Mosaics From Video Using Optical Flow Based on the Affine Transform:
This work was done as a final project for my CE 264 - Computer Vision course. This was a joint project with Colin Tracey. In this project my goal was to calculate the optical flow between two images based on the affine transform (a simplification of a full homography) and to use this information to create basic mosaics from a video stream. Colin then took these basic mosaics and used tessellation techniques to refine the mosaics to look smoother.
The following images were created using this video file.
(Download the video by clicking on this link, you may need to right-click and use "Save As" (4.5MB))
(Or download the smaller version (2 MB))
The top image on the left is the mosaic created by my optical flow estimation algorithm. As can
be seen there are many discontinuities and other undesired warping effects.
(Note: The image can be in color, we simply ignore the color channels until
the stitching phase).
The bottom image on the right is the image after Colin's tessellation algorithm. Note that the black regions around the edge of the image are caused by
camera motion. These could always be removed by cropping the image inside the boundry created by the black regions.
Hover the mouse over an image to see a larger view, or click the image to see the image at full resolution. (May not work in Internet Explorer)
We are quite pleased with the results. For more information on the optical flow estimation feel free to email me at the address given on the contact page.