This paper presents a high-speed corner detection method using machine learning, which can process live PAL video using less than 7% of the available processing time. Traditional feature detectors like Harris and SIFT are too slow for real-time applications. The proposed method, called FAST, is designed for speed and outperforms existing detectors in terms of repeatability and robustness to changes in viewpoint and scale. The method uses a decision tree built from machine learning to classify corners efficiently. It is tested on various datasets, including planar scenes, mazes, and bas-relief textures, and shows high repeatability. The FAST detector is particularly effective for n=9, which is the most reliable of the FAST detectors. The method is compared to other detectors, including Harris, DoG, and SUSAN, and is shown to outperform them in terms of speed and repeatability. The paper also discusses the importance of using non-planar scenes in evaluating corner detectors and highlights the limitations of the FAST detector, such as its sensitivity to noise and dependence on a threshold. The results demonstrate that the FAST detector is a fast, high-quality corner detector suitable for real-time applications.This paper presents a high-speed corner detection method using machine learning, which can process live PAL video using less than 7% of the available processing time. Traditional feature detectors like Harris and SIFT are too slow for real-time applications. The proposed method, called FAST, is designed for speed and outperforms existing detectors in terms of repeatability and robustness to changes in viewpoint and scale. The method uses a decision tree built from machine learning to classify corners efficiently. It is tested on various datasets, including planar scenes, mazes, and bas-relief textures, and shows high repeatability. The FAST detector is particularly effective for n=9, which is the most reliable of the FAST detectors. The method is compared to other detectors, including Harris, DoG, and SUSAN, and is shown to outperform them in terms of speed and repeatability. The paper also discusses the importance of using non-planar scenes in evaluating corner detectors and highlights the limitations of the FAST detector, such as its sensitivity to noise and dependence on a threshold. The results demonstrate that the FAST detector is a fast, high-quality corner detector suitable for real-time applications.