JP2000048204A - Forming method of solid model of actual size - Google Patents

Abstract

PROBLEM TO BE SOLVED: To represent the size and positional relation of each subject precisely and to easily actualize real computer graphics(CG) by measuring the distances between visual points and a subject and calculating motion parameters of the visual points regarding movements by using the actual size information. SOLUTION: A subject 100 is photographed by a camera 120 at positions before and after camera movement. A gaze point EFP on the subject 100 is indicated with the light of a laser pointer 110 on the image and the 2D position of the gaze point EFP on the image can be extracted. The distances from the camera positions before and after the movement to the gaze point EFP are measured by the laser range finder incorporated in the camera 120. Motion parameters of the visual point regarding the movement are calculated by using the actual size information. Then the three-dimensional shape of the subject is restored in 3D by using the obtained motion parameters.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of creating a three-dimensional model of a real object (subject) in a computer graphics (hereinafter, referred to as "CG"). The present invention also relates to a method of creating a three-dimensional model in which the actual size of a subject is brought as data and the positional relationship and size between the subjects are precisely restored.

[0002]

2. Description of the Related Art Conventionally, as a method of restoring the 3D shape of a subject, a stereo method of restoring the shape of the subject based on the principle of triangulation using two images having parallax has been known. .

The stereo method can be said to be an excellent method without requiring a clay model or the like using clay or the like as an essential condition. However, in order to obtain a three-dimensional model by this method, at least two camera positions must be precisely defined in advance. Must be set to Therefore, if this method is applied to a large object such as a building that requires a long distance to shoot, not only must the camera be precisely installed, but also the distance between the two cameras must be increased. However, in the worst case, the positional relationship between the cameras cannot be determined.

In response to such a problem, the present inventors disclosed in Japanese Patent Application Laid-Open No. Hei 9-81779, without precisely controlling the positional relationship between cameras, And a method of accurately estimating the positional relationship between the cameras before and after the movement (hereinafter referred to as “camera motion parameters”) and creating a realistic three-dimensional model using the camera motion parameters (motion stereo method). is suggesting. Then, following this method, the present inventors apply the above-described camera motion parameter estimation method, integrate the partial stereoscopic model of the subject, and completely reconstruct the subject in full 3D from all angles. A model creation method has also been proposed (Japanese Patent Laid-Open No. 9-237346).

[0005]

However, as a result of the present inventors' investigations for realizing a more realistic CG, it has been found that there is room for the following improvement. That is, in the above-mentioned motion stereo method, the size of the subject is only measured relatively. For example, the distinction between a giant located far away and a dwarf of the same shape located near is C.
There is a problem that the size and the positional relationship of each subject tend to be unclear on the CG when the subject is not clear on G and there are a plurality of subjects.

Further, in the above-described method of creating a complete three-dimensional model, a plurality of partial three-dimensional models are combined and integrated. At this time, if the size and the positional relationship of each partial three-dimensional model are not clear, the combination is difficult. There was also a problem.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems of the prior art. It is an object of the present invention to accurately represent the size and the positional relationship of each subject and to realize a real CG. It is an object of the present invention to provide a method for creating a real-size three-dimensional model that can be easily realized.

[0008]

Means for Solving the Problems The present inventors have made intensive studies to achieve the above object, and as a result, the object has been achieved by bringing the distance between the subject and the camera in an actual size by a simple method. The inventors have found that the present invention has been completed.

That is, in the method for creating a full-size three-dimensional model according to the present invention, a plurality of images taken while moving the viewpoint are used.
In creating a three-dimensional model in which the three-dimensional shape of the subject is reconstructed in 3D, the distance between the viewpoint and the subject before and after the movement of the viewpoint is measured when moving the viewpoint, and the movement is performed using the actual size information. The three-dimensional reconstruction of the subject is performed using the obtained motion parameters.

In a preferred embodiment of the method for creating a full-size three-dimensional model according to the present invention, a distance between the viewpoints before and after the movement of the viewpoint is actually measured, and the motion parameter of the viewpoint is taken into consideration in consideration of the actual size information. The calculation is characterized in that the accuracy of the viewpoint motion parameter can be improved.

In a preferred embodiment of the method for creating a full-size three-dimensional model according to the present invention, one of the feature points of the subject is selected as a gazing point and the other point is selected as a reference point. The method is characterized in that a distance between the viewpoint after the movement and the gazing point and the reference point is actually measured, and a motion parameter of the viewpoint is calculated in consideration of the actual size information.

Further, in another preferred embodiment of the method for creating a full-size three-dimensional model according to the present invention, any one of the feature points of the subject is selected as a gazing point, and the gazing point is selected from the plurality of images. Controlling the movement of the viewpoint so as to always be located at the center of the image, and controlling the movement of the viewpoint such that the Y axis of the coordinates defining the viewpoint points vertically upward before and after the movement of the viewpoint. As a feature, the degree of freedom of the motion parameters (H, R) can be reduced, and the accuracy of the motion parameters of the viewpoint can be improved.

In another preferred embodiment of the method for producing a full-size three-dimensional model according to the present invention, regarding the movement of the viewpoint, coordinates defining the viewpoint are coordinates before movement and coordinates after movement, respectively. By converting the movement of the viewpoint into a function of φ by converting the coordinates into degrees of freedom (φ), the movement parameter of the viewpoint is calculated by searching for φ in a round robin manner. H,
The degree of freedom of R) can be further reduced, and the accuracy of the viewpoint motion parameters can be further improved.

Still another preferred embodiment of the method for producing a full-sized solid model according to the present invention is characterized in that the distance between the viewpoint and the subject and / or the distance between the viewpoints is actually measured by a laser range finder. I do. The laser range finder can be actually measured with a simple operation, and there is no major trouble in the operation of the 3D reconstruction. Further, since the laser range finder is inexpensive, it is advantageous in cost.

Further, a preferred embodiment of the method for producing a full-size solid model according to the present invention is characterized in that the point of regard is pointed by a laser pointer. Is easily obtained, and the gazing point can be easily identified on the image. Therefore, the gazing point can be easily positioned at the center of the image, which is advantageous in executing the gazing point restriction.

Further, the information medium of the present invention is characterized by recording instructions to a computer for performing the above-described method.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, a method for producing a full-size solid model according to the present invention will be described in detail. For convenience of description of the present invention, first, the motion stereo method and estimation of camera motion parameters which have been already proposed by the present inventors will be briefly described.

[Motion Stereo Method] (Coordinate System) FIG. 1 shows an example of a coordinate system used in the motion stereo method. In Figure 1, the coordinate system X - Y - Z
Before the camera moves, the coordinate system X '- Y' - Z 'represents a coordinate system after moving the camera (viewpoint). Also, "Feature points"
Is a characteristic point of the shape of the subject, for example,
Indicates a projection or an inflection point of a corner on a subject surface. In the figure, the camera coordinate system (X - Y - Z) Z-axis direction is coincident with the optical axis of the camera, X-axis and Y-axis each image coordinate system (x - y) that is parallel . The origin of the image coordinate system is located at a focal distance f from the origin O of the camera coordinate system.

Under the conditions described above, the movement amount ( H , R ) of the camera is expressed by the following equation (1). P _i = H + R P ′ _i , (i = 1,..., N) (1) In equation (1), the translation vector ( H ) and the rotation matrix ( R ) are coordinate systems before camera movement. This is the amount seen from P _i = ( X _i ,
Y _i , Z _i ) is the 3rd of the i-th feature point in the coordinate system before camera movement.
D indicates a position vector, and N indicates the number of feature points. Note that the dashes indicate the amounts observed in the coordinate system after the movement of the camera.

The translation vector H is given by H = ( H _X ,
H _Y , H _Z ) ^t , and the rotation matrix R is represented by R = R _X R _Y R _Z ,
In this case, R _X , R _Y , and R _Z indicate the camera rotation matrices around the X , Y , and Z axes, and t indicates the transpose of the matrices. here,
R _X is represented by, for example, the following formula (A).

[0021]

(Equation 1)

(Estimation of Camera Motion Parameters (Relative)) The image coordinates (x _i , y _i ) of the feature points can be obtained from the 3D position vector by projective transformation by the following equation. x _i = f · X _i / Z _i , y _i = f · Y _i / Z _i By substituting this equation for P _i , the following equation (B) is obtained.

[0023]

(Equation 2)

Therefore, the equation (1) is transformed into the following equation (C).

[0025]

(Equation 3)

The formula _{(C), r i, r} i ' denotes the distance from the origin of the camera back and forth movement of the coordinate system to the i-th feature point, m _i, m _i' is at each coordinate system 5 shows a unit vector indicating a direction from an origin to a feature point. In general, estimating the motion parameters ( H , R ) by associating N feature points before and after the movement of the camera results in a problem of obtaining a motion parameter that gives a minimum value in the following equation (2).

[0027]

(Equation 4)

[0028] (2) _{where, | H, m i, R} m i '| time motion parameters (H, R) is a true value, a epipolar condition becomes zero. W _i represents the weight.

The camera motion parameters (hereinafter abbreviated as "motion parameters") have six degrees of freedom including translation and rotation. A distinction between a case where the translation is large for a large distant subject and a case where the translation is small for a small nearby subject cannot be made from the screen. Therefore, the magnitude (| H |) of the translation vector H of the motion parameter cannot be obtained from the two images having parallax. When searching for the minimum value of the equation (2), the value of | H | is assumed (for example, | H
Since | = 1, the relative motion parameters have five degrees of freedom.

In Japanese Patent Application Laid-Open No. Hei 9-81779 by the present inventors, in particular, taking advantage of the fact that the subject is always photographed before and after the movement of the camera (“gazing object restriction”), the above-described five degrees of freedom are used. Has been reduced to three degrees of freedom, and a method of searching the three degrees of freedom space by brute force has been proposed.

(Calculation of Motion Parameters (Absolute)) To obtain the magnitude of H (similarly, the actual size information of the subject), the distance from the camera to the subject and the base line length (camera) at the shooting site are required. The distance between them may be measured and substituted into the above-described relative motion parameters to calculate the absolute motion parameters. The method for creating a full-size three-dimensional model of the present invention described below is based on this.

[Method of Creating Actual Size Solid Model] The operation in the method of creating the actual size solid model of the present invention will be described in detail below.

Work at the Shooting Site (Designation of Point of View) FIG. 2 shows the work at the shooting site. Look at one point on the subject 100 (EFP: Eye)
Fixation Point, that is, a point of particular interest among the feature points (indicated by a circle in FIG. 2), and is designated by a laser pointer 110 which is an example of a marker. By such an instruction of the laser pointer, the distance from the camera before and after the camera movement to the gazing point can be easily obtained, and the gazing point can be easily identified on the image. This is advantageous for executing the “gaze point constraint”.

(Photographing of the subject) At the positions before and after the movement of the camera shown in FIG. 2, a gazing object constraint is performed so that the subject 100 falls within the angle of view, and under this constraint, the camera 120 photographs the subject 100. The point of interest EFP is indicated by the laser pointer light on the image, and the 2D position of the point of interest EFP on the image can be extracted.

(Measurement of distance) The distance from the camera position before and after the movement of the camera to the gazing point EFP is determined by the camera 120.
It measures with the laser range finder built in. At this time, it is preferable to measure the distance (base line length) between the camera positions before and after the movement of the camera.

Constraints on Motion Parameters By imposing the following constraints, the motion parameters can be easily and accurately estimated.

(Gaze Object Constraint) It is known that camera rotation without translation does not affect the 3D reconstruction accuracy of a subject. Image processing (standard conversion) for moving an arbitrary point on an image to the image origin by rotating the camera is also known. Here, before moving the camera,
If the camera motion is set such that the point of interest is always brought to the center of the image, a certain constraint (translation and rotation) is generated in translation and rotation. Further, by performing the standard conversion, it is possible to convert a continuous image taken while gazing at the subject so that the gazing point constraint can be used.

In the present invention, image processing is performed using standard conversion ( S , S ') such that one point (point of interest) on the subject is at the center of the screen (see FIG. 3). As a result, it is possible to similarly convert the image into a captured image by performing gazing point control. In FIG. 3, the relationship between the translation amount (H), the rotation amount (R), and the like represented by the original coordinate system and the coordinate system after the standard conversion is represented by the following equation (D).

[0039] H = S H, R = S R S 't m i = S m i, m i' = S 'm i' ··· (D)

In the equation (D), H , R , m , and m ′ represent a translation vector, a rotation matrix, and a unit vector indicating a direction to a feature point expressed using coordinates before standard transformation. FIG.
XYZ is a coordinate system before camera movement in which the optical axis of the camera points to the gazing point, and X'-Y'-Z 'is a coordinate system after camera movement. Since the optical axes (Z, Z 'axes) of the camera point at the gazing point (gaze point constraint), the following equations (3) and (4) are derived.

[0041]

(Equation 5)

[0042]

(Equation 6)

In equations (3) and (4), Z ₀ , Z ₀ ′
Indicates the distance from the origin to the point of interest in each coordinate system,
θ _X and θ _Y indicate rotation angles about the X and Y axes, respectively.

The motion parameters are reduced from six degrees of freedom to four degrees of freedom due to the above-described gaze point constraint. When the equations (3) and (4) are replaced by the equation (2), the epipolar function (D) is expressed by the following equation (5), and it can be seen that the epipolar function (D) has four degrees of freedom. D (H, R) = D (H, θ _Z ) (5) Of the remaining four degrees of freedom, three degrees of freedom are translational components (H),
One degree of freedom is each roll (θ _Z ) around the optical axis of the camera.

(Up-down Constraint) A condition in which the Y axis of the camera always points to the upside is called up-down constraint. This condition
This is a constraint condition that is effective when shooting an object on the ground. It can be easily realized by using a down swing used in surveying. Due to such a vertical constraint, the following equation (6) is established for the roll angle (θ _Z ) of the camera.

[0046]

(Equation 7)

3D Shape Estimation Using Polar Coordinate Display (Introduction of Polar Coordinates) The origin of the coordinate system after moving the camera is a sphere having a radius H centered on the origin of the coordinate system before moving the camera, and a radius Z centering on the gazing point. The sphere of ₀ 'exists on the intersecting circle (see FIG. 4). In this specification, this circle is referred to as a solution search circle.

FIG. 5 is a view of the solution search circle viewed from the X-axis direction. From FIG. 5, the following (7) and (8) hold. H ² = R ² + A ² (7) Z ₀ ′ ² = R ² + (Z ₀ -A) ² (8)

As shown in the following equations (9) and (10),
R and A are obtained from the above equations (7) and (8).

[0050]

(Equation 8)

[0051]

(Equation 9)

FIG. 6 is a view of the solution search circle viewed from the Z-axis direction. From FIG. 6, the following equations (11) to (13) hold. H _X = R cos φ (11) H _Y = R sin φ (12) H _Z = A (13)

Equations (11) to (13) are replaced by equation (3),
Substituting in (4) and (6), the following (14) to (16)
It can be seen that the equation holds.

[0054]

(Equation 10)

[0055]

[Equation 11]

[0056]

(Equation 12)

Then, equations (11), (12) and (13)
When Equations (14) and (15) are substituted into Equation (2), the epipolar function (D) is expressed by the following Equation (17), and it can be seen that the epipolar function (D) has two degrees of freedom. D (H, R) = D (φ, θ _Z ) (17)

(Solution Search Procedure) As described above, considering a search in the motion parameter space reduced to two degrees of freedom by the polar coordinate display, the search range is from φ _min to φ _max , Equation (16)
Θz + θz from θz + θz _min centered on θ _Z calculated by
It should be _max . Distance search is respectively [Delta] [phi, and [Delta] [theta] _Z,
Equation (17) is calculated by brute force and the true value is estimated to determine camera motion parameters.

In the present invention, if the 3D reconstruction is performed in accordance with the ordinary method using the motion parameters determined as described above, a precise and realistic three-dimensional model can be obtained in consideration of the actual size of the subject.

[0060]

EXAMPLES Hereinafter, the present invention will be described in more detail with reference to some examples, but the present invention is not limited to these examples. In the following description, a simulation experiment and a real image experiment were performed for each of the above-described gazing point method according to the motion stereo method as a comparative example, and the distance measuring method according to the method for creating a full-sized solid model of the present invention as an example. Was.

[Experiment Conditions] Experimental conditions common to a simulation experiment and a real image experiment which will be described later are listed.

(Subject and Each Feature (Point, Side, Angle)) As shown in FIG. 7, two orthogonal faces of the rectangular parallelepiped 200 were set as subjects. In the figure, "●" marks represent characteristic points, and are indicated by reference numerals "-" respectively. The “sides” connecting the feature points are denoted by reference numerals 1 to 12 (upper case), and the angles between the sides are denoted by reference numerals 1 to 22 (lower case). The length of sides 1, 6, 11 is 50.0m
m, the length of the other sides was 40.0 mm. Also,
The size of the angles 3, 8, and 20 is 180.0 °, and the other angles are 90.0 °.

(Camera internal parameters) The measurement results of the camera internal parameters of the camera used for photographing are shown below. Focal length (f): 25.0 mm Pixel conversion ratio (ε _x , ε _y ): (81.0, 80.0) (pixel / mm) Pixel center (i ₀ , j ₀ ): (273, 235) (pixel) The sampling is 512 × 512 pixels.

(Selection of Point of Gaze) The point of gaze was a feature point closest to the center of the image in the image before moving the camera. In FIG. 7, the feature point is the gazing point.

(Solution Search Range, etc.) In the implementation of the brute force calculation of the distance measurement method according to the present invention, the solution search range is set to φ _min , φ
_max is -90 degrees, 90 degrees, and θz _min and θz _max are -3.
0 degrees, 30 degrees, to determine the solution interval Δφ search, the [Delta] [theta] _z as 0.5 degrees. Workstation Ind for calculation processing
Using y, the processing program was created in C language. The calculation time required for estimating the motion parameters was about 7 seconds by the fixation point method (conventional method) and about 2 seconds by the distance measurement method (the present invention).

[Simulation Experiment] FIG. 8 shows an example of the motion parameters of the simulation experiment. XY
−Z is the coordinates before the movement of the camera. (H _X , H _Y , H _Z ) is (433.0, 0.0, 250.0) mm, and (θ
_X , θ _Y , θ _Z ) are (0.0, −60.0, 0.0) degrees. For the rotational component of the motion parameter, noise described below is used, and for the distance measurement, a gazing point method (comparative example) and a distance measuring method (example) are applied by adding an error to be described later. Average 3D reconstruction accuracy,
The variance was determined. The noise and the error are represented by a normal distribution function N (μ, σ) of an average (μ) and a standard deviation (σ).

(Noise of rotation components θ _X , θ _Y , θ _Z ) θ _X ,
The true value of θ _Y is defined as the average value (μ) obtained from the true values of (H _X , H _Y , H _Z ) by the gazing point constraint, and a standard deviation (σ) of 1.0 degree The distribution Nθ _X (0.0, 1..
0) and Nθ _Y (−60.0, 1.0) were added as noise. For example, if a noise of 5 degrees is added to θ _X , the gazing point on the image is shifted by about 150 pixels from the center of the image in the horizontal direction. If the noise is large, the point of gaze shifts greatly from the center of the image, and the subject cannot be captured at the angle of view.
Also, the true value of θ _Z is defined as the average value (μ) of the true value of (H _X , H _Y , H _Z ) obtained by the vertical constraint, and the standard deviation (σ) is 10.0 degrees. Normal distribution Nθ _Z (0.0, 1
0.0) was added as noise.

(Distance Measurement Error of Rangefinder) The distance measurement error of the rangefinder is expressed by the following equation. ＾ Z ₀ = Z ₀ + N _Z0 (0, σ) × Z ₀ ＾ Z ₀ ′ = Z ₀ ′ + N _{Z0 ′} (0, σ) × Z ₀ 'H = H + N _H (0, σ) × H

In the above equation, ＾ Z ₀ , ＾ Z ₀ ′, and ＾ H indicate values of Z ₀ , Z ₀ ′, and H in FIG. 8 measured by a distance meter having a distance measurement error. Here, H indicates the magnitude of the translation vector (H). Hereinafter, “σ” is referred to as a distance measurement error coefficient. In a typical laser range finder, a laser range finder has σ of about 0.01%, and in a measure such as Teslon fiber, σ is about 0.1 to 1.0%.

Estimation of Exercise Parameters (Example 1 and Comparative Example 1) (Comparative Example 1) Exercise parameters were estimated using the conventional gazing point method, and the results are shown in Table 1. In the present example, a distance value (＾ Z ₀ ) assuming the case where the distance (Z ₀ ) from the camera center to the gazing point is measured by the distance meter is substituted.

[0071]

[Table 1]

In Table 1, δT and δR indicate the average value and the variance of the difference between the estimated value and the true value of the translation (T) and the rotation (R) of the movement, respectively. Here, the Euclidean distance defined by the following equation was used for the difference between the matrices. Note that R and R ^{0 in the} expression indicate the estimated value and the true value of the basic matrix, respectively.

[0073]

(Equation 13)

From Table 1, the estimated error of the translation (T) of the motion is 20.0 mm or less in any case, and the estimated error of the rotation (R) of the motion is 0.04 or less in any case. You can see that.

(Example 1) The motion parameters were measured by applying the distance measuring method according to the present invention, and the obtained results are shown in Table 2.

[0076]

[Table 2]

From Table 2, it can be seen that in the ranging method, the accuracy of the translation (T) of the motion is proportional to the ranging accuracy (σ), and the accuracy of the translation estimation of this method depends on the accuracy of the range finder. You can see that. Further, the error of the rotation amount (R) of the movement is 0.03 or less in each case. When Table 1 and Table 2 are compared, it is clear that the distance measurement method of the present invention significantly improves the accuracy of the motion parameter estimation, particularly the accuracy of the translation amount (T), compared to the fixation point method.

3D Reconstruction Experiment "Reconstruction of Feature Points" (Example 2 and Comparative Example 2) The 3D reconstruction of the object shown in FIG. 7 was performed by applying the fixation point method and the distance measurement method, and the obtained result was obtained. FIG.
9A to 9F, the horizontal axis represents a feature point number, and the vertical axis represents a 3D restoration error. FIG. 9 (a)-
FIG. 9C shows the result by the fixation point method (Comparative Example 2), and FIGS. 9D to 9F show the result by the distance measurement method (Example 2). 9A and 9D are X-axis, FIGS. 9B and 9E are Y-axis, and FIGS. 9C and 9F are 3D in the Z-axis direction.
This shows the restoration error. In order to compare and show the restoration error when the ranging accuracy (σ) is 0.0% to 1.0%, (a),
In (b), (d) and (e), each is 1.0 mm, (c),
In (f), the display is shifted by about 5.0 mm. The error bars in each figure are the standard deviation values from 10 experiments.

In Comparative Example 2 to which the gazing point method is applied, the restoration error when σ is 0.0% to 0.1% is ±± in the X-axis direction.
0.27 mm, ± 0.12 mm in the Y-axis direction, and ± 2.18 mm in the Z-axis direction (see FIGS. 9A to 9C). On the other hand, in Example 2 according to the distance measuring method, σ is 0.0%
From 0.1% to ± 0.1% in the X-axis direction.
mm, ± 0.09 mm in the Y-axis direction. The restoration error in the Z-axis direction is ± 0.2 when σ is 0.0% and 0.01%.
± 0.57 in the range of 9 mm and σ is 0.1%
mm (FIGS. 9D to 9F).

In the second embodiment according to the distance measuring method, the accuracy is improved 2.3 times in the X-axis direction and 1.3 times in the Y-axis direction as compared with Comparative Example 2 according to the gazing point method. In particular, in the Z-axis direction, σ
Is 7.5% when σ is 0.0% and 0.01%, and σ is 0.1
In the case of%, the precision reaches 3.8 times. As described above, in the 3D restoration according to the comparative example 2 relating to the gazing point method, the accuracy in the depth (Z-axis) direction is lower than the accuracy in the horizontal (X, Y-axis) direction. The restored subject is distorted in the depth direction. Also,
It can be seen that there is a disadvantage that the distortion varies and the distortion is likely to occur (see FIGS. 9A to 9C).

On the other hand, in the second embodiment according to the present invention,
Even if the depth (Z-axis) direction and the horizontal (X, Y-axis) direction are compared, the difference in accuracy is almost the same when σ is 0.0% and 0.01% (FIG. 9 (e) ) To (f)). It can also be seen that the variation in distortion is smaller than in the case of the fixation point method (FIGS. 9A to 9F). That is, it can be seen that the 3D restoration by the distance measurement method according to the present invention has less distortion and the variation of the distortion than the 3D restoration by the fixation point method.

Object Shape Restoration Experiment "Side Restoration" (Embodiment 3 and Comparative Example 3) FIG. 10 is a graph showing the restoration accuracy with respect to the length of the side of the object shape, and FIG. FIG. 10 (b) is based on the distance measurement method (Example 3). The horizontal axis indicates the side number (see FIG. 7), and the vertical axis indicates the error of each side from the true value. Distance measurement accuracy (σ)
Are shifted by about 2.0 mm to show the restoration errors from 0.0% to 1.0% in comparison. Also,
The error bars in each figure show the standard deviation values from ten experiments.

In Comparative Example 3 relating to the gazing point method, σ is 0.0
% To 1.0%, the side length is restored within ± 0.95 mm, whereas in the third embodiment according to the distance measuring method, σ is within the same range and ± 0.30 mm. Has been restored. Therefore, in the restoration of the length of the side of the subject, the third embodiment
Is 3.2 times more accurate than Comparative Example 3. Comparative Example 3
Although the errors of the side numbers 3, 4, and 5 are large, when the feature points are 3D-reconstructed by the fixation point method,
5 and 6 due to a large restoration error (FIGS. 9 and 10).
reference). On the other hand, in the third embodiment, the restoration accuracy of each side is substantially constant regardless of the side number.

Object Shape Restoration Experiment “Restoration of Angle” (Embodiment 4 and Comparative Example 4) FIG. 11 is a graph showing the restoration accuracy with respect to the angle of the object shape, and FIG. Comparative Example 4), FIG. 11B shows the result of the distance measurement method (Example 4).
It is. The horizontal axis indicates the angle number (see FIG. 7), and the vertical axis indicates the error of each angle from the true value. In order to compare and show restoration errors when the distance measurement accuracy (σ) is 0.0% to 1.0%, each is shifted by about 4.0 degrees. The error bars in each figure show the standard deviation values of 10 experiments.

In Comparative Example 4 relating to the gazing point method, σ is 0.0
% In the range of 1.0% to 1.0%, the magnitude of the angle is restored within ± 2.04 degrees.
Is restored in the same range within ± 0.53 degrees. Therefore, in recovering the angle of the subject, the fourth embodiment is 3.8 times more accurate than the fourth comparative example. In Comparative Example 4, the errors of the angle numbers 6, 11, 12, and 13 are large. However, when the feature points are 3D-reconstructed by the fixation point method, the restoration errors of the feature point numbers 4, 5, and 6 are large. Due to the large size (see FIGS. 9 and 10). On the other hand, in the fourth embodiment, the restoration accuracy of each angle is substantially constant regardless of the angle number.

Further, from FIGS. 10 and 11, it can be seen from FIG. 10 and FIG. 11 that the object shape restoration by the distance measuring method according to the present invention has a smaller distortion and the variation of the distortion than the conventional object shape restoration by the fixation point method. I find it difficult.

[Real Image Experiment] As shown in FIG. 12, a subject 200 (see FIG. 7) is set on a table 210 inclined at 45 degrees with respect to a horizontal plane, and a manipulator 230 capable of controlling the translation and rotation of a camera 220. Installed above. The feature points of the subject were manually extracted by the operator from the image subjected to the image enhancement. Under these conditions, the following gazing parameters and the like were applied to estimate the following motion parameters and the like.

Estimation of Exercise Parameters (Embodiment 5 and Comparative Example 5) Table 3 shows the results of estimating exercise parameters by applying the fixation point method (Comparative Example 5) and the ranging method (Example 5).

[0089]

[Table 3]

As shown in the table, the translational motion parameter error is 17.10 mm in Example 5 according to the distance measurement method, compared to 74.99 mm in Comparative Example 5 according to the gazing point method.
It can be seen that Example 5 has significantly better accuracy.

3D Restoration Experiment "Reconstruction of Feature Points" (Embodiment 6 and Comparative Example 6) FIG. 13 is a graph showing 3D reconstruction errors of feature points in the X, Y, and X axis directions. ) Is based on the fixation point method (Comparative Example 6), and FIG. 13B is based on the distance measurement method (Example 6). The horizontal axis indicates the feature point number (see FIG. 7), and the vertical axis indicates the 3D restoration error.

In Comparative Example 6 relating to the gazing point method, the restoration error was ± 0.79 mm in the X-axis direction and ± 1.19 in the Y-axis direction.
mm, ± 6.10 mm in the Z-axis direction (FIG. 13
(See (a)). Since the error in the y-axis direction when indicating the feature point number 6 is large, the restoration error is large. On the other hand, in the sixth embodiment according to the distance measuring method, the restoration error is in a range of ± 0.79 mm in the X-axis direction, ± 1.20 mm in the Y-axis direction, and ± 0.36 mm in the Z-axis direction (FIG. b)).

When the embodiment 6 is compared with the comparative example 6, both have almost the same accuracy in the X-axis direction and the Y-axis direction, but the embodiment 6 has much higher accuracy in the Z-axis direction. Further, in Comparative Example 6, the accuracy in the depth (Z-axis) direction is the same as that in the horizontal (X, Y-axis).
It is worse than the direction accuracy (see FIG. 13A). On the other hand, in the sixth embodiment, the depth accuracy and the accuracy in the horizontal direction are almost the same (see FIG. 13B). Therefore, the distortion in the depth direction is smaller in the 3D restoration of the sixth embodiment than in the 3D restoration of the sixth embodiment. Thus, in the 3D restoration of the feature points, the result of the real image experiment showed the same tendency as the result of the simulation experiment.

Object Shape Restoration Experiment "Side Restoration" (Embodiment 7 and Comparative Example 7) FIG. 14 is a graph showing the restoration accuracy with respect to the length of the side of the object shape. Example 7), the dashed line is based on the fixation point method (Comparative Example 7).
The horizontal axis indicates the side number (see FIG. 7), and the vertical axis indicates the error of each side from the true value.

In Comparative Example 7 relating to the gazing point method, the length of the side is restored within ± 2.73 mm, whereas in Example 7 relating to the distance measuring method, the length of the side is ± 0. Example 7 is 3.7 times more accurate than Comparative Example 7 in the restoration of the side length within 73 mm. In Comparative Example 7, the errors of the side numbers 3, 4, and 5 are large.
When the feature points are restored in 3D by the fixation point method, the feature point number 4,
5 and 6 are caused by a large restoration error (FIG. 1
3, see FIG. 14). On the other hand, in the seventh embodiment, the restoration accuracy of the length of each side is almost the same regardless of the side number.

Object Shape Restoration Experiment "Restoration of Angle" (Embodiment 8 and Comparative Example 8) FIG. 15 is a graph showing the restoration accuracy relating to the angle of the object shape. The solid line is based on the distance measurement method (Eighth Embodiment). The dashed line indicates the result of the fixation point method (Comparative Example 8). The horizontal axis indicates the angle number (see FIG. 7), and the vertical axis indicates the error of each angle from the true value.

In Comparative Example 8 relating to the gazing point method, the magnitude of the angle is restored within ± 6.17 degrees, whereas in Example 7 relating to the distance measuring method, the magnitude of the angle is ± 1. Example 8 is 3.2 times more accurate than Comparative Example 8 in the angle restoration. In Comparative Example 8, the errors of the angle numbers 6, 11, 12, and 13 are large. However, when the feature points are 3D-restored by the fixation point method, the restoration errors of the feature point numbers 4, 5, and 6 become large. (See FIGS. 13 and 15). On the other hand, in the eighth embodiment, the restoration accuracy of each angle is almost the same without depending on the angle number.

As described above, it is apparent from FIGS. 14 and 15 that the object shape restoration by the distance measuring method according to the present invention has less distortion than the object shape restoration by the fixation point method.
In the reconstruction of the object shape related to the side length and the angle, the result of the real image experiment showed the same tendency as the result of the simulation experiment.

[0099]

As described above, according to the present invention, since the distance between the subject and the camera is brought to the actual size by a simple method, the size of each subject and the positional relationship thereof are precisely expressed. It is possible to provide a method of creating a full-size three-dimensional model capable of easily realizing a real CG.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram illustrating an example of a coordinate system used in a motion stereo method.

FIG. 2 shows an example of a method for creating a three-dimensional model according to the present invention.
It is explanatory drawing which shows the operation | work in a photography site.

FIG. 3 is an explanatory diagram showing gazing point control using standard conversion.

FIG. 4 is an explanatory diagram showing an example of a solution search circle.

FIG. 5 is an explanatory diagram of a solution search circle viewed from an X-axis direction.

FIG. 6 is an explanatory diagram of a solution search circle viewed from a Z-axis direction.

FIG. 7 is a perspective view showing a subject and respective feature amounts (points, sides, angles).

FIG. 8 is an explanatory diagram showing an example of a motion parameter of a simulation experiment.

FIG. 9 is a graph showing a 3D restoration error of a feature point in a simulation experiment.

FIG. 10 is a graph showing restoration accuracy with respect to the length of a side of a subject shape in a simulation experiment.

FIG. 11 is a graph showing the restoration accuracy relating to the angle of the object shape in a simulation experiment.

FIG. 12 is a perspective view showing an experimental environment of a real image experiment.

FIG. 13 is a graph showing a 3D restoration error of a feature point in a real image experiment.

FIG. 14 is a graph showing restoration accuracy with respect to the length of a side of a subject shape in a real image experiment.

FIG. 15 is a graph showing the restoration accuracy relating to the angle of the object shape in a real image experiment.

[Explanation of symbols]

 Reference Signs List 100 subject 110 laser pointer 120 camera 200 subject 210 units 220 camera 230 manipulator

 ──────────────────────────────────────────────────続 き Continued on the front page F-term (reference) 2F065 AA04 AA06 AA53 BB05 FF05 FF09 FF11 FF23 GG04 JJ01 JJ03 PP22 QQ00 QQ03 QQ21 QQ41 QQ42 UU05 5B046 EA09 FA15 FA18 GA01 GA09 5B057 AA16 CB01 CA02 CA01

Claims (9)

[Claims] When creating a three-dimensional model in which a three-dimensional shape of a subject is reconstructed by using a plurality of images taken while moving the viewpoint, the viewpoint before and after the movement of the viewpoint when the viewpoint is moved and photographed. The distance between the viewpoint and the subject is measured, a motion parameter of the viewpoint related to the movement is calculated using the actual size information, and the three-dimensional shape of the subject is 3D-reconstructed using the obtained motion parameter. How to create a real size solid model. 2. The solid object according to claim 1, wherein a distance between the viewpoints before and after the movement of the viewpoint is measured, and a motion parameter of the viewpoint is calculated in consideration of the actual size information. How to create a model. 3. One of the feature points of the subject is selected as a point of interest, and the other point is selected as a reference point, and the viewpoint before and after movement of the viewpoint, the point of interest, and the reference point are selected. 3. The method according to claim 1, further comprising: actually measuring a distance of the object and calculating the motion parameter of the viewpoint in consideration of the actual size information. 4. An arbitrary point among the feature points of the subject is selected as a gazing point, and the movement of the viewpoint is controlled such that the gazing point is always located at the image center of the plurality of images. The movement of the viewpoint is controlled so that a Y-axis of coordinates defining the viewpoint points vertically upward before and after the movement of the viewpoint.
How to create an actual size solid model described in the two sections. 5. The movement of the viewpoint, wherein the coordinates defining the viewpoint are coordinates before movement and coordinates after movement, respectively, and the coordinates after movement are converted into one-degree-of-freedom coordinates (φ). Is a function of φ, and this φ is searched by brute force to calculate the motion parameter of the viewpoint, and the method of creating a full-size solid model according to any one of claims 1 to 4, wherein 6. The method according to claim 1, wherein a distance between the viewpoint and the object is measured by a laser range finder. 7. The method according to claim 2, wherein the distance between the viewpoints is measured with a laser range finder. 8. The method according to claim 3, wherein the gazing point is indicated by a laser pointer. 9. An information medium characterized by recording an instruction to a computer for performing the method according to claim 1. Description: