System and method for reconstructing surface elements of solid objects in a three-dimensional scene from a plurality of two dimensional images of the scene

• US 6,094,198 A
• Filed: 01/27/1997
• Issued: 07/25/2000
• Est. Priority Date: 01/10/1994
• Status: Expired due to Fees

• Current Assignee: CogniTens Ltd. (Hexagon AB)
• Original Assignee: CogniTens Ltd. (Hexagon AB)
• Inventors: Shashua, Amnon
• Primary Examiner(s): Nguyen, Phu K.
• Assistant Examiner(s): Vo, Cliff N.
• Application Number:
  US08/787,345
• Time in Patent Office: 1,275 Days
• Field of Search: 345/419, 345/422, 345/420, 345/423, 345/424, 348/40, 348/42, 348/51
• US Class Current: 345/419
• CPC Class Codes: G06T 15/10 Geometric effects

Abstract

An arrangement for generating reconstructon information to facilitate reconstruction of three-dimensional features of objects in a scene based on two-dimensional images of the scene taken from a plurality of locations. The arrangement includes a plurality of elements including an epipole generating means, a homography generating means and a depth value generating means. The epipole generating means identifies the location of epipoles, that is, the coordinates in each image plane, in which the images were recorded, of the point of intersection of the line interconnecting the centers of projection of the image recorders that record the images. The homography generating means uses the epipoles and the coordinates in the respective images of selected reference points to generate a homography that relates the coordinates of all points in the images. Finally the depth value generating means uses the homography generated by the homography generating means and the coordinates of the projection of at least one other point, other than the selected reference points, in the images to generate, for generating a depth value representative of the distance of the other point relative to the location of at least one of the image recorders. Using the depth values generated for an number of points of objects in the scene and the coordinates of the projections of the points in at least one of the image planes, as determined from the images, the three-dimensional (Euclidean) structures of the objects can be determined.

Claims

1. A method for generating a reconstruction of a scene from two two-dimensional images of at least one object in a three-dimensional scene as recorded from a plurality of diverse locations, each location defining an image plane for one of said images, including:

A. identifying, for each image plane, coordinates defining an epipole location, the epipole location for each said image plane comprising a point of intersection between of a line interconnecting the locations and the respective image plane;

B. using the epipoles and coordinates in the respective image planes of selected reference points in the scene which define at least one reference plane to generate a homography that relates coordinates of projections of the selected respective points onto the image planes;

C. using the homography generated by the homography generating means and coordinates of the projection of at least one additional point in the respective image planes to generate a depth value; and

D using coordinates of projection of at least one point of the object onto the image planes and the depth value generated by the depth value generator for said at least one point for generating a reconstruction of said at least one point in the scene.