International Journal of Computer Applications (0975 – 8887)
Volume 64– No.13, February 2013
31
Real-Time Activity Recognition Technique for
Surveillance Camera and Implementation on Digital
Signal Processor
R.Karthikeyan
Assistant Professor
Department of Electrical and
Electronics Engineering
Jaya Engineering College
P.Mahalakshmi
Assistant Professor
Department of Electronics and
Communication Engineering
Sakthi engineering college
N.Gowri Shankar
Assistant Professor
Department of Electronics and
Communication Engineering
Meenakshi Sundararajan
Engineering College
ABSTRACT
Intelligent Video Surveillance allows users to easily monitor
the secure areas with surveillance cameras, and thus
eliminating the need for manual work and saves the huge
monitoring costs. A novel method for object tracking, based
on image segmentation is proposed to automatically recognize
activities around restricted area to improve safety and security
of the servicing area by multiplexing hundreds of video
streams in real time. Key component for the proposed system
includes background learning and updating, foreground
segmentation, features extraction, and decision-making
process. The proposed method uses adaptive background
subtraction techniques to handle illumination changes to
improve the performance of the video surveillance and video-
enable operations. The algorithms are simulated using
MATLAB tool to verify its stability in various conditions by
giving various input video samples and its output is taken as
benchmark for real time implementation in DSP processor.
Keywords
Video Analytics, image segmentation, tracking, video
surveillance, Background Subtraction
1. INTRODUCTION
Manual analysis of video is labor-intensive, fatiguing, and
prone to errors [1]. Additionally, research indicates that there
are severe limitations in the ability of humans to monitor
simultaneous signals [2].The ability to quickly search large
volumes of existing video or monitor real-time footage will
provide dramatic capabilities to transit agencies. Some
drawbacks of video analytic systems are their vulnerability to
environmental variables, such as detrimental lighting
conditions and weather. These adverse conditions can trigger
false alarms, which may become a source of frustration for the
user. Conversely, a human analyst may use judgment and
training to determine if an alarm should be raised for a wider
range of scenarios. Video analytic algorithms often are
sensitive to parameters and initial calibration. Event detection
performance typically depends on this calibration process. It
is difficult to achieve a good balance between event detection
and false alarms [3].
The entire paper is organized as follows: Section 2 introduces
all the procedure about background subtraction and temporal
differencing method. Section 3 described an improved
proposed method. Section 4 presents the implementing model
and Section 5 presents experimental results of the Real-Time
Activity Recognition Technique .Section V concludes the
research work.
2. BACKGROUND SUBTRACTION AND
TEMPORAL DIFFERENCING
A popular object segmentation strategy is background
subtraction. Background subtraction compares an image with
an estimate of the image as if it contained no objects of
interest [4]. It extracts foreground objects from regions where
there is significant difference between the observed and the
estimated image. Common algorithms include methods by
Heikkila and Olli [5], Stauffer and Grimson (Adaptive
Gaussian Mixture Model or GMM) [6], Halevy [7], and
Cutler [8]. A detailed general survey of image change
algorithms can be found in Radke et al [9]. GMM is one of the
most commonly-used methods for background subtraction in
visual surveillance applications for fixed cameras. A mixture
of Gaussians is maintained for each pixel in the image. As
time passes, new pixel values update the mixture of Gaussians
using an online K-means approach. The estimation update is
used to account for illumination changes, slight sensor
movement, and noise [10]. Nevertheless, transit surveillance
researchers continue to emphasize the importance of robust
background subtraction methods [11] and online construction
and adaptive background models. A large number of recent
background subtraction methods improve on prior existing
methods by modeling the statistical behavior of a particular
domain or by using a combination of methods.
3. PROPOSED METHOD
The algorithms are first simulated using MATLAB tool to
verify its stability in various conditions by giving various
input video samples and its output is taken as benchmark for
real time implementation. The real time implementation of the
algorithms is done by using the DSP processor. Figure 1
shows the simulation flow of the proposed system where the
input for the method is given to the MATLAB tool and output
is viewed in the system.
Figure 2 shows the simulation flow diagram of the real time
implementation using DM355 Texas instrument development
board. The real-time input is given to the board by using
analog surveillance camera. To verify the stability of the
system the board is connected to the PC and the algorithm is
verified by using various input samples downloaded from the
internet under various illumination conditions.
(19 páginas)
Manymanuals.com
Manymanuals.de
Manymanuals.fr
Manymanuals.it
Manymanuals.pl
Manymanuals.cz
Manymanuals.es
Manymanuals-pt.com
Comentários a estes Manuais