calculates the apparent motion of objects or features between consecutive frames in a video
- recovering motion from optical flow
- relies on 3 assumptions
conditions for solving
- Harris Corner Detector - we want to track corners in this equation
Iterative
- solve lucas-kanade
- warp l(t-1) towareds l(t) using estimated flow
- repeat until convergence
issues
What are the potential causes of errors in this procedure?
When our assumptions are violated
- Brightness constancy is not satisfied
- The motion is not small
- A point does not move like its neighbors
- window size is too large
- what is the ideal window size? (hyperparameter)
- find u bar and v bar?? → horn-schunk