Semantic Video Content Analysis for Security
Department of Computer Science,
Queen Mary University of London
There is a huge demand for fully automated semantic video content
analysis due to massive increase of video media in the last
decade. However, there is also a lack of effective analytical tools to
extract automatically the most relevant information in context and in
good time, especially when dealing with CCTV video data of public
space. Significantly, human attention span usually lasts no more than
15-20 minutes resulting in highly inconsistent and error-prone manual
based content labelling and extraction of CCTV video. Furthermore, the
lack of any structured script or embedded meta-data in security and
surveillance video, as is present in most commercial and entertainment
video, makes the task of automated semantic content analysis of such
video data extremely difficult.
In this talk, I will present recent results on activity event and
behaviour based video content analysis of security and surveillance
video. I will highlight that some of the fundamental problems in
security video content analysis are more than merely object tracking
and trajectory matching. I will address the problem of modelling and
recognising complex activities involving simultaneous movement of
multiple overlapped objects. Dynamic probabilistic graph models are
exploited for modelling the temporal relationships among a set of
different object temporal events and are used to profile and index
salient event and behaviour patterns captured in CCTV video, and for
the detection of atypical and abnormal behaviours. I will also briefly
discuss the problem of extracting and synthesising high-resolution
image patches of saliency in low-resolution CCTV content under motion
blur, especially in the context of face recognition in low-resolution
CCTV video.
*
Shaogang Gong is Professor of Visual Computation at Queen Mary,
University of London, elected a Fellow of the Institution of
Electrical Engineers, a member of the UK Computing Research Committee,
and Head of Queen Mary Computer Vision Research Group he founded in
1993. He received his DPhil in computer vision from Oxford in 1989
with a thesis on the computation of optic flow using second-order
geometric analysis. He is a recipient of a Queen's Research Scientist
Award in 1987, a Royal Society Research Fellow in 1987 & 1988, and a
GEC-Oxford fellow in 1989. He twice won the Best Science Prize of the
British Machine Vision Conferences (1999 and 2001) and once won the
Best Paper Award (2001) of the IEEE International Workshops on
Recognition, Analysis and Tracking of Faces and Gestures. He is the
principal author of a book on Dynamic Vision: From Images to Face
Recognition (Imperial College Press, 2000). His work focuses on visual
motion and video analysis with applications to the detection, tracking
and recognition of vehicles and human objects, activity profiling,
behaviour recognition and abnormality detection in CCTV & live
video. A current significant focus is in security for crime prevention
and detection funded by the MOD, EPSRC, DTI and industry.
