ISSN 2348 – 9928
IJAICT Volume -1, Issue-1, May 2014 Doi:01.0401/ijaict.2014.01.27 Published Online 05 (05) 2014
© 2014 IJAICT (www.ijaict.com)
Corresponding Author: Ms. K. Pavithradevi, Sri Guru Institute of Technology, Coimbatore, Tamilnadu , India 140
DETECTION OF SUSPICIOUS ACTIVITIES IN PUBLIC AREAS USING
STAGED MATCHING TECHNIQUE
Ms. K. Pavithradevi
PG Scholar,
Department of Computer Science and Engineering,
SriGuru Institute of Technology,
Coimbatore, Tamilnadu, India
Ms. S. Aruljothi
Assistant Professor,
Department of Computer Science and Engineering,
SriGuru Institute of Technology,
Coimbatore, Tamilnadu, India
Abstract—Detection of suspicious activities of human crowd
scenes in public areas using video surveillance has attracted an
increasing level of care. A framework that contains video data
receives from a fixed color camera installed at a particular location.
The noise from video frames is removed by using Gaussian filtering
with color and gamma correction. The foreground blob is extracted
from video frames using background subtraction method. The
framework obtains 3-D object level information by detecting and
tracking persons and luggage in the scene. Using staged matching
technique, the detection of merging and splitting in occlusion. The
actions of public are identified and clustered in a crowd scene by
using an adjacency matrix-based clustering and support vector
machine. The features are extracted from the frames using Gabor
algorithm and histogram of gradient. To predict the behaviors of
human crowd based on the model and then detect if any anomalies
of human crowd present in the scene that is relevant to security in
public areas. The experimental results are to demonstrate the
outstanding performance by using extensive dataset, fast object
tracking, low computational complexity and effective in detecting
anomalous events for uncontrolled environment of surveillance
videos.
Keywords—Crowd behavior, suspicious activities, anomalous
events, adjacent matrix-based clustering, support vector machine.
I. INTRODUCTION
The activities of human crowd behavior using surveillance
videos is an vital issue for public security, as it allows
detection of both anomalies and abnormal in human crowd
behavior being important surveillance applications.
Anomalous behavior recognition and video understanding are
core components in video surveillance system. The detection
of changes in human crowd, behaviors and anomalies in
imagery and video is a problem in machine vision. Lately
there has been much effort to devise automated real time high
accuracy video surveillance systems.
This practice is almost witness in large public areas such as
metro station and airport. The purpose of this paper is to
identify the activities of behaviors, anomalous events and
suspicious behavior of human crowd in public areas. The
framework that processes raw video data receives from a fixed
color camera at a particular location. The preprocessing stage
is done to removes noise from video frame using Gaussian
filtering with color correction and gamma correction to
improve the quality of the image. The background subtraction
method is used to subtract the background in each video frame
and extract the foreground objects as blobs. The human crowd
areas in video frames are notified after the extraction of
foreground blobs. The blobs are extracted in foreground that
as automatically finds human crowd and single areas. The
clustered objects are obtained by background segmentation
into semantic entities in the scenes.
The clustered objects are separated by using adjacent matrix
based clustering. The action and various anomalous events are
detecting by using support vector machine. After that
individual objects are completely modeling and tracking. A
complete semantic based recognition that depends on object
tracking has been visualized and extensively investigated.
These objects are tracked by using particle filtering with color
histogram, spatiogram and structural similarity index measure.
The color objects are tracked in 2-D and classified as being
either animate (people) or inanimate (object) in human crowd
scene. These objects are modeled by using spatio spectral
algorithm to estimate the pixel color and halt update of
occlusion stage. The objects are matching with blob by
comparing of color histogram and histogram intersection with
threshold. The matched object and blobs are move to feature
calculation.
The occlusion of unmatched blob and objects are separated by
using staged matching technique potential occlusion method to
detect merges and splits. The unmatched blobs are processes
into new object and recover the objects. The feature
calculation is based on the histogram of gradient and gabor
algorithm to form in the order of historical sequence by using
recorded dataset and local features of human crowd scenes