Apparatus for extracting image object in 3D image system and method thereof

• US 9,294,753 B2
• Filed: 11/28/2012
• Issued: 03/22/2016
• Est. Priority Date: 04/25/2012
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Lee, Su Woong, Lee, Jun Suk, Kang, Seok Bin, Lim, Suk Hyun
• Primary Examiner(s): Tseng, Charles
• Application Number:
  US13/687,800
• Publication Number:
  US 20130286162A1
• Time in Patent Office: 1,210 Days
• US Class Current: 1/1
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: H04N 13/189 Recording image signals; Reproducing recorded image signals

Abstract

Disclosed are an apparatus for extracting an image object in a 3D image system and a method thereof.

The apparatus for extracting an image object in a 3D image system includes: an image inputting unit receiving an image; a space setting unit setting virtual space at which the image object is positioned from the received image; and an object extracting unit extracting the image object from the set virtual space from the received image.

Claims

1. An apparatus for extracting an image object, comprising a non-transitory computer-readable medium tangibly embodying computer-readable instructions, the instructions when executed by a computer implementing:

an image inputting unit receiving an image;

a space setting unit setting virtual space at which the image object is positioned from the received image;

an object extracting unit extracting the image object from the set virtual space from the received image; and

an object correcting unit correcting an image including the extracted image object at least in part by calculating a probability associated with a type allocated to a pixel in the image, the type indicating a level of confidence relating to whether the pixel belongs to a foreground or a background of the image,

wherein;

the object correcting unit

uses visible light information,

allocates the type for each pixel for each block having a predetermined size in the image including the image object, and

calculates a probability that a pixel allocated as a type in which a foreground or a background is obscurely discriminated belongs to the foreground by using a difference in distance and a difference in color from an adjacent pixel, and corrects the image object according to the calculated probability, and

the space setting unit extracts XYZ coordinates of apexes and a center point of polyhedral space as the set virtual space and converts the extracted XYZ coordinates into projection coordinates, and the object extracting unit calculates positions of the center point and each pixel in the image using boundary plans constituting the set virtual space on a predetermined projection coordinate system according to the projection coordinates and determines whether each pixel belongs to a foreground.

2. The apparatus of claim 1, wherein:

the space setting unit receives a plurality of points and sets virtual space including one floor plane constituted by the plurality of received points.

3. The apparatus of claim 2, wherein:

the space setting unit receives the plurality of points, and when the number of points of which the depths are measured validly among points adjacent to the corresponding point for each of the plurality of received points is equal to or more than a predetermined number, sets one floor plane including the corresponding point and sets the virtual space including the one set floor plane.

4. The apparatus of claim 1, wherein:

the object correcting unit allocates −

1 to pixels judged as the background positioned in the virtual space, allocates 0 to pixels in which the foreground or the background is obscurely discriminated, and allocates 1 to pixels judged as the foreground.

5. The apparatus of claim 1, wherein:

the object correcting unit further

averages predetermined calculated probabilities of pixels adjacent to some other remaining pixels to estimate a probability with an average value.

6. The apparatus of claim 1, wherein:

the object correcting unit further

mixes a pixel value of a previous image and a pixel value of a current image that are positioned at a same position for each pixel of the image including the corrected image object at a predetermined ratio to correct the image object according to a mixing ratio.

7. A method for extracting an image object, comprising using a computer to execute instructions stored on a computer-readable medium, the instructions when executed performing operations including:

receiving an image;

setting a virtual space at which the image object is positioned from the received image;

extracting the image object from the set virtual space from the received image;

correcting an image including the extracted image object at least in part by calculating a probability associated with a type allocated to a pixel in the image, the type indicating a level of confidence relating to whether the pixel belongs to a foreground or a background of the image; and

correcting the image including the extracted image object for each block having a predetermined size,

wherein the setting the virtual space includes extracting XYZ coordinates of apexes and a center point of polyhedral space as the set virtual space and converts the extracted XYZ coordinates into projection coordinates, and the extracting the image object includes calculating positions of the center point and each pixel in the image using boundary plans constituting the set virtual space on a predetermined projection coordinate system according to the projection coordinates and determines whether the each pixel belongs to a foreground, and

wherein the correcting the image includes;

allocating the type for each pixel for each block having a predetermined size in the image including the image object;

calculating a probability that a pixel allocated as a type in which a foreground or a background is obscurely discriminated belongs to the foreground using a difference in distance and a difference in color from an adjacent pixel; and

correcting the image object according to the calculated probability.

8. The method of claim 7, wherein the setting virtual space includes:

receiving a plurality of points; and

setting the virtual space including one floor plane constituted by the plurality of received points.

9. The method of claim 8, wherein the setting virtual space includes:

receiving a plurality of points;

setting one floor plane including the corresponding point when the number of points of which the depths are measured validly among points adjacent to the corresponding point for each of the plurality of received points is equal to or more than a predetermined number; and

setting the virtual space including the one set floor plane.

10. The method of claim 7, wherein the correcting the image further includes:

averaging predetermined calculated probabilities of pixels adjacent to some other remaining pixels and estimating a probability with an average value.

11. The method of claim 7, wherein the correcting the image further includes:

mixing a pixel value of a previous image and a pixel value of a current image that are positioned at a same position for each pixel of the image including the corrected image object at a predetermined ratio and correcting the image object according to a mixing ratio.