ENPM 673, Robotics Perception
Homework: Optical Flow Estimation
For this project, you will compute and compare optical flow on some sequences of the
Middlebury Optical Flow dataset. You are to implement the Lucas-Kanade optical flow
algorithm. The details are on the Wikipedia page.
[Part 1] 30 pts
Your algorithm will be validated on the Middlebury grayscale dataset, which you can find here.
Each of these dataset items contains 8 frames of motion. Test your implementation of the Lucas
Kanade algorithm on the Grove and Wooden sets. Make quiver plots for your optical flow
computations. Store these frames as a video file called Grove_LK and Wooden_LK,
respectively.
Upload the code, a representative quiver plot from each sequence and the videos.
[Part 2] 30pts
After this, you compare your output with inbuilt MATLAB implementations of LucasKanade, Farneback and Horn-Schunck. Evaluate them on the same dataset mentioned in Part 1,
and write a report comparing how the different methods behave in textured regions, non-textured
regions, and at object boundaries.
