Camera Pose Estimation and 3-D Reconstruction

Computer Vision Toolbox™ provides a comprehensive set of foundational tools for estimating camera poses and reconstructing 3-D scenes from multiple views. These tools are essential for applications such as structure-from-motion, visual SLAM, augmented reality, and robotics. You can estimate camera poses using the essential matrix or fundamental matrix and also compute relative camera positions based on epipolar geometry. For calibrated cameras, you can use the perspective-three-point (P3P) algorithm or the camera projection matrix to help determine absolute or relative poses from 2-D to 3-D point correspondences. For more information, see estimateEssentialMatrix, estimateFundamentalMatrix and estrelpose.

To refine pose estimates and improve reconstruction accuracy, the toolbox supports bundle adjustment techniques for optimizing camera poses and 3-D point locations jointly or independently. For more information, see bundleAdjustment.

For 3-D reconstruction, you can triangulate points from multiple views and manage point-track correspondences. You can also convert between image and world coordinates, which supports projection and back-projection tasks. For more information, see pointTrack, triangulate and img2world2d.

The toolbox provides functionality to manage and organize image and point data using specialized objects. These structures enable pose graph creation and optimization for large-scale reconstruction pipelines. The toolbox also supports visualization of camera positions and reconstructed point clouds in 3-D space. For more information, see imageviewset, createPoseGraph and plotCamera.