Faster and Better: A Machine Learning Approach to Corner Detection

By Edward Rosten, Reid B. Porter, Tom Drummond. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008

2,005 citations recorded by the source corpus.

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. Three advances are described in this paper. First, we present a new heuristic for feature detection and, using machine learning, we derive a feature detector from this which can fully process live PAL video using less than 5 percent of the available processing time. By comparison, most other detectors cannot even operate at frame rate (Harris detector 115 percent, SIFT 195 percent). Second, we generalize the detector, allowing it to be optimized for repeatability, with little loss of efficiency. Third, we carry out a rigorous comparison of corner detectors based on the above repeatability criterion applied to 3D scenes. We show that, despite being principally constructed for speed, on these stringent tests, our heuristic detector significantly outperforms existing feature detectors. Finally, the comparison demonstrates that using machine learning produces significant improvements in repeatability, yielding a detector that is both very fast and of very high quality.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/a5d3a937e5ee43ab1542657ea2baf0c5cb139d6e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Pose Estimation for Augmented Reality: A Hands-On Survey (2016, neutral): “AGAST [77] further improved computational cost against FAST by dynamically changing the optimal conﬁguration of pixels for similarity measurement.”
PL-VIO: Tightly-Coupled Monocular Visual–Inertial Odometry Using Point and Line Features (2018, background): “If the number of tracked point features is less than a threshold after outlier rejection, new corner features which are detected by the FAST detector [38] will be added.”
Learning to Detect Good 3D Keypoints (2017, background): “As far as images are concerned, the most important contribution is probably FAST (Rosten et al. 2010), which is based on the Accelerated Segment Tests to detect corner-like features: the order of the tests is learned in a tree from a training set to speed-up detection time.”
Learning 6D Object Pose Estimation Using 3D Object Coordinates (2014, background): “More recently a trend has been to learn interest points [22,10], descriptors [30], and matching [14,19,1].”
Robust monocular SLAM in dynamic environments (2013, methodology): “Our framework is similar to PTAM, but we employ the invariant features (i.e. SIFT [20]) instead of FAST [33] with keyframe representation to perform global matching, which is more robust to fast movement and camera relocalization, and can be easily extended to handle a larger scale scene.”
Scene Coordinate Regression Forests for Camera Relocalization in RGB-D Images (2013, background): “Much work has focused on efficiency, scalability, and learning of feature detection [13, 26], description [4, 27, 34], and matching [18, 25, 29].”
Adaptive and Generic Corner Detection Based on the Accelerated Segment Test (2010, methodology): “The size of the arc angle controls the repeatability, as shown in [9], and the robustness against blur.”
Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters (2019, methodology): “Further works extended FAST by optimizing it [25], adding a descriptor [26] or orientation estimation [27].”
SVO: Semidirect Visual Odometry for Monocular and Multicamera Systems (2017, methodology): “We ﬁrst compare the proposed sparse image alignment algorithm against semi-dense and dense image alignment algorithms and investigate the inﬂuence of the patch size used in the sparse approach.”
BRIEF: Computing a Local Binary Descriptor Very Fast (2012, methodology): “In the case of BRIEF, any fast detector such as CenSurE [34] or FAST [35], [36] can be used.”
LF-Net: Learning Local Features from Images (2018, methodology): “However, many other approaches target some of their individual components, be it feature point extraction [31, 44], orientation estimation [50], or descriptor generation [36, 40, 41].”
TILDE: A Temporally Invariant Learned DEtector (2014, methodology): “We compare our method with the SIFT [20], SURF [3], and FAST-9 [28] detectors, using the same number of keypoints (300) for all methods.”
RIFT: Multi-Modal Image Matching Based on Radiation-Variation Insensitive Feature Transform (2019, methodology): “We compare the detector of RIFT with other six well-known feature detectors, including FAST [28], Harris [32], Brisk [56], SIFT [15], SURF [24], and MSER [33].”
High-Level Semantic Feature Detection: A New Perspective for Pedestrian Detection (2019, background): “Generally speaking, it mainly includes edge detection [4, 41], corner detection [37, 38], blob detection [33, 7] and so on.”
Recent progress in semantic image segmentation (2018, neutral): “…(Medioni andYasumoto 1987), SUSAN (Smith and Brady 1997), Features from Accelerated Segment Test (FAST) (Rosten and Drummond 2005), FAST- ER (Rosten et al. 2010), AGAST (Mair et al. 2010) and Multiscale AGAST (Leutenegger et al. 2011) Detector, Bag-of-visual-words (BOV) (Csurka et al.…”
A review of multimodal image matching: Methods and applications (2021, methodology): “The same concept, with a comparison of the center against the pixels along a circular pattern, was used in the famous FAST detector [144], which was further improved to enhance the repeatability [136] and robustness [145].”
Vision-based autonomous mapping and exploration using a quadrotor MAV (2012, methodology): “First, we extract and match FAST corners [21] and BRIEF descriptors [6] from the left cameras of both the reference and current frames.”