mapillary/OpenSfM
OpenSfM
OpenSfM is a computer vision library and structure-from-motion pipeline designed to reconstruct three-dimensional scenes and camera trajectories from overlapping images. It functions as a 3D reconstruction engine and photogrammetry toolkit, utilizing automated feature-based image matching and incremental bundle adjustment to derive spatial geometry.
The system distinguishes itself as a geospatial alignment tool, integrating GPS and inertial sensor data to align reconstructed 3D models with real-world geographic coordinates. It employs a hybrid Python and C++ execution model to manage large-scale image datasets through graph-based view clustering, which partitions data into manageable sub-graphs to reduce computational complexity.
The framework covers a broad range of photogrammetry capabilities, including multi-view stereo depth estimation for generating dense point clouds and web-based visualization for inspecting reconstructed models and camera paths.
Features
- Structure from Motion Pipelines - Converts overlapping photographs into accurate 3D point clouds and camera trajectories using computer vision algorithms.
- Computer Vision Libraries - Provides a hybrid Python and C++ framework for processing large-scale image datasets into spatial geometry.
- Sensor-Fused - Aligns local 3D coordinate systems with global geographic data by integrating GPS and inertial measurements.
- Image Pair Matching - Identifies corresponding points between images using local descriptors to establish geometric constraints.
- 3D Scene Reconstructions - Generates three-dimensional scenes and camera poses from multiple images using feature detection and matching.
- Geometric Feature Matching - Identifies corresponding points between images using local descriptors to establish geometric constraints.
- 3D Reconstruction Pipelines - Provides an automated pipeline for feature-based image matching and incremental bundle adjustment of camera parameters.
- Photogrammetry Data Pipelines - Implements automated image processing workflows to extract spatial geometry and camera parameters from large datasets.
- Bundle Adjustment Algorithms - Refines camera parameters and 3D point positions by iteratively minimizing reprojection errors.
- Geospatial Model Alignments - Integrates GPS and accelerometer data to align 3D models with real-world geographical coordinates.
- View Clustering - Implements view clustering to partition large image datasets into sub-graphs, reducing computational overhead during reconstruction.
- Multi-View Depth Estimators - Calculates pixel-wise depth maps from overlapping image pairs to generate dense point clouds.
- Multi-View Stereo Generators - Generates dense 3D geometry by calculating depth maps from overlapping image pairs using spatial consistency.
- Graph Partitioning Utilities - Organizes large image collections into manageable sub-graphs to reduce the computational complexity of reconstruction.
- Photogrammetry Orchestrators - Offers a toolkit for generating dense 3D point clouds using multi-view stereo depth estimation.
- Python-C++ Bindings - Combines high-level Python orchestration with performance-critical C++ numerical routines for efficient image processing.
- Structure From Motion - Python-based library for structure from motion.
- Structure From Motion - Python-based library for structure from motion pipelines.
