Juan Carlos San Miguel Avedillo and Jose
M. Martinez
email: {JuanCarlos.SanMiguel, JoseM.Martinez} at uam.es
Video Processing and Understanding Lab (http://www-vpu.ii.uam.es/)
Escuela Politecnica Superior (http://www.ii.uam.es/)
Universidad Autonoma de Madrid (http://www.uam.es/)
Abstract
Selected Test
Sequences
Results
(video clips)
In this work an approach for detecting video events in surveillance video is proposed. It is based on the use of feedback strategies between the signal processing stages of a typical video surveillance system to increase the performance of the event detection stage and to reduce the computational cost in easy/simple situations. Specifically, these feedback strategies are applied to change the output quality of the analysis stages as an increase/decrease in the level of detail of their analysis. Feedback information is based on performance measures computed in the different analysis stages of the system. Experimental results show that the proposed approach increases the detection reliability and reduces the computational cost as compared to the initially developed surveillance system across a variety of multiple video surveillance scenarios operating at real-time.
Experiments were carried out on several sequences from PETS2006 dataset [1], PETS2007 dataset [2], i-LIDS dataset for AVSS2007 [3], VISOR dataset [4], CANDELA dataset [5], WCAM dataset [6] and ChromaVSG dataset [7]. A more detailed description can be downloaded here.
Sequence | Cat. | Analysis Results | Improvement with feedback strategies | Scene Description | |||
No feedback strategies | Using Feedback strategies | Computation Cost Reduction | Score Improvement of True Positives | Rejection of False Positives | |||
Video Clip 1 |
1 |
Yes |
Yes | Yes | A single person leaves and object on the floor | ||
Video Clip 2 |
2 |
Yes | Yes | - | A single person enter in the scene and leaves a small object in the floor | ||
Video Clip 3 |
2 |
Yes | Yes | - |
A single person enter in the scene, picks an object and leaves the scene. |
||
Video Clip 4 |
2 |
Yes | Yes | No (due to their high initial probability) |
A single person enter in the scene, leaves an object and leaves the scene. |
||
Video Clip 5 |
3 |
Yes | Yes | Some |
Multiple people leaving and picking objects in the scene |
||
Video Clip 6 |
3 |
Yes | Yes | Some |
Multiple people leaving and picking objects in the scene |
||
Video Clip 7 | 4 | Yes | Yes | - |
Hall room: A single person enter in the scene, leaves an object and leaves the scene. Then another person enters and picks the object |
||
Video Clip 8 | 4 | Yes | Yes | Some |
Train station: A single person leaves an object. Then the object is abandoned. |
||
Video Clip 9 | 4 | Yes | - | Yes |
Train station: This scenario contains a person waiting for a train, the person temporarily places their briefcase on the ground before picking it up again and moving to a nearby shop. |
||
Video Clip 10 | 4 | Yes | - | Some | Train Station: This scenario contains a person waiting for a train, the person temporarily places their briefcase on the ground before picking it up again and moving to a nearby shop. | ||
Video Clip 11 | 4 | Yes | Yes | Some | Train station: This scenario contains a single person with ski equipment who loiters before abandoning the item of luggage. | ||
Video Clip 12 | 4 | Yes | Some | Some | Train station: This scenario contains two people who enter the scene together. One person places a rucksack on the ground, before both people leave together (without the rucksack). | ||
Video Clip 13 | 5 | Yes | Some | Some | Underground station: This crowded scenario contains a person that places an object, abandoning it few seconds later. |
References
[1] PETS2006,
Performance Evaluation of Tracking and Surveillance dataset, 2006
(available online at:
www.pets2006.org).
[2] PETS2007, Performance Evaluation of Tracking and Surveillance
dataset, 2007 (available online at:
www.pets2007.org).
[3] i-Lids dataset for the 4th IEEE Int. Conference on Advanced Video
and Signal Based Surveillance, 2007 (available online at:
http://www.elec.qmul.ac.uk/staffinfo/andrea/avss2007.html).
[4] VISOR, VIdeo Surveillance Online Repository, 2008 (available online
at
http://www.openvisor.org).
[5] The test visual material used in this publication has been provided
by the ITEA CANDELA project, (available online at:
http://www.multitel.be/~va/candela/).
[6] The test visual material used in this publication has been provided
with courtesy of Thales Security Systems within the scope of the IST
FP6 WCAM project, (available online at:
http://wcam.epfl.ch/).
[7] F. Tiburzi, M. Escudero, J. Bescós, and J. M.
Martínez, “A ground truth for motion-based
video-object segmentation,” in Proc. of IEEE Int. Conf. on
Image Processing (Workshop on Multimedia Information Retrieval), pp.
17–20, 2008 (dataset available at:
http://www-vpu.ii.uam.es/CVSG/.