Real Time Detection of Human and Non-Human Parts From Video Surveillance

REAL TIME DETECTION OF HUMAN AND NON-HUMAN PARTS
FROM VIDEO SURVEILLANCE

MS. LAXMI TYAPI
PG Student of ECE Department, Don Bosco Institute of Technology, Bangalore, Karnataka, India
Email id : tclaxmi13@gmail.com
MRS. SOWMYA K S
Associate Professor of ECE Department, Don Bosco Institute of Technology, Bangalore,
Karnataka, India
Email id : sowmyaks@yahoo.com
ABSTRACT: One of the effective tool of the surveillance is the video surveillance. It enables a
security to monitor wide range areas. The video surveillance systems are used to track the moving
objects and also to represent their behaviour patterns. In general, surveillance is to make a unit of
three different phases i.e., Object segmentation, Object classification and Object tracking. In this
paper, mainly object classification method is used to differentiate the human and non-human part
in a video. This paper also explains the different background subtraction techniques and finally
concluded for improved performances. This will be implemented in Intel atom processor using
OpenCV.
KEYWORDS: Video surveillance, object segmentation, object classification, object tracking,

contour formation, background subtraction
INTRODUCTION
Video surveillance is the most effective means of ensuring security and safety. Due to the increase
in terrorism, violence and illegal activities there is a need for identification technique using video
surveillance. Detecting person in real time surveillance is a challenging problem due to variations
in pose, lighting conditions and complexity of backgrounds. Object segmentation is to change the
representation of an image. Object tracking is to estimate the trajectory of an object. This paper
deals with object classification rather than object segmentation or object tracking.
The proposed work of this paper is as shown in the figure 1. First the given video is converted into
several frames. As this paper deals with the real time implementation conversion of video into
frames is not shown separately. The background subtraction of these frames is done to extract
particular object or human for further processing. Next contour formation or edge detection
operation is to be performed on background subtracted image. For object classification, some
typical poses of human is stored as a database. Then the image obtained from contour formation is
compared with the database and most likely pose or human is detected. The detection of human
and non-human part is differentiated like by representing a human part in a red color rectangle and
non-human part in a green color rectangle.
The proposed work of this paper mainly consists of different modules like background subtraction
and contour formation as shown in the figure 1.
Figure 1. Proposed work flow
The organization of the remaining part of the paper is as follows: Section 2 gives a literature
survey on human detection techniques. Section 3 gives an existing methodology for detecting
human and non-human parts. Section 4 gives explanation about result analysis of the proposed
work. Section 5 concludes the paper. This paper deals with the real time implementation.
RELATED WORK
This section provides a review on different methods for human detection and background
subtraction. Changickkim and Wonjun Kim [1], explains about FCH (Fuzzy Colored Histogram)
background subtraction technique. Background subtraction is one of the popular approach to detect
moving objects. This technique minimizes the color variations in the image. But in case of real
time applications histograms are not suitable [1]. Jae Kyu Suhr and Ho Gi Jung [2], explains
another technique for background subtraction called MoG (Mixtures of Gaussians) mainly for
Bayer pattern image. This technique classifies the object nearer to the observer. The accuracy of
the MoG technique is same for both RGB and gray images [2].
The term contour refers to locate the edges of the image. In [3] Michael Maire explains about
contour formation. Contour formation is similar to edge detectors but edge detection near junctions
is not much accurate [3]. In case of video surveillance, object classification is very important.
Because a video contains several objects like animals, vehicles, humans etc. If object classification
is based on the feature then it distinguishes the moving object like human or vehicle. In [4],
Masamitsu Tsuchiya and Hironobu Fujiyoshi explains about object classification in video
surveillance. It explains method to evaluate various features relation importance for object
classification.
METHODOLOGY
The workflow of this paper consists of video as input as shown in figure 1. The first step is to
convert the video input into different frames. Frames are images obtained from the video. These
frames are processed through different modules like background subtraction, contour formation
and object classification. The following section describes about these modules.
87
Background Subtraction
Background subtraction is one of the widely used preprocessing technique used in many
applications. The main goal of this module is to extract an object from background in an image.
There are different background subtraction techniques such as Frame Differencing method, Mean
filter method, Running Gaussian average and Background Mixture Models etc. The simple and
easy method is frame differencing method as shown in the figure 2.
Background
Model
Video Input Comparison of
Background
Video frames
Extracting Difference subtracted
with Background
Video frames model of frames image
No motion detected
Figure 2. Block diagram of background subtraction module
This process has background model or one frame as a reference frame, then the difference between
the current frame and background frame is obtained. If the obtained difference is greater than the
threshold value then the object is detected as a foreground mask. This technique can be
formulated as,
| Framei -Framei-1 | > Threshold (1)
Advantages of this technique are easy to implement, more robust to noise and change in light
conditions.
Contour Formation
Object classification needs contour formation as a preprocessing step. Contour defines the
boundary values of an object in a frame or image. Hence it is used to extract the general shape of
the object. It is also called as boundary detection or edge detection.
Object Classification
Object classification is last module of the proposed work. It is used to classify the detected object
type such as human or non-human. In this paper, object classification will be implemented using
Silhouette based method i.e., few object silhouettes are stored as the database to classify the
objects. Silhouette is a dark outline of some object against brighter background. The output of
contour formation is of same kind. After comparison of contour with Silhouettes some threshold
value need to be set so that differentiation of human part (above threshold value) and non-human
part (below threshold value) can be done easily.
RESULTS
The input is taken as a real time i.e., video input is taken from webcam. It is converted to frames
and background subtraction is also performed as shown in the following figures.
88
Figure 3 shows the input video from the webcam. Figure 4 indicates the mean background image.
It is taken as reference frame for background subtraction.
Figure 3. Input Figure 4. Mean background Image
Figure 5 is the foreground image i.e., it recognizes the object or human nearer to observer. It is
background subtracted image but not that clear. Figure 6 indicates the exact background subtracted
image.
Figure 5. Foreground Image Figure 6. Background Subtracted Image
CONCLUSION
The main goal of the proposed system was to classify the human and non-human parts in a real
time video. Frame differencing method is used to implement background subtraction. Contour
formation can be done using either canny edge detection or Moores algorithm. Once the
89
classification of human and non-human parts is done then the future work includes the identifying
or tracking the behavior of the human.
REFERENCES
ChangickKim, Wonjun Kim Background Subtraction for Dynamic Texture Scenes Using Fuzzy
Colour Histograms March 2012, IEEE Signal Processing Letters, VOL. 19, NO. 3.
Gen Li, Jae Kyu Suhr, Jaihie Kim and Ho Gi Mixture of Gaussians-Based Background
Subtraction for Bayer-Pattern Image Sequences March 2011, IEEE Transactions on Circuits
And Systems for Video Technology, Vol. 21, NO. 3, pp.365-370.
Charless Fowlkes, Michael Maire, Jitendra Malik and Pablo Arbelaez Using Contours to Detect
and Localize Junctions in Natural Images
Hironobu Fujiyoshi and Masamitsu Tsuchiya Evaluating Feature Importance for Object
Classification in Visual Surveillance Proc. Of the Ninth IEEE International Conference on
Computer Vision, Vol. 2, 2003.
B. Ku, H. Ko, S. Kang, H. Kim and D.K. Han Adaptive selection of model histograms in block-
based background subtraction 12th April 2012, Electronics Letters Vol. 48 No. 8.
Massimo Piccardi Background subtraction techniques: a review IEEE International Conference
on Systems, 2004.
90

Real Time Detection of Human and Non-Human Parts From Video Surveillance

Uploaded by

Copyright:

Available Formats

Real Time Detection of Human and Non-Human Parts From Video Surveillance

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Real Time Detection of Human and Non-Human Parts From Video Surveillance

Uploaded by

Copyright:

Available Formats

REAL TIME DETECTION OF HUMAN AND NON-HUMAN PARTS

FROM VIDEO SURVEILLANCE

KEYWORDS: Video surveillance, object segmentation, object classification, object tracking,

Figure 2. Block diagram of background subtraction module

Figure 3. Input Figure 4. Mean background Image

Figure 5. Foreground Image Figure 6. Background Subtracted Image

You might also like