Proceedings of the IEEE

International Conference on Automation and Logistics

Shenyang, China August 2009

Design of Automatic Target-Scoring System

of Shooting Game Based On Computer Vision
Xinnan Fan,Qianqian Cheng Penghua Ding,Xuewu Zhang
Computer and Information Institute Computer and Information Institute
Hohai University Hohai University
Changzhou,Jiangsu Province,China Changzhou,Jiangsu Province,China
[email protected]
H TH [email protected]
H

Abstract - This paper describes a machine the other hand, the system achieves athletes’
vision system with back lighting illumination and information and accurate results. What’s more,
friendly man-machine interface. Subtraction is the system improves the shooting efficiency,
used to segment target holes quickly and enjoyment and equity.
accurately. The oval obtained after tracing
boundary is processed by Generalized Hough II. THE OVERALL DESIGN OF THE
Transform to acquire the target's center. AUTOMATIC SCORING SYSTEM
Marked-hole's area, perimeter and moment
invariants are extracted as cluster features. The
A. The back lighting illumination scheme
auto-scoring software, programmed by Visual C++, The spotlight below the target stands out
has successfully solved the recognition of off-target bullet holes, making for separating and
and overlapped holes through alarming analyzing bullet holes[3], the structure of which is
surveillance and bullet tacking programs. The shown in Fig.1. Sheet iron protects the camera
experimental results show that, when the target is from being destroyed by bullet. The capture card,
distorted obviously, the system can recognize the DH-CG320, digitizes the achieved analog
overlapped holes on real time and also clusters signals through the CCD black and white camera,
random shape holes on the target correctly. The becoming the digital signals.
high accuracy, fast computing speed, easy
debugging and low cost make the system can be
widely used.
Index Terms - Moment Invariant, Image
Processing, Pattern Recognition, Generalized Hough
Transform, Auto-scoring System
I. INTRODUCTION
Nowadays, advanced technology and
equipment are more essential than hard training Fig. 1 back lighting illumination and target image
in acquiring excellent shooting score. Traditional acquisition device
artificial scoring system has seriously impeded B. The whole constitute of the auto-scoring
increasing our whole national shooting level system
sharply. Meanwhile with the development of The schematic plot of the automatic target
people’s living standard, it seems quite necessary scoring system is shown in Fig.2, which is
to study the automatic target-scoring system to divided into two modules. The front module
make the shooting game more interesting and consists of black and white camera, video server,
bring more economic benefit, because the live automatic target-scoring computer, changing
ammunition shooting game is a popular target machine and back lighting illumination
entertainment[1].Due to the shortcomings of device. The back module, locating beside the
photoelectric target and acoustoelectric judge’s stand, consists of server, printer, sound
localization target, in technology and cost,which box and liquid crystal display.
are too complicated to need professionals to The flow of the whole system is a loop
debug and operate on the spot, limit the useable process. After shooting ten times, athletes use
universality in small scale match and daily target scoring machine to change a new target
shooting training[2].On the one hand, the system and start a new round loop. Before the shooting
described in the paper uses plain target and match, the auto-scoring system software,
camera, which is absolutely superior to the two installed in target scoring machine, is turned on
above systems in the cost. With the appearance to recognize the bull’s eye and the ring’s radius
of new image processing technology and method, in the intact target image. Then athletes’
the accuracy generally satisfies the match’s personal information is put into the
demands, and it’s easy to operate and debug. On software[4].The auto-scoring software captures

978-1-4244-4795-4/09/$25.00 © 2009 IEEE

825
the perforated target,recognizes the holes and training, via video surveillance and voice
calculates the score automatically. Audiences can communication software, coach acquires the
enjoy the match and know the score via liquid shooting situation and directs in the real time,
crystal display. The server is in charge of the enhancing efficiency.
score statistic,printer and sound scoring. In daily

Fig.2 the overall structure of the auto-scoring system

III. THE DESIGN OF AUTOMATIC is shown in Fig.4(d).The approximate ellipse,

with the width of a single pixel，is achieved by
TARGET SCORING SOFTWARE
contour following[8], shown in Fig.4(e).
A. The flow chart of Software design
Automatic target scoring software is the
core of the whole system, the flow chart of
which is shown in Fig.3. The bullet hole, target’s
ring and bull’s eye are segmented and
recognized firstly. Then according to target
scoring’s computation method, all the gained
parameters are used to calculate scoring, which
are controlled by scoring statistic database[5].
N

Achieve perforated border target New Bullet? Y Subtraction

image
Automatic scoring calculation

Bullet hole
Achieve intact target
Score database control

recogniton
image

Preprocessing

Bull’ s eye
Black plate extract
recognition

Scoring ring
Ring thread extract recogniton

Fig.3 The flow chart of automatic scoring software

Fig.4 The flow chart of target plate’s preprocess and
B. Target preprocessing and achievement achievement
1) Black plate’s segment and achievement 2) Target rings’ segment and achievement
Noise reduces the target image’s quality, The image of median filter operation,
adding noise image shown in Fig.4(a).Median Fig.5(a), is processed by laplace transform to
filter is proved effective on noise-rejecting, but strengthen rings’ edge[9], shown in Fig.5(b).
the ring lines become vague[6], shown in Fig.4(b). Adaptive threshold segmentation filters noise
The aim is to extract the black plate, so the and uninterested numerical part and obtains
vague ring lines facilitate the black plate binary image, shown in Fig.5(c).Contour
extraction. The threshold value of adaptive extraction is used to obtain the rings’ edge and
threshold segmentation is choosed automatically, filter the black plate[10], but there is some
realizing rapid segmentation of target and numerical noises, shown in Fig.5(d). Thinning
background[7], shown in Fig.4(c). Some isolated operation helps to extract ring lines with the
black dot and line existing in the binary image, width of a single pixel, shown in Fig.5(e), which
interfere in extracting the black plate’s contour. makes it possible to get the rings’ radius.In
Erosion operation can filter out these isolated horizontal and vertical direction, there are some
points and lines, but the plate will be out of numerical noise, however, in other directions
shape. Dilation operation could well recover the there is nothing, so the image is whirled 45
shape,the result of erosion and dilation operation degree to obtain the ring radius in horizontal and

826
vertical direction. To identify bull’s eye of approximate
ellipse with the width of a single pixel via
Generalized Hough Transform can overcome the
former shortages, such as large computation and
vast memory occupancy[11].The obtained
ring,shown in Fig.7,operated as following steps:
Step one: From left to right scanning all
points through the first row, if the value of the
point scanned is 0, the coordinates of these
points are preserved. Then their midpoints are
calculated and preserved. In a similar way, all
the midpoints Xmi are calculated and preserved,
B B

and a vacant image D is initialized. In the same

position, the value of the midpoints is set as 0,
lots of which are collinear, and the line is L.
Step two: From top to bottom scanning all
Fig.5 Ring lines’ preprocess flow chart and achievement points through the first column, obtain all these
3) Bullet holes’ segment and achievement midpoints Ymj , reserved in image D. Lots of
B B

The method of subtraction filters the these midpoints Ymj are collinear, and the line is
B B

effection of old bullet holes relative to the new N. The node of line L and line N is just the bull’s
ones,only leaving behind background and new eye[12], the point O, shown in Fig.7.
holes. For the single aim, it’s easy to calculate
the threshold. Using subtraction method also
reduces the times of target changing. When
recognizing the new bullet holes, it is no need to
repeat identifying the old bullet holes. The
question ， “The more bullet holes you shoot, the
more slowly the system runs”is resolved. The
following target subtracts the former target and
at the same time,the rings, numbers and old Fig. 7 Method of getting bull’s eye
holes are filtered, obtaining the same gray of 2) The recognition of target rings
dark background, shown in Fig.6(c).Fixed Now,firstly,calculate the coordinate of
threshold segmentation causes lots of isolated bull’s eye (a,b). Secondly,scan the black pixel
plots, but erosion and dilation operations filter points through a column in fig.5(f) and stop the
them and keep the holes’ shape. scanning as soon as encountering the bull’s eye.
When the scan is completed, 11 points will be
m
E( x, y) = AND[I ( x + i, y + j) &T (i, j)] (1)
i , j=0 achieved, which are the points of ring lines.
m
(2) Thirdly,compute the distance between the points
D ( x, y ) = OR[I ( x + i, y + j ) & T (i, j )]
i , j =0 and the bull’s eye, which are the target rings[12].
Where T(i,j) is the structure element, I(x,y) 3). The recognition of single bullet hole
A defective bullet hole, a binary image, is
is a binary image. E(x,y) is obtained erosion
shown in Fig.8(a). Template matching method
image and D(x,y) is obtained expansion locates the defective bullet hole[13], and then
image. covers the defective bullet hole, shown in
Fig.8(b). The standard template consists of
points, whose gray value is 0. Erosion and
dilation operations filter these black spots.
Contour tracing extracts the edge of bullet hole ,
shown in Fig.8(c).

Fig.8 the flow chart of template matching method

Finally, Hough Transform is used to obtain the
Fig. 6 The flow chart of bullet holes’ preprocess and center of the hole.
achievement For the distance between the bullet’s
C The identification of target image innermost pixel and the bull’s eye is the shortest,
1) The recognition of bull’s eye by GHT

827
the following formula can be used to calculate shown as follow.
the coordinate of the bullet’s innermost pixel. 5
Dij 2 = ( Xi − X j )T ( Xi − X j ) = Xi − X j = ∑( Xi − X j )2 (11)
2

Where the point ( x i , y i ) is the bull’s eye and the i =1

point ( x j , y j ) is on the ring lines. When Dij 2 is small, the distance between
two samples is small. The smaller the distance is,
dc = ( xi − x j ) 2 + ( yi − y j ) 2 (3) the more similar between two samples. The
4) The recognition of multi-bullet holes distance of specimen and class is the distance of
specimen X and midpoint X W is shown as
（ ）
After shooting ten times, an unknown type P
P

bullet holes are obtained. The image with the follow.

unknown type bullet holes subtracts the intact 5
D 2 ( X ,W ) = ∑ X k − X k ( w)
2
target image to extract the ten bullet holes, and (12)
then these unknown holes are filtered and k =1
segmented. According to the judging methods of Finally minimum and maximum distance
eight connected domains, the binary image’s method is used to classify ten bullet holes.
connectivity is identified[14], shown in Fig.9, and Firstly one bullet point is set as the centeral point,
the connected domains are marked. The area, and then the distances between the center point
perimeter and moment invariant of the marked and other bullet holes are calculated. The
domains are extracted[15]. maximum distance is set as md, as the classified
reference threshold. Without artificial setting
value, the classified accuracy is improved,
because the setting threshold is one of the key
components to cluster analysis. If the distance
between one sample and the centeral point is less
than md/2, the sample will be classified in this
class,otherwise set a new centeral point.
Fig. 9 The judge method of eight connected domains
Images of seven moment invariants are
obtained by Papoulis uniqueness theorem. On IV. COMPUTE TARGET SCORING
condition that scaling factor is less than 2 and To improve the precision, the target scoring
the rotation is no more than 45 degrees, the is divided into two parts, integer and decimal.
seven moment invariants are invariable in According to the relative position of bullet holes
translation, rotation and the change of and ring lines, three different relative positions
proportional divisor. For the shape of bullet are existed. The point may be on the ring or
holes is simple, the anterior three moment between rings or outside the ring, so the position
invariants can represent primary features, so the of the innermost point should be firstly judged.
paper only chooses the anterior three ones as
bullet holes’ features[12]. The classification flow CN+1
chart of unknown type bullet holes is shown in CN‐1
bc
Fig.10. L
a

L
Fig.10 The classification flow chart of unknown type bullet
holes Pb

mpq = ∑ ∑ f (i, j)* i p * j q ; p, q = 0,1,2... (4) Fig. 11 The relative position of bullet holes and ring lines
i j Point b is the innermost point of bullet hole,
(5) shown in fig.11. Now, the innermost point,
upq = ∑ ∑ (i −i ') p ( j − j ')q f (i, j); p, q = 0,1,2... bull’s eye, and ring lines’ radius are known. The
i j
m1 0 m0 1
following algorithm can be used to compute the
重 心 ： ( i ', j ' ) =( , ) value of bullet holes.
m0 0 m0 0 (6)
1) Firstly,calculate the distance between the
upq p+q (7)
ηpq = u00
r; r =1+ , p + q = 2,3... innermost point b and the bull’s eye,point o.
2 Judge whether point b is on ring line by
H1 = η20 + η02 (8) comparing with ring lines’ radius. The integer N
of the score can be obtained.
H 2 = (η 20 − η 02 ) + 4η 2
11
2
(9)
2) Secondly, two points, the innermost point b
H 3 = (η 3 0 − 3η 1 2 ) + (3η 12 − η 0 3 )
2
(10) 2
and the bull’s eye o, define a line L. Judge
Euclidean distance is adapted to calculate whether the point is outside the rings,such as the
the distance between samples and the formula is point P, shown in Fig.11, is a node outside the

828
ring. Compare two segments' distance, op and ob. The back lighting illustration device makes
If op<ob, the ring value is 0. the bullet hole be the lightest part in the target,
3) Thirdly, the innermost point is must between so outside light can’t affect the accuracy of
two rings. The line L intersects two rings recognition. The designed system considers the
CN-1 and CN+1,, shown in Fig.11. The two nodes
B B B B abnormality and designs a new calculation
are point a and point c. Firstly the integer part N method to improve the accuracy, according to
is fixed, and then the following formula can fix that the relative position isn’t changed when the
the decimal part dpoint, so the whole value is
B B obtained target is abnormality. Off-target alarm
obtained by the following formula. program is set in video surveillance system. The
d ab target’s surrounding is the alarm area. If the
d po int = bullet is off-target, appearing in the alarm area,
d ab + d bc (13) the difference of two adjacent images isn’t zero,
Dscore = N + d po int and the alarming will inform us. If bullet holes
(14) are full overlapped, we firstly counter the total
Where dab represents the distance between point
B B holes’ number after the game. If the number is
a and point b,and dbc represents the diatance
B B not ten, we examine the video record by bullet
between point b and point c. tacking to locate the bullet and obtain the absent
bullet mark. If it’s the situation of off-target, the
V. EXPERIMENTAL RESULT’S ANALYSIS alarming will inform us.The consideration of
The table analyses a group marks of 10 meters two peculiar situations strengthens this system’s
air pistol and the comparison between automatic stability and reliability. Because of the low cost
and artificial target scoring is shown in table I. and easy operation , this system can overcome
the shortcomings of extant targeting systems and can
Table I：The comparison between automatic be used widely.
and artificial target scoring
(* represents overlapped bullet holes)
REFERENCE
Shoot Artificial Automatic Obvious
times target-scoring target-scoring abnormality [1] Chaoyang Lu, Xiquan Gao, Yumei Ding,
*the first “An automatic scoring system of shooting
10.5 10.45 10.41
shoot games using image recognition”, Computer
*the Engineering & Science, Vol. 23, no. 3, p.
second 10.4 10.46 10.42
shoot 11, 2001.
the third [2] Jun Zhang, Shuhua Yan, Yan Xu,
10.1 10.17 10.20
shoot “Automatic target-reading system
*the fourth development”, Laser & Infrared , Vol. 36,
7.9 7.87 7.84
shoot
No. 12, pp. 1152-1154, 2006.
the fifth
shoot
9.8 9.84 9.87 [3] Fenghua Jin, Lei Qin, Lu Jiang, Bin Zhu,
the sixth Yang Tao, “Novel separation method of
9.5 9.54 9.52
H

shoot black walnut meat from shell using

the seventh invariant features and a supervised
10.0 9.98 10.01
shoot
the eighth
self-organizing map”, Journal of Food
H

7.9 7.87 7.84 Engineering, Vol 88, Issue 1, pp. 75-85,

shoot
the ninth Sep 2008.
9.2 9.18 9.21
shoot [4] Jianbo Zhang, Chaoyang Lu, Xiquan Gao,
*the tenth
7.7 7.78 7.72 Yumei Ding, “An approach to improving
shoot the scoring precision of the automatic
From the above table, the obtained shooting scoring system for shooting sports”, Journal
results of the automatic target-scoring system of XIDIAN UNIVERSITY, Vol. 29, No. 3,
could accurate to 2 decimal places, but artificial pp: 324-327, Jun 2002.
target-scoring system could accurate to 1 [5] Yafei Shao, Li Zhang, Guowei Wu, “The
decimal place, so the system in the paper visual detection of bullet holes in a target”
improves the accuracy of target-scoring. When vol. 2, pp: 914 – 917, Aug. 2000.
bullet holes are overlapped, template matching [6] W.F. Xie, D J Hou, Q Song, “Bullet-hole
operator corrects and mends defective bullet image classification with support vector
holes. The system can still exactly and machines” IEEE Trans ,Signal Processing
automatically compute the score. When the Society Workshop, Vol. 1, pp: 318 – 327,
obtained images are obvious abnormality, the Dec. 2000.
system repeatedly recognizes the same target [7] Ming Ye, Deguang Zhou, Xiaomao Zhu,
image, the biggest error between them is 0.1, “Design of ball firing system of ball bore in
which still satisfy the match’s demands. self cover bore based on image recognition”,
VI. CONCLUSION Modern Manufacturing Engineering, No.8,

829
 p.100
[8] Mingyang Cheng, Chengchien Lee, “On
H H H H

real-time contour error estimation for

contour following tasks”, IEEE
trans, Advanced Intelligent Mechatronics,
H H

pp: 1047-1052, July 2005.

[9] R. C. Gonzalez and R. E. Woods, Digital
Image Processing, Second Edition, Bei Jing:
House of Electronics Industry, 2002.
[10] Doukovska, Lyubka, Kabakchiev, Christo,
“Performance of Hough Detectors in
Presence of Randomly Arriving Impulse
Interference”, pp: 1 - 4 ,May 2006.
[11] Nakanishi M, Ogura T, “Real-time line
extraction using a highly parallel Hough
transform board” ,IEEE Trans, Image
Processing, vol.2, pp: 582 - 585 ,Oct. 1997.
[12] Liru Yuan, “The Bullet-spot's Identifying of
Military Automatic Target-scoring System
Based On Image processing” , master thesis
of Chongqing university,2006.4
[13] Bin He, Tianyu Ma, Yunjian Wang, Visual
c++ the image processing, Wiley Computer
Press, 2001.4
[14] Shuying Yang, image pattern recognition:
picture processing implementation,
Tsinghua University Press. Beijing, 2005.7
[15] Mingyang Cheng, Chengchien Lee, “On
H H H H

real-time contour error estimation for

contour following tasks”, IEEE
Trans, Advanced Intelligent Mechatronics,
H H

pp: 1047-1052, July 2005.

830