Project 2: Structure from Motion
Due: With Final Exams
Mandatory Subpart.
1) Implement Factorization approach for orthographic camera.
And one of the following.
2) Recursive solution:
Basic Algorithm: Pentland 1995, Broida-Chellappa 1990(1991).
Extension: Particle filtering approaches (Gang Chellappa)
3) Fundamental Matrix based approach.
a) Estimating the fundamental matrix.
Some Implementation Details that are useful in addressing robustness of the algorithm.
Hartley, In defense of the 8-point Algorithm, ICCV 1995
Butterfield, Hoggs, An Evaluation of the Normalised 8-Point Algorithm, 1995
b) Implement Para-persp Factorization. [Costerior Kanade]
c) Implement Triggs factorization approach using the estimates of fundamental matrices from prev part.
Dataset: From Prof. Marc Polleyfeys' website.
Castle Sequence
Medusa Head
Medusa head .avi
Submission Guidelines
1) Report detailing main implementation steps (not necessarily in detail), a critique on the performance of each algorithm and failure modes , reasons for poor/superlative performance.
2) You are required to test on both datasets above. Feel free to test on other ones too.
3) Do remember to cite sources appropriately, including those involving the datasets
Feature Point Tracking
Feel free to use any feature point trackers. You are free to use any code for feature point tracking available online. KLT (Shi, Tomasi "Good Features to Track") is a popular tracker. Code for that is available in C/C++ as a part of the OpenCV libraries. Matlab may be available online.