Features from Accelerated Segment Test (FAST) for keypoints detection

Project described on this article

Implementation based on this paper

Principle:

• For a pixel p with intensity Ip, a circle of 16 pixels surrounding the pixel p is considered (Bresenham circle of radius 3):

• Given a threshold T, if N pixels (among the 16) are either above or below Ip plus or minus T, p is an interest point (N=12 in the first version of the algorithm)

• To accelerate the computation, first compare the intensity of pixels (1, 5, 9, 13) and compute the others afterwards only if at least three pixels of these four pixels satisfy the threshold criterion

images source: researchgate

Kanade–Lucas–Tomasi feature tracker [IN PROGRESS]

Optical flow algorithm to track key points found by the previous algorithm along a sequence.

Implementation explained in the same article in the second section.