Mostly just linked to CV notes…

  • 9.1 Explains that Feature extraction is about information reduction
  • 9.2 lists some common features and ways to extract them
  • 9.3 covers Line detection specifically
  • 9.4 talks about SIFT - just re-routing that to SIFT notes from CV
  • 9.5 explains that most of the things done in this could easily be done by NN but still useful to know these things for pre-processing and such

Take home lessons:

  1. Features are “interesting” information in sensor data that are robust to variations in rotation and scale as well as noise.
  2. Which features are most useful depends on the characteristics of the sensor generating the data, the structure of the environment, and the actual application.
  3. There are many feature detectors available some of which operating as simple filters, others relying on machine learning techniques.
  4. Lines are among the most important features in mobile robotics as they are easy to extract from many different sensors and provide strong clues for localization.