Kccsi 2012 a real-time robust object tracking-v2
Download as PPTX, PDF2 likes2,810 views
The document presents a novel object tracking method utilizing a Hamming distance-based gradient orientation pattern matching technique, which addresses challenges such as illumination changes and occlusion. The proposed method outperformed traditional techniques in tracking scenarios with partial occlusion and varying lighting conditions by introducing unit gradient vectors (UGVs) as a robust feature. Experimental results demonstrated the effectiveness of this approach in real-time applications.
![Sample Image Template
1st Derivative (Sobel Operator)
Step 1
gx1 gy1 gx2 gy2
Extract UGVs
feature Normalize
nx1 ny1 nx2 ny2
Threshold Threshold
Step 2 Absolute Diff. Absolute Diff.
Perform template
matching by using OR
Hamming distance Block at
position Sum
(x,y) Iterate (N-M-1) blocks
return [Best matching position (x,y)] 12](https://image.slidesharecdn.com/kccsi2012-areal-timerobustobjecttracking-v2-120229223130-phpapp02/85/Kccsi-2012-a-real-time-robust-object-tracking-v2-12-320.jpg){kind=tracking}
Kccsi 2012 a real-time robust object tracking-v2
- 1. A PRESENTATION OF A REAL-TIME ROBUST OBJECT TRACKING Prarinya Siritanawan (SIIT) Toshiaki Kondo (SIIT), Kanokvate Tungpimolrut (NECTEC), Itsuo Kumazawa (Tokyo Tech) Master of Information and Communication Technology for Embedded System Sirindhorn International Institute of Technology Presentation at International Advanced School on Knowledge Co-creation and Service Innovation 2012, Japan Advanced Institute of Science and Technology, March 1 1
- 2. OUTLINE • Introduction • Hamming Distance based Gradient Orientation Pattern Matching • Experimental Results • Conclusion • Question and Answer 2
- 3. INTRODUCTION “OBJECT TRACKING IS A DETERMINATION OF LOCATION, PATH AND CHARACTERISTICS OF AN INTERESTED OBJECT” Subhash Challa, Mark R. Morelande, Darko Musichki and Robin J. Evans, “Fundamentals of Object Tracking”, Cambridge University Press, 2011
- 4. INTRODUCTION Applications for object tracking • Video surveillance • Human-machine interface • Robot control • Air space monitoring • Weather monitoring • Cell biology Subhash Challa, Mark R. Morelande, Darko Musichki and Robin J. Evans, ”Fundamentals of Object Tracking”, Cambridge University Press, 2011 4
- 5. INTRODUCTION Major problems of visual tracking are caused by • Illumination change • Occlusion We focus on these problems • Computation time • Scaling • Rotation • Focus • Aperture 5
- 6. INTRODUCTION • Typical visual tracking and motion estimation techniques assume that lighting conditions are constant and minimal occlusion. • We proposed a new template matching technique. 6
- 7. INTRODUCTION • Template matching is the intensity-based technique for measuring the similarity between template and corresponding block of image. x Popular similarity metrics SAD I (i , j ) T (i , j ) Template y i j 2 SSD I (i , j ) T (i , j ) i j Match 7 Sample frame
- 8. INTRODUCTION • We obtain an array of SADs or SSDs after scanning the template over the entire image. Best matching position Fig. 1. Inverted SSD result. 8
- 9. INTRODUCTION • However SADs and SSD are sensitive to changing lighting conditions and occlusion. In order to develop a method that can provide the robustness to illumination change, a new template matching technique is used. • For illumination change problem, we introduced a robust feature called Unit gradient vector (UGVs). • To cope with the occlusion problem, we introduce the Hamming distance as a new matching method instead of SSD. 9
- 10. HAMMING DISTANCED BASED GRADIENT ORIENTATION PATTERN MATCHING 10
- 11. HAMMING DISTANCE BASED GRADIENT ORIENTATION PATTERN MATCHING • Hamming distance based Gradient Orientation Pattern Matching (P.Siritanawan & T.Kondo) – Template matching based technique using Hamming distance (HD) on Unit gradient vectors (UGVs). Fig. 2. Intensity image. Fig. 3. Unit gradient vectors in x and y direction 11
- 12. Sample Image Template 1st Derivative (Sobel Operator) Step 1 gx1 gy1 gx2 gy2 Extract UGVs feature Normalize nx1 ny1 nx2 ny2 Threshold Threshold Step 2 Absolute Diff. Absolute Diff. Perform template matching by using OR Hamming distance Block at position Sum (x,y) Iterate (N-M-1) blocks return [Best matching position (x,y)] 12
- 13. UNIT GRADIENT VECTORS • UGV is a robust feature against Illumination changes. Normal condition Intensity image Gradient vectors Unit gradient vectors Lighting change condition 13
- 14. UNIT GRADIENT VECTORS • The unit gradient vectors (UGVs) feature can be extracted through the following normalized equations where Ix and Iy are gradient of intensities in x and y direction is a small constant to prevent zero division 14
- 15. MATCHING METHOD • We introduce Hamming Distance (HD), 0 0 1 1 0 0 1 1 = 0 0 1 1 0 0 0 1 1st Pattern 2nd Pattern HD(x,y) = 8 HD counts the number of pixels that are not match 15
- 16. MATCHING METHOD • HD uses XOR but UGVs is not binary info. • We need to transform the non-binary image to be binary image using threshold absolute difference function (O.Pele & M.Werman), 16
- 17. MATCHING METHOD • Then the total distance of the block at position (x,y) is given by Fig. 4. Inverted HD result. 17
- 18. MATCHING METHOD Image Template Occluded Similarity Score features Image metric 1 4 2 6 1 4 0 0 0 0 2 6 SAD 6 3 4 4 6 3 0 0 0 0 4 4 Intensities = 32 5 6 6 2 5 6 0 0 0 0 6 2 2 3 5 3 2 3 0 0 0 0 5 3 Unit 0 0 1 1 gradient 0 0 1 1 HD = 8 vectors 0 0 1 1 (UGV) 0 0 1 1 Pixelwise voting 18
- 19. EXPERIMENTAL RESULTS 19
- 20. DEMONSTRATION 20
- 21. EXPERIMENTAL RESULTS Which is the ? ? Best matching best matching ? position ? ? peak found !! Fig. 5. Tracking results under irregular lighting with occlusion by (a) SSD on UGVs, (b) HD on UGVs, (c) and (d) are the distributions of the corresponding similarity measurements 21
- 22. CONCLUSION • A novel pattern matching technique combines the advantages of – Unit gradient vectors (UGVs) – Hamming distance metric (HD) • UGV is a robust feature against the time-varying lighting conditions. • Compared with conventional matching with SAD or SSD on intensity, HD yields better results in partial occlusion scenarios. (60-70% covered). • Efficient over the existing matching techniques on both synthetic and real image sequences. 22
- 23. PUBLICATION 1. Wattanit Hotrakool, Prarinya Siritanawan, and Toshiaki Kondo, “Real-time Gradient Orientation Pattern Matching”, International Conference on Embedded System and Information Technology, Chaing Mai, Thailand, 2010 2. Wattanit Hotrakool, Prarinya Siritanawan, and Toshiaki Kondo, “A Real-time Eye- tracking Method using Time-varying Gradient Orientation Patterns”, In proc. ECTI-CON, Thailand, 2010 3. Prarinya Siritanawan and Toshiaki Kondo, “Hamming Distance based Gradient Orientation Pattern Matching”, In proc. International Symposium of Artificial life and Robotics 17th, Chaing Mai, Oita, Japan, January 2012 4. Prarinya Siritanawan, Toshiaki Kondo, Kanokvate Tungpimolrut, Itsuo Kumazawa, “A visual tracking method using the Hamming distance”, In proc. International Conference on Information and Communication Technology for Embedded System 3rd, Bangkok, Thailand, March 2012 23
- 24. ACKNOWLEDGEMENT This research is supported by • National Research University Project of Thailand, Office of Higher Education Commission • Sirindhorn International Institute of Technology (SIIT) • Thailand Advanced Institute of Science and Technology (TAIST) • Tokyo Institute of Technology • National Electronics and Computer Technology Center (NECTEC) 24
- 25. THANK YOU FOR YOUR ATTENTION QUESTION AND ANSWER 25