my thesis defence. department of cse, ruet
TRANSCRIPT
![Page 1: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/1.jpg)
Heaven’s Light is Our Guide
Rajshahi University of Engineering & TechnologyDepartment of Computer Science & Engineering
Human Segregation from Detected Moving Objects Using Histogram of Oriented
GradientsPRESENTED BY
Md. Nahiduzzaman Rose
Roll No. 103034
Dept. of CSE, RUET.
SUPERVISED BY
Prof. Dr. MD. Rabiul IslamDept. of CSE, RUET
![Page 2: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/2.jpg)
Outlines Objective Related Works Proposed Methodology Implementation Experimental Result Limitation Further Works Conclusion References
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
2/28
![Page 3: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/3.jpg)
Objective
Video surveillance Distinguish human from other moving
objects Develop an advance and intelligent
security system
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
3/28
![Page 4: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/4.jpg)
Existing Works Human detection approach
Histogram of Oriented Gradients for Human Detection(HOG) [1]
Overview of this Method:
Fig.1. Original work [1]Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
4/28
![Page 5: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/5.jpg)
Existing Works (Cont’d)
Human detection approach Pedestrian detector using infrared images
and histogram of oriented gradients.[4] Pedestrian detector using histograms of
oriented gradients and SVM classifier.[5]
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
5/28
![Page 6: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/6.jpg)
Existing Works (Cont’d)
Moving object detection approach Improved adaptive Gaussian mixture model
for background subtraction.[2] Adaptive Background Mixture model for
Real-time tracking with shadow detection. using Mixture of Gaussian. [3]
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
6/28
![Page 7: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/7.jpg)
Proposed Methodology
Fig. 2. Procedure of propose method
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
7/28
![Page 8: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/8.jpg)
Implementation
Proposed methodology indicates two decent algorithm.
Mixture of Gaussian (MOG) for Background Subtraction.
Histogram of Oriented Gradients (HOG).
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
8/28
![Page 9: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/9.jpg)
Mixture of Gaussian (MOG)
MOG is common technique ,that has been used for background subtraction.
Every pixel's intensity values in the video can be modeled using a Gaussian mixture model.
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
Fig. 2. Sum of two Gaussian distribution [6]
09/28
![Page 10: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/10.jpg)
Mixture of Gaussian (Cont’d)
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients
Background model creation
08/08/2015
Fig. 3. Background model creation [8]
Original frame Background Model
10/28
![Page 11: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/11.jpg)
Histogram of Oriented Gradients
HOG is a type of feature descriptor. It uses global feature to describe a person. The whole person is used as a feature
vector.
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
11/28
![Page 12: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/12.jpg)
Histogram of Oriented Gradients (Cont’d)
How HOG descriptor works? HOG uses 64 x 128 detection window
A 64 x 128 detection window
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
Fig. 5. 64 x 128 detection window
12/28
![Page 13: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/13.jpg)
Steps of HOG
Feature extraction:For a 64 x 128 window,Compute gradient orientation and magnitude.Divide image into 16 x 16 ,50% overlap block.Each block consists 2x2 cells with size.Quantize the gradient orientation in 9 bin histogram.Concatenate histogram.
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
13/28
![Page 14: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/14.jpg)
Steps of HOG (Cont’d)
Blocks and Cells: Total 7 x 15=105 blocks Two adjacent block
overlaps each other by 50%.
Each block has 2x2 cells with size 8x8 pixel.
08/08/2015
BlocksCell
Fig 6-Blocks and cells[7]
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients
14/28
![Page 15: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/15.jpg)
Steps of HOG (Cont’d)
Calculating Gradient Vector Gradient vector is calculated for each pixels in
8x8 cells. Every pixel provides gradient magnitude and
orientation. So we will have 64 gradients in one cell (8x8
size).
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
15/28
![Page 16: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/16.jpg)
Calculating Gradient Vector
Notice the gray scale value of N4(P) of the desired (colored in “Blue”) pixel.
161
84
54
128
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
(a) (b)
Fig. 7. (a) Operating pixel(blue color), (b) Grayscale value of N4(P)
16/28
![Page 17: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/17.jpg)
Change in Y direction:161-54 = 107
Calculation Gradient of the pixel:
Change in X direction:128-84 = 44
161
54
128 84
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
Gradient orientation and magnitude (Cont’d)17/28
![Page 18: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/18.jpg)
Gradient orientation and magnitude (Cont’d)
Change in Y direction: 211-104 = 107
After Increasing Brightness by 50
Change in X direction:178-134 = 44
211
104
178 134
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
18/28
![Page 19: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/19.jpg)
Gradient orientation and magnitude(Cont’d)
Change in Y direction:242 - 81 = 161
After Increasing Contrast by 1.5
Change in X direction:192-126 = 66
242
81
192 126
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
19/28
![Page 20: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/20.jpg)
Quantizing Histogram Histogram contains 9
bins and ranges 0 to 180 degree.
Each bin is contributed by the magnitudes of the gradient vector.
Each histogram represents a cell of size 8x8.So one block contains 4 histogram.
20/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
OrientationM
agni
tude
Fig. 8. Quantizing histogram
![Page 21: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/21.jpg)
Quantizing Histogram (Cont’d)
Interpolation: Suppose an orientation of
85 degree. Split 85 between two
nearest neighbors bins. Distance of ‘85’ from bin
center ’70’ and ’80’ are 15 and 5 respectively.
So shares of the magnitude are 15/20=3/4 and 5/20=1/4 for 90 and 80 respectively.
21/28
85
1/43/4
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
Mag
nitu
de
Orientation
Fig. 9. Interpolation
![Page 22: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/22.jpg)
Final Descriptor Size
Concatenating Histogram: Every block with 4 histograms gives
4 x 9 bin = 36 components
Total 7 x 15= 105 blocks. So there are, 7x15x36= 3,780 components
Concatenation of histograms produces, 1D matrix of dimension 3,780
22/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
Fig. 10. Final feature vector
![Page 23: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/23.jpg)
Experimental Result23/28
MOG
Foreground detection
HOG
Detecting human
Segregated human and detected moving objects
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
a
c
b
Fig. 11. a. Input image ,b. Object detection, c. Final result [7]
![Page 24: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/24.jpg)
Limitations
Limitations of proposed method Very slow frame rate. Noise sensitive. False shape detection. Humans far from the camera (small size)
are hard to detect.
24/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
![Page 25: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/25.jpg)
Further Works
Access control of moving objects and tracking their positions.
Evaluate performance of proposed methodology.
25/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
![Page 26: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/26.jpg)
Conclusion
A method of distinguish human and also keep detecting the other moving objects is has been proposed here
Detection is always a challenging task. There are always room for improving the method.
26/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
![Page 27: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/27.jpg)
References[1] Dalal and Triggs: “Histograms of oriented gradients for human detection”. In Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on (Volume: 1)25-25 June 2005 Page(s):886 - 893 vol. 1
[2] Z Zivkovic: “Improved adaptive Gaussian mixture model for background subtraction”. Published in: Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on (Volume: 2) 23-26 Aug. 2004 Page(s):28 - 31 Vol.2
[3] KaewTraKulPong and R. Bowden: “An Improved Adaptive Background Mixture Model for Real-time Tracking with Shadow Detection”. Vision and Virtual Reality group, Department of Systems Engineering, Brunel University, and Middlesex, UB8 3PH.
[4] Suard, F. ; Centre Nat. de la recherche scientifique, INSA Rouen, Saint Etienne ; Rakotomamonjy, A. ; Bensrhair, A. ; Broggi : “A Pedestrian Detection using Infrared images and Histograms of Oriented Gradients” Intelligent Vehicles Symposium, 2006 IEEE Page(s):206 – 212
[5] Bertozzi, M.; Univ.a di Parma, Parma; Broggi, A; Del Rose, M.; Felisa, M.more authors: A Pedestrian Detector Using Histograms of Oriented Gradients and a Support Vector Machine Classifier. In: Intelligent Transportation Systems Conference, 2007. ITSC 2007. IEEE, Page(s):143 – 148
27/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
![Page 28: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/28.jpg)
References (Cont’d)[6] Gaussian mixture model image. Available at: http://www.mathworks.com/ matlabcentral/ fileexchange/screenshots/3544/original.jpg (Access on: 03-08-15)
[7] Output screenshot from experimental dataset video footage. Available at: http://www.dailymotion.com/video/x12l3e6_traffic-moves-along-a-busy-road-in bangkok_travel (Access on: 07-08-15)
[8] Background model creation output screenshot. Available at: http://code. opencv.org/issues/1931 (Access on: 07-08-15)
28/28
Human Segregation from Detected Moving Objects Using Histogram of Oriented Gradients 08/08/2015
![Page 29: My thesis defence. Department of CSE, RUET](https://reader033.vdocuments.net/reader033/viewer/2022051122/588aff9e1a28abf8548b6abb/html5/thumbnails/29.jpg)
Thank You.Any Question?