Method and system for recovery of 3D scene structure and camera motion from a video sequence

摘要

An improved method and a system are disclosed for recovering a three-dimensional (3D) scene structure from a plurality of two-dimensional (2D) image frames obtained from imaging means. Sets of 2D features are extracted from the image frames, and sets corresponding to successive image frames are matched, such that at least one pair of matched 2D features refers to a same 3D point in a 3D scene captured in 2D in the image frames. A 3D ray is generated by back-projection from each 2D feature, and the generated 3D rays are subjected to an anchor-based minimization process, for determining camera motion parameters and 3D scene points coordinates, thereby recovering a structure of the 3D scene.

主权项

1. A method of recovering a 3D scene structure from a plurality of image frames obtained from imaging means, the method comprising the steps of
extracting a first set of 2D features from a first image frame, extracting a second set of 2D features from a second image frame, matching the second set with the first set, such that at least one pair of matched 2D features refers to a same 3D point in a 3D scene, generating a 3D ray by back-projection from each 2D feature, subjecting the generated 3D rays to an anchor-based minimization process, for determining camera motion parameters and 3D scene points coordinates, thereby recovering a structure of said 3D scene, and processing the determined camera motion parameters using 3D ray intersection by minimizing a function based on the Euclidian distance between the 3D rays; wherein the anchor-based minimization process comprises selecting a 3D point in the 3D scene as an anchor, being a centre of rotation, such that the camera motion parameters can be expressed relative to the anchor point.