AST

The latest class of feature point extractors falls under the name of Accelerated Segment Test developed by Rosten. Currently, there are three slightly different versions of this algorithm.

Figure 5.5: FAST: the 16 pixels on the circumference of radius 3 on which to perform the continuity test.

The first version of Features from Accelerated Segment Test FAST (RD05) is probably the most intuitive: in this case, points are considered characteristic if they have a continuous sequence of $n$ pixels along a circumference of a given radius, all being either (more or less) luminous than the central pixel used as a reference for the gray tone. In the case of FAST-9, for example, the 16 pixels on the circumference of radius 3 are analyzed to check if there are 9 contiguous pixels that are all above or all below a certain threshold relative to the central pixel. In subsequent versions (RD06), the extraction is optimized through the use of decision trees trained to identify feature points that maximize the local amount of information. These trees consistently process the pixels on the circumference.

This approach has become typical in recent years, as the abundance of public datasets has led to widespread use of classifiers to construct stable feature point detectors. In fact, given primitives that describe the neighborhood of a point, the application of an optimization technique allows for the identification of those that exhibit greater stability for the specific task. The article by Rosten, among other things, provides an excellent survey of previous feature point extraction techniques.

In the latest variant (FAST-ER), the area to be analyzed is extended not only to the points on a circumference but to all the pixels in the vicinity of the central point.