5. Feature Detection and Matching
Track features using optical flow
Least sum of squares flow of gradient among frames
6. Must choose frame I’ that overlaps well with previous frame I
When tracking n features, once the successfully tracked
features n’ in the current frame is less than n/2, save I’
Key Frame Selection
7. Image Stitching
Homography H is a projective transformation that relates
image points x’ = Hx, x and x’ are image points in I and I’
From H and K, the camera rotation R is estimated
Image point x in the world space of the cylinder is X= K-1R-1x
where X= (x,y,z)
World coordinates mapped to UV texture coordinates, u =
arctan(x/z) and v = y/√(x2 + z2)
Texture map cylinder x = r*cos(u), y = v, z = r*sin(u), r is the
radius of the cylinder.