Structure-from-Motion

The aim was to construct 3D sparse reconstruction of a scene using a set of unordered images.

Softwares

• Conding was done on Python using the opencv library
• The visualisation was done on MeshLab

Algorithm

• Extract features between two images using fast features
• Estimate the relative pose of 2nd frame wrt 1.
• Triangulate common features found between two images to obtain 3D point.
• Append the point in a .ply file
• Repeat the above 4 steps for the series of images.

Dataset

The dataset was taken from here The calibration matrix for the same can be ontained from the respective data.mat files

• Results: