(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 9, No. 9, 2018

A Novel Approach for Background Subtraction using

Generalized Rayleigh Distribution
Pavan Kumar Tadiparthi1 Srinivas Yarramalle2 Nagesh Vadaparthi3
Associate Professor Professor Professor
Department of Information Technology, Department of Information Technology, Department of Information Technology,
MVGR College of Engineering, India GIT, GITAM University, India MVGR College of Engineering, India

Abstract—Identification of the foreground objects in dynamic mixture model based on the optimization of the GMM and
scenario video images is an exigent task, when compared to static Combining the spatial information. H.Zhou et al [2] have
scenes. In contrast to motionless images, video sequences offer proposed a foreground detection methodology in which the
more information concerning how items and circumstances authors have tried to improvise the codebook. Viswanth et al
change over time. Pixel based comparisons are carried out to [3] suggested and modeled an approach using non parametric
categorize the foreground and the background based on frame background modeling. In this approach, a single Spatio
difference methodology. In order to have more precise object Temporal Gaussian is used for modeling the back ground
identification, the threshold value is made static during both the pixels. However, this methodology fails as the adequate
cases, to improve the recognition accuracy, adaptive threshold
features are not obtainable from the section. Yuhan.L et al [4]
values are estimated for both the methods. The current article
also highlights a methodology using Generalized Rayleigh
proposed a robust back ground subtraction methodology based
Distribution (GRD). Experimentation is conducted using on the adaptive dictionary strategy and penalized splitting
benchmark video images and the derived outputs are evaluated approach. Lu yang et al [5] considered a pixel for modeling
using a quantitate approach. the background information in case of complex scenes. The
consideration pixel manipulates the distance between the
Keywords—Background subtraction; segmentation; generalized pixels and it was used for updating the back ground model as a
rayleigh distribution (GRD); quantitative evaluation; image substitute of local descriptors. Chen et al [6] proposed a model
analysis using varying learning rate and also adaptively selecting the
number of Gaussians. This model performs better in particular
I. INTRODUCTION cases of extraction of dynamic background information and
The most imperative characteristic of an intelligent vision sudden illumination variations. However, this model cannot
based inspection system is background subtraction, which is handle strong dynamic back ground and also fails in case of
considered to be a primitive step for object recognition and capturing the paused objects. Chien et al [7] proposed a
tracking. Typically, pixel by pixel comparison is practiced for foreground object detection method by using a threshold
either detection or tracking with a predefined object dataset. value. In this article, the authors have assumed that the camera
However, this procedure of searching and comparing against considered for capturing the videos are tolerant to noise and
each pixel requires a huge computational time and as an posses a zero –mean Gaussian distribution. But this
improvement to this approach, background subtraction assumption has affected the selection of the threshold. Lui et
methods are coined for the optimization of both search and al [8] proposed an approach based on the binary descriptors.
computational time. In many of the Human computer In this article, the authors have generated the back ground
interactive systems, background subtraction is considered instances using binary descriptors. The developed model has
during the pre-processing procedure to optimize the cost. As proved to be robust against lighting changes and dynamic
such, background subtraction has become a significant method back ground and is tested against the environmental changes.
and has deeply penetrated with strong roots in the area of Haung et al [10] proposed a method for back ground
computer vision. Since background modelling considerably modeling based on binary descriptors. However, this method
influences the performance on the whole vision system, it is can reduce the effect of noise and capable for the extraction of
imperative to make use of an excellent background subtraction rough shaped images from the foreground objects. Hedayathi
methodology. However, most of the background modelling et al [11] proposed a statistical frame work for back ground
techniques need to combat the challenges due to dynamic or subtraction, which acquired better performance in terms of
non-static backgrounds, unexpected or steady lighting segmentation.
changes; motion in the object and shade, Background
modelling methods should intelligently overcome such issues. Stephan Kopf‟s et al [15] has proposed a model for
To overcome these challenges, many models are presented in automatic scaling and cropping. The main limitation with
the literature [1-11], [13], [15], [16], [17], [18]. reference to this approach is that, this methodology needs pre-
identification of certain parameters which helps during the
II. LITERATURE REVIEW cropping of the selected regions. In the motion detection, the
Haiying et al [1] has proposed a modified Gaussian objects can be recognized only when the object is in moving
condition. Therefore, appropriate reorganization of objects can

506 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

be well planned, only if the back ground information from resolutions is highly difficult. Hence, to overcome this
these motion images is subtracted. This methodology is very disadvantage, the work presented by Michael Unger et al [19]
much useful for recognization of images acquired from has been extended by considering Generalized Rayleigh
surveillance cameras. Stauffer and Grim Son [18] have Distribution. The main advantage of proposed method is that,
presented an approach using pixel wise operation for the the generalization process allows estimating the back ground
identification of the back ground images. The main restraint of images in particular situations where the information is
this model is that, extracting the back ground pixel suppressed. Beyond the above mentioned advantage, the
information from the static camera is relatively difficult and model also includes,
therefore, this methodology leaves an unsolved issue about the
problem with respect to the images acquired from the static 1) It has the advantage over the other distribution, where
cameras. Tao Mei and Xian et al [16] have presented a model the degree of freedom can be easily obtained using the
for background subtraction, where image mosiacing is maximum value which is generally unique.
considered. The limitation of this approach is that, mosaicing 2) These distributions are maximum when it approaches
of background information is highly impossible. AL-Najdawi towards Y-axis.
et al [13] have utilized the kalman filter for the purpose of 3) The back ground information can easily be interpreted
seeking optimal estimation in tracking. D.Farin, P.de et al [17] using the Rayleigh distribution with the maximum value.
have proposed a model for the extraction of individual frames.
The limitation of this model is with respect to identification of Another limitation with respect to the Rayleigh
back ground pixels from the frames from the static cameras. distribution is that it can‟t handle specals having heavy tails.
D. Hari Hara Santosh et al [9] have utilized the Gaussian Therefore, the generalizations of the models help to overcome
Mixture models for the effective identification of the these limitations. Hence, as a contributing factor in this article,
foreground information. In this article, the authors have we propose a model based on generalized distribution to
addressed the concepts of object tracking using Blob Analysis. overcome the limitations of Rayleigh distribution. This
However, this methodology has its limitations while dealing distribution has an advantage to handle both high tail and low
with sudden, drastic lighting changes. tail impulsive noises.
To overcome these challenges many techniques were A continuous random variable is said to follow a
therefore planned, in particular by considering mostly the Rayleigh distribution if its probability density function (pdf) is
statistical frame works, with the very criteria that, the given by
efficiency of statistical models will be relatively high and
helps towards better identification of the object pixels more ( ) , (1)
appropriately. Based on this approach, many models have
been further developed using the mixture models, such as =0 , otherwise
Gaussian Mixture models and models based on the Rayleigh The Cumulative distribution function (cdf) is given by the
distribution. formula
The Rayleigh distribution cannot model the natural images
having specals, impulsive behaviour leading to heavy tails. ( ) , (2)
Therefore in this article, an attempt is made to overcome the =0 , otherwise
limitation highlighted and proposes a methodology by
considering a Generalized Rayleigh distribution (GRD). The Where , where , is the scale parameter.
main advantage of using the GRD is that, it is more accurate
in the reverberation regions. IV. DATA SET
The rest of the paper is organized as in Section III where To exhibit the proposed work, a Bench mark data set of
the Generalized Rayleigh distribution is considered. Section Video images from www.changedetection.net [12] has been
IV presents the details about data sets and Section V considered for experimentation. The dataset consists of 6
emphasizes the methodology. Section VI the experimentation different video categories with a total of 31 videos comprising
carried out is highlighted. The performance evaluation and the of 90,000 frames. These videos are mainly based on Baseline,
results derived are highlighted in Section VII. The Section Camera Jitter, Dynamic Background, Intermittent Object
VIII concludes the paper and describes the future scope. Motion, Shadow and Thermal.
III. GENERALIZED RAYLEIGH DISTRIBUTION V. METHODOLOGY
The early roots of considering Rayleigh Distribution for 1) Post processing: In order to extract the background
enhancing the Background Subtraction is considered by image, each of the inputs has to be first preprocessed by
Michael Unger et al [19]. In this article, the authors have considering the pixels having least deviation. The least
considered the model for Background Subtraction for deviation pixels have to be considered as background pixels.
unconstrained images acquired from cameras that are in
However, the main limitation in choosing the background
motion, as they have low resolutions and less distortion.
pixels is that, in particular situation, the foreground and
However, with the availability of sophisticated background shall share similar information with respect to
technological cameras, capturing the images with different color, size and orientation. Therefore, whenever we need to

507 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

estimate the background images, lighting conditions play a are estimated in line with the heuristics given by W. Jun-
vital role. This lighting condition, if interpreted exactly helps Qin[22]. Adaptive background subtraction the difference
to model the background images. In this article, the usage of between video frame at time t and the frame at time t-1. The
Generalized Rayleigh Distribution helps to overcome this optimal threshold is thereby estimated. In case of Adaptive
disadvantage because of its ability to handle low illumination frame difference fused method, the choice of the adaptive
images. threshold value is based on the difference value obtained by
2) Back ground subtraction: In background subtraction subtracting the background reference frame from the current
technique, motion objects were identified by deducting the frame and then these values are fused to get a unique threshold
present image from the background image. The initial frame value.
of the video progression was taken as reference image for
background frame. The present frame will be deducted from VI. EXPERIMENTATION
the considered background frame. The background pixel is Blob analysis is considered for the effective identification
decided on the basis of the resultant difference, i.e. if the of the background and foreground regions. Each pixel value is
output of the subtraction reference pixel value is greater than extracted based on threshold values obtained from background
the reference pixel value, then it is considered to be a subtraction method and frame difference methods and the
corresponding pixels are categorized into either background or
background pixel, else it is considered as a foreground pixel.
foreground. In general, pixels with minimum threshold values
3) Frame difference method: Here, we estimate the will be mostly considered as background pixels. The pixels
difference in values between two consecutive frames, „t ‟ and with high threshold values are given as inputs to the
„t-1‟. If the resultant value is better, then the value is taken to Generalized Rayleigh Distribution (GRD) presented in section
be the threshold value, and the pixel will be treated as III. Basing on the log likelihood estimation of the pixels, each
background pixel. pixel is categorized either as a back ground pixel or a
foreground pixel. The experimentation is carried out in matlab
The estimated threshold values, from both the cases are environment and the results obtained are shown below. In this
considered and are given as input to the model Generalized article, we have experimented with numerous ways of
Rayleigh Distribution proposed in section III of the article. estimating the background pixel; namely Fusion method,
The probability density functions (pdf) against each of the adaptive Background subtraction, Adaptive frame difference,
intensity values are given as input to the model and the Adaptive frame difference fused with Adaptive background
respective values are estimated. These values which are below subtraction. The significance of each of these methods are
the threshold value are considered as background information presented in section V. The results were also compared with
else they are considered as foreground information. the model based on Gaussian Mixture Model.
4) Fusion technique: In this article, two methods for VII. PERFORMANCE EVALUATION AND EXPERIMENTAL
estimating the background pixel, viz., background subtraction RESULTS
method and frame difference method are highlighted. In order to validate the model, we have considered the
However, each of these methods have their own limitations, performance for quantitative analysis are metrics Precision,
i.e., if we consider the background subtraction method, the Recall, Accuracy, F- Score, MSE, RMSE, FNR, FPR,
boundaries and contour will be intact, however the output PSNR[20].In order to validate the model, we have performed
result may be affected due to the noise parameter, in contrary, the experimentation with different frames 257,863,1005,1954
in frame difference method, result will have minimal impact respectively. The formulas for the identification of Precision,
due to noise, but in this case, the complete information Recall, Accuracy, F-Score. Recall is expressed in terms of the
regarding the boundaries and contours may be a bit number of allocated foreground pixels to that of actual
foreground pixels; and the evaluation outcome of this metric
influenced. To overcome these limitations, in this article, we showcase, the exact number of true foreground pixels that are
have considered the fusion concept using “AND” operation. classified as foreground pixels. Precision is defined in terms of
5) Adaptive background subtraction: The best possible the number of exact foreground pixels against the allocated
threshold value can be estimated using the adaptive threshold foreground pixels; it signifies the exactness of the pixels that
technique and it is estimated using the formula given by were classified as true foreground pixels against the allocated
foreground pixels. The performance of the model can be
F(x,y)=C(x,y)-R(x,y) (3)
justified by the value of calculated precision, if it is high, it
F(x, y) =1, if F(x, y) ≥ T signifies high performance. On the other hand, if method
allocates the majority of the pixels to background, the output
(where T is the Threshold value, ( using the methodology precision value may be high, but proportionally, the value of
proposed by N. Otsu[21] ) and zero otherwise. Here C(x, y) recall declines. To identify the trade-off between recall and
denotes the current frame, and R(x, y) represents the precision, F-measure is also considered. The other
considered reference background image, F(x, y) denotes, the performance metrics considered include; Mean Squared Error
deviation between the present frame and the reference frame. (MSE).Root Mean Square Error (RMSE), False Negative Rate
6) Frame differencing method and Adaptive frame (FNR) False Positive Rate (FPR) and Peak Signal to Noise
difference fused methods: Here the optimal threshold values Ratio (PSNR) [20]

508 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

The formulas for the calculation of the above metrics are classified as foreground, TN- the number of back ground
given by pixels classified as background.
Precision = TP / (TP+FP) (4) Experimentation is performed with the developed model,
by considering the data set presented in the section-IV. The
Recall = TP / (TP + FN) (5) results derived are presented in the following Fiig-1
Accuracy = TP+ TN / (TP+TN+FP+FN) (6) 1. Original frame. 2. Ground truth .3.Back ground
F-Score = (2 * Precision * Recall) / (Precision + recall) (7) subtraction.4.frame difference. 5 .Back ground subtraction and
frame difference .6.Adaptive back subtraction. 7. Adaptive
MSE = FP+FN / M *N (8) frame difference. 8. Adaptive back ground subtraction and
adaptive frame difference. 9. GMM. 10. GRD.
RMSE = √(MSE) (9)
We evaluated the different background modelling methods
FNR=FN/(TP+FN) (10) discussed in sectionV. The scenarios used to evaluate different
FPR=FP/(FP+TN) (11) methods thermal, baseline, dynamic background, shadow.
2
There are many videos for each scenario. We selected one
PSNR=10log10 (R /MSE) (12) typical frame work from each video. Fig 1 a-d are selected
from four categories in the CDNet 2014 dataset. Fig .1 (1)
show the original frame of the video and Fig 1. (2) are the
results of the ground truth data .Fig 1.(3)-(10)are the
foreground detection results of the state of the art background
modelling methods.Tabel I-IV presents nine performance
evaluation metrics of the eight back ground modeling methods
in the CDNet 2014 dataset.
The performance of the different methods can be
confirmed by the recall, precision,F-Score and other metrics.
For each evaluation metric, we give the results of the back
ground modeling methods in different Scenes via Figs.2 to 37.
Thermal: As shown on Fig.1 (a), their results are closer to
the ground truth data. Fig .5 indicates the F-Score are greater
than 50%
Base line: These videos contain a noise free static back
ground Fig.1 (b) shows foreground detection results of every
method. The proposed method (GRD) successfully detected
the foreground object. It can also observed that the F-Score of
each method in Fig.14 is very high, greater than 63%.
Dynamic background: As shown in Fig 1 (c), the
proposed method (GRD) is more effective than the other
methods when dealing with dynamic backgrounds. Fig .21
shows that their Recall is very high when compared with other
methods.
Fig. 1. Foreground Detection Results of Thermal, Baseline, Dynamic
Background, Shadow from the CDnet2014 Dataset. Shadows: The methods differ in the capability of
classifying shadow pixels as back grounds. As shown in Fig 1.
Where, TP-the number of Foreground pixels classified as (d) shows foreground detection results of every method. Fig
foreground, FN-the number of Foreground pixels classified as .32 shows that their F -score is very high when compared with
background, FP- the number of pixels of background pixels other methods.

509 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

TABLE I. EVALUATION METRICS OF DIFFERENT METHODS ON THREMAL VIDEO FROM CD NET DATASET

SUBTRACTION AND FRAME

ADAPTIVE BACKGROUND

ADAPTIVE BACKGROUND
FRAME DIFFERENCE

SUBTRACTION AND
ADAPTIVE FRAME

ADAPTIVE FRAME
Metrics\

BACKGROUND

BACKGROUND
SUBTRACTION

SUBTRACTION
Methods

DIFFERENCE

DIFFERENCE

DIFFERENCE

GMM

GRD
PRECISION 0.0625 0.0226 0.1225 0.0707 0.105 0.0473 0.0511 0.0362

RECALL 0.041 0.1287 0.0829 0.0923 0.076 0.0615 0.0643 0.1327

ACCURACY 0.97 0.8865 0.9719 0.9628 0.9724 0.9605 0.9607 0.9248

F-SCORE 0.0495 0.0385 0.0989 0.0801 0.0882 0.0535 0.057 0.0569

MSE 0.0166 0.0529 0.0158 0.0205 0.0153 0.0221 0.0213 0.0387

RMSE 0.129 0.23 0.1257 0.1431 0.1239 0.1488 0.1458 0.1967

FPR 0.0119 0.0999 0.0113 0.0217 0.0116 0.0229 0.0225 0.0615

FNR 0.959 0.8713 0.9171 0.9077 0.924 0.9385 0.9357 0.8673

PSNR 65.955 60.9316 66.1809 65.0508 66.3061 64.7143 64.8912 62.2877

TABLE II. EVALUATION METRICS OF DIFFERENT METHODS ON BASELINE VIDEO FROM CD NET DATASET
BACKGROUND SUBTRACTION AND

ADAPTIVE FRAME DIFFERENCE

SUBTRACTION AND ADAPTIVE

BACKGROUND SUBTRACTION

ADAPTIVE BACKGROUND

ADAPTIVE BACKGROUND

Metrics\
FRAME DIFFERENCE

FRAME DIFFERENCE

FRAME DIFFERENCE

Methods
SUBTRACTION

GMM

GRD
PRECISION 0.0339 0.0171 0.0366 0.04 0.0411 0.0234 0.0227 0.0609

RECALL 0.0568 0.2112 0.2123 0.1059 0.1066 0.0253 0.094 0.0681

ACCURACY 0.9585 0.7764 0.907 0.9441 0.9482 0.9662 0.9745 0.9662

F-SCORE 0.0425 0.0317 0.0625 0.0581 0.0594 0.0243 0.0366 0.0643

MSE 0.0323 0.0617 0.0672 0.0425 0.0407 0.0267 0.011 0.0255

RMSE 0.1798 0.2483 0.2593 0.206 0.2017 0.1635 0.1048 0.1596

FPR 0.0267 0.2136 0.0827 0.0421 0.0387 0.0177 0.0209 0.0182

FNR 0.9432 0.7888 0.7877 0.8941 0.8934 0.9747 0.906 0.9319

PSNR 63.0694 60.2637 59.8884 61.886 62.0724 63.8943 67.7608 64.1051

510 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

TABLE III. EVALUATION METRICS OF DIFFERENT METHODS ON DYNAMIC BACKGROUND VIDEO FROM CD NET DATASET

BACKGROUND SUBTRACTION AND

ADAPTIVE FRAME DIFFERENCE

SUBTRACTION AND ADAPTIVE

BACKGROUND SUBTRACTION

ADAPTIVE BACKGROUND

ADAPTIVE BACKGROUND

FRAME DIFFFERENCE
Metrics\

FRAME DIFFERENCE

FRAME DIFFERENCE
Methods

SUBTRACTION

GMM

GRD
PRECISION 0.23 0.0111 0.038 0.0104 0.0078 0.0635 0.013 0.0238

RECALL 0.0922 0.1207 0.0421 0.0765 0.1159 0.0153 0.013 0.3333

ACCURACY 0.9896 0.8929 0.9848 0.9224 0.8574 0.9891 0.7324 0.7972

F-SCORE 0.1316 0.0204 0.0399 0.0183 0.0146 0.0246 0.033 0.0444

MSE 0.0051 0.0337 0.0072 0.0341 0.0485 0.0053 0.004 0.005

RMSE 0.0711 0.1836 0.085 0.1846 0.2203 0.0727 0.025 0.0268

FPR 0.0027 0.0999 0.008 0.0696 0.1358 0.002 0.2676 0.1962

FNR 0.9078 0.8793 0.9579 0.9235 0.8841 0.9847 0.566 0.6667

PSNR 71.1247 62.8896 69.5814 62.8415 61.3044 70.9357 80.1485 79.6116

TABLE IV. EVALUATION METRICS OF DIFFERENT METHODS ON SHADOW VIDEO FROM CD NET DATASET
ADAPTIVE FRAME DIFFERENCE

SUBTRACTION AND ADAPTIVE

BACKGROUND SUBTRACTION

BACKGROUND SUBTRACTION

ADAPTIVE BACKGROUND

ADAPTIVE BACKGROUND
AND FRAME DIFFERENCE

FRAME DIFFFERENCE
FRAME DIFFERENCE

Metrics\
Methods
SUBTRACTION

GMM

GRD
PRECISION 0.1182 0.0517 0.0438 0.0117 0.0641 0 0.0081 0.022

RECALL 0.2734 0.159 0.2796 0.0686 0.0994 0 0.0342 0.0897

ACCURACY 0.9893 0.9832 0.9709 0.9663 0.9889 0.993 0.9737 0.9768

F-SCORE 0.165 0.0781 0.0757 0.0199 0.0779 0.078 0.0131 0.0353

MSE 0.0099 0.0156 0.0265 0.0303 0.0104 0.0066 0.0239 0.0214

RMSE 0.0993 0.1247 0.1626 0.1741 0.1022 0.0815 0.1546 0.1463

FPR 0.0079 0.0131 0.0261 0.0292 0.0069 0.002 0.0215 0.019

FNR 0.7266 0.841 0.7204 0.9314 0.9006 1 0.9658 0.9103

PSNR 68.2231 66.2475 63.9405 63.348 67.9736 69.942 64.3793 74.849

511 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

PRECISION F-SCORE
0.14 0.1225 0.12
0.105 0.0989
0.12 0.1 0.0882
0.1 0.0801
0.08 0.0625 0.0707 0.08
0.04730.0511 0.05350.0570.0569
0.06 0.0362 0.06 0.0495
0.0226 0.0385
0.04 0.04
0.02
0 0.02
0

Fig. 2. Precision of Different Methods on Thermal Data Set.

Fig. 5. F-Score of Different Methods on Thermal Data Set.

RECALL
0.14 0.1287 0.1327 MSE
0.12 0.06 0.0529
0.0923
0.1 0.0829 0.076 0.05
0.08 0.06150.0643 0.0387
0.04
0.06 0.041
0.03 0.0205 0.0221
0.0213
0.04 0.0166 0.0158 0.0153
0.02 0.02
0 0.01
0

Fig. 3. Recall of Different Methods on Thermal Data Set.

Fig. 6. MSE of Different Methods on Thermal Data Set.

ACCURACY
0.98 0.97 0.9719 0.9724
0.9628 0.9607
0.9605 RMSE
0.96 0.23
0.25
0.94 0.9248 0.1967
0.92 0.2
0.8865 0.1431 0.14880.1458
0.9 0.15 0.129 0.1257 0.1239
0.88
0.86 0.1
0.84 0.05
0

Fig. 4. Accuracy of Different Methods on Thermal Data Set. Fig. 7. RMSE of Different Methods on Thermal Data Set.

512 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

FPR PRECISION
0.12 0.0999 0.07 0.0609
0.1 0.06
0.08 0.0615 0.05 0.0411
0.06 0.0366 0.04
0.04 0.0339
0.04 0.0217 0.02290.0225 0.02340.0227
0.0119 0.0113 0.0116 0.03
0.02 0.0171
0 0.02
0.01
0

Fig. 8. FPR of Different Methods on Thermal Data Set.

Fig. 11. Precision of Different Methods on Base Line Video Frames.

FNR
0.98 0.959
0.96 0.93850.9357 RECALL
0.94 0.91710.90770.924 0.25
0.92 0.21120.2123
0.9 0.8713 0.2
0.88 0.8673
0.86 0.15 0.10590.1066 0.094
0.84 0.1 0.0568 0.0681
0.82 0.0253
0.05
0

Fig. 9. FNR of Different Methods on Thermal Data Set.

Fig. 12. Recall of Different Methods on Base Line Video Frames.

PSNR ACCURACY
67 65.955 66.1809 66.3061
66 65.0508 64.7143
64.8912 1.2
0.97450.9662
65
1
0.9585 0.9070.94410.94820.9662
64 62.2877
63 0.7764
62 60.9316 0.8
61 0.6
60
59 0.4
58
0.2
0

Fig. 10. PSNR of Different Methods on Thermal Data Set.

Fig. 13. Accuracy of Different Methods on Base Line Video Frames.

513 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

F-SCORE FPR
0.07 0.0625 0.0643 0.25 0.2136
0.05810.0594
0.06 0.2
0.05 0.0425
0.0366 0.15
0.04 0.0317 0.0827
0.03 0.0243 0.1
0.04210.0387
0.02 0.05 0.0267 0.01770.02090.0182
0.01 0
0

Fig. 17. FPR of Different Methods on Base Line Video Frames.

Fig. 14. F-Score of Different Methods on Base Line Video Frames.

FNR
MSE 1.2
0.9432 0.97470.9060.9319
0.08 1 0.8941
0.8934
0.0672 0.7888
0.7877
0.07 0.0617
0.8
0.06
0.05 0.04250.0407 0.6
0.04 0.0323 0.0267 0.0255 0.4
0.03
0.02 0.011 0.2
0.01 0
0

Fig. 18. FNR of Different Methods on Base Line Video Frames.

Fig. 15. MSE of Different Methods on Base Line Video Frames.

PSNR
RMSE 70 67.7608
0.3 0.2593 68
0.2483 66 63.0694 63.8943 64.1051
0.25 0.2060.2017 64 62.0724
61.886
0.2 0.1798 0.1635 0.1596 62 60.2637
59.8884
60
0.15 0.1048 58
0.1 56
54
0.05
0

Fig. 16. RMSE of Different Methods on Base Line Video Frames. Fig. 19. PSNR of Different Methods on Base Line Video Frames.

514 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

PRECISION F-SCORE
0.23 0.1316
0.25 0.14
0.12
0.2 0.1
0.15 0.08
0.06 0.0399 0.0444
0.1 0.04 0.0204 0.0183 0.0246
0.0635 0.0146
0.038 0.02 0
0.05 0.0111 0.01040.0078 0.0238 0
0
0

Fig. 23. F-Score of Different Methods on Dynamic Background.

Fig. 20. Precision of Different Methods on Dynamic Background.

MSE
RECALL 0.06
0.3333 0.0485
0.35 0.05
0.3 0.04 0.0337 0.0341
0.25 0.03
0.2 0.02
0.15 0.09220.1207 0.1159 0.0072 0.0053
0.0765 0.01 0.0051 7.17E-04
6.33E-04
0.1 0.0421
0.05 0.0153 0 0
0

Fig. 24. MSE of Different Methods on Dynamic Background.

Fig. 21. Recall of Different Methods on Dynamic Background.

RMSE
ACCURACY 0.25 0.2203
1.2 0.9896 0.9848 0.9891 0.2 0.1836 0.1846
1 0.8929 0.9224
0.8574 0.7972
0.7324
0.8 0.15
0.6 0.1 0.0711 0.085 0.0727
0.4
0.05 0.0250.0268
0.2
0 0

Fig. 22. Accuracy of Different Methods on Dynamic Background. Fig. 25. RMSE of Different Methods on Dynamic Background.

515 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

FPR PRECISION
0.3 0.2676 0.14 0.1182
0.25 0.1962 0.12
0.2 0.1
0.1358
0.15 0.0999 0.08 0.0641
0.1 0.0696 0.06 0.0517
0.0438
0.05 0.0027 0.008 0.002 0.04 0.022
0 0.02 0.0117 0.0081
0
0

Fig. 26. FPR of Different Methods on Dynamic Background.

Fig. 29. Precision of Different Methods on Shadow.

FNR
1.2
RECALL
0.9579 0.9847 0.2796
0.8793 0.9235
1 0.9078 0.8841 0.3 0.2734
0.8 0.6667 0.25
0.6 0.2 0.159
0.4 0.15 0.0994 0.0897
0.1 0.0686
0.2 0.0342
0 0.05
0 0
0

Fig. 27. FNR of Different Methods on Dynamic Background.

Fig. 30. Recall of Different Methods on Shadow.

PSNR ACCURACY
90 80.1485
79.6116
80 71.1247 69.5814 70.9357 1 0.993
62.8896 62.8415
61.3044 0.995 0.9893 0.9889
70 0.99 0.9832
60 0.985
50 0.98 0.9768
0.9709 0.9737
40 0.975
0.97 0.9663
30 0.965
20 0.96
10 0.955
0 0.95

Fig. 28. PSNR of Different Methods on Dynamic Background. Fig. 31. Accuracy of Different Methods on Shadow.

516 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

F-SCORE FPR
0.18 0.165 0.035 0.0292
0.16 0.03 0.0261
0.14
0.12 0.025 0.0215
0.019
0.1 0.0781
0.0757 0.0779 0.02
0.08 0.0131
0.06 0.0353 0.015
0.0199 0.01 0.0079 0.0069
0.04 0.0131
0.02 0 0.005 0.002
0
0

Fig. 32. F-Score of Different Methods on Shadow.

Fig. 35. FPR of Different Methods on Shadow.

MSE FNR
0.035 0.0303 1.2 1
0.03 0.0265
0.0239 1 0.841 0.9006 0.9658
0.9314 0.9103
0.025 0.0214
0.02 0.8 0.7266 0.7204
0.0156
0.015 0.0099 0.0104 0.6
0.01 0.0066 0.4
0.005 0.2
0 0

Fig. 33. MSE of Different Methods on Shadow. Fig. 36. FNR of Different Methods on Shadow.

RMSE PSNR
0.2 72
0.18 0.16260.1741 0.15460.1463 69.942
0.16 70 68.2231 67.9736
0.14 0.1247
0.12 0.0993 0.1022 68 66.2475
0.1 0.0815 66 64.849
64.3793
0.08 63.9405
63.348
0.06 64
0.04 62
0.02
0 60

Fig. 34. RMSE of Different Methods on Shadow. Fig. 37. PSNR of Different Methods on Shadow.

517 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 9, No. 9, 2018

VIII. CONCLUSION AND FUTURE WORK [9] D. Hari Hara Santosh, P. Venkatesh, P. Poornesh, L. Narayana Rao, N.
Arun Kumar “Tracking Multiple Moving Objects Using Gaussian
In this article, a model based on Generalized Rayleigh Mixture Model” in International Journal of Soft Computing and
Distribution (GRD) is presented for the tracking of video Engineering (IJSCE) ISSN: 2231-2307, Volume-3, Issue-2, May 2013
images. The model is compared with that of the existing [10] Huang, Guo-Hao, and Chun-Rong Huang. "Binary invariant cross color
models based on GMM using the metrics Precision, Recall, descriptor using galaxy sampling." ICPR, pp. 2610-2613, 2012
Accuracy, F-Score, MSE ,RMSE ,FNR, FPR, PSNR. The [11] Hedayati, M., Wan Mimi Diyana Wan Zaki, and Aini Hussain. "A
results are derived and presented in the table -I to table -IV qualitative and quantitative comparison of real-time background
subtraction algorithms for video Surveillance applications." Journal of
and Figures 1-37. From the tables,, it can be clearly observed Computational Information Systems 8.2, pp. 493-505, 2012.
that the proposed model perform well with respect to all the [12] Goyette, N., Jodoin, P.-M., Porikli, F., onrad, J., Ishwar,
parameters and the results , when compared to the existing P.:changedetection.net: a new change detection benchmark dataset. In:
model based on the GMM, showcases better performance Proceedings of IEEEWorkshop on Change Detection at CVPR-2012, pp.
accuracy. This method can be applied to the practical 16–21 (2012)
situations, in particular to the areas of medical imaging, in [13] Al-Najdawi N., Tedmori S., Edirisinghe E., and Bez H., “An Automated
particular situations, where the patient tracking is necessary. Real-Time People Tracking System Based on KLT Features Detection,”
The International Arab Journal of Information Technology, vol. 9, no. 1,
The model developed can be further extended in a distributed 2012.
network scenario, where the images in a distributed [14] Bouwmans, T.: Recent advanced statistical background modeling for
environment are to be tracked. This methodology leverages foreground detection: a systematic survey. Recent Pat. Comput.Sci. 4(3),
the data intensive frameworks in distributed environment. 147–176 (2011)
REFERENCES [15] Stephan Kopf, Fleming Lampi, Thomas King, and Wolfgang Effelsberg,
“Automatic scaling and cropping of videos for devices with limited
[1] Xia, Haiying, Shuxiang Song, and Liping He. "A modified Gaussian screen resolution,” in MULTIMEDIA ‟06: Proceedings of the 14th
mixture background model via spatiotemporal distribution with shadow annual ACM international conference on Multimedia, New York, NY,
detection. “Signal, Image and Video Processing 10.2, pp. 343-350, 2016. USA, 2006, pp. 957–958, ACM.
[2] H.Zhou, G.Su and X.Jiang, ―”Dynamic foreground detection based on [16] Tao Mei, Xian-Sheng Hua, He-Qin Zhou, Shipeng Li, and Hong-Jiang
improved Codebook model” The Imaging Science Journal, Vol. 64, Zhang, “Efficient video mosaicing based on motion analysis,” Proc.
Issue 2, January 2016. ICIP 2005, vol. 1, 2005
[3] A. Viswanath, R. Kumari Behra, V. Senthamilarasu, K. Kutty, [17] D. Farin, P. de with, and Wolfgang Effelsberg, “Video-object
―”Background Modeling from a Moving Camera”, Elsevier Second segmentation using multi-sprite background subtraction,” in Conference
International Symposium on Computer Vision and the Internet, Vol. 58, on Multimedia and Expo (ICME), 2004.
pp. 289-296, 2015
[18] Chris Stauffer and W.E.L. Grimson, “Adaptive background mixture
[4] Luo, Yuhan, and Hong Zhang. "Sparse Learning for Robust Background models for real-time tracking,” Proceedings of theIEEE Computer
Subtraction of Video Sequences." International Conference on Society Conference on Computer Vision and Pattern Recognition, vol. 2,
Intelligent Computing. Springer International Publishing, 2015. pp. 246–252, 1999
[5] Yang, Lu, et al. "Pixel-to-model background modeling in crowded [19] Michael Unger, Mark Asbach, and Peter Hosten “Enhanced background
scenes."2014 IEEE International Conference on Multimedia and Expo subtraction using global motion compensation and mosaicing”, 978-1-
(ICME). IEEE, 2014 4244-1764-3/08/ IEEE pp.2708-2711, 2008.
[6] Chen Z, Ellis T (2014) “A self-adaptive Gaussian mixture [20] Nagesh Vadaparthi, Srinivas Y, P Suresh Varma, P S Sitharama Raju,
model.”Comput Vis Image Underst122:35- 46.doi:10.1016/ j. cviu. “Hierarchical Clustering technique for Medical Image Segmentation
2014. 01.004 Based on Finite Skew Gaussian Mixture Model and EM Algorithm”,
[7] Chien, Shao-Yi, et al. "Video object segmentation and tracking The International Arab Journal of Information Technology, 15(5),
framework with improved threshold decision and diffusion distance." September 2016
IEEE Transactions on Circuits and Systems for Video Technology 23.6, [21] N. Otsu;” A threshold selection method from gray-level histograms”.
pp. 921-934, 2013. Automatica, 1975, Vol. 11, Pg. 23- 27.
[8] Liu, Wan-Chen, et al. "Real-time binary descriptor based background [22] W. Jun-Qin. 2012; “An adaptive frame difference method for human
modeling." 2013 2nd IAPR Asian Conference on Pattern Recognition. tracking”. Advances in information sciences & services sciences, 2012,
IEEE, 2013. Vol. 4, No. 1, Pg. 381-387

518 | P a g e
www.ijacsa.thesai.org