2011 Fourth International Joint Conference on Computational Sciences and Optimization

The technology of image processing used in Automatic Target-Scoring System

Cuiliu Ye, Hong Mi

Pattern Recognition and Intelligence Institute
Xiamen University
Xiamen, Fujian Province, China
[email protected]

Abstract—Military Automatic Target-scoring System can avoid points[2]. The rectification diagram is shown in figure 1, the
the defects in efficiency, reliablity, safety, which exist in basic idea is as follow.
traditional counting system made by manpower, it can a)According to certain transformation, choose some
implement fast and precise counting. We take Chest silhouette appropriate coordinate of the tamplate datum points from the
as studying object, detail each selected method of image data measured with the Chest silhouette, then save them to
processing and image recognition in the Auto-counting System. the database of the target-scoring system.
In addition,the results of all algorithms that MATLAB b)Choose the corresponding coordinate of the
imitated come to the army’s need. The work flow of this Auto- observation datum point on the intact image.
counting system can be composed of these phases: Target
c)Calculate the transformation matrix by computer with
image Rectification, Bullet-spot’s extraction, Bullet-spot’s
Recognition and Scoring.
certain coordinate transform type.
d)At last, start the geometric correction useing the
Keywords- Image processing; Military Automatic Target- corrected transformation matrix on the following distorted
scoring System ;Image recognition; Geometric rectification image of traget.

I. INTRODUCTION
In military training, most of the Traditional counting in
firing practice is made by manpower, and it has some
defects in efficiency, safety, etc.. Meanwhile, its detection
accuracy depends on the target sheet which is influenced by
many environmental factors. With the development of
people’s living standard, it seems quite necessary to study
the automatic target-scoring system to make the shooting Figure 1. Automatic target-scoring project of this thesis
game more efficiency. On the one hand, the Automatic The image preprocessing including Median Filter, Image
Target-scoring System can achieve athletes’ information Binary and Image edge Extraction must be done before
and accurate results. What’s more, it can automatically selecting the datum points.
count and print the firing score to ensure the reliable A. The original target map preprocessing
accurate impact point report[1].
In this paper, a scheme of automatic target-scoring 1) Median Filter
system is presented, which is based on digital image Noise reduces the tartet image’s quality, adding noise
processing and recognition technology. This system can image shown in Figure 2(a).To meet the requirements of the
realize the function of automatic target-scoring and data system, we choose the method of median filter[3]. The
management (statistics, record, analysis)etc., by Target median filter is popular because of its noise reduction
image Rectification, Bullet-spot’s extraction, Bullet-spot’s without blurring properties. It is proved effective on noise-
Recognition and Scoring. rejecting, shown in Figure 2(b).
2) Image Binary
II. TARGET IMAGE RECTIFICATION Image Binary is an important foundation of the image
Nowadays, the images collected with an optical lens processing. To do this, it converts the input image to
always exist the problem of geometric distortion because of grayscale format. And then converts this grayscale image to
the three-dimensional relative position between the camera binary by thresholding. It is much better than grey image.
and the objects, lens distortion and other factors. During What’s more, the image binary algorithms have a fast
target detection based on image analysis and pattern operation speed, and the results are remarkable. The image
matching we need to correct the geometric distortion of binary is shown in Figure 2(c).
image to achieve effective digital image. 3) Image edge Extraction
We present a geometric calibration method for lens The aim is to extract the black plate, so the vague ring
camera by applying polynomial warping and bilinear lines facilitate the black plate extraction. The threshold value
interpolation, and discuss the selection pattern of control of adaptive threshold is choosed automatically, realizing

978-0-7695-4335-2/11 $26.00 © 2011 IEEE 349

DOI 10.1109/CSO.2011.287
rapid segmentation of target and background, shown in Thus, the undetermined coefficients are evaluated by
Figure 2(d). least square method based on minimum square fitting error
Image edge Extraction takes intensity or a binary image I sum, that is,
as its input, and returns a binary image BW of the same size L n n −i

as I, with 1's where the function finds edges in I and 0's min A ε x = ¦ ( xk −¦¦ aij uk i vkj ) 2 (4)
k =1 i =0 j = 0
elsewhere. Extract characters with Roberts edge detection. L n n −i
We can find the datum point easily in the picture shown in min B ε y = ¦ ( yk −¦¦ bij uk i vkj ) 2
k =1 i =0 j =0
Figure 2.
In the case, we have A = {aij } 䯸 B = {bij } . So that we
i, j i, j
can solve equations with undetermined coefficients a ,b.
Get the coordinate (x,y) of the corrected image by
transform each of the coordinate of distorted image. If the
coordinates fall precisely on the coordinates of distorted
image, then take the gray value of the point (u, v) as the
value of point (x, y); If the coordinates land on the ring line,
we need to interpolation for gray values; else we can use the
closest pixel instead .
The image after geometric rectification is shown in
Figure 2(f).
III. BULLET-SPOT’S EXTRACTION

A. The calculation of the bull’s eye, ring-radius and

background template
First, Since the effective ring of Chest silhouette is
irregular, we can not read the target value directly by the
distance between the bomb hole and the target center. The
Figure 2. the flow chart of target plate’s preprocess and Rectification judgment this article adopted is achieving the target value
by judging whether the bullet hole is on the background
B. Image geometric distortion rectified template or in the effective ring region. So we need to
Images with severe geometrical distortions, which are calculate of the bull’s eye, ring-radius and background
induced by camera lens and position, must be corrected template.
before extracting quantitative measurements. The monomial Some isolated black dot and line existing in the binary
can only correct the linear distortion, such as the rigid image, interfere in extracting the black plate’s contour.
transformation. The quadratic polynomial can not revise the Erosion operation can filter out these isolated points lines,
distortion. We demonstrate that, for common optical lens, but the plate will be out of shape. Dilation operation could
images with geometric distortions can be rectified with cubic well recover the shape. The result of erosion and dilation
polynomial warping method using only a few control points. operation is shown in Figure 3(a).
This method is simple and practical[4]. Contour extraction is used to obtain the ring’s edge and
Let f(x,y) and f(u,v) represent the original and the filter the black plate. In addition, make lines through the
distorted images, respectively, the nonlinear transformation bull’s eye in horizontal and vertical direction useing Roberts
between the two coordinates is : edge detection which make it possible to get the fings’
( x, y ) = T (u, v) (1) radius, shown in Figure 3(b). Take 4 points in each loop,
and then find out the average distance of the 4 points to the
Let aij and bij be the Undetermined Coefficient, then the bull’s eye which is the radius of each ring.
approximation by nonlinear polynomial is then given by Erosion and dilation operation filter the lots of isolated
n n −i plots and obtain the background template which is achieved
x = ¦¦ aij u i v j by inverting color, shown in Figure 3(c).
i =0 j =0 (2)
n n −i B. Bullet-spot’s extraction
y = ¦¦ bij u i v j
i =0 j =0 The bullet-spot’s segmentation is one of the key
When n=3 which is the polynomial coefficient. techniques in the automatic traget-scoring system. The
First, define the correspondence between the reference image subtraction is the kernel in technologies used in this
points: dissertation and key to reduce Bullet-spot’s Segmentation
( x1 , y1 ) ↔ (u1 , v1 ) algotithm. It simplifies the method and the bullet-spot’s
( x2 , y2 ) ↔ (u2 , v2 ) (3) extraction process, the system efficiency is increased.
The method of subtration filters the effection of old
#
bullet holes relative to the new ones, only leaving behind
( x L , y L ) ↔ ( u L , vL )

350
background and new holes. Using subtraction method also characteristics of the image[5]. Therefore, moment invariants
reduces the times of target changing. When recognizing the matching method combined with Euclidean distance are used
new bullet holes, it is no need to repeat identifying the old in bullet-spot’s recognition.
bullet holes. The following target subtracts the former target Moment function is used in the recognition of bullet
and at the same time, obtaining the same gray of dark holes. The algorithm of the 7 center of the moment
background, shown in Figure 3(d). invariants is shown as follow.
A defective bullet hole, a binary image, is shown in We define f(x,y) as two-dimensional discrete image
Figure 3(e). Template matching method locates the functions, that the moment of (p+q) order is given by.
defective bullet hole. We use an algorithm which combines m pq = ¦¦ x p y q f ( x, y ) (5)
the simple statistic method with Roberts edge detection to x y
extract characters from the complex background with noises Then the corresponding center of the moment invariant
and uneven gray levels, also adopts mathematical is given by
morphology to filter the noises of binary image, shown in u pq = ¦¦ ( x − x0 ) p ( y − y0 ) q f ( x, y ) (6)
Figure 3(f). x y

Where
m10 m
x0 = , y0 = 01 (7)
m00 m00
In the two-dimensional image, x0 represents the center
of gravity in the horizontal direction, y0 represents the
center of gravity in the vertical direction. As the moment
invariant is stable in recognition rate and in rotation, the
center moment of each older can be normalized with the
center moment of zero-older which can be obtained by the
following formula.
u pq
η pq = r
u00 ˄8˅
( p + q + 2)
r= , p + q = 2,3,"
2
The following formula can be used to calculate the
center moment.
φ1 = η 20 + η02
Figure 3. The flow chart of Bullet-spot’s extraction (9)
φ2 = (η 20 − η 02 ) 2 + 4η112
IV. BULLET-SPOT’S RECOGNITION AND SCORING φ3 = (η30 − 3η12 ) 2 + (3η12 − η03 ) 2
Image recognition is the final stage of the Automatic Image of moment invariants are obtained by Papoulis
Target-Scoring System. It contains bullet-spot’s recognition uniqueness theorem. For the shape of bullet holes is simple,
and scoring which is the most important. Bullet-spot’s the anterior three moment invariants can represent primary
Recognition mainly adopts Euclidian distance features, so the paper only chooses the anterior three ones as
distinguishable standard and Moment invariants method. In bullet holes’ features.
order to scoring, it first to obtain bull’s-eye and wreath Ediclidean distance is adapted to calculate the distance
line’s radius which are obtained in Bullet-spot’s extraction, between samples and the formula is shown as follow.
then to establishes corresponding region of each point ring n

and last to score by comparing bullet-spot heart’s place with D( x, y) =| | X- Y| | = ¦ (φ − φ )

i =1
i i
' 2
(10)
all regions.
When D(x,y) is small, the distance between two samples
A. Bullet-spot’s Recognition is small. The smaller the distance is, the more similar
The moment function is widely used in feature extraction between two samples. Finally, the following method is used
a pattern classification of image recognition respectively. to classify the bullet holes.
Using Hu moment features to identify targets is an important After one shoot, the target map compared with the
method for image identification. Through the nonlinear previous target image may find the following three cases:
combination of the geometric moments, we can achieve the two images are exactly the same; the later one has one more
invariant moments ever the images for translation, rotation hole; the later one has one more hole even with some noise.
and scale invariant. The moment invariant is stable in According to the relative position of bullet holes and ring
recognition rate and in rotation which is also of great anti- lines, three different relative positions are existed. The point
interference that can effectively reflect the essential

351
may be on the ring or between rings or outside the ring, so
the position of the innermost point should be firstly judged.
Therefore, the algorithm of bullet-spot’s recognition is
as follow.
a) If we extract zero region after region labeling, judge
the result as undershoot.
b) If we extract 1 region after region labeling, then
calculate the center of the region as the coordinate of the
bullet hole.
c) If we extract more than 1region after region labeling,
calculate the three moment invariants of each region, take
the shortest Euclidean distance as the bullet hole region, then
calculate the center of this region as the coordinate of the
bullet hole..
B. Bullet-spot’s Scoring Fig.4 The simulation interface of automatic target-scoring system.
To realize bullet-spot’s scoring, we mainly by comparing
the center of the bullet-spot and the background template and VI. CONCLUSION
compute the distance between the points and the bull’s eye, This paper is devoted to the research on automatic
which are the target rings. We will get exactly the same scoring system for shooting games, which is one of the
image as the background template if the bullet hole is outside applications of digital image processing. The paper focuses
the region of the background, else the white area in the on studying the Geometric Rectification and filtering to
background template will be the bullet hole region. reduce the interference from noise. The main method
Secondly, the results of the previous step which will be includes Median Filter, Image Binary, Image Subtraction,
performed exclusive or operation with the background etc.. We adopted cubic multinomial rectification technology
template. If the bullet hole is in the target area, the resulting which not only can correct optical lens’s distortion but also
image is completely black, else a white area of bullet hole has good rectification results for perspective and incline
will be got in the image. distortion. Among with the immense progress in the science
Finally, you will know whether the bullet hole is in the and technology of computer, the technology of image
target by judging the color of the image. If you get black processing is extending the applicable field of it. Because of
image, you miss the target, else calculate the distance the low cost and easy operation, this system can overcome
between the center coordinates of the bullet hole and the the shortcomings of extant targeting systems and can be
bull’s eye, then get the bullet-spot’s scoring according to the used widely.
radius of each ring.
REFERENCES
V. EXPERIMENT
[1] Chunlei Cui, “The research of Image Recognition Technologies used
This article presents the experimental interface of using in Military Automatic Target-Scoring System”, master thesis of
MATLAB to realize the automatic target-scoring system Dalian Maritime University, 2004.2
which is shown in Figure 4 using the above flow to process [2] Juyang Weng, Paul Cohen㧘and Marc Hernion,“ Camera calibration
pictures. First input the image data into computer memory, with distortion models and accuracy evaluation”IEEE,
TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE
then a clear image of higher signal-to-noise ratio will be INTELLIGENCE,VOL 14, NO.10,OCTOBER 1992.965-980
obtained through the stages of preprocessing of images
[3] DONZALO R.ARCE.Theoretical Analysis of The Max/Median
including Median Filter, Image Binary, Image Edge Filter.IEEE-TRANS;ASSP-36㧘N0.5㧘P826㧘1988.
Extration, Geometry Rectification and Image Subtraction. [4] Shizhong Liao, Peihuan Gao, Yiwang Su, “ A Geometric
Next, find the location of the bullet-spot via pattern Rectification Method for Lens Camera”,Journal of Imagee and
matching. And then match the image of bullet-spot with the Graphics,Vol.5(A),No.7,July 2000.
standard template, the results of shooting will be shown in [5] Xiangpei Qin, Image Processing and interface Programming Using
the target images. The system succeed in scoring it MATLAB, Electronic Industry Press. 2009.3.
automatically with the comparatively high accuracy of 90%. [6] Liru Yuan, “The Bullet-spot’s Identifying of Militaru Automatic
The obtained shooting results of the automatic target- Target-scoring System Based On Image processing”,master thesis of
Chongqing University,2006.4.
scoring system could satisfy the match’s demands. When
bullet holes are overlapped, template matching operator [7] Haifeng Chen, “Study of Automatic Scoring System for Shooting
Sports Based on Image Processing”, master thesis of Nanjing
corrects and mends defective bullet holes. The system can University,2005.3:
still exactly and automatically compute the score. [8] Zhihua Wei, Sijin Xin, “Design of an Automatic Target Record
System”,Journal of WUT,Vol 27,No.2, Apr 2005.
[9] Min Yao, “Digital Image Processing”, Mechanical industry
press.2006.

352