Object Tracking By Online Discriminative Feature Selection Algorithm

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 12 | Dec-2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 717
Object Tracking By Online Discriminative Feature Selection Algorithm
Ms.Ashwini Narake1 , Mrs.Pradnya Kharat2, Ms. Rupali Ghodake3
1,3 Lecturer M.E.(ENTC), Dept. of Electronics and Telecommunication Engineering ,Agppi college, solapur,
Maharashtra, India.
2Lecturer M.E.(Digital system), Dept. of Electronics and Telecommunication Engineering ,Agppi college, solapur
Maharashtra, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Most tracking-by-detection algorithms train
discriminative classifiers to separate target objects from their
surrounding background. In this setting, noisy samples are
likely to be included when they are not properly sampled,
thereby causing visual drift. The multiple instances learning
(MIL) learning paradigm has been recentlyappliedtoalleviate
this problem. However, important prior information of
instance labels and the most correct positive instance (i.e., the
tracking result in the current frame) can be exploited using a
novel formulation much simpler than an MIL approach.Inthis
paper, it shows that integrating such prior information into a
supervised learning algorithm can handle visual drift more
effectively and efficiently than the existing MIL tracker. It
present an online discriminative feature selection algorithm
which optimizes the objective function in the steepest ascent
direction with respect to the positive samples while in the
steepest descent direction with respect to the negative ones.
Therefore, the trained classifier directly couples its score with
the importance of samples, leading to a more robust and
efficient tracker. Numerous experimental evaluations with
state-of-the-art algorithms on challenging sequences
demonstrate the merits of the proposed algorithm.
Keywords: steepest ascent, steepest descent, optimize,
Robust, trained classifier.
1. Introduction
Object tracking has been extensively studied in
computer vision due to its importance in applications such
as automated surveillance, video indexing, traffic
monitoring, and human-computerinteraction,tonameafew.
While numerous algorithmshave been proposed during the
past decades, it is still a challenging task to build a robust
and efficient tracking system to dealwithappearancechange
caused by abrupt motion, illumination variation, shape
deformation, and occlusion .It has been demonstrated that
an effective adaptive appearance model plays an important
role for object tracking .
A simple and effective online discriminativefeatureselection
(ODFS) approach which directly couples the classifier score
with the sample importance, thereby formulating a more
robust and efficient tracker than state-of-the-art algorithms
and 17 times faster than the MIL Track method (both are
implemented in MATLAB).
It is unnecessary to use bag likelihood loss functions for
feature selection as proposed in the MIL Track method.
Instead, it can directly select featureson theinstancelevelby
using a supervised learning method which is more efficient
and robust than the MIL Track method. As all the instances,
including the correct positive one can be labeled from the
current classifier, they can be used for update via self-taught
learning. Here, the most correct positive instance can be
effectively used asthe tracking result of the current framein
a way similar to other discriminative models.
2. Problem Statement
Most tracking by detection algorithms train discriminative
classifier to separate target object from their surrounding
background .In this setting ,noisy samples are likely to be
included when they are not properly sampled their by
causing visual drift . Our proposed work is to;
 Design an algorithm to eliminate this problem.
 To analyze our tracker performancewithnumerous
experimental results and evolutions on challenging
video or image sequences in terms of efficiency,
accuracy and robustness.
3. Theoretical Framework
Block diagram of tracking:
Fig3.1: Block Diagram of Tracking
Sample
Set of
Image
patches
Track
Location
Feature
Extractio
n
Select
Feature
Apply
classifier to
each feature
vector
Apply
ODFS
algorith
m
Update
classifier
parameters

3.1 Tracking by Detection:
Figure illustrates the basic flow of algorithm. The
discriminative appearance model is based a classifier which
estimates the posterior probability given a classifier, the
tracking by detection process is asfollows.Letthelocationof
sample at frame. The object location where assume the
corresponding sample and then densely crop some patches
within a search radius centering at the current object
location and label them aspositive samples. Then, randomly
crop some patches from set and label them as negative
samples. It utilizes these samples to update the classifier.
When the frame arrives, it crop some patches with a large
radius surrounding the old object location frame.
Next, apply the updated classifiertothesepatchesto
find the patch with the maximum confidence. The locationis
the new object location in the frame. Based on the newly
detected object location, tracking system repeats the above
mentioned procedures.
 Assume the‘t’ frame arrives crop some patches at
radius alpha (α) is considered as positive samples.
 Again we crop some patches randomly from set
where radius is α<ς<β is considered as negative
samples.
 α= 4ς= [2α] &β=[1.5α] for t-th frame.
 Next we again crop some patches at radius
ς (gamma)= 25 when (t+1)th frame arrives.
 Like this generate two set of image patches.
 Find the tracking location with help of It (x) € R2i.e
by calculating radius of 4 pixel.
3.2 Haar like feature extraction:
Fast computation of Haar-like features: One of the
contributions of Viola and Jones was to use summed area
tables, which they called integral images.Integralimagescan
be defined as two-dimensional lookup tables in the formofa
matrix with the same size of the original image. Each
element of the integral image contains the sum of all pixels
located on the up-left region of the original image (in
relation to the element's position). This allows computing
sum of rectangular areasin the image at anypositionorscale
using only four lookups:
Fig3.2: Finding the sum of the shaded rectangular area
Sum=I(C) +I (A)-I (B)-I (D).
Where points A, B,C,D belong to the integralimageI,
as shown in the figure. Each Haar-like feature may need
more than four lookups, depending on how it was defined.
Viola and Jones's 2-rectangle features need six lookups, 3-
rectangle features need eight lookups, and 4-rectangle
featuresneed nine look ups.Lien hart and Maydt introduced
the concept of a tilted (45°) Haar-like feature. This was used
to increase the dimensionality of the set of features in an
attempt to improve the detection of objects in images. This
wassuccessful, as some of these featuresare abletodescribe
the object in a better way. For example, a 2-rectangle tilted
Haar-like feature can indicate the existence of an edge at
45°.Messom and Barczak extended the idea to a generic
rotated Haar-like feature. Although the idea sounds
mathematically sound, practical problemspreventedtheuse
of Haar-like features at any angle. In order to be fast
detection algorithms use low resolution images causing
rounding errors. For this reason rotated Haar-like features
are not commonly used.
3.3 Classifier Construction and Update:
In this sample is represented by a feature vector
where each feature is assumed to be independently
distributed as MILTrack and then the classifier can be
modeled by A Naive Bayes classifier is a weak classifier with
equal prior. Next, the classifier is a linear function of weak
classifiers and uses a set of Haar-like features (15) to
represent samples. The conditional distributions and in the
classifier are assumed to be Gaussian distributed as the MIL
Track method with four parameters. The parameters are
incrementally estimated and N is the number of positive
samples. In addition update and with similar rules. It can be
easily deduced by maximum likelihood estimation method
where learning rate to moderate the balance between the
former frames and the current one.It should be noted that
parameter update method is different from that of the MIL
Track method and it can be update equations are derived
based on maximum likelihood estimation. For online object
tracking, a feature pool with M > K featuresis maintained.As
demonstrated in online selection of the discriminative
features between object and background can significantly
improve the performance of tracking. The objective is to
estimate the sample with the maximum confidence from as
with K selected features. However, directly select K features
from the pool of M features by using a brute force method to
maximize the computational complexity with combinations
is prohibitively high (set K = 15 and M = 150 in experiments)
for real-time object tracking. An efficient online
discriminative feature selection method which is a
sequential forward selection method where the number of
feature combinations is MK, thereby facilitating real-time
performance.
Hk=log( k
k=1p(fk(x)|y=1)P(y=1))/
( k
k=1p(fk(x)|y=0)P(y=0))=∑k
k=1 (x),
Where
=log ((p (fk(x) |y=1)/ (p (fk(x) |y=1)),

 Where p is posterior probability of feature vector
f(x).
 Y= binary variable of gray scale image having value
[0,1] .i.e. 0= black ,1=white it is y axisof imagepatch
 Weal classifier is phi of k.
It is estimate the posterior probability i.e. confidence map
function of feature vector
C(x) =P(y=1|X) = (hk(X),
 X=sample set, Sigma=sigmoid function.
(Z)=1/1+e-Z
 Confidence map function is Gaussian distributed .It
is plot of expected value to the variance.
3.4 Bag Likelihood with Noisy-OR Model:
The instance probability of the MIL Track methodis
modeled by i indexes the bag and j indexes the instance in
the bag, and is a strong classifier. The weak classifier is
computed by and the bag probability based on the Noisy-OR
model. The MIL Track method maintains a pool of M
candidate weak classifiers and selects K weak classifiers
from this pool in a greedy manner using the following
criterion. Each weak classifier is composed of a featureisthe
bag likelihood function and it is a binary label.TheselectedK
weak classifiers construct the strong classifier as .The
classifier is applied to the cropped patches in the new frame
to determine the one with the highest response as the most
correct object location. It show that it is not necessary to use
the bag likelihood function based on the Noisy-OR modelfor
weak classifier selection and it can select weak classifiersby
directly optimizing instance probability via a supervised
learning method as both the most correct positive instance
(i.e. the tracking result in current frame) and the instance
labels are assumed to be known.
3.5 Principle of ODFS:
The confidence map of a sample being the target is
computed, and the object location is determined by thepeak
of the map. Providing that the sample space is partitioned
into two regions it defines a margin as the average
confidence of samples in minus the average confidence of
samples. In the training set, assume the positive set consists
of N samples and the negative set is composed of L samples.
Each sample is represented by a feature vector. A weak
classifier pool is maintained using objective is to select a
subset of weak classifiersfrom the poolwhichmaximizesthe
average confidence of samples in while suppressing the
average confidence of samples. Therefore, maximize the
margin function use a greedy scheme to sequentially select
one weak classifier from the pool to maximize. A classifier
constructed by a linear combination of the first weak
classifiers. Note that it is difficult to find a closed form
solution of the objective function in Furthermore,althoughit
is natural and easy to directly select that maximizes
objective function in the selected is optimal only to the
current samples, which limits its generalization capability
for the extracted samples in the new frames. An approach
similar to the approach used in the gradient boosting
method to solve which enhances the generalization
capability for the selected weak classifiers. The steepest
descent direction of the objective function of in the (N+L)
dimensional data space at the inverse gradient (i.e., the
steepest descent direction) of the posterior probability
function, its generalization capability is limited.
Friedman proposes an approach to select that
makes most parallel to when minimizing objective function
in.The selected weak classifier ismost highlycorrelatedwith
the gradient over the data distribution,therebyimprovingits
generalization performance. In this work, instead selectthat
is least parallel to as maximize the objective function. Thus,
choose the weak classifier with the followingcriterionwhich
constrains the relationship between Single Gradient and
Single weak Classifier (SGSC) output for each sample.
However, the constraint between theselectedweakclassifier
and the inverse gradient direction is still too strong in
because is limited to the small pool.
In addition, both the single gradient and the weak
classifier output are easily affected by noise introduced by
the misaligned samples, which may lead to unstable results.
To alleviate this problem, relax the constraint and with the
Average Gradient and Average weak Classifier (AGAC)
criteria in a way similar to the regression tree method i.e.,
take the average weak classifier output for the positive and
negative samples, and the average gradientdirectioninstead
of each gradient direction for every sample. However, this
pooled variance is easily affected by noisy data or outliers.
Which means the selected weak classifier tendsto maximize
while suppressing the variance thereby leading to more
stable results. In this a small search radiusis adoptedtocrop
out the positive samples in the neighborhood of the current
object location, leading to the positive samples with very
similar appearances. Therefore, the ODFS criterionbecomes
it is worth noting that the average weak classifier output
computed from different positive samples alleviates the
noise effects caused by some misaligned positive samples.
Moreover, the gradient from the most correct positive
sample helpsselect effective features that reducethesample
ambiguity problem. In contrast, other discriminativemodels
that update with positive features from only one positive
sample are susceptible to noise induced by the misaligned
positive sample when drift occurs. If only one positive
sample (i.e., the tracking result) is used for feature selection
in this method, the single positive feature selection (SPFS)
criterion it present experimental results to validate why the
proposed method performs better than the one using the
SPFS criterion. When a new frame arrives, it updates all the
weak classifiers in the pool in parallel and select K weak
classifiers sequentially from using the criterion.

3.6 ODFS algorithm:
 Extract the features with these two sets of samples
by the ODFS algorithm.
 ODFS algorithmuse center location error in pixelto
quantitatively compare which is calculated by the
Gaussian distribution plot of frame per second to
the position of error pixel.
Fig3.5: Gaussian distribution plot of frame per second
to the position of error pixel.
3.7 Relation to Bayes Error Rate:
In this the optimization problem in is equivalent to
minimizing the Bayes error rate in statistical classification.
The Bayes error rate is the class conditional probability
density function and describes the prior probability. In
experiments, the samples in each set are generated with
equal probability i.e., maximizing the proposed objective
function is equivalent to minimizing the Bayes error rate.
4. Architectural Design:
4.1 Algorithm:
Algorithm1: Object tracking algorithm:
1. Input: video frame.
2. Sample a set of image patches.
3. Track the location at the frame.
4. Extract feature for each sample.
5. Apply classifier to each feature vector.
6. Find tracking location.
7. Sample two set of image patches.
8. Extract feature with these two sets of samples by
the ODFS algorithm and update classifier
parameters.
9. Output: tracking location and classifier parameter.
Algorithm2: Feature selection algorithm:
1. Input: dataset.
2. Update weak classifier pool.
3. Update the average weak classifier outputs.
4. Update inverse gradient.
5. Correlate gradient and classifier.
6. Calculate average weak classifier output.
7. Normalize classifier.
8. Output : strong classifier and confidence map
function.
ODFS tracker selects 15 features for classifier
construction which is much more efficient than the MIL
Track method that sets K = 50. The number of candidate
features M in the feature pool is set to 150, which is fewer
than that of the MIL Track method (M = 250).Use radius of 4
pixels for cropping the similar positive samples in each
frame and generate 45 positive samples. A large can make
positive samples much different which may add more noise
but a small generates a small number of positive samples
which are insufficient to avoid noise. The inner and outer
radii for the set that generates negative samples are set.
Set the inner radius larger than the radius to reduce the
overlaps with the positive samples which can reduce the
ambiguity between the positive and negative samples. We
use the same generalized Haar-like features as which can be
efficiently computed using the integralimage.Eachfeaturefk
is a Haar-like feature computed by the sum of weighted
pixels in 2 to 4 randomly selected rectangles. For
presentation clarity, in we show theprobabilitydistributions
of three selected features by our method. The positive and
negative samples are cropped from a few frames of a
sequence. The results show that a Gaussiandistributionwith
an online update using is a good approximation of the
selected features. As the proposedODFStrackerisdeveloped
to addresses issues of MIL based tracking methods. We
evaluate it with the MIL Track on 16 challenging video clips,
among which 14 sequences are publicly available and the
others are collected on our own. In addition, seven other
state-ofthe-art learning based trackers are also compared.
For fair evaluations, we use the original source or binary
codes in which parameters ofeach method are tunedforbest
performance. The 9 trackerswe compare with are: fragment
tracker (Frag), online AdaBoost tracker (OAB) ,Semi-
Supervised Boosting tracker(SemiB) ,multiple instance
learning tracker (MILTrack) Tracking-Learning
detection(TLD)method, Struck method 1-tracker visual
tracking decomposition(VTD) method and compressive
tracker (CT) .We fix the parameters of the proposed
algorithm for all experiments to demonstrate its robustness
and stability. Since all the evaluated algorithmsinvolvesome
random sampling except we repeat the experiments 10
times on each sequence, and present the averaged results.
Implemented in MATLAB, our tracker runs at 30 frames per
second (FPS) on a Pentium Dual-Core 2:10 GHz CPU with
1:95 GB RAM.
Our source codes and videos are available at We use a
radius of 4 pixels for cropping the similar positivesamplesin
each frame and generate 45 positive samples. A large can
make positive samples much different which may add more
noise but a small generates a small number of positive
samples which are insufficient to avoid noise. The inner and

outer radii for the set X that generates negative samples are
set as 8 and 38, respectively. Note that we set the inner
radiuslarger than the radiusto reduce the overlapswith the
positive samples, which can reduce the ambiguity between
the positive and negative samples. Then, we randomlyselect
a set of 40 negative samples from the set X, which is fewer
than that of the MIL Track method (where 65 negative
examples are used). Moreover, we do not need to utilize
many samples to initialize the classifier whereas the MIL
Track method uses 1000 negative patches. The radius for
searching the new object location in the next frame is set as
= 25 that is enough to take into account all possible object
locations because the object motion between two
consecutive frames are often smooth, and 2000 samplesare
drawn, which is the same as the MIL Track method.
Therefore, this procedure is time-consuming if we use
more features in the classifier design. Our ODFS tracker
selects 15 features for classifier construction which is much
more efficient than the MIL Track method that sets K = 50.
The number of candidate features M in the feature poolisset
to 150, which is fewer than that of the MIL Track method (M
= 250). We note that we also evaluate with the parameter
settings K = 15; M = 150in the MILTrack method but find it
does not perform well for most experiments. The learning
parameter can be set as 0.80 to 0.95. A smaller learning rate
can make the tracker quickly adapts to the fast appearance
changes and a larger learning rate can reduce the likelihood
that the tracker drifts off the target. Good results can be
achieved by fixing 0.93in our experiments.
4.2 Experimental set up and results
All of the test sequencesconsist of gray-levelimages
and the ground truth object locations are obtained by
manual labels at each frame. We use the center location
error in pixels as an index to quantitatively compare 10
object tracking algorithms. In addition, we use the success
rate to evaluate the tracking results This criterion is used in
the PASCAL VOC challenge and the score is defined as score
= area (Intersection)/area (G union T), where G is the
ground truth bounding box and T is the tracked bounding
box. If score is larger than 0.5 in one frame, then the result is
considered a success. Shows the experimental results in
terms of center location errors, and presents the tracking
results in terms of success rate. Our ODFS based tracking
algorithm achieves the best or second best performance in
most sequences, both in terms of success rate and center
location error. Furthermore, the proposed ODFS based
tracker performs well in termsof speed (only slightlyslower
than CT method) among all the evaluated algorithms on the
same machine even though other trackers (except for the
TLD, CT methods and `1-tracker) are implemented in C or
C++ which is intrinsically more efficient than MATLAB. We
also implement the MIL Track method in MATLAB which
runs at 1.7 FPS on the same machine. Our ODFS-based
tracker (at 30 FPS) is more than 17 timesfaster than theMIL
Track method with more robust performance in terms of
success rate and center location error. The quantitative
results also bear out the hypothesis that supervisedlearning
method can yield much more stable and accurate results
than the greedy feature selection method used in the MIL
Track algorithm as we integrate known prior (i.e., the
instance labels and the most correct positive sample) into
the learning procedure.
4.3 Flow chart:
Fig 4.1: Flow chart of ODFS algorithm
6. Results:
6.1. Result analysis table of all types of databases.
Table of success rate:-
The table6.1 gives the result for different database like
David, Panda, Tiger and My database (real time database).
The result is calculated for every database in terms of
success rate. And it also gives average success rate of all
databases.

Database Type
Success Rate in
percentage%
David 88.62
Panda 82.75
Tiger 88.75
Mydatabase 85.62
Average success
rate 86.44
Table 6.1: Result analysis of all type databases.
Analysis table:-
The table 6.2 gives the success rate of all databases. In the
table the effect like brightness, saturation, contrast and
Noisy is added to database. The effect addedis10%and90%
i.e. minimum and maximum level of effect is applied and
under this all effect result of database is calculated in terms
of success rate given in table.
Database
type
Effect applied Success rate in
percentage %
Brightness 10% 85.5
Brightness 90% 81.62
Contrast 10% 85.75
My
database Contrast 90% 87.37
Saturation 10% 87.25
Saturation 90% 85
Noisy 86.87
David
Contrast 10% 75.50
Contrast 90% 88.12
Noisy 83.54
Panda
Brightness 10% 84
Contrast 10% 79.37
Contrast 90% 78.75
Saturation 90% 82
Noisy 83.87
Table 6.2: Result analysis of all type databases with
applied effect.
Changed in predefined parameter:-
The table 6.3 gives the changed predefined parameterwhich
is set fixed for standard database. The fixed parameter is
varied from maximum to minimum. Here only one database
is used i.e. (David database). And the parameter of database
is varied and success rate is calculated for each and every
varied parameter.
Database
type
Parameter
name
Values Range of
values
Success
rate (%)
David Learning
rate
Predefined
value
0.93 88.62
Minimum
value
0.61 87.12
Maximum
value
0.90 88.25
David Image size Predefined
value
320x240 88.62
Minimum
value
314x235 89.75
Maximum
value
620x480 90.25
David Database
value
Predefined
value
[120 55
75 95]
88.62
Minimum
value
[100 50
70 90]
91.87
Maximum
value
[200 100
90 100]
98.50
David Number of
feature
pool
Predefined
value
150 88.62
Minimum
value
50 78.50
Maximum
value
200 89
David Number
selected
feature
Predefined
value
15 88.62
Minimum
value
10 78.25
Maximum
value
50 92
David Searching
window
size
Predefined
value
25 88.62
Minimum
value
15 87.37
Maximum
value
50 92
Table 6.3: Result analysis of all type databases with
changed parameter.
Result:- The result displays the GUI windows for different
database.
Fig 6.1: The GUI windows for different database.

The figure 6.1 gives the graphical user interface of object
tracking. With help of this GUI we can combine &displays all
the result like tracking, plot & parameter calculation.GUI
helpsus to displays result combinely. The figure 6.1 gives18
pushbotton which help us to call all the respected file and
display the result with help of GUI.
Result:- The result displays the tracking ,plot and
parameter value for david database.
#462
Fig 6.2: A tracking result of David database.
The figure 6.2 gives tracking result of David database after
complete tracking. The red rectangle indicated as searching
window. 462# indicates number of iteration. The tracking
indicates with help of red window.
Fig 6.3: The plot of object tracking parameter.
The figure 6.3 gives tracking result of David database after
complete tracking. The tracking result is displayed in terms
of plot. The figure 6.3 gives plot of positive feature,
histogram of positive feature, negative feature, histogramof
negative feature, maximum probability, histogram of
maximum probability, classifier parameter plot and
histogram of classifier.
Fig 6.4: The tracking parameter value for david
database.
The figure 6.4 gives parameter value like number of feature
pool, number of selected feature pool , initial state i.e value
of grounding box, area , score value calculated by PASCAL
VOC method and successrate of david database. In figure6.4
total time of execution is also displayed.
Result:- The result displaysthe tracking ,plotandparameter
value for panda database.
#31
Fig 6.5: A tracking result of Panda database.
The figure 6.5 gives tracking result of Panda database after
complete tracking. The red rectangle indicated as searching
window. 31# indicates number of iteration. The tracking
indicates with help of red window.

The figure 6.6 gives tracking result of Panda database after
complete tracking. The tracking result is displayed in terms
of plot. The figure 6.6 gives plot of positive feature,
histogram of positive feature, negative feature, histogramof
negative feature, maximum probability, histogram of
maximum probability, classifier parameter plot and
histogram of classifier.
Fig 6.7:The tracking parameter value for Panda
database.
The figure 6.7 gives parameter value like number of feature
pool, number of selected feature pool, initial state i.evalueof
grounding box, area , score value calculated by PASCAL VOC
method and success rate of Panda database. In figure 6.7
total time of execution is also displayed.
Result:- The result displaysthe tracking ,plotandparameter
value for realtime database.
#200
Fig 6.8: A tracking result of Real time database (My
database).
The figure 6.8 gives tracking result of Realtimedatabase(My
database) after complete tracking. The red rectangle
indicated as searching window. 200# indicates number of
iteration. The tracking indicates with help of red window.
The figure 6.9 givestracking result of Realtimedatabase(My
database) after complete tracking. The tracking result is
displayed in terms of plot. The figure 6.9 gives plot of
positive feature, histogram of positive feature, negative
feature, histogram of negative feature,maximumprobability,
histogram of maximum probability, classifierparameterplot
and histogram of classifier.
Fig 6.10:The tracking parameter value for Real time
database (My database)
The figure 6.10 givesparameter value likenumberoffeature
pool, number of selected feature pool, initial state i.evalueof
grounding box, area , score value calculated by PASCAL VOC
method and success rate of Real time database (My
database).In figure 6.10 total time of execution is also
displayed.
7. Advantages & Application:
7.1 Advantages:
 Scale and pose variation .
 Heavy occlusion avoided.
 Abrupt motion, rotation and blur.
 Cluttered background and abrupt camera shake.
 Large illumination change.

7.2 Application:
 Automated surveillance.
 Video indexing.
 Traffic monitoring.
8. Conclusion:
In online discriminative features selection (ODFS) method
for object tracking which couples the classifier score
explicitly with the importance of the samples. ODFS method
selects features which optimize the classifier objective
function in the steepest ascent direction with respect to the
positive samples while in steepest descent direction with
respect to the negative ones. This leadsto a more robustand
efficient tracker without parameter tuning. The tracking
algorithm achieves real time performance with MATLAB
implementation on a Pentium dual-core machine.OurODFS-
based tracker is faster than the other tracking method with
more robust performance in terms of success rate. The
tracker achieves favorable performance when compared
with several state of the art algorithms.
References:
1. “Robust online appearance models for visual
tracking,” IEEE Trans. Pattern Anal. Mach. Intell.A
Jepson, D. Fleet, and T. El-Maraghi,vol.25,no.10,pp.
1296–1311, 2003.
2. “Support vector tracking,” IEEETrans.PatternAnal.
Mach Intell.S. Avidan, vol. 29, no. 8, pp. 1064–1072,
2004.
3. “Online selection of discriminative tracking
features,” IEEE Trans. Pattern Anal. Mach. Intell.,R.
Collins, Y. Liu, and M. Leordeanu, vol. 27,no. 10, pp.
1631–1643, 2005.
4. “Real-time tracking via online boosting,” In British
Machine Vision Conference,H. Grabner, M. Grabner,
and H. Bischof, pp. 47–56, 2006.
5. “Semi-supervised on-line boosting for robust
tracking,” In Proc. Eur. Conf. Comput.Vis.H.Grabner,
C. Leistner, and H. Bischof, pp. 234–247, 2008.
6. “Robust object tracking with online multiple
instance learning,” IEEE Trans. Pattern Anal. Mach.
Intell. B. Babenko, M.-H. Yang, and S. Belongie, vol.
33, no. 8, pp. 1619–1632, 2011.
7. “Real-time compressivetracking,”InProc.Eur.Conf.
Comput.K. Zhang, L. Zhang, and M.-H. Yang, Vis.,
2012.
8. “Real-time visual tracking via online weighted
multiple instance learning,” Pattern Recognition,K.
Zhang and H. Song, vol. 46, no. 1, pp. 397–411,
2013.
9. “Online selection of discriminative features using
bayes error rate for visual tracking,” Advances in
Multimedia Information Processing-PCM 2006,D.
Liang, Q. Huang, W. Gao, and H. Yao, pp. 547–555,
2006. 6
10. “Online selecting discriminative tracking features
using particle filter,” in Proc. IEEEConf.Comput.Vis.
Pattern Recognit.Y. Wang, L. Chen, and W. Gao, vol.
2, pp. 1037–1042, 2005.
11. “Gradient feature selection for online boosting,” in
Proc. Int. Conf. on Comput. Vis.,X. Liu and T. Yu, pp.
1–8, 2007.
12. “Robust object tracking via active feature selection
”Kaihua Zhang, Lei Zhang ,Member ,IEEE, Ming-
Hsuan Yang ,Senior Member ,IEEE, and Member,
IEEE, Qinghua Hu.11, November 2013. 1957.

Object Tracking By Online Discriminative Feature Selection Algorithm

More Related Content

What's hot (18)

Similar to Object Tracking By Online Discriminative Feature Selection Algorithm (20)

More from IRJET Journal (20)

Recently uploaded (20)

Object Tracking By Online Discriminative Feature Selection Algorithm