One of the significant application of computer vision is Stabilizing a video that was captured from a jittery or moving platform. One way to stabilize a video is to track a prominent feature in the image and utilize it as an anchor point to cancel out all perturbations relative to it. This technique, however, must be bootstrapped with knowledge of where such a salient feature remains in the first video frame. The paper presents method of video stabilization that works without any such erstwhile knowledge. The method is built on the basis of Random Sampling and Consensus (RANSAC) and adding few additions to the existing methodologies. It instead automatically investigates for the "background plane" in a video sequence, and utilizes its observed distortion to precise for camera motion. All the simulations have been performed using MATLAB tool.
Biology for Computer Engineers Course Handout.pptx
VIDEO STABILIZATION USING SIFT AND RANSAC
1. APPLICATION OF FEATURE POINT
MATCHING TO VIDEO STABILIZATION
By
Nikhil Prathapani
Student Member, IEEE
2. INTRODUCTION
Along with advancement, digital video has introduced new
problems like video noising, video de-stabilization and video
jitter.
In order to overcome these problems, new techniques like
video enhancement and video stabilization have been
proposed.
Of the proposed video stabilization techniques, all most all of
them require prior knowledge of prominent frame.
But the proposed technique is based on RANSAC*, SSD and
SIFT, it does not require any erstwhile knowledge of prominent
*Tordoff, B; Murray, DW. "Guided sampling and consensus for motion estimation."European Conference n Computer Vis
2002.
3. Why estimate visual motion?
Visual Motion can be annoying
Camera instabilities, jitter
Measure it; remove it (stabilize)
Visual Motion indicates dynamics in the scene
Moving objects, behavior
Track objects and analyze trajectories
4. Getting six parameters
SIFT algorithm – Find corresponding pairs
At time k
It needs three pairs to determine a unique solution
Y X A
6. The fundamental matrix F
C C’
T=C’-C
R
p p’
TRp'p
Two reference frames are related via the extrinsic parameters
7. The fundamental matrix F
The fundamental matrix is the algebraic
representation of epipolar geometry
The fundamental matrix satisfies the condition that
for any pair of corresponding points x↔x’ in the two
images
0Fx'xT
0lxT
8. RANSAC (Random Sampling and Consensus )
repeat
select minimal sample (8 matches)
compute solution(s) for F
determine inliers
until (#inliers,#samples)>95% or too many times
compute F based on all inliers
9. SSD (sum of squared differences) surface – textured area
13. ANALYSIS AND RESULTS
Step 1- Reading frames from a movie file
Step 2- Collecting Salient Points from Each Frame
SIFT
Step 3- Selecting Correspondences Between P
SSD
Step 4-. Estimating Transform from Noisy
Correspondence
RANSACaffine transform will be a 3-by-3 matrix:
[a_1 a_3t_r;
a_2 a_4t_c;
0 0 1]
The parameters ‘a’ define scale, rotation, and sheering
effects of the transform, while the parameters ‘t’ are
translation parameters.
14. Step 5- Transform Approximation and Smoothing
Step 6- Run on the Full Video
Raw input mean and Corrected sequence
mean images
PSNR MSE
RAW INPUT MEAN 22.5406 3.62
CORRECTED
SEQUENCE MEAN
25.5725 3.59
15. CONCLUSION
The paper presents a comprehensive and thorough
approach to video stabilizing videos using MATLAB.
This kind of novel approach to video stabilizing [6, 7, 8]
without prior knowledge of prominent features in the
frames has targeted many applications in the fields of
motion estimation, remote sensing, and airborne
applications
16. REFERENCES
[1]P. A. Keller, The cathode-ray tube: technology, history and applications,Palisades Press,
1991, ISBN 0963155903.
[2]W. C. O’Mara, Liquid crystal flat panel display: manufacturing science andtechnology,
Van Nostrand Reinhold, 1993, ISBN 0442014287.
[3]J. Hutchison, “Plasma display panels: the colorful history of an Illinois tech-nology”, ECE
alumni news, university of Illinois, vol. 36(1), 2002.
[4]C. Poynton, Digital video and HDTV algorithms and interfaces, MorganKaufmann, 2003,
ISBN 1558607927.
[5] Tordoff, B; Murray, DW. "Guided sampling and consensus for motion
estimation."European Conference n Computer Vision, 2002.
[6] Lee, KY; Chuang, YY; Chen, BY; Ouhyoung, M. "Video Stabilization using Robust
Feature Trajectories." National Taiwan University, 2009.
[7] Litvin, A; Konrad, J; Karl, WC. "Probabilistic video stabilization using Kalman filtering
and mosaicking." IS&T/SPIE Symposium on Electronic Imaging, Image and Video
Communications and Proc., 2003.
[8] Matsushita, Y; Ofek, E; Tang, X; Shum, HY. "Full-frame Video Stabilization." Microsoft®
Research Asia.CVPR 2005.
20. Any Queries?
For research articles, papers and projects in the fields
of Image Processing and Nanoelectronics,
you can connect to my research profile:
http://jntuhcej.academia.edu/NikhilPrathapani
21. SIFT detector proposed considers local image
characteristic and retrieves feature points that are
invariant to image rotation, scaling, translation, partly
illumination changes and projective transform.
The scale-invariant feature extractor detects feature
points through a staged filtering approach that
identifies stable points in the scale-space.
Scale Invariant Feature Transform
22. Why Features?
A brief yet comprehensive representation of the
image
Can be used for:
Image alignment
Object recognition
3D reconstruction
Motion tracking
Indexing and database search
More…
23. Desired Feature Properties
• Robustness => Invariance to changes in illumination, scale,
rotation, affine, perspective
• Locality => robustness to occlusion and clutter.
• Distinctiveness => easy to match to a large database of
objects.
• Quantity => many features can be generated for even small
objects
• Efficiency => computationally “cheap”, real-time performance