IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
International Journal of Research in Information Technology (IJRIT)
www.ijrit.com
ISSN 2001-5569
Object Tracking Based On Illumination Invariant Method and Kalman Filter Himani Gupta1, Rohit Garg2 1
M.Tech, Electronics and Communication Engineering Department SKIET, Kurukshetra
[email protected] 2 Sr.Lecturer, Electronics and Communication Engineering Department SKIET, Kurukshetra
[email protected]
ABSTRACT: In computer vision application, object detection is fundamental and most important steps for any video analysis. Although several works aimed at detecting and tracking of objects in video sequences have been reported but due to fast illumination change in a visual surveillance system, many are not suitable to fast illumination changes or dynamic background. The errors occur in the detection stage which can be removed by a combination of illumination invariant method with Kalman filter for object extraction. For illumination invariance, normalized DCT is used as it can clearly differentiate between luminance and reflectance. There is another prominent problem of shadow effect which can be solved by using some nonparametric methods. A study has been carried out to know the merits and demerits of the methods or techniques used in detection and tracking the objects in video frames. Keyword: Background subtraction, Tracking, Illumination, Image segmentation, DCT.
I. Introduction Video surveillance systems have long been in use to monitor security sensitive area and history of video surveillance consists of three generations. Object detection is used in video surveillance and commonly used techniques for object detection are background subtraction, statistical models, temporal differencing and optical flow. Public and commercial security, Smart video data mining, Law enforcement, Military security are some scenarios that handle the smart surveillance systems. 1.1 Object Detection Distinguishing foreground objects from the stationary background is both a significant and difficult research problem. It is commonly used in video surveillances, vehicle auto-navigation, motion capture in sports, child care applications and many more[1]. It is a general idea that if an object is changing its position with respect to a point in the space, then it is considered to be moving. Rest scene is said to be the background[2]. Almost the entire visual surveillance systems first step is detecting foreground objects. This creates a focus of attention for higher processing levels such as tracking, classification and behaviour understanding and reduces computation time considerably. There are two major issues in visual target tracking system and they are :Variations in light illumination Misinterpretation of shadow as background
Himani Gupta, IJRIT
57
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
To solve these issues some nonparametric methods are used and a new algorithm developed. Following points are the thrust areas during the object tracking:It has been found that illuminance invariant method can detect the change in illumination but limitations are it is not adaptive for handling the scaling and orientation of the target and [3] many algorithms are available to track the moving object in static and dynamic conditions but occlusion which can be overcome by the use of Kalman filter. So a combination of illumination invariant method with Kalman filter for object extraction is used. False background detection can be due to illumination variation. Intensity of Light on human body can be varied depending upon the angle of motion. Darker area of human body can be misinterpreted as background and results in false object detection. To avoid this false detection an intensity illumination invariant method[4] based on discrete cosine transform is proposed along with Kalman filter. The Kalman filter is a recursive estimator [8],[10]. This means that only the estimated state from the previous time step and the current measurement are needed to compute the estimate for the current state. In contrast to batch estimation techniques, no history of observations and/or estimates is required.
II. Proposed Work:
Object tracking is done with Kalman filter. In object detection it is necessary that object should be clearly subtracted from the background as if there will be false detection then tracking will not be precise[5]. Illumination can be divided into two parts: reflection and luminance. The darker area of body represents the reflectance part whereas brighter area represents the luminance. Energy confined is more in luminance part as compared to reflectance. So to make uniform illumination, energy confined into luminance must be reduced. Normalization of Discrete cosine transform is used for this purpose.[11] Mathematically the illumination can be represented in terms of reflectance and luminance as: Illumination = reflectance* luminance To reduce the computational complexity these are represented in the logarithmic form as: Illumination = log (reflectance) + log (luminance) Discrete cosine transform is used to counter the luminance. DCT breaks the image into two different frequency components: low frequency and high frequency. Low frequency component contains high energy and can be considered as luminance part of image whereas reflectance is constituted by high frequency component as it contains the low energy The discrete cosine transform (DCT) helps separate the image into parts (or spectral sub-bands) of differing importance (with respect to the image's visual quality). The DCT is similar to the discrete Fourier transform which transforms a signal or image from the spatial domain to the frequency domain[11].
Figure1: Discrete cosine transform
Himani Gupta, IJRIT
58
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
The general equation for a 2D (N by M image) DCT is defined by the following equation:
=
=
=
# !
(2 + 1) (2 + 1) 2 2
1
√
, = 0
"√2 ! ,1 ≤ ≤ − 1 # !
1
√
, = 0
"√2 ! ,1 ≤ ≤ − 1
and the corresponding inverse 2D DCT transform is.:
=
=
(2 + 1) (2 + 1) 2 2
(2 + 1) (2 + 1) 2 2
The low frequency component of DCT contains the maximum information about an image whereas high frequency component only has fine details as shown in a desert fig. 2.1(a) & (b)
Himani Gupta, IJRIT
59
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
LF
HF 250
100
200
200
100 20
200 150
300 400
100
500
300 0
400 500
50
600 700
Original Image
40
0 200 400 600 800 1000
(a) Figure2: (a) Low frequency coefficients, (b) high
-20
600 700
-40 200 400 600 800 1000
(b) frequency coefficient
As shown in the above fig. that after converting the original image into gray scale low frequency coefficient shows the maximum information whereas high frequency coefficient depicts nothing [6].But in illumination invariant method normalization of DCT coefficients is done. The low frequency component contains a DC component having highest energy and then decreasing. So to reduce the luminance and make the image intensity uniform the low frequency coefficients pixel values are reduced. For this purpose first 20 pixels are divided by 50 and their pixels values are reduced. Not any information loses by this, so DC coefficient is increased by 10 % or 20%.
Figure3: Flow chart for illumination invariant method
Himani Gupta, IJRIT
60
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
As shown in fig. the image thus obtained after making it illumination invariant is pass to the kalman filter where object tracking and background subtraction take place. Kalman Filter The Kalman filter, also known as linear quadratic estimation (LQE), is an algorithm that uses a series of measurements observed over time. The Kalman filter can be written as a single equation, however it is most often conceptualized as two distinct phases: "Predict" and "Update". In what follows, the notation represents the estimate of the time n given observations up to, and including at time m ≤ n. The state of the filter is represented by two variables:
), (*|*) a posteriori state estimate at time k given observations up to and including at time k;
-*|* a posteriori error covariance matrix (a measure of the estimated accuracy of the state estimate). Predict Predicted (a priori) state estimate
), , + * /* *|* = .* )*|*
(1)
Predicted(a priori) estimate covariance 1 0 -0 *|* = .* -*|* .* + 2*
(2)
Update Innovation or measurement residual
3 4* = 5* − ), *|* 6*
(3)
Innovation (or residual) covariance
7* = 6* -*|* 6*1 + 8*
(4)
Optimal Kalman gain
9* = -*|* 6*1 7*
(5)
Updated (a posteriori) state estimate
)*|* ,= 3 49 * * + ), *|*
(6)
Updated (a posteriori) estimate covariance
-*|* = (1 − 9* 6* )-*|*
(7)
Fk is the state transition model which is applied to the previous state xk−1;
Himani Gupta, IJRIT
61
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
Bk is the control-input model which is applied to the control vector uk; wk is the process noise which is assumed to be drawn from a zero mean multivariate normal distributionwith covariance Qk. Hk is the observation model Qk is the Gaussian process Rk is the covariance of the observation noise
Figure 5: Flow chart of Kalman filter with background Subtraction. III. Results In background subtraction or in image acquisition toolbox of MATLAB 7.8 is used to capture a real time video. 10 frames at the interval of 1 sec have been set and 10 RGB frames are at the output captured by laptop’s webcam. Before processing further each frame is converted into gray scale as gray scale color level also contains all information but reduces the computational complexity as shown in fig. 6.1
Himani Gupta, IJRIT
62
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
Figure 6.1: Gray scale conversion of real time frames
200
200
100
100
0
0
200
200
200
100
100
100
0
0
0
200
200
200
100
100
100
0
0
0
200 100 0
Figure6.2: An illumination invariant image A histogram is used in fig. 6.3 which shows a comparision of original image and an altered image After logarithmic output of low frequency coefficient and after DCT(Discrete cosine transform) and IDCT (Inverse Discrete cosine transform) we get a normalised IDCT for illumination invariant image shown in fig. 6.4
Himani Gupta, IJRIT
63
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
2
4 Histogram of Original Image x 10
4 Histogram of altered Image x 10
1.8 2 Number of Pixels in a tone
Number of Pixels in a tone
1.6 1.4 1.2 1 0.8 0.6
1.5
1
0.5
0.4 0.2 0
0 Tonal Variation 0
100
Tonal Variation 200
0
100
200
Figure 6.3: Histogram of original and altered image
Figure 6.4: Normalized IDCT for illumination invariant
Himani Gupta, IJRIT
64
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
Figure 6.5: Object tracked Background subtracted
Figure 6.6: Final image of background removal Notice the high illumination area behind the object. This high illumination area is made uniform by proposed illumination invariant method as shown in fig. 6.2. This invariant image is passed to Kalman filter with initialisation parameters of Kalman filter and this filter removed the background and foreground object was tracked as shown in fig. 6.6 250 Object Motion for original frame Object Tracking for original frame Object Motion for Compensated frame Object Tracking for Compensated frame
Y Position of track ing rectangle
200
150
100
50
0
-50
1
2
3
4
5 6 Number of frames
7
8
9
10
Figure 6.7: Graph shows tracking and motion of both the images.
IV. Conclusion & Future Work In the proposed work illumination variation problem in object detection is solved out. Kalman filter is used to track the motion of the object. Basically the problem lies with the foreground subtraction as the Kalman filter uses the smallest rectangle coordinates which are predicted and updated. If there is illumination variance then false object detection may take place and tracking will not be precise. That’s why we develop an algorithm which is a combination of illumination invariance and Kalman filter which gives rise to correct tracking of object. This also reduces the time consumed in operation. A graph is also taken which shows the comparison between the motion and tracking of original and compensated image.
Himani Gupta, IJRIT
65
IJRIT International Journal of Research in Information Technology, Volume 2, Issue 8, August 2014, Pg. 57-66
There are still some limitations of this work like if background is having noise and very much dynamic like waving of trees leaves, then that may impose challenge to my object tracking work. For that purpose adaptive background subtraction or object tracking might not work well.
Refrences [1]Vinayak G Ukinkar, Makrand Samvatsar, “Object detection in dynamic background using image segmentation: A review,” IJERA, ISSN:2248-9622, Vol. 2, Issue 3, May-Jun 2012, pp.232-236 [2]Deepak Kumar Rout, Sharmistha Puhan, “Video Object Detection in Dynamic Scene using Inter-Frame Correlation based Histogram Approach,” International Journal of Computer Applications (0975 – 8887), Volume 82 – No 17, November 2013,pp19-24 [3]Hemavathy R, Dr.Shobha G, “Object Detection and Tracking under Static and Dynamic environment:AReview,”International Journal of Advanced Research in Computer and Communication Engineering, Vol. 2, Issue 10, October 2013pp4095-4100 [4]Kalyan Kumar Hati, Pankaj Kumar Sa, and BanshidharMajhi, “IntensityRange Based Background Subtraction for Effective Object Detection,” IEEE Signal Processing Letters, Vol. 20, No. 8, August 2013,Pp759-762 [5]Bo Liu, Yan Lin, Guan Guan, “A Method of Multi-scale Edge Detection for Underwater Image,” Journal of Information & Computational Science 10: 2(2013) 345–354 [6]CHI Jian-nan, ZHANG Chuang, ZHANG Han, LIU Yang, YAN Yan-tao, “Approach of Moving Objects Detection in Active Video Surveillance,” Joint 48th IEEE Conference on Decision and Control and 28th Chinese Control Conference Shanghai, P.R. China, December 16-18, 2009 [7]ParisaDarvishZadehVarcheie, Michael Sills-Lavoie and Guillaume-AlexandreBilodeau, “A Multiscale RegionBased Motion Detection and Background Subtraction Algorithm,”Sensors 2010, ISSN1424-8220,1041-1061 [8]Caius SULIMAN, Cristina CRUCERU, Florin MOLDOVEANU, “Kalman Filter Based Tracking in an Video Surveillance System,” 10th International Conference on DEVELOPMENT AND APPLICATION SYSTEMS, Suceava, Romania, May 27-29, 2010 [9]PranamJanney And Glenn Geers, “A Robust Framework For Moving-Object Detection And Vehicular Traffic Density Estimation,” Arxiv:1402.0289v1 [Cs.CV] 3 Feb 2014 [10]HarshaVarwaniHeenaChoithwani, “Understanding various Techniques for Background Subtraction and Implementation of Shadow Detection” IJCTAVol 4 (5),822-827 [11]Virendra P. Vishwakarma, Sujata Pandey Member IEEE, and M. N. Gupta , “A Novel Approach for Face Recognition Using DCT Coefficients Re-scaling for Illumination Normalization,” 15th International Conference on Advanced Computing and Communications © 2007 IEEE
Himani Gupta, IJRIT
66