEG-4 (StreamClip encoder, low compression (1600 kbs) |
| 25fps medium |
MPEG-4 (StreamClip encoder, medium compression (400 kbs) |
| 25fps high |
MPEG-4 (StreamClip encoder, highly compressed (80 kbs) |
| 5 fps medium |
MPEG-4 (StreamClip encoder, medium compression (400 kbs) |
| 5 fps high |
MPEG-4 (StreamClip encoder, highly compressed (80 kbs) |
| 1 fps high |
MPEG-4 (StreamClip encoder, highly compressed (80 kbs) |
Table 2: Encoding operating points of i-LIDS recordings for challenge
For further Information please contact:
Dr James Orwell
Digital Imaging Research Centre
Faculty of SEC, Kingston University
Penrhyn Road
Kingston upon Thames
Surrey KT1 2EE
Email j.orwell@kingston.ac.uk
Tel +44 (0) 20 8417 2858
|
