October 14, 2008 | Edward Rosten, Reid Porter, and Tom Drummond
The paper presents a new heuristic for feature detection and a machine learning-derived feature detector that can process live PAL video using less than 5% of the available processing time, outperforming other detectors like the Harris detector (115%) and SIFT (195%). The detector is generalized to optimize for repeatability with minimal loss in efficiency. A rigorous comparison of corner detectors based on repeatability in 3D scenes shows that the heuristic detector significantly outperforms existing detectors, demonstrating the effectiveness of machine learning in improving repeatability and detector quality. The paper also discusses previous work on corner detectors, including edge-based, greylevel derivative-based, and direct greylevel detectors, and reviews methods for evaluating feature detectors. The FAST detector and its enhanced version, FAST-ER, are introduced, along with their implementation details and performance metrics. The results show that the FAST-ER detector achieves high repeatability and speed, making it suitable for real-world applications.The paper presents a new heuristic for feature detection and a machine learning-derived feature detector that can process live PAL video using less than 5% of the available processing time, outperforming other detectors like the Harris detector (115%) and SIFT (195%). The detector is generalized to optimize for repeatability with minimal loss in efficiency. A rigorous comparison of corner detectors based on repeatability in 3D scenes shows that the heuristic detector significantly outperforms existing detectors, demonstrating the effectiveness of machine learning in improving repeatability and detector quality. The paper also discusses previous work on corner detectors, including edge-based, greylevel derivative-based, and direct greylevel detectors, and reviews methods for evaluating feature detectors. The FAST detector and its enhanced version, FAST-ER, are introduced, along with their implementation details and performance metrics. The results show that the FAST-ER detector achieves high repeatability and speed, making it suitable for real-world applications.