Faster and Better: A Machine Learning Approach to Corner Detection

Corner detection is used for a large variety of computer vision tasks, such as object detection and recognition, tracking and image registration and stitching. The repeatability and efficiency of a corner detector determines how likely it is to be useful in a real-world application. The repeatability is important because the same scene viewed from different positions should yield features which correspond to the same real-world 3D locations.  The efficiency is important because this determines whether the detector combined with further processing can operate at frame rate.

This presentation is about the FAST corner detector. Firstly this consists of a new heuristic which can be optimized for speed effectively using machine learning so that it can process live video using  less than 3% of the CPU budget on a modern machine.  Secondly, by using the definition of good corner detection, we generalize the FAST detector so that we can optimize it for repeatability with little loss of efficiency.  Finally, a thorough experimental comparison demonstrates that the new detector produces significant improvements in repeatability, yielding a detector that is both very fast and very high quality.