KR101318323B1 - Method and device for detecting object - Google Patents

Abstract

PURPOSE: An object detecting method and device can evaluate similarity of extracted arcs and detect an object similar to a model sketch in a test image. CONSTITUTION: An arc extracting unit (100) extracts a test arc and a model arc containing a feature vector from a test image and a model sketch. A matching unit (200) matches similarity of the test arc and the model arc extracted from the arc extracting unit. The matching unit adjusts an endpoint of the test arc to maximize the similarity of the test arc for the model arc. An object detection unit (300) evaluates a similarity value of the test arc adjusted in the matching unit. The object detection unit detects an object similar to the model sketch in the test image. [Reference numerals] (100) Arc extracting unit; (110) Arc generating unit; (120) Arc extending unit; (130) Specific vector extracting unit; (200) Matching unit; (300) Object detection unit

Description

Object detection method and apparatus {METHOD AND DEVICE FOR DETECTING OBJECT}

The present invention relates to a method and an apparatus for detecting an object capable of receiving a test image and a model sketch and detecting an object similar to the model sketch in the test image.

Background Art Conventionally, various image matching apparatuses are known as apparatuses for performing matching on the basis of image information. For example, a correlation value is calculated by comparing a registered image with a comparison image to be compared in a predetermined positional relationship, and an image matching apparatus or a correlation value for collating the registration image and the matching image based on the correlation value is generated. At this time, an image matching apparatus is known which generates a correlation value by a pixel-by-pixel operation.

Among the object detection methods through image matching, Korean Patent Registration No. 10-1122944 discloses an image matching method and an image matching device and a program.

However, when detecting a circular object using the image matching method, the circular arc extraction is not accurate, and there is a problem that an incorrect circular object is detected because there is no separate correction process.

In addition, there is a problem that it is difficult to extract the feature vector based on the arc for image contrast.

Accordingly, an object of the present invention is to take an image (test image and model sketch), extract an arc containing a feature vector through decomposition and merging, and evaluate the similarity of the extracted arcs to find an object similar to the model sketch in the test image. The present invention provides a method and apparatus for detecting an object.

According to an aspect of the present invention, an object detecting method includes extracting a model arc and a test arc including a feature vector from a model sketch and a test image inputted by an arc extracting unit, and a model arc and a test unit extracted from the arc extracting unit by a matching unit. Matching the similarity of the arcs, adjusting the end points of the test arcs to maximize the similarity of the test arcs to the model arcs, and the object detector evaluating the similarity values of the test arcs adjusted by the matching part to obtain the similarity to the model sketch. Detecting the object in the test image.

The method may further include generating an arc by dividing and merging the pixel chain, and generating an arc by dividing and merging the pixel chain, wherein the arc extension is a pixel adjacent to the arc generated by the arc generator. And recovering the damaged arc by extending the circular arc, and including the feature vector extracting unit extracting the feature vector of the extended arc.

According to another aspect of the present invention, an object detecting apparatus includes: an arc extracting unit configured to extract a model arc and a test arc each including a feature vector from an input model sketch and a test image; A matching unit for matching the similarity between the model arc extracted from the circular arc extracting unit and the test arc, and adjusting the end point of the test arc to maximize the similarity of the test arc with respect to the model arc; And an object detector for evaluating a similarity value of the test arc adjusted by the matching unit to detect an object similar to the model sketch in a test image.

As a result, the object detecting method and apparatus according to the present invention receives an image (test image and model sketch), extracts an arc including a feature vector through decomposition and merging, and evaluates the similarity of the extracted arcs to sketch the model. Similar objects can be detected quickly and easily in the test image.

1 is a block diagram of an object detecting apparatus according to an embodiment of the present invention
2 is a flowchart of an object detection method according to an embodiment of the present invention.
3 is a flowchart of an arc extraction unit in an object detection method according to an embodiment of the present invention.
4 and 5 are exemplary views for explaining an object detection method according to an embodiment of the present invention.
6 and 7 are exemplary diagrams for explaining arc extension and feature vector extraction in the object detection method according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings.

1 and 2, a detection apparatus including an object detecting method according to an exemplary embodiment of the present invention includes an arc extracting unit 100, a matching unit 200, and an object detecting unit 300.

An object detecting apparatus according to an embodiment of the present invention is an arc extraction unit 100 for extracting a model arc and a test arc containing a feature vector from the input model sketch and the test image, respectively, and extracted from the arc extraction unit 100 The similarity value of the matching unit 200 and the matching test arc adjusted by the matching unit 200 are adjusted to match the similarity between the model arc and the test arc, so that the similarity of the test arc with respect to the model arc is maximized. And an object detector 300 to detect an object similar to the model sketch in the test image.

In the object detection method according to the embodiment of the present invention, the arc extraction unit 100 extracts a model arc and a test arc including a feature vector from the input model sketch and the test image, respectively (S100).

The matching unit 200 adjusts an end point of the test arc so that the similarity of the test arc with respect to the model arc is the maximum by matching the similarity between the model arc and the test arc extracted by the arc extracting unit 100 (S110).

The object detector 300 detects an object similar to the model sketch in the test image by evaluating the similarity value of the test arc adjusted by the matching unit 200 (S120).

The circular arc includes a straight line or a portion thereof, the radius of which is infinity or the arc of curvature '0'.

Referring to FIG. 4, by this, the object detecting apparatus according to the present invention is a model sketch (Applogo, Bottle, Similar objects (giraffe, cup, duck).

The model sketch and the test image are input to the arc extraction unit 100 through separate input means (camera, scanner, etc.).

1 and 3, the arc extracting unit 100 includes an arc generating unit 110, an arc extending unit 120, and a feature vector extracting unit 130.

The arc generator 110 detects an edge from an image (model sketch, test image), extracts an edge pixel chain, and divides and merges the pixel chain to generate an arc (S101).

The arc extension unit 120 restores the damaged arc by extending the arc to neighboring pixels around the arc generated by the arc generating unit 110 (S102).

The feature vector extractor 130 extracts a feature vector of an arc extending from the arc extension 120 (S103).

The arc generator 110 detects an edge in an image through an edge extraction method disclosed in M Maire, P. Arbelaex, C. Fowlkes, and J. Malik, "Using contours to detect and localize junctions in natural images". Obtain a chain.

In step S101,

The arc generating unit 110 divides the pixel chain in units of pixels, calculates an error ratio (E (X; λ): error ratio) for the pixel chain through Equation 1, and uses a merging rule using the error rate. To merge the pixels to create an arc segment.

X: pixel chain

λ: parameter

N: number of divided pixels

: 픽셀 (x_i,y_i) 와 원 f(x,y;λ)=0 사이의 유클리드(Euclidean) 거리

Euclidean distance between pixel (x _i , y _i ) and circle f (x, y; λ) = 0

Lx: the maximum distance between two pixels in a pixel chain (X)

The N _min and E _max are defined by the user as the minimum value of the divided pixel and the maximum value of the error rate.

로 정의하고, 상기 픽셀 체인을 원형 방정식

로 표현한다. In order to calculate the error rate,

And define the pixel chain as a circular equation

.

In this case, λ = (a, b, c, d) is a parameter, and the parameter is defined by Equation 2 below.

)(only,

)

If the number N of divided pixels does not exceed N _min defined by the user, the merge error is prevented by setting the maximum error rate E _max defined by the user.

),The merging rule assumes that the pixel chain is divided into M pieces (

),

1. If a pair of adjacent arc segments (X _j , X _{j + 1} ) satisfy N _j + N _{j +1} <N _min , merge the pair of arc segments.

2. If a pair of adjacent arc segments (X _j , X _{j + 1} ) satisfy the minimum min (N _{j ,} N _{j +1} ) <N _min , merge the pair of arc segments.

3. If the pair of arc segments satisfy Equation 3, merge the pair of arc segments.

이다. X _{j , j + 1} = X _{j J} X _{j +1 The} merged arc segments Xj and Xj + 1 and the weighted average error (E _s (X _i , X _j )) is

to be.

Referring to FIG. 5, when extracting an arc of an image through the arc generator 110, an accurate arc may not be extracted due to inaccuracy of edge detection and interference of other lines.

Accordingly, the arc extension 120 extends the arc segment generated by the arc generator 110 and the sub arc segment neighboring to generate an accurate arc.

For example, the arc segments positioned below in FIG. 5A are arcs extracted through the arc generator 110, and the arc arcs 120 may be used to generate an accurate arc as described above.

The arc extension 120 is divided into an arc segment to be extended and a sub arc segment adjacent to the arc segment, and the arc segment is a sub arc segment by using a gap ratio, a cost function, and an extension rule. Selectively combine to create an extended arc.

The gap ratio is expressed by Equation 4.

Y: arc adjacent to arc X

: 원호 X와 Y 사이의 최소거리

: Minimum distance between arc X and Y

: 원호 X와 Y 의 곡선의 거리(둘레)

Is the distance (circumference) of the curve between arcs X and Y

는 원호 X와 원호 Y 사이의 최소갭이고(도6 c 참조), 상기

는 원호 X와 원호 Y 사이의 최대갭이며(도6 d 참조), 상기 갭비율은 수학식 4와 같이 나타낼 수 있다. remind

Is the minimum gap between arc X and arc Y (see Fig. 6c),

Is the maximum gap between the arc X and the arc Y (see Fig. 6d), and the gap ratio can be expressed as Equation 4.

The cost function is (5).

: weighting factor

weighting factor

는 갭비율의 가중요소(weighting factor)로서 사용자에 의해 임의적으로 설정된다. remind

Is arbitrarily set by the user as the weighting factor of the gap ratio.

The cost is a weighted sum of the error ratio and the gap ratio.

The extension rule is as follows. If the following extension rule is satisfied, arc X is extended.

는 문턱값(thresholds)으로써 사용자에 의해 정의된다( 단,

). remind

Is defined by the user as thresholds (

).

, 원호 'X'의 둘레 길이를

로 나타내고(도6 a 참조), 원호 X의 진행방향의 끝점에서 서브 아크 세그먼트를 찾을 수 있도록 Lx/2 반경을 설정한다(도6 b 참조). 각각의 서브 아크 세그먼트 또는 서브 아크 세그먼트에 연결된 그룹은 원호 X에 연장된 영역에 존재한다.Referring again to FIG. 6, the length having the maximum value between two points (pixels) in arc X

, The circumference of the arc 'X'

(See Fig. 6a), the Lx / 2 radius is set so that the sub-arc segment can be found at the end point of the traveling direction of arc X (see Fig. 6b). Each sub-arc segment or group connected to the sub-arc segment is in the region extending to arc X.

Uncertain arcs in the image are reproduced into the correct arc through the arc extension of the arc extension 120 (see FIGS. 5B, C).

Referring to FIG. 7, an arc may be represented by two end points and directions. If the end point is represented by "e" with a vector [x, y, θ], as shown in Fig. 7 (a), the arc has a shape of " e1, e2 ".

The feature vector extractor 130 defines an end point vector for a relative position and a direction between two end points by Equation 6.

d: Euclidean distance of two endpoints

L: Size of the arc contained at the end point e1 (see Fig. 5b)

In addition, the feature vector extractor 130 defines a primary feature vector (or descriptor u _i ) using one arc _ai and both endpoints by Equation 7 (see FIG. 7C).

In addition, the feature vector extractor 130 defines one set of circular arcs (a _i ) and a set of neighboring circular arcs (N _i ) by Equation (8).

D _ij : distance between a _i and a _j

In addition, the feature vector extractor 130 defines a secondary feature vector b _ij , which is one neighboring arc viewed by one arc, by Equation 9 (see FIG. 7D).

)와, 테스트 원호(

)를 전달받으며, 두 원호를 매칭하여 유사도를 측정한다. The matching unit 200 is a model circular arc (from the arc extraction unit 100)

) And the test arc (

) And measure the similarity by matching the two arcs.

)과, 테스트 모델의 끝점(

)의 유사도를 수학식 10으로 정의한다. The matching unit 200 is the end point of the model arc (

) And the endpoint of the test model (

The similarity of) is defined by Equation 10.

σ _{uv ,} σ _Ф : parameters

, 이차 특징벡터 (b, b')의 유사도

, 그리고 두 원호 a_i, a'_i 에 대해 각각의 일차 특징벡터 및 이웃 원호간의 이차 특징벡터들을 모두 고려한 전체 유사도 S(a_i, a'_i)를 수학식 11로 정의한다. The matching unit 200 has a similarity degree between the primary feature vectors (u, u ') of two arcs.

, Similarity of quadratic feature vectors (b, b ')

For the two arcs a _i and a ' _i , the total similarity S (a _i , a' _i ) considering both the primary feature vectors and the secondary feature vectors between neighboring arcs is defined by Equation 11.

: 두 원호(

,

)의 일차 특징벡터간의 유사도.

: Two arcs (

,

Similarity between the primary feature vectors of

: 두 원호(

,

)의 이차 특징벡터간의 유사도.

: Two arcs (

,

Similarity between the quadratic feature vectors of

By measuring similarity using the secondary feature vector and the primary feature vector, the similarity between the arcs in the test image may be represented as a score.

Using the above equation, the matching unit 200 adjusts the end point of the test image arc to maximize the similarity between the model arc and the test arc.

The arc detector 300 extracts object candidates included in a test image by identifying similarities between the arcs measured by the matcher 200.

In this case, the object candidates use a Hugh transform algorithm.

In addition, the arc detection unit 300 is A.J. Model among object candidates using graph matching algorithm disclosed in Stoddart, M. Petrou, and J. Kittler, "Onthe foundations of probabilistic relaxation with product support," J. Math. Imaging Vis., (1998). An object similar to the sketch is found, and the similarity of the object is detected to detect the object.

The arc detector 300 may obtain an approximate position and size of an object to be detected by extracting object candidates, and may detect an accurate object among the object candidates through the graph matching algorithm.

The present invention is not necessarily limited to these embodiments, as all the constituent elements constituting the embodiment of the present invention are described as being combined or operated in one operation. In other words, within the scope of the present invention, all of the components may be selectively operated in combination with one or more. In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. In addition, such a computer program may be stored in a computer-readable medium such as a USB memory, a CD disk, a flash memory, etc., and read and executed by a computer, thereby implementing embodiments of the present invention. As the storage medium of the computer program, a magnetic recording medium, an optical recording medium, a carrier wave medium, or the like may be included.

Furthermore, all terms including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined in the Detailed Description. Terms used generally, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and variations without departing from the essential characteristics of the present invention. In addition, the embodiments disclosed in the present invention are not intended to limit the scope of the present invention but to limit the scope of the technical idea of the present invention. Accordingly, the scope of protection of the present invention should be construed according to the claims, and all technical ideas within the scope of equivalents should be interpreted as being included in the scope of the present invention.

100: arc extraction unit
200: matching unit
300: arc detection unit

Claims (15)

Extracting a model arc and a test arc including a feature vector from the model sketch and the test image inputted by the arc extracting unit (S100);
Matching unit matching the similarity between the model arc and the test arc extracted from the arc extraction unit, adjusting the end point of the test arc to the maximum similarity of the test arc to the model arc (S110) and
And detecting (S120) an object similar to the model sketch in a test image by evaluating a similarity value of the test arc adjusted by the matching unit.
The method of claim 1,
In step S100, an arc generating unit detects an edge from an image (model sketch, test image) to extract an edge pixel chain, and divides and merges the pixel chain to generate an arc (S101). );
Restoring the damaged circular arc by extending the circular arc to neighboring pixels centering on the circular arc generated by the circular arc generating unit (S102);
And a feature vector extracting unit for extracting a feature vector of an arc extended by the feature vector extracting unit (S103).
The method of claim 2, wherein in step S101,
The arc generating unit divides the pixel chain in units of pixels, calculates an error ratio (E (X; λ): error ratio) for the pixel chain through the following equation, and merges pixels using a merge rule using the error rate. (merge) to generate an arc segment.

X: pixel chain
λ: parameter
N: number of divided pixels

Euclidean distance between pixel (x _i , y _i ) and circle f (x, y; λ) = 0
Lx: the maximum distance between two pixels in a pixel chain (X)
Claim 4 has been abandoned due to the setting registration fee. The method of claim 3, wherein
The merging rule is that the pixel chain is divided into M pieces (

),
1. If a pair of adjacent arc segments (X _j , X _{j + 1} ) satisfy N _j + N _{j +1} <N _min , merge the pair of arc segments, or,
2. If a pair of adjacent arc segments (X _j , X _{j + 1} ) satisfy a minimum of min (N _{j ,} + N _{j +1} ) <N _min , merge the pair of arc segments, or;
3. The method of detecting an object, wherein the pair of arc segments are merged when the pair of arc segments satisfy the following equation.

The method of claim 2, wherein in step S102,
The arc extension is divided into an arc segment to be extended and a sub arc segment adjacent to the arc segment, and the arc segment selectively combines the sub arc segment using a gap ratio, a cost function, and an extension rule. And extending the arc to generate the object.
Claim 6 has been abandoned due to the setting registration fee. The method of claim 5, wherein
The gap ratio is an object detection method characterized in that the following equation.

Y: arc adjacent to arc X

: Minimum distance between arc X and Y

: Length of curve of arc X and Y
Claim 7 has been abandoned due to the setting registration fee. The method of claim 5, wherein
And the cost function is the following equation.

weighting factor
Claim 8 was abandoned when the registration fee was paid. The method of claim 5, wherein
The extension rule is as follows, and the object detection method characterized in that extending the arc X if the following extension rule is satisfied.

: Thresholds,

Claim 9 has been abandoned due to the setting registration fee. The method of claim 2, wherein in step S103,
And the feature vector extracting unit defines an end point vector for a relative position and direction between two end points by the following equation.

d: Euclidean distance of two endpoints
L: Size of the arc contained at the end point e1 (see Fig. 5b)
Claim 10 has been abandoned due to the setting registration fee. The method of claim 2, wherein in step S103,
An object detecting method, characterized in that for defining the equation for the primary feature vector (u _i) Using the positive end the feature vector extracting unit a circular arc (a _i).

Claim 11 was abandoned when the registration fee was paid. The method of claim 2, wherein in step S103,
And the feature vector extracting unit defines one circular arc (a _i ) and a set of neighboring arcs (N _i ) by the following equation.

D _ij : distance between a _i and a _j
Claim 12 is abandoned in setting registration fee. The method of claim 2, wherein in step S103,
And a second feature vector (b _ij ), which is one neighboring arc viewed by one arc, is defined by the following equation.

The method of claim 1, wherein in step S110,
The matching part end point of the model arc (

) And the endpoint of the test model (

) Similarity is defined by the following equation.

σ _{uv ,} σ _Ф : parameters
The method of claim 1, wherein in step S110,
The matching part has a similarity degree between the primary feature vectors (u, u ') of two arcs.

, Similarity of quadratic feature vectors (b, b ')

, And to the two circular arc (a _i, a _'i) the total degree of similarity S (a _i, a consideration of all of the second feature vector between each of the primary characteristic vector and the neighboring arc for' _i) is characterized in that defined in formula Object detection method to perform.

: Two arcs (

,

Similarity between the primary feature vectors of

: Two arcs (

,

Similarity between quadratic feature vectors of
An arc extraction unit for extracting a model arc and a test arc each including a feature vector from the input model sketch and the test image;
A matching unit for matching the similarity between the model arc extracted from the circular arc extracting unit and the test arc, and adjusting the end point of the test arc to maximize the similarity of the test arc with respect to the model arc; And
And an object detector for evaluating a similarity value of the test arc adjusted by the matching unit to detect an object similar to the model sketch in a test image.

Images