Noname manuscript No. (will be inserted by the editor) EPnP: An Accurate O(n) Solution to the PnP Problem
Citations in the corpus
Pix3D: Dataset and Methods for Single-Image 3D Shape Modeling (2018, methodology): “To solve Equation 1, we first calculate a rough 3D pose using the Efficient PnP algorithm [35] and then refine it using the Levenberg-Marquardt algorithm [36, 39], as shown in Figure 2.”
Structure-from-Motion Revisited (2016, background): “A large number of 2D-3D correspondences provides this estimation with redundancy [34], while a uniform distribution of points avoids bad configurations and enables reliable estimation of intrinsics [41].”
ORB: An efficient alternative to SIFT or SURF (2011, methodology): “We apply rBRIEF to object recognition by implementing a conventional object recognition pipeline similar to [19]: we first detect oFAST features and rBRIEF descriptors, match them to our database, and then perform PROSAC [7] and EPnP [16] to have a pose estimate.”
Real-Time Seamless Single Shot 6D Object Pose Prediction (2017, methodology): “BB8 [25] is a 6D object detection pipeline made of one CNN to coarsely segment the object and another to predict the 2D locations of the projections of the object’s 3D bounding box given the segmentation, which are then used to compute the 6D pose using a PnP algorithm [16].”
A Direct Least-Squares (DLS) method for PnP (2011, methodology): “, based on successive Linear Matrix Inequality (LMI) relaxations [15], transformation to a Semi-Definite Program (SDP) [23], or a geometric transformation of the problem [16].”
6-DoF object pose from semantic keypoints (2017, methodology): “As a simple baseline, the following greedy approach was implemented: the maximum response locations in the heatmaps were selected as 2D keypoint locations and the standard PnP problem was solved by the EPnP algorithm [4] to”
The MOPED framework: Object recognition and pose estimation for manipulation (2011, methodology): “In the case of point-based features, this is known as the Perspective-n-Point (PnP) problem (Fischler and Bolles 1981), for which many solutions are available, both closed-form (Lepetit et al. 2008) and iterative (Dementhon and Davis 1995).”
Segmentation-Driven 6D Object Pose Estimation (2018, methodology): “Given these filtered 2D-to-3D correspondences, we obtain the 6D pose of each object using the RANSAC-based version of the EPnP algorithm of [20].”
On Face Segmentation, Face Swapping, and Face Perception (2017, methodology): “Similarly to [13], we use these correspondences to estimate 3D pose, computing 3D face rotation, R ∈ R 3 , and translation vector t ∈ R 3 using the EPnP solver [25].”
OpenMVG: Open Multiple View Geometry (2016, methodology): “– Absolute pose from pairs of 3D-2D matching points by different algorithms, P3P (Perspective from 3 Points) [16], DLT (Direct Linear Transform) [6] (6 pairs), ePnP [15] (n pairs).”
StarMap for Category-Agnostic Keypoint and Viewpoint Estimation (2018, methodology): “It has been well explored by traditional techniques that solve for transformations between corresponding points in the world and image views; this is known as the Perspective-n-Point Problem [13,18].”
DeepV2D: Video to Depth with Differentiable Structure from Motion (2018, methodology): “We unroll a single iteration of Perspective-n-Point (PnP) (Lepetit et al., 2009; Li et al., 2012) performing Gauss-Newton update over SE3 perturbations to minimize geometric reprojection error.”