Robust 3D Reconstruction by Feature Tracking over Large Displacements
Aldo Cumani · 2002
Abstract: Accurate recovery of geometric structure from an image sequence strongly depends upon two contrasting requirements: numerical conditioning, needing large image disparities, and ease of matching, which needs small ones. This work discusses a solution for an actively controlled observer (a camera on the end-effector of a robot arm) using feature tracking of image features along the camera trajectory. Restricting the scene to objects with straight line edges allows easy assessment of the reconstruction accuracy. Key-Words: 3D reconstruction – Feature matching – Trifocal tensor – Reconstruction accuracy 1