Pankaj Daga: March 2015

The prototype for the image mosaicing workflow has been created using Python. It seems that SIFT features give the most reasonable results.

Areas for making novel contributions have also been identified as:

• Use of Bayesian methods to perform outlier detection int the feature set. Current techniques using RANSAC/MAPSAC/XSAC might be too heuristic and there is scope for improvement.
• The use of perspective projection/homography as the transformation model seems inadequate. Alternate model including the infinitesimal plane approximation or piecewise homographies need to be investigated.
• Feature tracking using Bayesian filtering methods could help with 3D scene reconstruction. This could be something interesting for Marcel to focus on during his first year PhD.

Pankaj Daga