KR101150871B1 - Method and apparatus for calculating fatigue of stereoscopic image based on horizontal disparity and vertical disparity - Google Patents

Abstract

Disclosed are a method and apparatus for estimating a horizontal shift characteristic and a vertical shift characteristic of a stereo image and calculating a fatigue degree based on the horizontal shift and vertical shift of feature points of the stereo image.

Description

Method and apparatus for calculating fatigue image of stereo image based on horizontal and horizontal variation {method and apparatus for calculating fatigue of stereoscopic image based on horizontal disparity and vertical disparity}

The present invention relates to a method and apparatus for estimating fatigue of a stereo image, and more particularly, to a method and apparatus for estimating fatigue of a stereo image based on a horizontal shift and a horizontal shift of a stereo image.

With the development of image technology, research on 3DTV (3D TV) is being done a lot after high definition HDTV. In order to improve the quality of 3DTV, research on high-quality contents that can give a visually comfortable feeling considering human visual characteristics is necessary.

Two major problems caused by the undesirable effects of 3DTV are depth distortion and fatigue. Depth distortion is a distortion that occurs when a user experiences a predetermined scene in a stereo image and the depth of the actual scene is different. Therefore, methods for reducing distortion through fine adjustment of camera parameters have been studied.

Fatigue occurs when a user watching a stereo image is constantly trying to resolve the visual contradictions of the stereo image. Fatigue can be caused by a number of factors, such as camera configuration, viewing environment and image characteristics.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide a method and apparatus for more accurately calculating fatigue of a stereo image, and to provide a computer-readable recording medium having recorded thereon a program for executing the method.

According to an aspect of the present invention, there is provided a method of estimating fatigue of a stereo image, including extracting feature points of the stereo image; Estimating horizontal and vertical shifts of the feature points of the feature points; Generating at least one value representing a horizontal shift characteristic of the stereo image and at least one value representing a vertical shift characteristic based on the horizontal shift and the vertical shift of the feature points; And estimating the fatigue degree of the stereo image based on at least one value representing the generated horizontal shift characteristic and at least one value representing the vertical shift characteristic.

According to another embodiment of the present invention, at least one value representing the horizontal shift characteristic is the horizontal shift in a sum of values obtained by multiplying a horizontal shift of feature points having a relatively large horizontal shift among the feature points and a weight based on a spatial frequency. Is a first value obtained by subtracting a sum of multiplying horizontal shifts of relatively small feature points by a weight based on spatial frequency, and values multiplied by horizontal shifts of feature points having a relatively large horizontal shift among the feature points and weights based on spatial frequency. The sum of the second values.

According to another embodiment of the present invention, at least one value representing the vertical shift characteristic sets the vertical coordinate of the feature points to x and sets the normalized vertical shift generated by normalizing the vertical shift of the feature points to y. Then, a and b of y = ax + b estimated by the least squares method.

An apparatus for estimating the fatigue level of a stereo image according to an embodiment of the present invention for solving the technical problem, extracts feature points of the stereo image, estimates the horizontal shift and vertical shift of the feature points of the feature points, and then A variation estimating unit for generating at least one value representing a horizontal variation characteristic of the stereo image and at least one value representing a vertical variation characteristic based on the horizontal variation and the vertical variation of the stereo image; And a fatigue estimator for estimating the fatigue degree of the stereo image based on at least one value indicating the generated horizontal shift characteristic and at least one value indicating the vertical shift characteristic.

In order to solve the above technical problem, the present invention provides a computer-readable recording medium having recorded thereon a program for executing the fatigue estimation method of the stereo image.

According to the present invention, it is possible to estimate the fatigue degree of the stereo image more accurately, and to encode the stereo image of the low fatigue degree in consideration of the fatigue degree when encoding the stereo image. In addition, when decoding the stereo image, the stereo image may be reproduced by a reproduction method that reduces fatigue, or a warning related to the fatigue may be displayed to the user.

1 illustrates an apparatus for estimating fatigue degree of a stereo image according to an exemplary embodiment of the present invention.
2A and 2B show display coordinates of a stereo image.
3 illustrates keystone distortion modeling according to an embodiment of the present invention.
4 is a flowchart illustrating a fatigue estimation method of a stereo image according to an embodiment of the present invention.

Hereinafter, with reference to the drawings will be described embodiments of the present invention;

1 illustrates an apparatus for estimating fatigue degree of a stereo image according to an exemplary embodiment of the present invention.

Referring to FIG. 1, an apparatus 100 for estimating fatigue of a stereo image according to an exemplary embodiment of the present invention includes a variation calculator 120 and a fatigue calculator 120.

The variation calculator 110 extracts feature points to be used for fatigue estimation in the stereo image. Dense matching methods have already been proposed for calculating the variation in left and right eye images for all pixels of a stereo image. However, the dense matching method is not suitable for a method for calculating the variation in a specific image, such as an image having a small texture area. In addition, since the dense matching method assumes that only horizontal shift of a stereo image exists, it is not suitable for measuring vertical shift in consideration of keystone distortion caused by a stereo camera.

Therefore, a sparse disparity estimation method is used to calculate a fatigue degree of a stereo image. To this end, the variation calculator 110 selects feature points to be used for estimating fatigue in a stereo image (SIFT). Extract with. SIFT is an algorithm for extracting feature points that are invariant in size and rotation, and can provide robust matching at affine distortion or three-dimensional viewpoint.

When the feature points are extracted, the variation calculator 120 calculates the variation of the feature points extracted from the stereo image. The variation of the feature points in the stereo image may be defined as in Equation 1 below.

Disp _ih means horizontal shift in image coordinates, and Dis _iv means vertical shift in image coordinates. X _ir , X _il , Y _ir, and Y _il mean x and y coordinates of the same feature point in the left eye image and the right eye image.

However, since the variations according to Equation 1 are variations that do not consider the depth value changing according to the viewing environment, each variation may be used to measure the variation of the feature points. Angular variation refers to the difference between the angle of both eyes relative to the three-dimensional position of the feature point and the angle of both eyes relative to the position where the feature point is projected onto the display plane. This will be described in detail with reference to FIGS. 2A and 2B.

2A and 2B show display coordinates of a stereo image.

Referring to FIG. 2A, when a user views a stereo image from the front, a feature point exists at a position 230 on a Z axis vertically transverse to the center of both eyes 210 and 211 and the center of the display plane 220. , The angular variation of the feature point is between the angle β of both eyes 210 and 211 relative to the three-dimensional position 230 of the feature point and the angle α of both eyes 210 and 211 relative to the position 231 where the feature point is projected onto the display plane. Can be defined as a difference. This may be expressed as Equation 2 below.

Disp _a means the angular variation of the feature point, I means the distance between both eyes 210 and 211, and V means the distance between both eyes 210 and 211 and the display plane 220. Z _d means the distance between the three-dimensional position 230 of FIG. 2A and the position 231 at which the feature point is projected onto the display plane.

If the location of the feature point is more generalized in three dimensions, it may be as shown in FIG. 2B. Referring to FIG. 2B, when the user views a stereo image from the front, the angle β of the binoculars 210 and 211 with respect to the three-dimensional position 240 of the feature point and the position 241 at which the feature point is projected on the display plane 220 are shown. The difference between the angles α of both eyes 210 and 211 relative to If this is expressed using the inner product of the vector, Equation 3 below.

이고, 각도 α는

이며, 각변이 Disp_a는 각도 β와 각도 α의 차이를 의미한다. (논문 및 직무 발명서에는 각도 β가 벡터 c와 벡터 d의 내적에서 계산되는 것으로, 각도 α가 벡터 a 및 벡터 b의 내적에서 계산되는 것으로 기재되어 있습니다. 논문의 기재가 옳은지 상기 기재가 옳은지 확인바랍니다.)The vector a is a vector from the left eye position 210 to the left eye projection position 250 of the feature point, that is, P _dl = (X _dl , Y _dl , Z _dl ), and the vector a is the right eye of the feature point from the right eye position 211. It is a vector to the projection position 251, ie, P _dr = (X _dr , Y _dr , Z _dr ). When the viewer's position, i.e., the center of both eyes 210 and 211, is P _v = (X _v , Y _v , Z _v ), the position 210 of the left eye is P _vl = (X _vl , Y _vl , Z _vl ) = (X _v -I / 2, Y _v , Z _v ), and the position of the right eye 211 is P _vr = (X _vr , Y _vr , Z _vr ) = (X _v + I / 2, Y _v , Z _v ). Angle β is

And the angle α is

Each variation Disp _a means a difference between an angle β and an angle α. (Thesis and job invention documents that the angle β is calculated from the dot product of the vectors c and vector d, and the angle α is calculated from the dot products of the vectors a and vector b. Confirm whether the article is correct or not. wish.)

When the disparity calculated by Equation 3 is Disp _a , the disparity calculator 110 calculates the horizontal and vertical shifts of the feature points from the disparity. Since angular variation Disp _a includes horizontal angular variation and vertical angular variation, it is possible to calculate horizontal angular variation and vertical angular variation in angular variation Disp _a , respectively. Assume that the viewer's position is P _v = (0, 0, -Z _v ) and Z _v = Z _vl = Z _vr , average the y-axis coordinates to find the horizontal angular shift, and find the vertical angular shift. In order to average the x-axis coordinates, the horizontal angular shift and the vertical angular shift can be estimated as in Equation 4 below.

Disp _ah means the horizontal angular shift of the feature point, and Disp _av means the vertical angular shift of the feature point. (Please make sure that the vectors c and d are the same as the vectors c and d in Equation 3.)

When the angular variation is calculated for the feature points, the variation calculator 110 may reduce the matching error by filtering the feature points extracted by the SIFT according to a predetermined algorithm. For example, the feature points extracted by the SIFT may be filtered according to a random SAmple consensus (RANSAC) algorithm to remove feature points in which a matching error occurs.

(1) Estimation of horizontal shift characteristics

When the horizontal shift is calculated for each of the feature points as shown in Equation 4, the horizontal shift characteristic of the stereo image may be estimated based on the horizontal shift of the feature points.

(i) the extent of horizontal variation

In order to simplify and estimate the horizontal shift characteristic of the stereo image, the maximum horizontal shift and the minimum horizontal shift among the horizontal shifts of the feature points may be used.

Range _h1 is F (l) means feature points where the horizontal shift is relatively large among the feature points, and f (s) means feature points where the horizontal shift is relatively small among the feature points. According to Equation 5, the range _h1 is a value obtained by subtracting the horizontal variation of the feature points having the relatively small horizontal shift from the sum of the horizontal variation of the feature points having the relatively large horizontal shift among the feature points. For example, when all fn feature points are listed in the order of increasing horizontal displacement, f (l) may be feature points whose horizontal shift is included in the top 5% to 15%, and f (s) is the top 85% to It may be feature points included in 95%.

(ii) the extent of the horizontal shift and the maximum horizontal shift, taking into account the position of the feature points;

The maximum horizontal variation as well as the range of the horizontal variation may be used to estimate the horizontal variation characteristic of the stereo image.

Since the relatively large horizontal variation among the horizontal variations greatly affects the fatigue, Max _h2 , which is the sum of the horizontal variations of the feature points f (l) having the large horizontal variation, can also be used to estimate the horizontal variation of the stereo image. . However, Max _h2 only for feature points whose absolute value of the horizontal shift is greater than or equal to a predetermined threshold, that is, th _ah. And Calculate Range _h2 .

(iii) the extent of the horizontal shift and the maximum horizontal shift, taking into account the spatial frequency and the position of the feature points;

Experimental evidence has shown that fatigue increases easily as the spatial frequency (eg, cpd: cycle per degree) of stereo images increases. Therefore, by multiplying the above-described equation (6) based on the spatial frequency, the range of the horizontal shift and the maximum horizontal shift can be calculated.

The spatial frequency may be generated by applying a Sobel operation to the stereo image. As a result of the Sobel calculation, the generated Sobel (f) is a value representing a spatial frequency as values corresponding to each feature point. The weight S (f) is calculated by scaling Sobel (f) according to the maximum size of the pixel value, ie 255, and multiplying by the constant λ. Then, based on the weight S (f), Max _h3 , which is the sum of the horizontal shifts of the feature points having the large horizontal level, and Range _h3 , which is the range of the horizontal shifts, are calculated.

(2) Estimation of vertical shift characteristics

(i) the range of vertical variation

In order to simplify and estimate the vertical shift characteristic of the stereo image, the maximum and minimum values among the horizontal shifts of the feature points may be used.

Range _v1 is The range of vertical variation means f (l) is a feature point where the vertical shift is relatively large among all the feature points, and f (s) refers to feature points where the vertical variation is relatively small among the feature points. According to Equation 8, Range _v1 is a value obtained by subtracting the vertical variation of the feature points having a relatively small vertical shift from the sum of the vertical variation of the feature points having a relatively large vertical shift. For example, when all fn feature points are listed in order of increasing vertical displacement, f (l) may be feature points whose vertical variation is included in the top 5% to 15%, and f (s) is the top 85% to It may be feature points included in 95%.

(ii) the range of vertical shifts and the maximum vertical shifts taking into account the position of the feature points;

The maximum vertical shift as well as the range of vertical shift may be used to estimate the vertical shift characteristic of the stereo image.

및

중 더 큰 값이 더 큰 피로를 야기하므로,

및

를 비교하여, Max_v2를 결정하는 것이 수학식 6과 상이하다.(특징점들 중 수평 변이가 큰 특징점들의 수평 변이의 합, 즉 Max_v2를 계산하는 방법이 수학식 6과 상이한 이유를 아주 상세히 설명해주시기 바랍니다.)In the case of vertical variation, both relatively large and relatively small variations affect fatigue. Therefore, the absolute value of the sum of the vertical shifts of the feature points f (l) having a relatively large vertical shift and the sum of the vertical shifts of the feature points f (s) having a relatively small vertical shift are given. In comparison, a larger absolute value can be set to Max _v2 and used to estimate the vertical shift characteristic of the stereo image. However, Max _v2 only for feature points whose absolute value of the vertical shift is greater than or equal to a predetermined threshold, that is, th _ah. And Range _v2 . However, in the case of vertical variation

And

Of the larger values cause more fatigue,

And

The comparison, it is different from the equation (6) to determine the Max _v2. (Feature points the sum of the horizontal variation, the horizontal variation of the large characteristic point of, i.e., the method of calculating the Max _v2 very detailed description of Equation (6) with different reasons Please.)

(iii) keystone distortion modeling

Most vertical shifts are caused by camera settings. Keystalk distortion occurs due to separation of the camera, convergence distance and focal length. Since the vertical shift caused by the keystone distortion can be scaled by the vertical coordinate of the specific point, normalization can be performed.

The linear equation 320 y = ax + b for a plurality of feature points may be estimated by a least square method. Set X _i (f), the horizontal coordinate of the feature point, to x in the linear equation (320), set Normalized_Disp _iv (f) to y in the linear equation (320), and then for a plurality of feature points. Applying the least squares method, the linear equation 320 for the feature points can be estimated.

3 illustrates keystone distortion modeling according to an embodiment of the present invention.

Referring to FIG. 3, normalized vertical variations 310 to 316 caused by keystone distortion are normalized to normalized vertical variations 330 to 333 on the line according to the linear equation 320 estimated by the least squares method. . At this time, the vertical shifts whose vertical coordinate is less than or equal to the predetermined value th are not normalized in order to minimize an error caused by normalization. The normalized vertical shift is calculated by the following equation (11).

In Equation 10, col denotes a vertical size of an image, and Yi (f) denotes a vertical coordinate of a feature point. By normalizing the vertical shifts according to Equation 10, the linear equation 320 for the feature points is estimated, and the slope a and y-axis intercepts b of the linear equation 320 are used as the vertical shift characteristics.

When the variation estimator 110 generates at least one value representing the horizontal variation of the stereo image and at least one value representing the vertical variation characteristic through the equations 5 to 10 described above, the fatigue estimation unit 120 may determine the stereo image. Fatigue can be estimated by Fatigue = c ₁ D _h + c ₂ D _v + c ₃ or Fatugue = c ₁ D _h1 + c ₂ D _h2 + c ₃ D _v1 + c ₄ D _v2 + c ₃ . D _h is a value indicating a horizontal shift characteristic of a stereo image and D _v is a value indicating a vertical shift characteristic of a stereo image.

For example, D _h may correspond to Range _h1 according to Equation 5. In addition, D _h1 And D _h2 may correspond to Max _h2 and Range _h2 according to Equation 6, or may correspond to Max _h3 and Range _h3 according to Equation 7.

Similarly, D _v may correspond to Range _{v 1} according to Equation 8. Also, D _v1 And D _v2 may correspond to Max _v2 and Range _v2 according to Equation 9, or may correspond to a and b of linear equation 320 obtained through keystone distortion modeling.

c1, c2 and c3 By analyzing a plurality of different stereo images and applying one-dimensional linear regression analysis to the data obtained as a result of analysis, c1, c2 and c3 can be calculated as in the following equation (12).

D _h (p _k ) is a value indicating the horizontal shift characteristic of the k-th stereo image and D _v (p _k ) is a value indicating the horizontal shift characteristic of the k-th stereo image. d _k is the subjective fatigue value of the viewer for the k-th stereo image. n represents the total number of stereo images used in the experiment. The subjective fatigue value of the viewer is estimated through experiments for a plurality of stereo images, and the correlation between the estimated subjective fatigue value and the value indicating the horizontal shift characteristic and the value indicating the horizontal shift characteristic obtained through Equations 6 to 11 Regression analysis can be used to find C matrices.

When D _h and D _v correspond to a plurality of values as calculated in Equations 6, 7, 9 and 10, c1, c2, c3, c4 and c5 may be calculated as follows.

In Equation 12, D _h1 (p _k ) and D _h2 (p _k ) correspond to Max _h2 and Range _h2 of Equation 6 as a plurality of values representing horizontal shift characteristics of the k-th stereo image, or Equation 7 May correspond to Max _h3 and Range _h3 , respectively.

In addition, D _v1 (p _k ) and D _v2 (p _k ) are a plurality of values representing vertical shift characteristics of the k-th stereo image and correspond to Max _v2 and Range _v2 of Equation 9, respectively, or are obtained through keystone distortion modeling. Correspond to a and b of the linear equation 320, respectively.

Fatigue of stereo image Fatigue = c ₁ D _h + c ₂ D _v + c ₃ Or, if estimated by Fatugue = c ₁ D _h1 + c ₂ D _h2 + c ₃ D _v1 + c ₄ D _v2 + c ₃ , modify the stereo image based on this to improve the quality of service (QoS) that the viewer feels. Can be. In the encoding process of the stereo image, the fatigue degree of the stereo image is estimated in advance, and if it is determined that the fatigue degree is large, the depth may be reduced by adjusting the depth to reduce the fatigue degree, or the stereo image may be encoded by adjusting the camera settings and the camera parameters. . In addition, in the decoding process of the stereo image, before the stereo image is displayed, fatigue of the decoded stereo image may be estimated in advance to adjust depth or camera parameters. When displaying a stereo image, a warning message may be displayed to the viewer in advance about fatigue.

4 is a flowchart illustrating a fatigue estimation method of a stereo image according to an embodiment of the present invention.

Referring to FIG. 4, in operation 410, the fatigue estimation apparatus 100 according to an exemplary embodiment extracts feature points of an input stereo image. As described above, the feature point of the stereo image may be extracted using the SIFT. In addition, by filtering the extracted feature points through the RANSAC algorithm, feature points that cause an error in the fatigue estimation may be excluded from the fatigue estimation.

In operation 420, the fatigue estimation apparatus 100 estimates the horizontal variation and the vertical variation of the feature points extracted in operation 410. As described above, the horizontal shift and the vertical shift of the feature points in the image coordinates may be estimated, and the horizontal angular shift and the vertical angular shift may be estimated.

In operation 430, the fatigue estimation apparatus 100 may determine at least one value indicating a horizontal shift characteristic of the input stereo image and at least one value indicating the vertical shift characteristic based on the horizontal shift and vertical shift of the feature points estimated in step 420. Create

The above-described values with respect to Equations 5 to 10, that is, Range _h1 , Max _h2 , Range _h2 , Max _h3 , Range _h3 , Range _v1 , Max _v2 , Range _v2 , a and b may be generated.

In operation 440, the fatigue estimation apparatus 100 estimates the fatigue degree of the stereo image based on at least one value indicating the horizontal shift characteristic and the at least one value indicating the vertical shift characteristic generated in step 430.

Based on at least one value representing the horizontal shift characteristic of the stereo image and at least one value representing the vertical shift characteristic Fatigue = c ₁ D _h + c ₂ D _v + c ₃ Alternatively, the fatigue of the stereo image may be estimated by calculating Fatugue = c ₁ D _h1 + c ₂ D _h2 + c ₃ D _v1 + c ₄ D _v2 + c ₃ .

As described above, although the present invention has been described by way of limited embodiments and drawings, the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible. Accordingly, the spirit of the invention should be understood only by the claims set forth below, and all equivalent or equivalent modifications will fall within the scope of the invention. In addition, the system according to the present invention can be embodied as computer readable codes on a computer readable recording medium.

For example, a fatigue estimation apparatus according to an exemplary embodiment of the present invention may include a bus coupled to respective units of the apparatus as shown in FIG. 1, and at least one processor coupled to the bus. . It may also include a memory coupled to the bus for storing instructions, received messages or generated messages and coupled to at least one processor for performing instructions as described above.

The computer-readable recording medium also includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of the recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device and the like. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Claims (17)

Extracting feature points of a stereo image;
Estimating horizontal and vertical shifts of the feature points of the feature points;
Generating at least one value representing a horizontal shift characteristic of the stereo image and at least one value representing a vertical shift characteristic based on the horizontal shift and the vertical shift of the feature points; And
Estimating the fatigue degree of the stereo image based on at least one value representing the generated horizontal shift characteristic and at least one value representing the vertical shift characteristic,
The at least one value representing the horizontal shift characteristic may be equal to a horizontal shift of feature points having a relatively small horizontal shift from a sum of values obtained by multiplying a horizontal shift of feature points having a relatively large horizontal shift among the feature points and a weight based on a spatial frequency. A first value obtained by subtracting a sum of values multiplied by a weight based on a spatial frequency, and a second value obtained by adding a value obtained by multiplying a horizontal variation of feature points having a relatively large horizontal shift among the feature points and a weight based on a spatial frequency;
The at least one value representing the vertical shift characteristic is set by a vertical coordinate of the feature points to x, a normalized vertical shift generated by normalizing the vertical shift of the feature points to y, and then estimated by a least squares method. A fatigue estimation method of a stereo image, comprising a and b of y = ax + b. delete delete delete delete The method of claim 1, wherein the at least one value representing the vertical shift characteristic is
And a third value obtained by subtracting a sum of feature points having a relatively small vertical shift from a sum of vertical shifts of feature points having a relatively large vertical shift among the feature points. 7. The method of claim 6, wherein at least one value representing the vertical shift characteristic is
An absolute value for the sum of the vertical shifts of the feature points having a relatively high vertical shift among the feature points, and a fourth value greater than the absolute value for the sum of the vertical shifts of the feature points for which the vertical shift among the feature points is relatively small; A fatigue estimation method of a stereo image, characterized in that. delete Extracting feature points of the stereo image, estimating a horizontal shift and a vertical shift of the feature points of the feature points, and based on the horizontal shift and the vertical shift of the feature points, at least one value representing a horizontal shift characteristic of the stereo image and A variation estimator for generating at least one value representing the vertical variation characteristic; And
A fatigue estimator configured to estimate a fatigue degree of the stereo image based on at least one value indicating the generated horizontal shift characteristic and at least one value indicating the vertical shift characteristic;
The at least one value representing the horizontal shift characteristic may be equal to a horizontal shift of feature points having a relatively small horizontal shift from a sum of values obtained by multiplying a horizontal shift of feature points having a relatively large horizontal shift among the feature points and a weight based on a spatial frequency. A first value obtained by subtracting a sum of values multiplied by a weight based on a spatial frequency, and a second value obtained by adding a value obtained by multiplying a horizontal variation of feature points having a relatively large horizontal shift among the feature points and a weight based on a spatial frequency;
The at least one value representing the vertical shift characteristic is set by a vertical coordinate of the feature points to x, a normalized vertical shift generated by normalizing the vertical shift of the feature points to y, and then estimated by a least squares method. An apparatus for estimating fatigue of a stereo image, comprising a and b of y = ax + b. delete delete delete delete 10. The method of claim 9, wherein at least one value representing the vertical shift characteristic is
And a third value obtained by subtracting a sum of feature points having a relatively small vertical shift from a sum of vertical shifts of feature points having a relatively vertical vertical shift among the feature points. 15. The method of claim 14, wherein at least one value representing the vertical shift characteristic is
An absolute value for the sum of the vertical shifts of the feature points having a relatively high vertical shift among the feature points, and a fourth value greater than the absolute value for the sum of the vertical shifts of the feature points for which the vertical shift among the feature points is relatively small; The fatigue estimation device of the stereo image, characterized in that. delete A computer-readable recording medium having recorded thereon a program for executing the method of any one of claims 1, 6 and 7.

Images