Project 5: Tracking and Structure from Motion
Vibhu Ramani
December 12th, 2011
Brief Summary
- We were given a series of images from a video.
- Using Harris corner detector we detected certain keypoints in the first image which we could track over all the images.
- Using the Kannade-Lucas-Tomasi tracker on these keypoints we recreated the Structure.
- The keypoints were tracked over the frames by computing the optical flows. We used a window size of 15x15
- We eliminated the affine ambiguity by using the metric transformation in the "A Sequential Factorization Method for Recovering Shape and Motion from Image Streams, Morita and Kanade, 1997."
Original set of images
Initial points selection using the Harris Corner Detector.
Some of the points during the course of the video go out of scope.
For any point that came near the boundry (7px) of the image
we stopped tracking it and avoided using it to recreate the structure.