In this post is treated the first step for the reconstruction, considering different options to get a set of points to reconstruct.

Keypoints

The selection of the points for the reconstruction is the key to get a solid mesh of points, easily findable by the matching algorithm and speeding up the process of the reconstruction.

Among the several methods considered for the point selection are:

• Edge filters: Edge filtering is the easiest way to obtain a reduced set of points with enough variance to find them on a correspondent image. Some of the most known edge filters are Sobel, Robers or Canny (shown below).
• Lines or corners: Some methods employ only corner points easily findable, like the Harris detector.
• Feature selection: Some algorithm like SIFT or SURF use features to find special points easily distinguished.
• EVERY POINT!: This is the last option, viable when we want a dense reconstruction of the scene, but considerably slower than the others.