7: Feature Extraction
( \newcommand{\kernel}{\mathrm{null}\,}\)
A robot can obtain information about its environment by both active (e.g., ultra-sound, light, and laser) or passive sensing (e.g., acceleration, magnetic field, or cameras). There are only few cases where this information is directly useful to a robot. Before being able to arrive at semantic information such as “I’m in the kitchen”, “this is a cup” or “this is a horse”, is identifying higher-level features.
The goal of this chapter is to introduce a series of standard feature detectors such as
- the Hough-transform to detect lines, circles and other shapes,
- numerical methods such as least-squares, split-and-merge and RANSAC to find high-level features in noisy data,
- Scale-invariant features.