iui

structurefrommotion.pdf

Lowe, D. G., “Distinctive Image Features from Scale-Invariant Keypoints”, International Journal of Computer Vision, 60, 2, pp. 91-110, 2004.

klt-featuretracking.pdf

http://www.ces.clemson.edu/~stb/klt/

http://cmp.felk.cvut.cz/~svoboda/SelfCal/

M. Pollefeys, L. Van Gool, M. Vergauwen, F. Verbiest, K. Cornelis, J. Tops, R. Koch, Visual modeling with a hand-held camera, International Journal of Computer Vision 59(3), 207-232, 2004. [IJCV][pdf]

epipolargeometry.pdf

Richard I. Hartley (June 1997). “In Defense of the Eight-Point Algorithm“. IEEE Transaction on Pattern Recognition and Machine Intelligence 19 (6): 580—593.

Chojnacki, W., Brooks, M. J., van den Hengel, A., and Gawley, D. 2003. Revisiting Hartley’s Normalized Eight-Point Algorithm. IEEE Trans. Pattern Anal. Mach. Intell. 25, 9 (Sep. 2003), 1172-1177

In this homework, I would like to work on analysis of graph-theoretic clustering methods, including basic Eigen vector method, minimum cuts and normalized cuts. Algorithms will be implemented first for both numeric data and images and further followed by comparisons and discussions. Related issues, such as problems caused by identical Eigen values in basic Eigen vector method, will be analized through some designed experiments.

In this homework, I want to go further with the mean-shift algorithm. I want to implement mean-shift tracking algorithm for tracking an object in a video sequence.

The idea of mean-shift tracking is that given an object, we can use the distribution of the value in gray (color) level to describe it. the object can be expressed as a PDF located at x.And an object located at y can  be expressed as another PDF.

So to track an object means to find an optimal y which can minimize the distance between the two PDFs.

HW2 - View Synthesis using Fundamental matricesand Homography matrices:

Approach 1:

=========

1. Using the camera calibration tool box get the intrinsic camera parameters.

2. Capture two views of a scene. (Reference views)

3. Using a test/calibrating scene, compute the Fundamental matrices, F1 and F2, relating each of the reference views to the synthesised view by establishing feature correspondences.

4. For the actual scene, establish Feature correspondences between the two Reference views (Ref Image1 and Ref Image 2).

5. For every point ‘x’ in Ref Image1 get the epipolar line in the synthesised view by L1 = F1x

6. For the corresponding point in Ref Image 2 get the epipolar line in the synthesised view by L2= F2x

7. Get the intersection of the lines L1 and L2 to get x’ in the synthesised view.

8. Combine adjacent groups of 4 feature points to establish local homographies and map the non-feature points into the synthesised vie..

Approach 2:

=========

Similar to Approach 1, except that the epipolar line is formed from two points projected on the synthesised plane, using the homographies of two planes in the scene.

The Homography matrix takes the place of the Fundamental matrix.

The criteria for evaluation would be the MSE between the actual and the synthesised view of the scene.