Global Abnormal Behaviour Detection Using a Network of CCTV Cameras
Introduction
Multiple camera tracking has attracted much attention
recently however, not much work has been done on modelling global
behaviours of objects monitored by a network of CCTV cameras with
non-overlapping views and little effort has been taken to tackle the
challenging problem of detecting abnormal global behaviours, which are
only meaningful and recognisable when observed over space and time
across multiple camera views. We propose a novel framework, which
consists of object tracking across camera views, global behaviour
modelling based on unsupervised learning, and probabilistic abnormality
inference.The effectiveness of the framework is demonstrated with experiments
on real-world surveillance video data.
Our approach
Using the global trajectories obtained from the training phase, we
obtain similarities between global tracks using spatial and velocity
information and we use spectral clustering methods in order to group
global tracks into clusters. For each global cluster, we learn its
semantic scene model. This model can be viewed as a likelihood of a
point on a trajectory at a certain global position. Then, for each
global track in the testing phase, we can look at the ensamble of points
which make up its trajectory, find their corresponding likelihoods from
the semantic model for each cluster and determine if the trjectory is
likely to fit well with at least one global cluster. If this is the
case, then that global tracjectory is classified as being normal but if
the global track does not match well with any of the global clusters,
then it is classified as an abnormal track.
Results
The following two figure present the test trajectories in the global scenes. The green global test
trajectories in were detected as normal tracks while the red global test
trajectories were detected as anomalous tracks:
The following three figures show respectively: clustered global tracks from the office to the foyer camera, clustered global tracks from the foyer to the office camera and a schematic of the office and foyer cameras.
Demos
Foyer.avi
Office.avi
Publicatons
Emanuel E. Zelniker, Shaogang Gong, Tao Xiang, 'Global Abnormal
Detection Using a Network of CCTV Cameras', submitted to Visual
Surveillance 2008.
|
