Journal of Signal and Information Processing, 2013, 4, 144-149
doi:10.4236/jsip.2013.43B025 Published Online August 2013 (http://www.scirp.org/journal/jsip)
New Approach in Processing of the Infrared Image
Sequence for Moving Dim Point Targets Detection
Mohamed Abdo M., Li Hongzuo
School of Electronics and Information Engineering, Changchun University of Science and Technology, Changchun city, China.
Received May, 2013.
ABSTRACT
The development of an efficient moving target detection algorithm in IR-image sequence is considered one of the most
critical research fields in modern IRST (Infrared Search and Track) systems, especially when dealing with moving dim
point targets. In this paper we propose a new approach in processing of the Infrared image sequence for moving dim
point targets detection built on the transformation of the IR-image sequence into 4-vectors for each frame in the se-
quence. The results of testing the proposed approach on a set of frames having a simple single pixel target performing a
different motion patterns show the validity of the approach for detecting the motion, with simplicity in calculation and
low time consumption.
Keywords: IR Image Sequence Processing; Statistical Processing; Dim Point Target Detection
1. Introduction
The detection and tracking of pixel-size moving targets
in optical or infrared images have been an active research
area for a many years. The pixel-size target is produced
when distance between target and imaging system (visual
camera or forward looking infrared—FLIR) is long
enough. In these circumstances, the target in each frame
of images only occupies several pixels, even one pixel.
Early work in IR search and tracking systems utilized
algorithms that initially attempted to detect the target
spatially in each image, and then used temporal associa-
tions for target tracking. Such as the unified framework
for IR target detection [1], the correlation based algo-
rithm [2], wavelet based algorithm [3], mathematical
morphology based algorithm [4], image processing tech-
niques [5], maximum local contrast [6] and pattern
analysis [7], etc. These algorithms are also known as
“detect before track” algorithms. Although the “detect
before track” algorithms were adequate for applications
where the targets were bright compared with the back-
ground, they performed poorly with dim targets in severe
clutter. In long range surveillance, the target occupies
few pixels or even single pixel, and it can be easily con-
taminated by noise and evolving clutter.
As for the case of the detection of moving dim point
target in images is quite difficult because the target in-
tensity in images is low due to the energy transmission
loss through long distance, that is, the signal-to-noise
ratio (SNR) is low. Another difficulty of detecting mov-
ing dim point target in low SNR images is that it is not
easy to accumulate the target energy due to the small size
and the motion of the target. Because of these reasons, it
has been realized that moving dim point target cannot be
detected on the basis of single frame image (DBT algo-
rithms). We have to deal with image sequence processing
instead of single frame image processing to get better
detection results. Thus the “track before detect” algo-
rithms were proposed. The “track before detect” algo-
rithm is a temporal based algorithm which uses multiple
frames to incorporate temporal as well as spatial infor-
mation. A lot of techniques and algorithms was devel-
oped within this concept of processing as 3-D matched
filtering [8,9], Velocity filter banks [10,11], multistage
IIR filter [12,13], Temporal Profile Based detection [14],
Temporal filters [15,16], triple temporal filter (TTF) [17],
bilateral TTF [18], dynamic programming [19, 20], se-
quential detection [21], parallel spatial and temporal fil-
tering [4,22] and Probabilistic data association (PDA)
[23], etc.
The main problem that faces any of these algorithms is
caused from the huge number of data handling which
cause either the computational complexity or the time
consumption which for the realization and implementa-
tion for real time processing it either cost money or for
compensation it comes on the expense of the accuracy.
This paper is organized as follows. In Section 2, we
outline temporal profile of IR image sequence and it’s

New Approach in Processing of the Infrared Image Sequence for Moving Dim Point Targets Detection 145
constitutes. In Section 3, we introduce the proposed new
approach and describe its basic steps. In Section 4, a
simple practical example of the approach implementation
is presented. These results allow us to evaluate the per-
formance of the validity of the approach. Finally the
conclusion and plans for future work are outlined in Sec-
tion 5.
2. Temporal Profile Model
By using a focal plane array (FPA) detector to constantly
monitor a scene, each pixel will produce a temporal pro-
file over a short period of time. The temporal profile in-
dicates the variation of the pixel values in this period of
time. When a target moves across the pixel, a pulse-like
shape disturbance is created on the temporal profile. The
width of the pulse will be inversely proportional to the
target velocity. Its height above (or depth below) the
background depends on its differential intensity with
respect to the background.
The pixels that see clear sky or other features constant
in time will have temporal profiles that usually behave
like a constant mean value plus white noise. Stationary or
very large slow moving clutter will also appear as a
slowly varying mean plus the same random noise process.
Pixels affected by cloud edges or other difficult clutter
features will have less regular temporal behaviors. A
pixel affected by a small moving target will have a
pulse-like shape on the temporal profile, which is distinct
from that of the cloud clutter and clear sky [14, 18, 24].
2.1. Static Background
Pixels seeing static background or slow moving objects
such as clear sky and the inner portions of cloud have
approximately constant intensities. Intensity variation is
often caused by random noise. Therefore, the temporal
profile of pixels seeing a static background can be mod-
eled as a constant plus a low level of random noise. The
random noise can be effectively modeled by a Gaussian
distribution.



ItC wt (1)
where x(t) is the intensity value of pixel at time t, C is the
constant value, w(t) is the random noise assumed to be
Gaussian with zero mean and variance .
2
s
σ
2.2. Cloud Edge
Pixels seeing a cloud edge have the temporal profile
modeled by a first-order Markov model.
 
ItIt 1nt (2)
where n(t) is assumed independent with normal density,
zero mean, and variance .
2
c
σ
2.3. Target
Pixels that see a target have intensities that are distinct
from those of the cloud clutter and clear sky. The intensi-
ties temporal profile of small targets and the background
are different: either colder or hotter than the surroundings.
As the target moves across these pixels, there will be a
disturbance signal on the temporal profile. The width and
height of the disturbance signal is related to the target
velocity and intensity respectively. Therefore, the tem-
poral profile of the target can be modeled by su-
per-imposing a disturbance signal on the background.




ItBtTt

(3)
where T(t) is the disturbance signal generated when tar-
get move across the starring pixel. B(t) is the background
intensity related to the position where the target is lo-
cated. If the target appears on a static background, then
B(t) is the intensity of the static background, else it is
cloud edge.
T(t) can be modeled as an independent Gaussian signal
with higher variance and mean value reflecting the tem-
perature of the target and can be represented as follows:


Tt cxt


  (4)
where, c is the background intensity, Δt is time of target
entering and exit the starring pixel, x() is normal
Gaussian function which exists during Δt, with mean μ
and variance . f(t) has a constant value of c when the
target does not exist. The value of mean μ is either higher
or lower relative to background intensity c to reflect the
temperature difference between target and the surround-
ings. The variance is higher than the process noise
variance for static object to reflect the disturbance signal
caused by target moving across the starring pixel.
t
2
t
σ
2
t
σ
So in general the observation model of infrared image
sequence can be described as follows:




fx,y,k
Bx,y,kTx,y,kNx,y,kwith target
Bx,y,kNx,y,kwithout target
 




(5)
where


fx,y,k represents the intensity of infrared
image sequences,


Bx,y,k is intensity of background
clutter, which is the main component in sequences,


Nx,y,k is the intensity of noise generating in focal
plane, and


y,kTx, is the target components when it
appears. x,y in above expressions respectively represent
the spatial coordinate in focal plane (x=l,2,…,M,
y=l,2,…,N for an image of size MxN) and k is temporal
coordinate in frames ( k=l,2,..,K).
3. The Proposed Approach
A single IR-image frame contains N rows, M columns

New Approach in Processing of the Infrared Image Sequence for Moving Dim Point Targets Detection
146
and D Diagonals (same for both right and left diagonals),
Where D is the number of diagonals in a single frame
(Figure ), which can be simply calculated by
D=(M+N)-1 (6)
The proposed approach mainly perform a transforma-
tion of the every single IR-image frame of N rows by M
columns pixels is converted into 4-vectors, Horizontal
Vector with (1xM) cells, Vertical Vector with (1xN) cells,
Right Diagonal Vector with (1xD) and the Left Diagonal
Vector with (1XD) cells. Each cell of those vectors is the
standard deviation value of the corresponding vertical
columns pixel, horizontal row pixels, Left Diagonal pix-
els and Left Diagonal pixels respectively. Figure and
Figure shows an illustration for the process of creating
the 4-vectors (Horizontal - Vertical – Right Diagonal –
Left Diagonal).
The standard deviation values are calculated using the
population standard deviation
 
2
V
v1 Ix,y,v μ
σV

 (7)
where the is the pixel intensity, x, y are the
pixel position, v is the pixel frame number, V is the total
number of pixels (V=N for rows and diagonals, M for
columns),

Ix,y,v

is the average of the pixels to have their
standard deviation calculated.
Figure 1. Sample of FPA Pixels and the description of the
Columns, Rows, Left Diagonals and Right Diagonals.
During the calculation of the Diagonals standard de-
viation some diagonals will contain less number or pixels,
so to normalize the standard deviation value for all di-
agonals we perform zeroing for shorter diagonals as ex-
plained in Table (Figure ), where the Added Zeroing
pixels number different from diagonal to another ac-
cording to the number of existing diagonal pixels.
Table 1. Diagonal Standard deviation calculation.
Population standard deviation
# of
pixels Before Zeroing After Zeroing
shortest
diagonal 1
 
2 1
v1 Ix,y,v μ
1


 
2 N
v1 Ix,y,vμ
N


longest
diagonal N
 
2 1
v1 Ix,y,vμ
N




Figure 2. Illustration of Horizontal/Vertical Vectors crea-
tion process.
Figure 3. Illustration of Right/Left Diagonal Vectors crea-
tion process.
2
N
v1 Ix,y,v μ
N


Figure 4. Diagonals Zeroing.

New Approach in Processing of the Infrared Image Sequence for Moving Dim Point Targets Detection
147
the frames a linear motion through four paths vertical
(path 1) , horizontal (path 3), and two diagonal (path 2-4)
as shown in Figure. It was clear that the Motion of the
target will be recognized in at least three of the 4-vectors
corresponding to the frame. This change in the 4-vectors
will correspond to the disturbance created by the target
passing through the corresponding column/row/diagonal.
The effect of motion is shown in Figure , Figure ,
Figure and Figure , it’s also very noticeable that the
target motion disturbance will only be fixed in the case
of target motion in a path normal to the vector (marked
with (X) on the figures).
Figure 5. Testing Target Paths.
4. Experimental Results
We applied the proposed new approach on a 95 frames
each of (15x20) pixels, where a target is moving across
Figure 6. Target moving in a vertical path (normal to the Horizontal Vector).
Figure 7. Target moving in a Horizontal path (normal to the Vertical Vector).
Figure 8. Target moving in a right diagonal path (normal to the left Diagonal Vector).

New Approach in Processing of the Infrared Image Sequence for Moving Dim Point Targets Detection
148
Figure 9. Target moving in a left diagonal path (normal to the right Diagonal Vector).
5. Conclusions
In this paper we propose a new approach in processing
the of the Infrared image sequence for moving Dim Point
targets detection built on the transformation of the
IR-image sequence into 4-vectors for each frame in the
sequence. Where these 4-vectors represent that standard
deviation of columns, vectors and two diagonals of each
image frame pixels. Thus any point on the IR-image can
be presented by two different pairs of coordinate either
axial or diagonal. We tested the proposed approach on a
set of 100 frames with different moving target’s behavior
to prove the ability of it to adapt and response to different
dim small targets motion patterns. The test results show a
great performance beside its calculations and low time
consumption which make them valid to be used in real
time detection.
REFERENCES
[1] D. S. Chan, “Unified Framework for IR Target Detection
and Tracking,” Proceedings of SPIE 1698, Signal and
Data Processing of Small Targets 1992, 1992, pp. 66-76.
doi:10.1117/12.139403
[2] R. J. Liou and M. R. Azimi-Sadjadi, "Dim Target Detec-
tion using High Order Correlation Method," Aerospace
and Electronic Systems, IEEE Transactions on, Vol. 29,
1993, pp. 841-856. doi:10.1109/7.220935
[3] G. Boccignone, A. Chianese and A. Picariello, "Small
Target Detection using Wavelets," in Pattern Recognition,
1998. Proceedings. Fourteenth International Conference
on, Vol. 2, 1998, pp. 1776-1778.
[4] J. F. o. Rivest and R. Fortin, "Detection of Dim Targets in
Digital Infrared Imagery by Morphological Image Proc-
essing," Optical Engineering, Vol. 35, pp. 1886-1893,
1996. doi:10.1117/1.600620
[5] T. J. Patterson, D. M. Chabries and R. W. Christiansen,
"Image Processing For Target Detection Using Data From
A Staring Mosaic Ir Sensor In Geosynchronous Orbit,"
Optical Engineering, Vol. 25, pp. 251-258, 1984.
[6] Y. Jihui, K. Yongjin, L. Boohwan, K. Jieun and C.
Byungin, "Improved Small Target Detection for IR Point
Target," in Infrared, Millimeter, and Terahertz Waves,
2009. IRMMW-THz 2009. 34th International Conference
on, 2009, pp. 1-2.
[7] N. C. Mohanty, "Computer Tracking of Moving Point
Targets in Space," Pattern Analysis and Machine Intelli-
gence, IEEE Transactions on, Vol. PAMI-3, pp. 606-611,
1981. doi:10.1109/TPAMI.1981.4767153
[8] I. S. Reed, R. M. Gagliardi and H. M. Shao, "Application
of Three-Dimensional Filtering to Moving Target Detec-
tion," Aerospace and Electronic Systems, IEEE Transac-
tions on, vol. AES-19, 1983, pp. 898-905.
doi:10.1109/TAES.1983.309401
[9] I. S. Reed, R. M. Gagliardi and L. B. Stotts, "Optical
Moving Target Detection with 3-D Matched Filtering,"
Aerospace and Electronic Systems, IEEE Transactions on,
vol. 24, 1988, pp. 327-336. doi:10.1109/7.7174
[10] A. D. Stocker and P. D. Jensen, "Algorithms and Archi-
tectures for Implementing Large-velocity Filter Banks,"
Proceedings of SPIE 1481, Signal and Data Processing
of Small Targets 1991, 1991, pp. 140-155.
[11] W. B. Kendall, A. D. Stocker and W. J. Jacobi, "Velocity
Filter Algorithms for Improved Target Detection and
Tracking with Multiple-Scan Data," Proceedings of SPIE
1096, Signal and Data Processing of Small Targets 1989,
pp. 127-139, 1989.
[12] H. E. Rauch, W. I. Futterman and D. B. Kemmer, "Back-
ground Suppression and Tracking with A Staring Mosaic
Sensor," Optical Engineering, 1981, Vol. 20, pp.
201103-201103. doi:10.1117/12.7972672
[13] E. T. Lim, S. D. Deshpande, C. W. Chan and R.
Venkateswarlu, "Dim Point Target Detection in IR Im-
agery using Multistage IIR Filter," Proceedings of SPIE
4025, Acquisition, Tracking, and Pointing XIV, 2000, pp.
194-202.
[14] D. Liu, J. Zhang and W. Dong, "Temporal Profile Based
Small Moving Target Detection Algorithm in Infrared
Image Sequences," International Journal of Infrared and
Millimeter Waves, Vol. 28, 2007, pp. 373-381.
[15] J. Silverman, J. M. Mooney and C. E. Caefer, "Tracking
point targets in cloud clutter," Proc. SPIE 3110, 10th
Meeting on Optical Engineering in Israel, 1997, pp.
496-507. doi:10.1117/12.281362
[16] A. P. Tzannes and D. H. Brooks, "Temporal Filters for
Point Target Detection in IR Imagery," Proceedings of
SPIE 3061, Infrared Technology and Applications XXIII,
1997, pp. 508-520.

New Approach in Processing of the Infrared Image Sequence for Moving Dim Point Targets Detection 149
[17] J. M. Mooney, J. Silverman and C. E. Caefer, "Point Tar-
get Detection in Consecutive Frame Staring Infrared Im-
agery with Evolving Cloud Clutter," Optical Engineering,
vol. 34, pp. 2772-2784, 1995.
doi:10.1117/12.210757
[18] W. Zhang, J. Gong, Q. Hou, and C. Bian, "Point Target
Detection Based on Nonlinear Spatial-temporal Filter in
Infrared Image Sequences and Its Analysis," Proc. SPIE
8558, Optoelectronic Imaging and Multimedia Technol-
ogy II, 2012, pp. 85582E-85582E.
doi:10.1117/12.2000626
[19] Y. Barniv, "Dynamic Programming Solution for Detect-
ing Dim Moving Targets," Aerospace and Electronic
Systems, IEEE Transactions on, Vol. AES-21, 1985, pp.
144-156. doi:10.1109/TAES.1985.310548
[20] J. F. Arnold and H. Pasternack, "Detection and Tracking
of Low-observable Targets through Dynamic Program-
ming," Proc. SPIE 1305, Signal and Data Processing of
Small Targets 1990, 1990, pp. 207-217.
[21] S. D. Blostein and T. S. Huang, "Detecting Small, Mov-
ing Objects in Image Sequences using Sequential Hy-
pothesis Testing," Signal Processing, IEEE Transactions
on, Vol. 39, 1991, pp. 1611-1629. doi:10.1109/78.134399
[22] S. Pohlig, "Spatial-temporal Detection of Electro-optic
moving Targets," Aerospace and Electronic Systems,
IEEE Transactions on, Vol. 31, 1995, pp. 608-616.
doi:10.1109/7.381909
[23] A. Hamdulla and L. Xingke, "High-Resolution Bayes
Detection of Dim Moving Point Target in IR Image Se-
quence Using Probabilistic Data Association Filter," in
Computer Science and Software Engineering, 2008 In-
ternational Conference on, 2008, pp. 365-368.
[24] B. Liu, Z. Shen, Z. Li and Z. Sun, "New Method of De-
tecting Moving Dim-point Targets in Image Sequences,"
Proc. SPIE 1954, Signal and Data Processing of Small
Targets 1993, 1993, pp. 167-172.doi:10.1117/12.157796