Method for segmenting and estimating a moving object motion using a hierarchy of motion models

• US 5,734,737 A
• Filed: 07/13/1995
• Issued: 03/31/1998
• Est. Priority Date: 04/10/1995
• Status: Expired due to Term

• Current Assignee: Wi-LAN Inc.
• Original Assignee: Daewoo Electronics (Government of South Korea)
• Inventors: Jung, Hae-Mook, Kim, Seong-Dae, Choi, Jae-Gark, Lee, Si-Woong, Cho, Soon-Jae, Chang, Gyu-Hwan
• Primary Examiner(s): Razavi, Michael T.
• Assistant Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/502,228
• Time in Patent Office: 992 Days
• Field of Search: 382/236, 382/107, 382/173, 348/402, 348/407, 348/413, 348/416
• US Class Current: 382/107
• CPC Class Codes: G06T 7/254 involving subtraction of images
• CPC Class Codes: H04N 19/53 Multi-resolution motion estimation; Hierarchical motion estimation

Abstract

A method, for use in an object-oriented analysis-synthesis coder, for segmenting an image signal into moving objects contained in a current frame and describing each motion of the moving objects with a set of motion parameters using a multi-stage segmentation comprises the steps of: (a) comparing the current frame with its preceding frame to detect a moving area in the current frame; (b) further comparing the current frame with the preceding frame to detect a motion vector for each pixel in the moving area, i.e., a motion vector field; (c) first segmenting the motion vector field by using a two dimensional translational patch model; (d) secondly segmenting the first stage segmenting results by using a planar patch model; and (e) thirdly segmenting the second-stage segmenting results by using a parabolic patch model.

Claims

1. A method, for use in an object-oriented analysis-synthesis coder, for segmenting an image signal, said image signal containing moving objects therein, of a current frame into regions according to the moving objects therein and describing each motion of the moving objects with a set of motion parameters based on a hierarchy of motion models, which comprises the steps of:

(a) comparing the current frame with a preceding frame to detect a moving area in the current frame;

(b) further comparing the current frame with the preceding frame to detect a motion vector for each pixel in the moving area, to thereby produce a motion vector field;

(c) segmenting the motion vector field into translational patches, each of the translational patches having a two dimensional translation motion defined, using a set of two motion parameters, as;

space="preserve" listing-type="equation">u(x,y)=a.sub.1

space="preserve" listing-type="equation">v(x,y)=a.sub.2

and

(d) merging the translational patches into planar patches, each of the planar patches being represented with a planar patch model described with a set of six motion parameters as;

ti u(x,y)=a₁ +a₂ x+a₃ y

space="preserve" listing-type="equation">v(x,y)=b.sub.1 +b.sub.2 x+b.sub.3 y

and

(e) grouping the planar patches into parabolic patches, each of the parabolic patches being represented with a parabolic patch model described with a set of twelve motion parameters as;

space="preserve" listing-type="equation">u(x,y)=a.sub.1 +a.sub.2 x+a.sub.3 y+a.sub.4 x.sup.2 +a.sub.5 y.sup.2 +a.sub.6 xy

space="preserve" listing-type="equation">v(x,y)=b.sub.1 +b.sub.2 x+b.sub.3 y+b.sub.4 x.sup.2 +b.sub.5 y.sup.2 +b.sub.6 xy

wherein u(x,y) and v(x,y) are an x-directional and a y-directional components of a motion vector located at a pixel position (x,y) in the motion vector field, respectively, and a₁ to a₆ and b₁ to b₆ are real numbers.

2. The method in accordance with claim 1, which further comprises the steps of:

(f) motion compensating the previous frame with each set of the twelve motion parameters to provide a motion-compensated frame;

(g) subtracting the motion compensated frame from the current frame to produce an error signal for each pixel; and

(h) comparing the error signal with a predetermined value to extract a motion compensation fail region.