International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________
Algorithms of Distance, color & Shape Detection for 2-D Images

1
V.K.PATIL 2Gajanan Alaspure 3Trupti Jakkulwar 4Ganesh Pawar, 5Aishwarya Bajoriya, 6Sampada Bulbule
1
Asst.Prof,Department of E&TC AISSMSs IOIT,Pune, Maharashtra, India
2,3,4,5,6
Students,Department of E&TC AISSMSs IOIT,Pune, Maharashtra, India

AbstractThis paper represents our work on non contact distance calculation method. It is based on pixel variation of 2D images.
This can be done by referencing two points in the image frames. The methodology of distance calculation starts with capturing of
particular object image from any distances. Again capture another image of same object with some known distance. If an object
goes far from the camera the object has a reduced size in the image. As the object moves closer to the camera the object becomes
comparatively bigger. Thus we get the variation of pixel size between two images. Using the pixel variation we can calculate the
distance of the object from the camera .This paper focuses on recognition of 2D shapes of objects such as square, circle, triangle,
rectangle and ellipse as well as colors of objects. The algorithm was simulated and developed using MATLAB.

Keywords2D Images, Image processing, MATLAB, Extent, Bounding Box, Object detection, Digital Camera, Shape and color
recognition, non contact distance measurement.
__________________________________________________*****_________________________________________________

I. Introduction general observation is that if an object goes far from the

Ultrasonic based and laser based techniques are most camera the object has a reduced size in the image. As the
commonly used methods for non-contact distance object moves closer to the camera the object becomes
measurement [1]-[2]. Difficulties in using these techniques comparatively bigger [2].From this we get the variation of
are that the accuracy of these techniques heavily depends on pixel size between two images.
the surface reflectivity of the object whose distance is to be This paper also includes shape detection and color detection.
calculated. These create problems for the object having poor For shape detection bounding box is used. Depending upon
surface reflectivity such as objects made up of woods etc. the extent which is simply a ratio of area of an object to the
Hence it causes in loss of accuracy for that object. area of a bounding box, shape is easily get detected. For
Recording an image of the particular object during color detection, color of an object gets compared with the
measurement is also an issue for the above mentioned standard colors.
techniques [1]-[2]. The remaining part of this paper is discussed as follows.
Alternately image based techniques have been proposed for Section II contains relationship between pixel variation and
distance measurement purpose that used digital camera also distance of the image. Section III involves the brief view of
called as CCD camera. This technique however requires two our proposed method. Calculations are derived in section IV.
cameras at different positions for capturing an image of the Section V describes all the applications related to our topic
object under test. It requires pattern recognition or image i.e. distance calculation and shape detection etc. Section VI
analysis of whole image frame for the sake of extracting contains experimental results. Lastly, conclusion is in
features from the images for obtaining distance of that section VII.
object from the digital camera. Therefore it requires large
amount of storage capacity and high speed DSP processors. II. RELATIONSHIP BETWEEN PIXEL VARIATION COUNT AND
So the disadvantages by using these methods are processing DISTANCE OF THE IMAGE
speed, establishment cost and complexity [1]. From the
above mentioned techniques, we can conclude that image We are assuming that surfaces or objects are perpendicular to
analysis was not suitable with real time measurements. the optical axis of digital camera. In the previous researches,
Because of the problems and difficulties in the above it was proved that, on image frame scan line distance and
mentioned methods, accurate and reliable measurements pixel counts are closely related [3]-[4]. Computing actual
were not always guaranteed in real world applications [2]. distance of object decomposed into computing the depth and
lateral distance of the object. It has been seen that object
Corresponding author height is closely related to the pixel height of the object [2].
1
[email protected] The Figure 1 shows that a single digital camera captures the
To overcome the problems and difficulties encountered via image of the same object twice. Let us assume that it capture
existing image-based distance measuring methods, this the first image at a photographic distance h1. Now
paper presents distance measurement method based on pixel repositionised the camera and again capture another image of
number variation in images of digital cameras by the same object i.e. from another distance h2. Let hs be the
referencing to two arbitrarily designated points in image scan line i.e .horizontal scan line. We are assuming that
frame rather than laser projected spots in an image [2]. A distance L between reference points will remains fixed [2].

72
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________

Fig.1. Schematic diagram of distance calculation of object from camera using pixel variation

So the relation between distances and pixel count can be given as:

1
= (1)
1

2
= (2)
2

(1) 2
= (3)
2 (1)

Because there is a displacement, h=h1-h2,i.e. from the camera movement along the photographic directions [2]. Hence we can
write as:

(1) 2 1+
= = (4)
2 (1) 2+

But h1=h2+h. putting this value in equation 4 as:

1
2 = (5)
2 (1)

In a Similar way, if camera moves in forward direction then photographic distance becomes:

2
1 = (6)
2 (1)

In the above equation, value of h- hs is known to us so that the distance is calculated [2].

73
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________
III. ANGLE AND DISTANCE MEASUREMENT FOR THE method. Here we include angle measurement for the objects
OBJECT having oblique plane,. The accuracy using angle
measurement will also be greater.
Distance measurement by using the above mentioned
technique is only applicable for the objects that are exactly Fig.2 shows the schematic of measurement of object distance
perpendicular to the optical origin of the camera. It is on an oblique plane based on pixel variation.
possible to calculate distance for the objects which are
perfectly perpendicular by using the above mentioned

Fig.2. Schematic diagram of distance calculation of object on oblique plane

As shown above there exists an angle i.e. m between the perpendicular and oblique plane. In the figure, point B is much far as
compare to point A. Hence it is not possible to calculate distance by using above mentioned method [2].
To solve the above mentioned method, concept of virtual planes used. Consider the two virtual plane VP_A and VP_B [2].
Therefore the distance between virtual planes and camera can be given by:

2
1 = (7)
1 2

1
2 = (8)
1 2

2
2 = (9)
1 2

74
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________
1
2 = (10)
1 2

Here h1(B) and h1(A) are respectively the distances from camera 1 to virtual planes VP_B and VP_A [2]. Similarly h2(B) and
h2(A) are respectively the distances from camera 2 to virtual planes VP_B and VP_A. From (7) and (10),

1 1 = 2 2
= + . (11)

= tan 1 (12)
+

In the above equation k is nothing but the distance between virtual points .dA and dB can be expressed as follows:

21 2
= (13)
2 1

21 2
= (14)
2 1

The distance between points A and B can be represented to examine the position of an real world object. This distance can be
represented as D(AB):

= + (15)

Thus using the above equation it is possible to calculate the distance of an oblique plane.

IV. SHAPE DETECTION OF THE OBJECT

The process of shape detection starts by capturing the image

from any distance with the help of digital camera. We shall
consider that the acquired image is in RGB format which is
true color format for an image [5]. The next step is to convert
RGB image into the black and white image. To convert the
image into black and white it should be first converted into
2D gray scale image. The gray scale image is nothing but
matrix that holds luminance value of the image [5]. We
obtain the luminance values by combining RGB values using
NTSC standard equation given by:
Fig.3: Bounding box
Y=0.11B+0.59G+0.3R
(7) The dimensions of the bounding box are variable. The
dimensions i.e. width and height are adjusted according to
This image is then converted into black & white image by the the shape of the object [5]. If the object has square shape then
process called thresholding [5]. The threshold is set and the there exists a bounding box of equal height and width. But if
luminance of each pixel is compared. The values greater than the object is rectangular shaped then exists a bounding box
threshold are replaced with logical one i.e. white and the having width greater than height. After forming the bounding
values lesser than one are replaced with logical zero. The box next step is to find an extent for the object which is
image is now two dimensional arrays with binary elements. nothing but ratio of area of an object to the area of the
Boundaries of an object are recognized by first setting a bounding box [5].
single pixel on object background as starting point and Following flowchart shows for shape detection and color
moving on clockwise or anti-clockwise direction searching detection [5].
for the other object pixel [5]. The value of this extent is different for different shapes. Its
In this way the objects boundary can be found. After finding value is 1.000 for square and rectangle and less than 1 for
the boundary area can be easily calculated by summing the circles.
pixels within the boundary extent. Next step is to find the Hence the conclusion is made from this is that if the value of
bounding box of an object. The bounding box is an extent is much less than 1.000 then it is either circle or
imaginary rectangle that completely encloses the object. ellipse. The difference between circle and ellipse is that the
Figure 3 shows the concept of bounding box [5]. eccentricity of circle is zero and for ellipse it is non zero.
Hence if there is a non-zero value of eccentricity then it must
75
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________
be ellipse and if not then it must be a circle. But if the extent V. COLOR DETECTION OF THE OBJECT
is equal to 1.000 then it is either square or rectangle. The
decision of whether the given object is square or rectangle is The first step in determining the object color is to initialize
taken on the basis of object dimension i.e. the object with all the necessary colors i.e. white, black, blue, red and green
equal sides is square and with different sides is rectangle. For in MATLAB. These colors are initialized in terms of their R,
triangles, we know that the area taken by the triangle in the G and B values. After initializing the color it will first check
bounding box is much less. Hence if the extent of the object for the color of centre pixel of an object. Hence it is
is less than 0.66 then it must be a triangle. So in this way we necessary to have the object color same as that of centre
can differentiate between the above mentioned objects. color. For this there is a function available in the MATLAB.
After checking the centre pixel color it will compare that
color values with the above initialized color. Resultant values
are again checked and the color that has same values will be
the color of an object. Hence by using the concept of the
additive mixing number of colors can be found.

Observation Table

Object
Shape Camera
movement Actual Observed
Color distance Distance Distance
Triangle Orange 10 46 48.21
Ellipse Orange 5 56 61.211
Rectangle Yellow 15 60 62.109
Square Blue 8 40 41.339
Circle Green 10 46 47.465

Fig.5: All Shapes used for Experimental Work

Fig.4: Flowchart for shape detection and color detection Fig.6: Setup

76
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 4 Issue: 5 72 - 77
_______________________________________________________________________________________________
ACKNOWLEDGMENT [2] Chen-Chien James Hsu, Ming-Chin Lu, Yin-Lu yu on
Distance measurement based on pixel variation of CCD
Our work is done under the guidance of Prof. V. K. Patil. Images.
We would like to thank her for the suggestion to perform [3] Carullo A, Ferraris F, Graziani S. Ultrasonic distance
sensor improvement using a two level neutral network.
shape detection, color detection and distance calculation IEEE Transaction on Instrumentation and measurement.
on 2D images. We also like to thanks AISSMS IOIT, PUNE [4] C.C. Hsu and M.C. Lu, , Angle and distance
for providing the resources. measurement of distant objects based on pixel variation
of CCD images. May 36, 2010, pp. 318322.
REFERENCES [5] Song KT, Tang WH. Environment perception for a
mobile robot using double ultrasonic sensor and a CCD
[1] Chen-Chien James Hsu, Ming-Chin Lu and Yin-Lu camera.
Yu,Angle and distance measurement of objects on an [6] G. Borgefors, Distance transformations for arbitrary
oblique plane based on pixel number variation of CCD dimensions Dig. Image Processing, Graphics, and
images computer vision, no. 3, val 27, pp. 321345, September
1984. oblique plane based on pixel number variation of
CCD images

77
IJRITCC | May 2016, Available @ http://www.ijritcc.org
________________________________________________________________________________________________________