openMVG/openMVG
OpenMVG
openMVG es una librería y toolkit de visión artificial para geometría de múltiples vistas. Sirve como framework para la estructura a partir del movimiento (SfM) y la reconstrucción de escenas 3D, proporcionando las herramientas necesarias para recuperar nubes de puntos 3D y poses de cámaras a partir de colecciones de imágenes 2D.
La librería implementa pipelines de SfM tanto globales como incrementales. Utiliza algoritmos geométricos para calcular la estimación de la pose de la cámara y la localización de imágenes, empleando el ajuste de haces (bundle adjustment) de Levenberg-Marquardt para refinar las coordenadas 3D y los parámetros de la cámara minimizando el error de reproyección.
El proyecto cubre una amplia gama de capacidades de recuperación espacial, incluyendo la extracción de características de imagen, correspondencia de pares y recuperación 3D basada en triangulación. También incluye sistemas para organizar datos de reconstrucción y exportar información de escenas para su uso en herramientas de visualización externas.
Features
- Computer Vision Libraries - Provides a comprehensive library for processing visual data to perform 3D reconstruction and camera pose estimation.
- Structure from Motion Pipelines - Implements full structure-from-motion pipelines to generate 3D point clouds and camera poses from 2D images.
- Stereo Triangulation - Determines the 3D position of points by intersecting rays from multiple cameras observing the same visual feature.
- Scene Reconstruction - Creates 3D point clouds and camera positions from sets of 2D images using structure from motion.
- Local Feature Matching - Identifies shared visual landmarks between images by comparing local descriptors to establish geometric constraints.
- Image Pair Matching - The library calculates pairwise correspondences between images to identify shared visual content.
- Multi-View Geometry Solvers - Uses mathematical relationships between images to triangulate 3D landmarks and spatial layouts.
- Camera Pose Estimators - Calculates camera orientations and positions from imagery to enable triangulation-based 3D recovery.
- Image Feature Engineering - Maintains point-based features and associated descriptors for all images in a dataset to enable matching.
- Image Localizations - Provides tools to calculate the position and orientation of new images relative to an existing 3D scene.
- Bundle Adjustment Algorithms - Employs Levenberg-Marquardt bundle adjustment to refine 3D coordinates and camera parameters by minimizing reprojection error.
- Epipolar Geometry - Analyzes the geometric relationship between corresponding points in image pairs to calculate relative camera motion.
- 3D Reconstruction Pipelines - Provides a sequential pipeline for extracting image features, matching pairs, and estimating 3D structures.
- Global Solvers - Computes all camera poses and 3D points simultaneously by solving a large-scale linear system from all image matches.
- Incremental Solvers - Builds 3D scenes by adding cameras and points one by one and refining the model through local bundle adjustment.
- Robotics Algorithms - Multiple view geometry library for 3D vision.
- Structure From Motion - Library for multiple view geometry and 3D reconstruction.
- Matching and Verification - Library for multiple view geometry and feature matching.
- Structure From Motion - Modular library for multiple view geometry and reconstruction.
