Keypoint recognition using randomized trees

By Vincent Lepetit, Pascal Fua. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006

783 citations recorded by the source corpus.

In many 3D object-detection and pose-estimation problems, runtime performance is of critical importance. However, there usually is time to train the system, which we would show to be very useful. Assuming that several registered images of the target object are available, we developed a keypoint-based approach that is effective in this context by formulating wide-baseline matching of keypoints extracted from the input images to those found in the model images as a classification problem. This shifts much of the computational burden to a training phase, without sacrificing recognition performance. As a result, the resulting algorithm is robust, accurate, and fast-enough for frame-rate performance. This reduction in runtime computational complexity is our first contribution. Our second contribution is to show that, in this context, a simple and fast keypoint detector suffices to support detection and tracking even under large perspective and scale variations. While earlier methods require a detector that can be expected to produce very repeatable results, in general, which usually is very time-consuming, we simply find the most repeatable object keypoints for the specific target object during the training phase. We have incorporated these ideas into a real-time system that detects planar, nonplanar, and deformable objects. It then estimates the pose of the rigid ones and the deformations of the others

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/aa6557c658aed10ff303aa90fe8d0952332a43c0

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

Visual simultaneous localization and mapping: a survey (2012, methodology): “Other researches like (Lepetit and Fua 2006; Grauman 2010; Kulis et al. 2009; Özuysal et al. 2010) use learning strategies to determine similarity between features.”
Crater Detection Based on Gist Features (2015, methodology): “We adopted the random forest for feature selection and classification of craters samples, since it runs efficiently on large datasets and can handle thousands of input variables without variable deletion [24], [25].”
Decision Forests for Classification, Regression, Density Estimation, Manifold Learning and Semi-Supervised Learning (2011, background): “Classification forests have also been employed successfully in a number of practical applications [23, 54, 74, 83, 100].”
Fast Keypoint Recognition Using Random Ferns (2010, methodology): “We will demonstrate that the approach we propose is even faster than the trees [20], more scalable, and just as reliable.”
Scene Coordinate Regression Forests for Camera Relocalization in RGB-D Images (2013, background): “[18] learn to match particular discrete features in images with random forests.”
Evaluation of Interest Point Detectors and Feature Descriptors for Visual Tracking (2011, methodology): “Many of their building blocks are derived from or inspired by ideas mentioned above (Winder and Brown 2007, Winder et al. 2009). proposed a new approach to robust keypoint recognition: they formulate keypoint matching as a classification problem using Randomized Trees as classifiers .”
Detailed 3D Representations for Object Recognition and Modeling (2013, methodology): “They have been successfully used to train detectors for interest points [32], [30].”
Iterated Second-Order Label Sensitive Pooling for 3D Human Pose Estimation (2014, methodology): “Kinect[16] used fast feature extraction based on depth differences inspired by [23], to infer a dense part labeling of the human body from depth images.”
Decision Forests: A Unified Framework for Classification, Regression, Density Estimation, Manifold Learning and Semi-Supervised Learning (2012, positive): “Random decision forests where used in [63] for image classiﬁcation and in [60] for keypoint tracking in videos.”
Egocentric Shopping Cart Localization (2018, neutral): “SIFT or SURF [12], [13]).”
Image Classification using Random Forests and Ferns (2007, methodology): “Recent implementations of random forests [17, 24] have used quite simple pixel level tests at the nodes (for reasons of speed).”
Hough-based tracking of non-rigid objects (2011, methodology): “They are used for various tasks, such as key-point recognition [22] or semantic segmentation [31].”
Optimised KD-trees for fast image descriptor matching (2008, methodology): “Our KD-tree method is related to randomized trees, as used in [9], based on earlier work in [1].”
Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning (2018, background): “Both 2D and 3D keypoint detection are long-standing problems in computer vision, where keypoint inference is traditionally used as an early stage in object localization pipelines [27].”
Tracking-Learning-Detection (2012, background): “Similarly as in [60], [61], and [62], the pixel comparisons are generated offline at random and stay fixed in runtime.”
Comparative Evaluation of Hand-Crafted and Learned Local Features (2017, methodology): “For example, PCA-SIFT [22] uses principal component analysis (PCA) to embed a gradient image of a patch while Lepetit and Fua [25] embed patches using a random forest.”
Real time head pose estimation with random regression forests (2011, methodology): “Random forests [2] have become a popular method in computer vision [11, 10, 20, 14, 12] given their capability to handle large training datasets, high generalization power, fast computation, and ease of implementation.”