Method for classifying an object in a moving picture

• US 6,606,412 B1
• Filed: 08/27/1999
• Issued: 08/12/2003
• Est. Priority Date: 08/31/1998
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Maeda, Junji, Echigo, Tomio, Nakano, Hiroki
• Primary Examiner(s): Dastouri, Mehrdad
• Application Number:
  US09/384,743
• Time in Patent Office: 1,446 Days
• Field of Search: 382/190, 382/191, 382/209, 382/217, 382/218, 382/224, 382/103, 382/277, 382/278, 382/279, 382/280, 382/118, 382/132, 382/232, 382/240, 382/248, 382/249, 348/155, 348/700, 375/240.12, 375/240.18, 375/240.19, 700/17
• US Class Current: 382/224
• CPC Class Codes: G06T 7/20 Analysis of motion motion estimation for coding, decoding, compressing or decompressing digital video signals H04N19/43, H04N19/51
• CPC Class Codes: G06V 10/449 Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
• CPC Class Codes: G06V 10/507 Summing image-intensity values; Histogram projection analysis
• CPC Class Codes: G06V 40/23 Recognition of whole body movements, e.g. for sport training

Abstract

The invention provides a method for classifying the motion of an object such as a human being in a moving picture. A template is prepared in advance, which includes the Gabor wavelet expansion coefficients of an object image in a plurality of frames of a video image sequence representing each of a plurality of different reference motions of an object in a moving picture. Then, processing is performed to obtain the Gabor wavelet expansion coefficients of an object image in a plurality of frames of a video image sequence representing an unknown motion of the object. Matching factors are calculated based on the expansion coefficients for the unknown motion and the expansion coefficients for the reference motions in the template, and finally the unknown motion is classified based on the matching factors.

Claims

1. A method for classifying an object in a moving picture, comprising:

preparing a template that includes Gabor wavelet expansion coefficients of an image of said object in a plurality of frames of a video image sequence representing each of a plurality of different reference motions of said object;

obtaining Gabor wavelet expansion coefficients of an image of said object in a plurality of frames of a video image sequence representing an unknown motion of said object;

calculating matching factors between said unknown motion and said reference motions based on said Gabor wavelet expansion coefficients for said unknown motion and said reference motions in said template; and

classifying said unknown motion based on said matching factors,

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of selected sampling points in said object image, and

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of scale transformation levels, and a number of said sampling points is set to a different number for each of said levels.

2. The method as set forth in claim 1, wherein said Gabor wavelet expansion coefficients are obtained based on a Gabor wavelet function.

3. The method as set forth in claim 1, wherein said Gabor wavelet expansion coefficients are obtained with a coordinate origin being set to approximately a center of said object.

4. The method as set forth in claim 1, wherein said calculating said matching factors is performed by assigning a predetermined weight to said Gabor expansion coefficients according to said scale transformation levels.

5. The method as set forth in claim 1, wherein said Gabor wavelet expansion coefficients are obtained at a plurality of rotation positions with each said sampling point being a center of rotation.

6. The method as set forth in claim 5, wherein said Gabor wavelet expansion coefficients are obtained based on Eq. (11), $\begin{array}{c}(T_{m,l}\end{array}$

s)

(x0,y0)=α

m

∑

y=-q/2+1q/2

∑

x=-p/2+1p/2

s

(x,y)

ψ

l

(α

m

x-x0,α

m

y-y0),(11)

where

7. A method for classifying an object in a moving picture, comprising:

preparing a template that includes Gabor wavelet expansion coefficients of an image of said object in a plurality of frames of a video image sequence representing each of a plurality of different reference motions of said object;

obtaining Gabor wavelet expansion coefficients of an image of said object in a plurality of frames of a video image sequence representing an unknown motion of said object;

calculating matching factors between said unknown motion and said reference motions based on said Gabor wavelet expansion coefficients for said unknown motion and said reference motions in said template; and

classifying said unknown motion based on said matching factors,

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of selected sampling points in said object image,

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of rotation positions with each said sampling point being a center of rotation,

wherein said object is a human body, and a central wavelength 2π

/u₀ of said Gabor wavelet is about two times a width of said human body, and

wherein said Gabor wavelet expansion coefficients are obtained based on Eq. (19), $\begin{array}{cc}\left(T_{m,l}<br><br>s\right)<br><br>\left(x_0,y_0\right)={\mathrm{\alpha <br><br>}}^m<br><br>\underset{y=-q/2+1}{\overset{q/2}{\sum <br><br>}}<br><br>\underset{x=-p/2+1}{\overset{p/2}{\sum <br><br>}}<br><br>s<br><br>\left(x,y\right)<br><br>{\mathrm{\psi <br><br>}}_l<br><br>\left({\mathrm{\alpha <br><br>}}^m<br><br>x-x_0,{\mathrm{\alpha <br><br>}}^m<br><br>y-y_0\right),& \left(19\right)\end{array}$

where

8. A method for classifying an object in a moving picture, comprising:

preparing a template that includes Gabor wavelet expansion coefficients of an image of said object in a plurality of frames of a video image sequence representing each of a plurality of different reference motions of said object;

extracting an image of said object in a plurality of frames of a video image sequence representing an unknown motion of said object;

obtaining Gabor wavelet expansion coefficients of said extracted object image;

calculating matching factors between said unknown motion and said reference motions based on said Gabor wavelet expansion coefficients for said unknown motion and said Gabor wavelet expansion coefficients for said reference motions in said template; and

classifying said unknown motion based on said matching factors,

wherein said Gabor wavelet expansion coefficients for said reference motions and unknown motion are obtained at a plurality of selected sampling points in said object image with a coordinate origin being set to approximately a center of said object, and obtained at a plurality of predetermined rotation positions with each said sampling point being a center of rotation, and

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of scale transformation levels, and a number of said sampling points is set to a different number for each of said levels.

9. The method as set forth in claim 8, wherein said Gabor wavelet expansion coefficients are obtained at a plurality of scale transformation levels, and said calculating said matching factors is performed by assigning a predetermined weight to said Gabor wavelet expansion coefficients according to said levels.

10. A method for classifying a pattern of an object, comprising:

preparing a template including Gabor wavelet expansion coefficients of a reference pattern image representing a plurality of different reference patterns of said object;

obtaining Gabor wavelet expansion coefficients of an unknown pattern image representing an unknown pattern of said object;

calculating matching factors between said unknown pattern and said reference patterns based on said Gabor wavelet expansion coefficients for said unknown pattern and said Gabor wavelet expansion coefficients for said reference patterns in said template; and

classifying said unknown pattern based on said matching factors,

wherein said Gabor wavelet expansion coefficients for said reference patterns and unknown pattern are obtained based on Eq. (15) at a plurality of selected sampling points of said pattern image with a coordinate origin being set to approximately a center of said pattern image, and

wherein said Gabor wavelet expansion coefficients are obtained at a plurality of scale transformation levels, and a number of said sampling points is set to a different number for each of said levels, $\begin{array}{cc}\left(T_{m,l}<br><br>s\right)<br><br>\left(x_0,y_0\right)={\mathrm{\alpha <br><br>}}^m<br><br>\underset{y=-q/2+1}{\overset{q/2}{\sum <br><br>}}<br><br>\underset{x=-p/2+1}{\overset{p/2}{\sum <br><br>}}<br><br>s<br><br>\left(x,y\right)<br><br>{\mathrm{\psi <br><br>}}_l<br><br>\left({\mathrm{\alpha <br><br>}}^m<br><br>x-x_0,{\mathrm{\alpha <br><br>}}^m<br><br>y-y_0\right),& \left(15\right)\end{array}$

where