Apparatus and method for recognizing three-dimensional shape of object

• US 4,982,438 A
• Filed: 09/01/1989
• Issued: 01/01/1991
• Est. Priority Date: 06/02/1987
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Anjyo, Kenichi, Usami, Yoshiaki, Hamada, Nobuhiro, Yanaka, Masao, Nio, Misato
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Couso, Jose L.
• Application Number:
  US07/403,858
• Time in Patent Office: 487 Days
• Field of Search: 382/1, 382/8, 382/22, 382/23, 382/25, 382/28, 382/48, 382/49, 382/57, 356/2, 356/3, 356/8, 356/11, 356/12, 250/558, 901/47, 364/516, 364/518, 364/522
• US Class Current: 382/154
• CPC Class Codes: G06T 2207/10012 Stereo images
• CPC Class Codes: G06T 7/55 from multiple images
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: H04N 13/239 using two 2D image sensors having a relative position equal to or related to the interocular distance H04N13/243 takes precedence
• CPC Class Codes: H04N 13/243 using three or more 2D image sensors
• CPC Class Codes: H04N 2013/0081 Depth or disparity estimation from stereoscopic image signals

Abstract

A three-dimensional shape of an object is recognized by developing enveloping surfaces of the object by a cone-silhouetting method, wherein the object is picked up by a binocular three-dimensional vision method by making use of geometrical restraint conditions derived by the cone-silhouetting method. The binocular parallax is increased to thereby ensure the recognition of the three-dimensional shape with improved accuracy. Local concave portions which can not be recognized by the cone-silhouetting method by itself are recognized with higher accuracy.

Claims

1. An apparatus for recognizing a three-dimensional shape of an object, comprising:

a camera system for picking up images of said object from a plurality of points of vision;

first arithmetic means for arithmetically deriving geometrical restraint conditions imposed on position and/or shape of said object from the outputs of said camera system, said geometrical restraint conditions being a set of enveloping surfaces that contact at least one point of said object and envelop said images of said object; and

second arithmetic means for arithmetically determining whether a position of intersection between viewing lines extending from two points of vision of said camera system satisfies said derived geometrical restraint conditions, to thereby determine arithmetically three-dimensional coordinates of a corresponding point on said object based on a position of intersection which satisfies said geometrical restraint conditions as well as said geometrical restraint conditions.

2. An apparatus for recognizing a three-dimensional shape of an object according to claim 1, wherein said first arithmetic means determines a closed region formed of a set of planes circumscribing said object in terms of a set of enveloping surfaces of said object, to thereby allow the property that said object exists inside of said set of enveloping surfaces which are to be the content of said geometrical restraint conditions.

3. An apparatus for recognizing a three-dimensional shape of an object according to claim 2, wherein a presumed existence region of a cone-like shape has a bottom shape corresponding to that of in object image on a picked-up picture and a vertex at a point of vision in each direction of vision, and a logical product portion common to said region in all the directions of vision is determined as a common presumed existence region, the shape of said common presumed existence region being handled as the set of enveloping surfaces.

4. An apparatus for recognizing a three-dimensional shape of an object according to claim 3, wherein the shape of said presumed existence region is described in terms of a voxel model which represents presence or absence of spatial lattice points, said common presumed existence region being determined by performing a logical product determining operation for each elementary voxel, to thereby allow the set of enveloping surfaces of the object to be described in terms of said voxel model.

5. An apparatus for recognizing a three-dimensional shape of an object according to claim 1, wherein said second arithmetic means performs a determination of correspondence of pixels and said correspondence points on the basis of an assumption that respective correspondence points correspond to pixels of a picked-up image, and wherein an evaluation function for selecting an optimum correspondence from candidates which satisfy said geometrical restraint conditions is established as a sum of differences in brightness of pairs of pixels, each pair having correspondence relation therebetween, a corresponding point on the object being determined by selecting such correspondence which minimizes said evaluation function.

6. An apparatus for recognizing a three-dimensional shape of an object according to claim 1, further including display means for displaying the results of an arithmetic operation performed by said first arithmetic means and/or the results of an arithmetic operation performed by said second arithmetic means.

7. A method of recognizing a three-dimensional shape of an object, comprising the steps of:

obtaining a general shape of the object to be recognized in terms of a set of enveloping surfaces that contact at least one point of said object and envelop images of said object;

picking up said images of said object by means of left and right camera systems disposed horizontally;

identifying regions of said picked-up images over which said object is projected as object images;

extracting brightness values of pixels from regions of said object images on scanning lines of said images of said left and right camera systems;

determining straight lines interconnecting positions of said pixels of said object images and a position of a point of vision in a real three dimensional space independently for each of said images of said left and right camera systems, respectively;

determining arithmetically as candidates for selection those combinations of pixels of said images of the left and right camera systems which have associated straight lines intersecting each other, said intersections being located within a set of enveloping surfaces;

selecting such a combination of pixels from said candidates in which a sum of differences in brightness values between left and right pixels put into correspondence relations, respectively, is a minimum; and

correcting a shape defined by said set of enveloping surfaces by utilizing information of intersections derived from said selected combinations to thereby establish a final three-dimensional shape of said object.

8. A method of recognizing a three-dimensional shape of an object according to claim 7, further including the steps of:

picking up images of said object by a camera system in a plurality of viewing directions;

determining a presumed existence region of a cone-like shape having a bottom plane in a shape corresponding to an image of said object and a vertex corresponding to the point of vision in each of said viewing directions; and

determining as a common presumed existence region a logical product region common to presumed existence regions in all the viewing directions, wherein said common presumed existence region is regarded to be of a general shape of said object defined by said set of enveloping surfaces.