You are on page 1of 10

REAL TIME DETECTION

OF MOVING HUMAN
ABSTRACT:

Energy saving becomes a significant issue in recent years. There are two
conventional methods Of using the infrared ray sensor and CCD camera to detect human
movement. The method of using the infrared ray sensor, however, frequently causes
malfunction problems. On the other hand, the method which uses CCD camera requires
accompanying high price equipments. To save electric energy, this paper proposes an
algorithm to recognize human presence with USB Web camera. The proposed algorithm
detects human movement using the circle detection and morphological methods. Under
various experiments, the proposed method shows better performance in terms of the
recognition rate and speed.
1. INTRODUCTION

As the energy usage increases with the economic growth and human life
improvement, we are confronted with the depletion and environmental contamination of
resources. Korea imports most of energy from overseas, so it is inevitable to prepare
realistic plan for energy saving. Therefore, our government has been managing the
technology development programs such as tax reform, high efficiency resource
designation, and ESCO (Energy Service Company) projects to encourage energy
savings. Although the government has been striving for effective energy policy,
people generally lack in effectson energy saving. Specially wasting of energy frequently
happens in our everyday life. For example,the waste of energy in schools, general office
and public areas which are not related with private expenses is considerable. Some
business offices and schools are often exposed to leave lighting fixtures and air
conditioning even though the working day is over.
The objective of this paper is to reduce wastefull electric energy inside a building or
room. To save the energy, this proposed algorithm counts the number of person's
presence inside the building. If the number of recognized human is zero, then the
algorithm automatically turn off the lights. On the other hand, if human enters the
building, then automatically turn on the lights.
The research of human detection has been carried out for reducing the waste of electric
energy. There are two conventional methods using the infrared ray sensor and CCD
camera to detect human movement. The method of using the infrared rays sensor is
widely adopted. It senses the infrared ray from the human body and decides the presence
of human. This method can respond within 10~100 [msec] to human movement with high
response quality but it can not make continuous output signal because instantaneous
change of temperature is usually neutralized with the circumference charge. So it often
causes the malfunction of the electric devices.Recently the researches about detection of
the moving object with CCD camera based on computer vision technique have been
performed. But they usually focus on the unmanned observation for real time object
tracking, intelligence traffic system and autonomous robot mobile system rather than
human body detection.Also, they have the problem of requiring expensive image camera
for precise detection and expensive system for fast calculation speed of the image
processing .
On the other hand, we propose an efficient algorithm for human body detection
through USB web camera. The proposed method works with the pre-processed binary
image in a real-time. It recognizes the incoming and outgoing of the human by detecting
the head of the person through the circle detection. Under various
experiments, the proposed method shows better performance in terms of the recognition
rate and speed.
2.HUMAN DETECTION
ALGORITHM USING AN USB WEB CAMERA
2.1 System configuration
Digital images are captured by 1.3 Mega pixel CMOS web camera. This camera has
the specification of 1280x960 Max resolution, 30f/s Max frame, and USB 2.0 linkage to a
computer.
Fig.1 shows the configuration of the proposed system.It detects human body with real-
time image acquired by the web camera installed above an exit door. Fig. 2 shows the
flowchart of proposed algorithm. It counts the number of people in a designated space
after pre-processing of the real-time image from the camera.

Fig.1 System configuration


Fig 2. Flowchart of the proposed algorithm

2.2 Preprocessing process


The image acquired from the CMOS
camera requires preprocess to detect the
circular shape of human heads effectively.
First of all, acquired color image should be
converted to a simple gray image because it
needs a lot of time for color image processing.
At the next step, it will be enhanced against
lightning or background noises by median
filter and FFT filter. The median filter shown
in Eq.(1) is used to reduce noise by replacing
pixel data with middle brightness data around
pixel data.
f(x,y)=median(s,t)∈Sxy{g(s,t)} (1)
The FFT filter eliminates high frequency noises as a low pass filter. The image after
filtering process is converted to a binary image. Thresholding value is important for
binary image conversion as shown in Eq. (2). We decide a thresholding value to get
proper circular shape heads since we focus only on circular shape heads among human
body for human detection.
S (x, y) = { 1 if f(x,y) >=T
{0 if f(x,y)< T (2)
However, it is difficult to acquire an accurate circular shape of a head due to the effect of
illumination. It needs the Dilate and the Fill-Hole process with acquired binary image
.Specifically, it is converted to binary image with the critical value as shown in Eq.
(2).Although it is converted, circular shape of head area would be disconnected due to
reflection of illumination, so the disconnected pixel area needs to be expanded and
connected with the Dilate scheme. Finally, we can obtain circular shape of head by the
Fill-Hole peration. Mathematical formula of Dilate scheme is “f⊕b”.
The Fill-Hole operation fills the inside of hole connected pixels by Dilate using
surrounding pixel values. Fig. 3 shows the image from each pre-processing step. The
speed of image processing falls remarkably if camera resolution is high. The optimal
resolution was selected as 160x120 for correct detection and high speed calculation with
the resolution.

(a) Color image (b) Gray image

(c) After Median Filtering (d) After FFT & IFFT

(e) Binary image (f) Inverted binary image


(g) Dilate (h) Fill Hole
Fig. 3 Acquired images from pre-process

2.3 Circle detection


Several studies have been carried out to identify the human. This paper proposes an
efficient algorithm for human detection based on the image processing of detecting from
the image of web camera. In the case of the human image,a circular head image is
identified after several pre-processes as shown in Fig 3.
Various methods are used such as the smallest square law, clustering technique,
evolution algorithm and Hough transform to discriminate circular image. This paper uses
Heywood factor to discriminate the shape of an image. This method to detect a circle
measures maximum length and minimum length crossing center point of image, then
calculates circle aspect.According to the previous research, Heywood factors are shown
in Table 1 for various shapes shown in Fig 5. If the factor value is close to 1, the image
shape is close to a circle as shown in Table 1. Therefore the Heywood factor value is
chosen from the acquired image data by following pre-processing. After that, the
presence of a circle is decided with pre-defined critical value to detect the presence of a
human.

Fig. 4 Two lines crossing the center of an image

Fig. 5 Area with Heywood factor

Table1: Heywood factor according to the various shape


3. EXPERIMENT AND RESULT Aspect ratio Shape
In order to discriminate between the shape factor
incoming and outgoing of human, we set Theoretica measured error measured
two detection regions sectored with line 1 l
circle 1.000 0.97 3.0 0.01
and line 2 shown in Fig. 6. The detection
square 0.707 0.71 0.4 0.19
sequences of incoming consist of two
star N/A 0.60 N/A 0.26
steps. In the step 1, a circle is detected in B cross 0.447 0.48 7.4 0.34
region but no circle in A region. In the step explosion - 0.49 N/A 0.36
2, a circle is detected in A region but no
circle is B region. When an event satisfies above two steps, we can consider it as an
incoming.
Similarly, the reverse sequence can be considered as an outgoing event. These are
summarized in Table 2. Fig .7 and Fig. 8 shows the trajectory of human depicted with
circle for the cases of incoming and outgoing, respectively.
The test is carried out to observe the usefulness of the proposed method with 30 people
who income and outgo in the daytime. The test results show 100% recognition rate from
the image of entering and exiting. For the recognition rate of wearing a hat, the result
shows that the person wearing the black hat is recognized correctly but the person
wearing the white hat is not recognized properly.

Table 2 Incoming and outgoing logic

Incoming
Detection Step 1 Step 2
region
A region False true
B region True false
Outgoing
Detection Step 1 Step 2
region
A region True false
B region False true
Fig.6 Detection Area

Fig. 7 Circle Detection of Human Movement


(Incoming)

Fig. 8 Circle Detection of Human Movement


(Outgoing)

4. CONCLUSION
This paper proposes a method which detects human body and distinguishes the
incoming and outgoing of human movement by the circle detection method. In the
experiments, the proposed method shows excellent performance in terms of the
recognition rate and speed. For the further studies, we have a plan for developing more
advanced human detection method in the case where people wear hats and are exposed to
external illumination changes.

5. BIBILIOGRAPHY
[1] Kang-mo Lim, Joo-shin Lee, Moving Object Detection Algorithm for
Surveillance System, Journal of Korea Information and Communications
Society, Vol.30, No. 1C, pp.44-53, 2005.
[2] Jung-Hwan Ko, Sung-Il Kim, Eun-Soo Kim,Autonomous Mobile Robot
System`Using Adaptive Spatial Coordinates Detection Scheme
Based on Stereo Camera. Journal of Korea Information and Communications
Society, Vol. 31,
No. 1C, pp. 26-35, 2006.
.
Line 1
L

You might also like