JP7464134B2 - Shape modeling device and shape modeling method - Google Patents

Description

The present disclosure relates to a shape modeling device and a shape modeling method that reconstruct the surface shape of a structure based on three-dimensional point cloud data of the structure obtained by image measurement, laser measurement, etc.

Measurement using images and laser scanners is an easy way to measure the shapes of structures such as tunnels, roads, bridges, columnar installations, and buildings, as well as the layout of indoor components and machine parts. In image-based measurement, feature points in multiple images taken at different positions are matched to measure the position of the object to be measured using the principle of triangulation. In addition, an RGB-D camera that simultaneously obtains color images (RGB) and depth images (Depth) can also measure the position of the subject. Measurement using laser scanners includes a method that uses a fixed three-dimensional laser scanner and a method that uses mobile surveying known as a mobile mapping system (MMS). In these measurements, the measurement results are obtained as a set of points with three-dimensional coordinates called a point cloud on the surface of the object to be measured.

One method for reconstructing the shape of an object from this point cloud is to find the surfaces of the object from the point cloud and express the shape of the structure as a collection of surfaces. For example, one way to find planes is to divide the point cloud into each surface to which the point cloud belongs, and then use the least squares method to find the point cloud data belonging to each surface.

When the target structure covers a wide area or has a complex shape, it is not possible to measure the entire structure in a single measurement. In such cases, the overall shape is determined by using image measurement results from multiple viewpoints, or laser scanner measurement results from multiple positions, or by aligning a point cloud of driving measurements taken by an MMS over multiple runs, or a plane determined from the point cloud.

When detecting aging changes from measurement results, the current measurement results are aligned with the point cloud or plane of past measurement results that serve as a reference for comparison. For this alignment, there is a technique that divides the spatial region of the object into voxels, sets representative points and error distributions of the representative points within the voxels, and performs alignment taking errors into account.

Furthermore, Patent Document 1 discloses a technology in which surfaces are found from different point clouds, a rigid transformation is calculated under the condition that the sum of squares of the distances between points belonging to each surface of one surface and the corresponding surface of the other surface is minimized, and the entire one point cloud is transformed using this rigid transformation to align it with the other point cloud.

Publication No. WO2012/141235 JP 2004-272459 A Publication No. WO2007/69724

Point clouds obtained by three-dimensional measurement have errors in their coordinate values. For example, when using stereo image measurement, the error of points measured at positions far from the cameras will be large compared to the distance between the two cameras. With laser scanners, the accuracy and density of point clouds vary depending on the model. Also, errors become large when there is little diffuse reflection from metals, when the intensity of light returning to the laser scanner is low from black objects, when the reflection intensity is too strong due to specular reflection, or when the angle of incidence of the laser pulse on the surface is large. In this way, the measurement accuracy of each point cloud data varies depending on the distance from the measuring instrument, the relative attitude, and the condition of the surface. It is generally said that there is a deviation of several centimeters in the coordinate values of measurement results using MMS.

Errors in point cloud data also affect the plane reconstructed from the point cloud data. The plane will contain errors in its orientation and position. In Patent Document 2, the error in the reconstructed plane caused by errors in the point cloud is unknown. The variance of the residual error in fitting the points to the plane is used as the criterion for determining the range of the plane, but this variance is a numerical value that represents the degree of variation in the distance between each point and the plane, and does not represent the degree of error in the plane.

In Patent Document 3, the error of the voxels is represented, but the error of the plane is not represented. Even if a plane is reconstructed from voxel-formatted data, the error of the plane cannot be represented. In addition, voxel-formatted data must be created and all of that data must be stored.

In Patent Document 1, when aligning surface models, all point cloud information belonging to the surface must be retained. In addition, it is not possible to know the degree of error that the plane has.

In neither case is it possible to evaluate the error in the surface derived from the point cloud, or at least not accurately.

The objective of this disclosure is to provide a shape modeling device and a shape modeling method that assign a reliability, which is an index showing the degree of error, to a surface determined from a point cloud, and that precisely aligns surfaces determined by different measurements based on that reliability.

The shape modeling device disclosed herein comprises a data input/output means for inputting point cloud data and outputting data of a surface model, a surface approximation means for determining a surface shape that approximates the point cloud data, a reliability calculation means for calculating the reliability of the surface shape, and a reliability assignment means for assigning reliability to the surface model.

The shape modeling method disclosed herein is a shape modeling method for restoring the shape of an object from point cloud data obtained by measuring the object, and includes an input process for inputting point cloud data, a surface approximation process for determining a surface shape that approximates the point cloud data, a reliability calculation process for calculating the reliability of the surface shape, a reliability assignment process for assigning reliability to a surface model, and a data output process for outputting data of the surface model.

According to this disclosure, it is possible to obtain a shape modeling device and a shape modeling method using an approximate surface model to which a reliability indicating the likelihood is assigned.

1 is a diagram illustrating a configuration of a shape modeling device according to a first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 1 is a diagram illustrating point cloud data used in a shape modeling device according to a first embodiment of the present disclosure. FIG. 4 is a diagram illustrating point cloud accuracy data used in the shape modeling device according to the first embodiment of the present disclosure. FIG. 3A to 3C are diagrams illustrating surface model data used in the geometric modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4A to 4C are diagrams illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. 4 is a flowchart illustrating an operation of the shape modeling device according to the first embodiment of the present disclosure. FIG. 11 is a diagram illustrating a configuration of a shape modeling device according to a second embodiment of the present disclosure. FIG. 11 is a diagram illustrating an operation of the shape modeling device according to the second embodiment of the present disclosure. FIG. 11 is a diagram illustrating an operation of the shape modeling device according to the second embodiment of the present disclosure. FIG. 11 is a diagram illustrating an operation of the shape modeling device according to the second embodiment of the present disclosure. 11 is a flowchart illustrating an operation of a shape modeling device according to a second embodiment of the present disclosure. FIG. 13 is a diagram illustrating a configuration of a shape modeling device according to a third embodiment of the present disclosure. 13A to 13C are diagrams illustrating the operation of a shape modeling device according to a third embodiment of the present disclosure. 13A to 13C are diagrams illustrating the operation of a shape modeling device according to a third embodiment of the present disclosure. 13A to 13C are diagrams illustrating the operation of a shape modeling device according to a third embodiment of the present disclosure. 13A to 13C are diagrams illustrating the operation of a shape modeling device according to a third embodiment of the present disclosure. 13A to 13C are diagrams illustrating the operation of a shape modeling device according to a third embodiment of the present disclosure. 13 is a flowchart illustrating an operation of a shape modeling device according to a third embodiment of the present disclosure.

Embodiment 1.
FIG. 1 is a diagram showing an example of a configuration of a shape modeling device 1 according to a first embodiment of the present disclosure. A data input/output unit 2, which is a data input/output means, is connected to a data server 4 that stores data via a computer network 3 such as the Internet or an intranet, reads point cloud data 5 and point cloud accuracy data 6 stored in the data server 4, and outputs surface model data 7 to which reliability has been assigned. A surface approximation unit 8, which is a surface approximation means, obtains a surface on which the point cloud data 5 is placed using the least squares method. A reliability calculation unit 9, which is a reliability calculation means, calculates an error matrix in the surface fitting using the least squares method performed by the surface approximation unit 8 as a reliability that is an error index of the surface model. A reliability assignment unit 10, which is a reliability assignment means, assigns the reliability calculated by the reliability calculation unit 9 to the surface model calculated by the surface approximation unit 8 to obtain surface model data 7 having reliability that is an error index. A drawing unit 11, which is a drawing means, draws a shape represented by the obtained surface model using a three-dimensional computer graphics technique and displays it on a display device 12. The contents to be drawn by the drawing section 11 are specified or changed by receiving operational input from an input device 13 such as a mouse or keyboard at an operation input section 14 .

As shown in Figure 2, the point cloud data 5 is a collection of measurement points 18 having three-dimensional coordinate values obtained by measuring the target structure 15 using a laser scanner or the like. When modeling the shape of the structure 15 from the point cloud data 5, a reliability indicating the accuracy of the approximation is calculated for the approximation surface 16 approximated from the point cloud data 5.

Here, the point cloud data 5 is assumed to be grouped in advance according to the surface to which it belongs. For example, a method of grouping is known in which normal vectors are calculated from nearby points and points with similar directions are grouped together, or a plane extraction method using RANSAC. For example, each point may have information indicating the surface to which it belongs, or each point may be stored as different array data for each surface to which it belongs.

Next, the coordinate system to be used will be explained. A Cartesian coordinate system is used as the world coordinate system. The Cartesian coordinate system may be a planar rectangular coordinate system as specified by the Ministry of Land, Infrastructure, Transport and Tourism, or a coordinate system with an arbitrary origin. The units are meters, for example. In the following, a right-handed coordinate system will be explained, with the X and Y axes in a horizontal plane and the Z axis pointing vertically upward. Separately from this, a local coordinate system (hereinafter referred to as the "approximate surface coordinate system"), which will be described later, is defined on each approximate surface 16.

Of the measurement points 18 that make up the point cloud data 5, the i-th point is designated Pi, and its coordinate value in the world coordinate system is designated pi = (Xi, Yi, Zi). The number of the surface to which this point belongs is designated ki. If the point is noise or the like and does not belong to any surface, this is distinguished by setting, for example, ki = -1. Figure 3 is an example of the data format of the point cloud data 5, where the X coordinate value Xi, Y coordinate value Yi, Z coordinate value Zi of point Pi, and the number ki of the surface to which point Pi belongs are designated.

Point cloud accuracy data 6 is data that represents the measurement accuracy of each point, and may be a value calculated by taking into account the theoretical accuracy at the time of image measurement, the accuracy of the laser scanner, or the reflection intensity value and the angle of incidence. The point cloud accuracy data 6 is the value of the standard deviation of these. An example of the data format of the point cloud accuracy data 6 is shown in FIG. 4. This is an example in which the point cloud accuracy is the value σi of the standard deviation of the error for each point. This accuracy value may be included and stored in the point cloud data 5. In addition, the point cloud accuracy data 6 may be set to a predetermined value for each number of the same surface. Alternatively, if the point cloud data 5 is measured with the same laser scanner, a predetermined value may be used for all points. Of course, the value may be changed for each point depending on the measurement conditions.

The surface model data 7 is data representing an approximation surface 16 that represents the shape of the object restored from the point cloud data 5. It is composed of data that has information on a reference point 17, which is a point on the approximation surface 16, and a normal vector 19, or that can be converted into these. For example, if the surface is represented by a polygon model, it may be composed of a sequence of vertices of the polygon. The surface model 22 adds a reliability S, which is an index of the error of the surface, to the geometric information of this approximation surface 16. The reliability is the variance-covariance matrix of parameters that represent the orientation and position of the surface 16. The orientation of the approximation surface 16 is represented by two parameters that determine the direction of its normal vector 19. The position of the approximation surface 16 is represented by one parameter that indicates the amount of movement of the reference point 17 in the direction of the normal vector 19. FIG. 5 shows an example of the data format of the surface model data 7 representing a portion of a surface model 22, where the reference point 17 is Q, the coordinate value of Q is q = (qx, qy, qz), the normal vector 19 is a unit vector n = (nx, ny, nz), and the reliability of the approximation surface 16 is S.

The approximation of the point cloud to a plane in the surface approximation unit 8 will be explained using Figure 6. Note that in Figure 6, the measurement points 18 are simplified and only a portion of them are drawn, but in reality they are widely distributed on the surface. As mentioned above, the surface to which each measurement point 18 belongs is given in advance. We consider the measurement point 18 that belongs to one of these surfaces, the Kth surface. Let BK be the set of points Pi where ki = K. The plane of the approximation surface 16 is reconstructed from the points included in this BK using the least squares method.

In the following, to explain the points included in BK as an example, there are N points included in BK, and this jth point is denoted as PKj, where j = 1, 2, ..., N (natural number). The approximation surface 16 using PKj is obtained by the least squares method weighted by the accuracy σKj of each measurement point 18. In the following, for this Kth surface, the approximation surface 16 is denoted as Ω, its reference point 17 is denoted as Q, and the coordinate of reference point Q in the world coordinate system is denoted as q. Furthermore, the normal vector 19, which is a unit vector of the approximation surface 16, is represented as n in the world coordinate system.

Here, we explain the approximate surface coordinate system defined based on the approximate surface 16. First, the center of gravity of the measurement point PKj is set to Q, and the normal vector of the surface obtained by the least squares method in the world coordinate system is set to n, giving the initial approximate surface 16.

The x-axis and y-axis of the approximate surface coordinate system are taken within the approximate surface 16, and the z-axis is taken in the direction of the normal vector n. Here, the x-axis is taken horizontally. However, if the approximate surface 16 is horizontal, the x-axis is taken parallel to the X-axis of the world coordinate system. This approximate surface coordinate system will also be explained as a right-handed system.

If the approximation surface Ω is not horizontal, then if ex is a horizontal unit vector in the x-axis direction, then ex is as follows.

Here, × represents the vector product of vectors. The unit vector ey in the y direction on the approximation surface Ω is as follows:

The unit vector ez in the z direction is aligned with the direction of the normal vector n.

The approximate surface coordinate system is a coordinate system based on ex, ey, and ez.

When the approximation surface Ω is horizontal, ex, ey, and ez are set parallel to the X, Y, and Z axes, respectively, to form the approximation surface coordinate system. In this case, the z axis is also parallel to the normal vector n.

Let the coordinate value of point PKj in the world coordinate system be (XKj, YKj, ZKj). If the coordinate value in the approximate surface coordinate system is pLocalj = (xj, yj, zj), this can be found as follows. Here, ・ represents the dot product of vectors.

The weighted least squares method in the approximate surface coordinate system is as follows. In approximate surface coordinates, the equation of the plane is expressed as follows.

Note that since the xy plane is placed on the current approximation surface Ω, the equation for the current Ω is z = 0. The parameters are r = (a, b, c), which is three degrees of freedom. The observation model in the approximation surface coordinate system is as follows.

d _j is the difference in z-value between the point and the plane. In this case, for point PKj,

σKj is the point cloud accuracy data value of point PKj. Using this Jacobian matrix J and weighting matrix W, the parameter r is calculated as follows:

The surface represented by these parameters r = (a, b, c) becomes the approximate plane calculated using the weighted least squares method for PKj. If the weighted least squares method is used in the initial calculation of the approximate surface, it is expected that r = (0, 0, 0). However, if weighting is not applied in the initial calculation of the approximate surface, or due to numerical calculation errors, r may not be (0, 0, 0). In such cases, the approximate surface Ω is updated to the plane determined by Equation 5 using this r. This is done by setting Q to the point q + cez, and changing the normal vector n to the following:

If an update is made, the approximation surface coordinate system is set on the updated approximation surface Ω, and the Jacobian matrix J is calculated again. If the magnitude of r is small, the update of the approximation surface Ω may be omitted.

As described above, the surface approximation unit 8 approximates and reconstructs the plane Ω from the point cloud data PKj belonging to the Kth surface using the weighted least squares method.

We will now explain how the reliability S is calculated by the reliability calculation unit 9. In conjunction with the above weighted least squares method, the error matrix S of parameters r = (a, b, c) is obtained as follows:

This S is the variance-covariance matrix for the values of parameters a, b, and c. If the variances of parameters a, b, and c are expressed as σa2, σb2, and σc2, and the covariances as σab2, σbc2, and σca2, S is as follows. This error matrix S is the reliability.

The reliability S will be explained in detail below. a and b represent the inclination of z with respect to x and y, respectively, for the surface. Furthermore, c is the z coordinate of the intersection with the z axis. Here, the xy plane is placed on the approximation surface Ω. However, if surface 20, which is slightly different from Ω and expressed by equation 5, is considered as Ω', a and b will have small values on that surface. Let n' be the normal vector 40 of surface Ω', and 41 be the intersection with the z axis. As shown in Figure 7, if the angles of normal vector n' with the yz plane and zx plane are θ and φ, respectively, then we get the following.

Since these values are small on the surface Ω', they can be approximated as follows.

From the above, among the parameters shown in S above, a and b are equal to θ and φ respectively, and represent the orientation of the plane.

The other parameter c is the z coordinate of the intersection of plane Ω' and the z axis. This represents the position of the plane. From the above, the parameters (a, b, c) of the plane expressed in the approximate surface coordinate system represent the orientation and position of the plane. Therefore, the variance-covariance matrix shown in S above is an index that represents the variation in the orientation and position of the approximate surface Ω.

This S is assigned to the surface data as a reliability. The reliability S represents the degree of error in the orientation and position of the approximate surface Ω. The smaller the value of this element, the more accurately the approximate surface Ω has been obtained.

The probability density function f in abc space represented by this variance-covariance matrix S can be expressed as follows if it is a three-dimensional normal distribution.

The origin of the abc space represents the approximation surface Ω, and a point (a, b, c) corresponds to a surface that is displaced in orientation or position from Ω. If the range over which the probability density function spreads in the abc space is small, the accuracy of the approximation surface Ω is high. Figure 8 shows the spread of error due to σa, σb, and σc. Figure 9 shows an example of an isosurface 21 of f in the abc space. This is an ellipsoid.

The parameter error represented by S is equivalent to the parameter error of the Hough transform when a plane is represented by a three-dimensional Hough transform. This is explained below. In the Hough transform, a point (x, y, z) on a plane is represented by parameters Θ, Φ, and ρ as follows:

Note that the axes are not taken in the way that is usually used in explaining the Hough transform, but in a way that matches the approximate surface coordinate system described above. The parameters Θ and Φ are the direction in the zx plane and the elevation angle relative to the zx plane of the normal vector 19 of the approximate surface 16, respectively, and ρ is the distance from the origin of the plane. As shown in Figure 10, according to the Hough transform, one plane is represented by a point in the parameter space (Θ, Φ, ρ).

When the normal vector of a surface coincides with the z-axis or the angle it makes with the z-axis is small, as in surface 20 expressed in the approximate surface coordinate system, both Θ and Φ are small values. In this case, we have the following:

This has the same format as Equation 5 above, although the signs change. Θ corresponds to -a, Φ corresponds to -b, and ρ corresponds to c. Therefore, the above S can be converted into the variance-covariance matrix SH of the parameters when the approximation surface Ω is expressed in the approximation surface coordinate system using the Hough transformation. Therefore, SH can also be used as the reliability.

Next, the reliability assignment unit 10 will be described. The reliability assignment unit 10 generates data in which reliability S is assigned to the approximation surface Ω. This generates a surface model 22 having the approximation surface 16 and reliability S.

Next, the drawing by the drawing unit 11 will be described. The drawing unit 11 draws an image 23 of the target structure 15 by reflecting the reliability S of the approximation surface 16. For example, the maximum eigenvalue that is an index of the variation of the variance-covariance matrix S, which is the reliability S, is set as the evaluation value E of the reliability S, and the color or shading of the approximation surface 16 is changed depending on the magnitude of E. Alternatively, the sum or product of the eigenvalues, or one element of S may be set as the evaluation value E. In addition to the color of the drawing, the drawing form such as the type of hatching of the surface or the color and thickness of the outline may be changed. In addition, the isosurface 21 showing the degree of variation of the parameters as shown in FIG. 9 may be drawn so as to be displayed together. A drawn image 23 as shown in FIG. 11 is obtained by the surface 35 drawn by changing the shading or hatching depending on the evaluation value E. As a result, for example, a surface with low reliability can be visualized and represented as a surface that requires re-measurement.

When a polygon model is obtained, the approximation surface 16 is drawn by drawing the polygon model. Alternatively, when only the coordinate q and normal vector n of the reference point Q are obtained in the above operation, the drawing unit 11, for example, obtains the projection point of the measurement point 18 onto the approximation surface Ω, and draws the approximation surface 16 using that point. The approximation surface Ω may be drawn by representing the convex hull or circumscribing rectangle of the projection point on the approximation surface Ω. The approximation surface 16 may be converted into a polygon model by, for example, obtaining the intersection line between the approximation surfaces after modeling, and the polygon model may be drawn.

Alternatively, the point Pi may be drawn without drawing the approximation surface, and the color or shape of the point Pi may be changed depending on the evaluation value E of the approximation surface to which the point belongs. Alternatively, if the structure 15 is managed by a development diagram, such as a bridge or tunnel, the color or hatching of the corresponding surface on the development diagram may be changed depending on the evaluation value E of the approximation surface, and the image 23 may be drawn as a development diagram.

The image 23 drawn by the drawing unit 11 is displayed on the display device 12. The operation input unit 14 is also configured to allow the input device 13 to select the evaluation value E, change the drawing method, or cancel it.

Below, an example of the operation of the shape modeling device 1 of embodiment 1 of the present disclosure is explained using the flowchart of Figure 12.

In step ST1 of Figure 12, the surface approximation unit 8 reads out the point cloud data 5 and point cloud precision data 6 belonging to the same surface through the data input/output unit 2. The point cloud data 5 is assumed to have been classified in advance by surface, and here data corresponding to one surface is read out.

In step ST2, the surface approximation unit 8 uses the least squares method to find an approximation surface Ω that approximates the point cloud data 5.

In step ST3, the reliability calculation unit 9 uses the approximation surface Ω obtained by the surface approximation unit 8 to calculate the variance-covariance matrix S representing the reliability S, which is an index of the error of the approximation surface 16, using the above-mentioned method.

In step ST4, the reliability assignment unit 10 adds the reliability S calculated by the reliability calculation unit 9 to the approximate surface Ω to obtain the surface model data 7. Therefore, the data of the surface model includes the reliability S added to the data of the approximate surface Ω.

In step ST5, the reliability assignment unit 10 outputs the surface model data 7 to the data server 4 through the data input/output unit 2.

In step ST6, the surface approximation unit 8 determines whether there are any unprocessed surfaces, and if there are, returns to step ST1 and continues processing on those surfaces. If there are no unprocessed surfaces, the process proceeds to step ST7.

In step ST7, the drawing unit 11 reads the surface model data 7 through the data input/output unit 2.

In step ST8, the drawing unit 11 draws a drawing image 23 in which the surface drawing method is changed depending on the reliability S.

In step ST9, the drawing unit 11 displays the drawing image 23 on the display device 12.

In the shape modeling device 1 according to the first embodiment of the present disclosure described above, the surface model 22 is constructed by adding reliability S to the normal vector and reference point of the approximation surface, but in addition to this, information on the basis vectors indicating each coordinate axis of the approximation surface coordinate system may also be added to the surface model 22.

Although the orientation of the approximate surface was calculated as the angle of inclination with respect to the x and y coordinate systems of the approximate surface as described above, the method of taking the angle is not limited to this. For example, the x and y axes of the approximate surface coordinate system may be taken so that the covariance σab is 0, and the variance-covariance matrix S may be given together with the vectors indicating the axes.

Although the above variance-covariance matrix S is given as the reliability S, it may be configured to give the eigenvalues and eigenvectors of the variance-covariance matrix S. If the value of each covariance is smaller than the value of the variance, it may be configured to give the variance values σa, σb, and σc instead of the variance-covariance matrix S. It may also be configured to give the inverse matrix of the variance-covariance matrix S as the reliability S. Alternatively, it may be configured to give the reliability in a form that allows the variance-covariance matrix S to be derived. It may also be configured to give the variance-covariance matrix SH related to the parameters of the Hough transformation as described above.

Although the surface model is configured to be represented by normal vector 19 and reference point 17, other representation formats are acceptable as long as they are capable of representing a surface. Also, a convex hull of the projected points of point cloud 5 onto approximate surface 16 may be added to represent the edges and boundaries of the surface. Alternatively, the model may be configured to hold numbers and symbols representing the corresponding surfaces in design data such as drawings.

The device is also configured to read data on a data server 4 connected to a computer network 3 via the data input/output unit 2, and to store the data on the data server 4, but this is not limited to this. The destination of data input/output may be, for example, a storage device provided in the shape modeling device 1, or an external storage device or storage media that is detachable from the shape modeling device 1.

Although the configuration is such that reliability is assigned to planes, reliability may also be assigned to curved surfaces. For example, this can be achieved by approximating a curved surface with a plane to generate an approximation surface 16, or by defining curved surface coordinates on a curved surface. Furthermore, in the above embodiment, classification has already been performed as to which surface the point cloud data 5 belongs to. If a surface has already been determined in the classification process, in the above step ST2, the surface approximation unit 8 may be configured to use the determined surface to create the approximate surface Ω.

In addition, the surface approximation by the weighted least squares method performed by the surface approximation unit 8 is not limited to the above procedure of obtaining an initial approximation surface in the world coordinate system and then recalculating it in the approximation surface coordinate system. Furthermore, the drawing unit 11 may generate the image 23 as data in CAD format.

As described above, the configuration of the shape modeling device 1 makes it possible to obtain the reliability of the approximation surface 16 from the point cloud data 5 based on the error in its orientation and position, and to assign this reliability to the surface model 22. In addition, by assigning this reliability to the surface model 22, it is possible to clearly indicate the degree to which the approximation surface 16 from the point cloud data 5 has an error in its orientation and position.

The device is provided with a data input/output means for inputting point cloud data and outputting data of a surface model, a surface approximation means for determining a surface shape approximating the point cloud data, a reliability calculation means for calculating the reliability of the surface shape, and a reliability assignment means for assigning reliability to the surface model. In other words, the device is a shape modeling method for restoring the shape of an object from point cloud data obtained by measuring the object, and includes an input process for inputting point cloud data, a surface approximation process for determining a surface shape approximating the point cloud data, a reliability calculation process for calculating the reliability of the surface shape, a reliability assignment process for assigning reliability to the surface model, and a data output process for outputting data of the surface model. This makes it possible to obtain a shape modeling device and a shape modeling method using an approximate surface model to which a reliability indicating the likelihood has been assigned.

In addition, since reliability is the degree of error in the orientation and position of the surface model, by assigning a reliability representing the error in the orientation and position to an approximate surface (surface model) from point cloud data, it is possible to clearly indicate the degree of error the approximate surface has.

In addition, since the reliability is an error matrix in surface approximation using the least squares method of points belonging to the surface shape, by giving a reliability representing the error in the orientation and position of an approximated surface (surface model) from point cloud data as an error matrix using the least squares method, it is possible to clearly indicate the degree of error the approximated surface has.

Furthermore, the system is equipped with a drawing means that changes the drawing format of the surface represented by the surface model depending on the reliability, so that the degree of error in the approximate surface can be visualized and confirmed based on the reliability of the approximate surface (surface model) from the point cloud data.

Embodiment 2.
In the first embodiment of the present disclosure, the reliability indicating the degree of error of the surface approximating the point cloud data 5 is calculated, but the reliability may also be used to integrate surface models approximated from different point cloud data.

Figure 13 is a diagram showing an example of the configuration of a shape modeling device 1 according to embodiment 2 of the present disclosure. The integration unit 24, which is an integration means, integrates two surface models having reliability indicating the same surface into one likely surface model. The reliability integration unit 25, which is a reliability integration means, calculates the reliability of the integrated surface model. Note that in the figure, items with the same reference numerals are the same or equivalent, and this is common throughout the entire specification and all figures of the drawings. Furthermore, the forms of the components shown in the entire specification are merely examples and are not limited to these descriptions.

When measuring a structure 15, there are cases where measurements are taken from multiple positions using a fixed laser scanner, and cases where measurements are taken with an MMS on multiple travel routes. In this case, the same surface may be measured at multiple positions or during travel. The point cloud data will not overlap as is due to different measurement conditions and errors at each time. The same is true for surfaces approximated from the point cloud data. However, a single surface cannot be represented by multiple surface shapes (orientations and positions). For this reason, one reliable shape must be determined from the multiple shapes. In this case, the most probable surface is found and integrated based on each approximate surface 16 and its reliability, and this is used as the shape (surface model) representing that surface.

The operation of the integrating unit 24 for integrating surfaces will now be described. As shown in FIG. 14, for surface models representing the same surface obtained from the first point cloud data 5 and the second point cloud data 26, the respective approximate surfaces are taken as approximate surface 22Ω1 and approximate surface 27Ω2. Also, the respective reliability levels are taken as S1 and S2. A surface is estimated from Ω1 and Ω2 using the maximum likelihood method, and this estimated surface Ωe is taken as estimated surface 28 representing that surface.

Because Ω1 and Ω2 are measurements of the same surface, the angle between their normal vectors 19n1 and 29n2, both of magnitude 1, is sufficiently small. Here, the approximate surface coordinate system of Ω1 is used as the reference. In this case, the orthogonal basis of the approximate surface coordinate system of Ω1 is set to ex, ey, and ez as shown above. ez is set to match n1. If the parameters of Ω2 are expressed as (a2, b2, c2) in the approximate surface coordinate system of Ω1, it is as follows. This is shown in Figure 15.

In the approximate surface coordinate system of Ω1, Ω1 has (a, b, c) = (0, 0, 0). As above, Ω2 has (a, b, c) = (a2, b2, c2). From this, the maximum likelihood method can be used to find the estimated values (ae, be, ce) of the plane as follows:

The surface 28 represented by (ae, be, ce) is called Ωe. This is shown in Figure 16.

We will explain how to calculate the reference point 30qe and normal vector 31ne in the world coordinate system of the estimated surface Ωe. (ae, be, ce) are parameters in the approximate surface coordinate system of the approximate surface Ω1. If the normal vector of the surface of size 1 represented by (ae, be, ce) is ne, then this is as follows.

The reference point qe is expressed as follows. Ωe can be expressed by ne and qe.

Next, the operation of the reliability synthesis unit 25 will be described. The variance-covariance matrix Se of Ωe after synthesis is expressed as follows:

This Se is calculated by the reliability synthesis unit 25. This Se is added to Ωe as its reliability by the reliability assignment unit 10 to form a surface model.

An example of the operation of integrating surface models in the shape modeling device 1 according to the second embodiment of the present disclosure will be described below with reference to the flowchart in Fig. 17. It is assumed that the generation of the surface model from the first and second point cloud data has already been performed by the operation according to the first embodiment of the present disclosure.

In step ST201 of FIG. 17, the integration unit 24 reads out a set of surface models representing the same surface through the data input/output unit 2. Respective surface models are assigned reliability S1 and S2 for the approximate surfaces Ω1 and Ω2.

In step ST202, the integration unit 24 determines the normal vector ne and reference point qe of the surface Ωe that integrates Ω1 and Ω2 using the maximum likelihood method using the reliabilities S1 and S2 as described above.

In step ST203, the reliability synthesis unit 25 calculates the variance-covariance matrix Se representing the reliability of surface Ωe as described above.

In step ST204, the reliability assignment unit 10 assigns reliability Se to the surface Ωe to form a surface model.

In step ST205, the confidence assignment unit 10 outputs and stores the surface model representing the surface Ωe through the data input/output unit 2. This completes the operation of generating a surface model of the surface Ωe by integrating Ω1 and Ω2 into one.

In the operation of the shape modeling device 1 of the second embodiment of the present disclosure, the operation of integrating two sets of surface models into one surface model 28 has been described. In the case where a first surface model set is composed of a plurality of surface models obtained from the first point cloud data 5, and a second surface model set is composed of a plurality of surface models obtained from the second point cloud data 26, the above operation is performed on the surfaces commonly included in the first and second surface model sets. A surface model included in only one of the first and second surface model sets may be added to the surface model obtained in this way to form an integrated surface model set. In this way, the common parts of the first and second surface model sets can be integrated with a high degree of certainty to obtain a union. Furthermore, the contours of each surface model may be generated or modified according to the integrated surface model set by finding the intersection between adjacent surface models.

In addition, if the reliability evaluation value of the integrated surface model is lower than a predetermined value, it may be configured to determine that the surface requires re-measurement.

In addition, if the difference between Ω1 and Ω2 is large, the integration unit 24 may be configured to determine this and not perform integration. This is done, for example, as follows. The parameters of Ω2 expressed in the approximate surface coordinate system of Ω1 are (a2, b2, c2), which represent the difference from Ω1 to Ω2. This is expressed as the Mahalanobis distance D using the variance-covariance matrix, which is the reliability of Ω1.

If the value of D is greater than a predetermined value, for example, 2 or 3, it is determined that Ω2 is too different from Ω1, such as being measured incorrectly due to factors other than random error, and is not integrated with Ω1. It is also possible to switch Ω1 and Ω2 and make the same determination.

Furthermore, following the operation of integrating face models in the shape modeling device 1 of embodiment 2 of the present disclosure described above, the drawing unit may be configured to draw the faces by changing the color or hatching depending on the reliability after integration, the Mahalanobis distance between Ω1 and Ω2, and whether or not integration of Ω1 and Ω2 has been performed.

Note that the operation of the second embodiment of the present disclosure shown in the flowchart of FIG. 17 does not require point cloud data. Therefore, if a surface model is obtained, the operation of integrating the surface models can be executed.
According to such a configuration of the geometric modeling device 1, it is possible to integrate surface models obtained by different measurements into a probable surface model according to the reliability thereof, without using the original point cloud data.

In this way, by integrating the two surface models using their reliability, the surface can be represented by a reliable surface model.

As described above, the device includes a data input/output means for inputting point cloud data and outputting data of a surface model, a surface approximation means for determining a surface shape that approximates the point cloud data, a reliability calculation means for calculating the reliability of the surface shape, and a reliability assignment means for assigning reliability to the surface model. In other words, the device is a shape modeling method for restoring the shape of an object from point cloud data obtained by measuring the object, and includes an input process for inputting point cloud data, a surface approximation process for determining a surface shape that approximates the point cloud data, a reliability calculation process for calculating the reliability of the surface shape, a reliability assignment process for assigning reliability to the surface model, and a data output process for outputting data of the surface model. This makes it possible to obtain a shape modeling device and a shape modeling method using an approximate surface model to which a reliability indicating the likelihood has been assigned.

The device also includes an integration means for integrating different point cloud data representing substantially the same surface (common surface, identical surface) into one likely surface by weighting the reliability of the data. In other words, the data input/output means inputs a first surface model and a second surface model representing the same surface and having reliability, and includes an integration means for obtaining one surface model representing the target surface from the first surface model and the second surface model, and the integration means obtains the shape of the surface model to be integrated by weighting the reliability of each parameter representing the surface shape of the first surface model and the second surface model. In this way, the first and second surface models representing the same surface obtained by different measurements are integrated by the reliability representing the degree of error, thereby making it possible to represent the surface with a more likely surface model.

Furthermore, the reliability calculation means includes a synthesis means for calculating the reliability of the first and second point cloud data representing substantially the same surface (common surface, identical surface) as an error matrix in surface approximation by the least squares method, and setting the inverse matrix, which is the sum of the first inverse matrix of the error matrix by the first point cloud data and the second inverse matrix of the error matrix by the second point cloud data, as the reliability of the surface model. In other words, the reliability calculation means includes a reliability synthesis means for generating the reliability of the integrated surface model from the reliability of the first surface model and the reliability of the second surface model, and the reliability calculated by the reliability calculation means is an error matrix in surface approximation by the least squares method of the points belonging to the surface, and the reliability synthesis means synthesizes the reliability of the integrated surface model as an inverse matrix of the sum of the inverse matrix of the first error matrix and the inverse matrix of the second error matrix, which are the reliability of the first surface model and the second surface model, respectively. This makes it possible to synthesize and assign reliability to a surface model obtained by integrating the first and second surface models representing the same surface obtained by different measurements using a reliability that represents the degree of error.

Embodiment 3.
In the second embodiment of the present disclosure, two surface models representing the same surface are integrated into a likely surface model using the reliability of the surface models. The reliability may be used to convert the set of surface models obtained from the second point cloud data 26 so as to match the set of surface models obtained from the first point cloud data 5.

Figure 18 is a diagram showing an example configuration of a shape modeling device 1 according to embodiment 3 of the present disclosure. For two sets of surface models obtained by different measurements of substantially the same structure, a transformation calculation unit 32, which is a transformation calculation means, calculates a transformation that aligns the position of a second set of surface models with a first set of surface models that serves as a reference. Next, a model transformation unit 33, which is a model transformation means, transforms the second set of surface models so that they overlap with the first set of surface models that serves as a reference. Furthermore, a point cloud correction unit 34, which is a point cloud correction means, transforms the second point cloud data 26 that is the source of the second surface models to match the transformation of the surface models.

Here, a set of surface models refers to the surface shape of a structure being made up of multiple surface models, and the surface models for the structure being considered as a set of predetermined units. As explained in embodiment 1, multiple surface models can be obtained from point cloud data of each surface for substantially the same structure, and a reliability level can be assigned to each surface model. The first and second surface model sets refer to surface models calculated from point cloud data obtained by different measurements for substantially the same structure, and differentiated by measurement unit, with reliability levels assigned to each surface model.

Hereinafter, the approximate surface 16 represented by the m-th surface model in the first surface model set A1 serving as a reference is defined as Ω1m. For example, it is a surface model that is approximately restored from the first point cloud data 5 measured by a camera, a laser scanner, or by traveling measurement. Alternatively, it may be a surface model generated from design data such as design drawings. In addition, it may be a surface model that serves as a reference for managing the target structure 15, such as a surface model measured by a more precise device or a surface model measured at the time of completion of the structure 15. The surface 36 represented by the surface model in the second surface model set A2 corresponding to the element Ω1m of A1 is defined as Ω2m. This surface model of A2 is a surface model composed of a surface model that approximates the second point cloud data 26 measured by a camera, a laser scanner, or by traveling measurement.

As shown in Figure 19, surface model set 43 for A2 and surface model set 42 for A1 do not normally overlap. In this figure, surface model set 42 simulates a situation in which measurements are taken with a measuring instrument such as a laser scanner placed in the middle right of structure 15, which is a bridge, in the illustration. Surface model set 43 simulates a situation in which measurements are taken with a measuring instrument placed in the front left of structure 15 in the illustration. Surfaces 47 that are hidden by the measuring instrument and therefore cannot be modeled because no measurement points can be obtained, are shown hatched. Surface model set 42 is also shown with a dashed line to distinguish it.

Because the two do not overlap, it is not possible to use A2's surface model set 43 to complement A1's surface model set 42 or to detect parts that have been deformed from A1. The overall position of A2's surface model set must be adjusted so that it matches A1's surface model set. For example, if there is an error in the installation position of the camera or laser scanner, the point cloud data will have systematic errors. This is corrected by a rigid transformation consisting of a three-dimensional rotation transformation and translation. Therefore, alignment can be performed by comparing the surface model sets A1 and A2 and determining the rigid transformation that best matches the positions of the two.

Point cloud data 5 also contains random errors. For this reason, each Ω2m of A2 also contains errors, and for example, two orthogonal surfaces in A1 are calculated slightly out of alignment in A2. Therefore, a single transformation cannot precisely match all the surface models of A1 and A2. In such cases, the reliability is used to perform alignment so that the surface models overlap most appropriately, that is, so that the surface models with higher reliability are prioritized.

In the following, the reliability of Ω1m and Ω2m are S1m and S2m, respectively. Also, suppose that there are M faces that are included in both A1 and A2, where m ranges from 1 to M.

We will explain how to find the transformation. First, using Figure 20, consider matching the surface model Ω2m to Ω1m. In the approximate surface coordinate system of Ω1m, Ω2m is expressed as parameters (a2m, b2m, c2m). If the orthogonal basis of the approximate surface coordinates of Ω1m is ex, ey, ez, the normal vector of Ω2m is n2, and the reference points 17 of Ω1m and Ω2m are q1m and q2m, respectively, we get the following. Note that the subscript m is omitted in n2, ex, ey, and ez.

Here, as mentioned above, a2m can be regarded as the rotation angle of normal vector n2 around the y-axis. Similarly, b2m can be regarded as the rotation angle of normal vector n2 around the x-axis. In other words, to make Ω2m coincide with Ω1m, Ω2m is rotated by a2m around the y-axis of the approximate surface coordinate system of Ω1m, rotated by -b2m around the x-axis, and translated by -c2m along the z-axis. Note that the sign of the rotation differs depending on the coordinate axes. If the unit vectors on the x-axis and y-axis of the approximate surface coordinate system of Ω1m are ex = (uX, uY, uZ) and ey = (vX, vY, vZ), respectively, then if the absolute values of a2m and b2m are both small, the rotation transformation Rm for superimposing Ω2m on Ω1m can be expressed as follows.

First, the -b2m rotation around ex is expressed by Rodrigues' rotation formula as follows.

Similarly, the rotation of a2m around the y-axis is expressed as follows:

Therefore, Rm becomes:

Furthermore, the parallel translation tm is expressed as follows:

This transformation allows Ω2m to be made to match Ω1m. This rotation transformation Rm and translation tm are transformations that match the mth surface model. Due to measurement errors of each surface, this transformation will differ for each surface model. A uniform transformation for the whole is determined by a weighted average based on this reliability.

As described above, when a surface is estimated using the maximum likelihood method from the surface models obtained from S1m and S2m, the reliability Sm, which is the variance-covariance matrix of the estimated surface, is as follows.

Using this, we set the weight wm as follows:

The numerator of wm is the square of the Mahalanobis distance of κm, and the denominator is the square of the magnitude of κm. wm is a coefficient that converts the square of the magnitude of κm into the square of the Mahalanobis distance. By weighting in this way, it is possible to prioritize surface models with small variance and high accuracy when aligning.

The rigid transformation G of the second surface model set A2 is expressed by a rotation matrix R and a translation vector t. The rotation matrix R is determined as follows, assuming that the rotation angle of each Rm is small.

The translation vector t is determined as follows, where I is the identity matrix.

This gives us G as follows:

Here, Γ is the coordinate value of the world coordinate system. As shown in FIG. 21, the normal vector of Ω2m and the reference point 17 included in A2 are transformed by transformation G as follows:

n2m* is the transformed normal vector 44, and q2m* is the transformed reference point 45. The transformed ones are indicated with an *. In Figure 21, the transformed surface 38 is also indicated as Ω2m*.

This transformation G is performed on all surface models of A2, including surface models 48 that do not have a surface corresponding to A1, as shown in Figure 22. In this figure, surface models 49 in the surface model set of A1 that are not modeled in A1's surface model set 42 and are complemented by A2's surface model 48 are shown in a thick solid line. This allows A2 to be matched to A1 with priority given to highly reliable surface models, and surfaces that cannot be surface modeled in A1 can be complemented by the A2 surface model. As a result, surface 47, which cannot be modeled in A1's surface model set 42 in Figure 22 and is represented by hatching, is no longer there.

In some cases, A1 is a surface model generated from design data such as design drawings, or a surface model that has long been used as a standard, and so on. In such cases, for example, wm is calculated by using S2m instead of Sm in the above formula 32.

Furthermore, the point cloud correction unit 34 can convert the second point cloud data 26 in the same way to match the conversion of this surface model, and correct it so that it matches the surface model set A1. For example, when it becomes necessary to refer to the original point cloud data, no inconsistency with the converted surface model occurs. If the coordinate value of the measurement point 39 in the second point cloud data 26 is p, this conversion converts it into point 46 represented by G(p) below.

This is shown in Figure 23. Note that, for simplicity, the measurement points 39 and the converted points 46 are shown only partially on the surface model.

An example of the operation of the shape modeling device 1 according to the third embodiment of the present disclosure will be described below with reference to the flowchart in Fig. 24. It is assumed that the reliability of the surface models of the first and second surface model sets has already been obtained by the operation according to the first embodiment of the present disclosure.

In step ST301 of Figure 24, the conversion calculation unit 32 reads out information on two sets of surface models, A1 and A2, through the data input/output unit 2. Reliability is assigned to surface model 22, which is an element of the A1 set, and surface model 27, which is an element of the A2 set.

In step ST302, the transformation calculation unit 32 calculates the rotation transformation Rm and the translation vector tm and their weight wm for superimposing Ω2m on Ω1m for the mth pair of corresponding surface models, as described above.

In step ST303, the transformation calculation unit 32 calculates a rotation matrix R and a translation vector t representing the transformation G to be applied to all elements of A2 from each Rm, tm, and wm.

In step ST304, the model transformation unit 33 transforms the normal vector 29 and reference point 37 of each surface model of element A2 using transformation G.

In step ST305, the model conversion unit 33 outputs the surface model data 7 of the converted element A2 to the data server 4 via the data input/output unit 2.

Furthermore, in step ST306, the point cloud correction unit 34 reads out the second point cloud data 26 via the data input/output unit 2.

In step ST307, the point cloud correction unit 34 applies the transformation G to each point of the second point cloud data 26 as described above to convert its coordinate values.

In step ST308, the point cloud correction unit 34 outputs and stores this converted second point cloud data 26 via the data input/output unit 2.

In the shape modeling device 1 according to the third embodiment of the present disclosure, each surface model 27 of the element A2 is converted once in steps ST302 to ST304. However, steps ST302 to ST304 may be repeated multiple times. The repetition may be performed a predetermined number of times, or until the rotation angle of the rotation matrix R or the absolute value of the translation vector t becomes smaller than a predetermined value.

Furthermore, following the operation of converting the surface model of the shape modeling device 1 according to the third embodiment of the present disclosure, the drawing unit may be configured to draw the surface by changing the color or hatching of the surface using the weight based on the reliability after the integration as an evaluation value. This makes it easy to see which surface was given the emphasis in the conversion.

In addition, if the reliability evaluation value of the integrated surface model is lower than a predetermined value, it may be configured to determine that the surface requires re-measurement.

In addition, if only a surface model set is obtained and there is no original point cloud data, or if correction of the coordinates of the original point cloud data is not required, steps ST306 to ST308 may be omitted. In this way, if a surface model set is obtained, alignment between surface model sets can be performed even if there is no point cloud data.

With this configuration of the shape modeling device 1, it is possible to convert a pair of surface models obtained by different measurements so that one is accurately superimposed on the other, utilizing their reliability, without using the original point cloud data.

By aligning two pairs of surface models in this way using their reliability, it is possible to transform and align the surface models with the highest degree of certainty.

In the above-mentioned first to third embodiments of the present disclosure, the reliability of the surface model is configured to use the error matrix in the approximation by the least squares method. The reliability is not limited to this as long as it indicates the nature of the error of the point cloud data used to approximate the surface model. For example, the number of point cloud data used to approximate the surface model may be configured to be the reliability. It is said that the standard error of the estimate is inversely proportional to the square root of the number of samples. Therefore, the more the number of points of the point cloud data, which is the number of samples, the smaller the error of the approximate surface 16, and this can be used as the reliability indicating the accuracy of the surface. The reliability may also be given as the square root of the number of points of the point cloud data used to approximate the surface model or its reciprocal. Alternatively, a value obtained by dividing the reciprocal of the square root of the number of points by the accuracy of the image processing or laser scanner when measuring the point cloud data may be given. It may also be given as a 3×3 matrix whose diagonal components are the reciprocal of the square root of the number of points, or the reciprocal of the square root of the number of points multiplied by the accuracy of the point cloud data. In this way, the formulas shown in the second and third embodiments can be applied as they are, and the respective operations can be executed.

In addition, when the point density is constant, the number of points is proportional to the area, so the area of the approximation surface 16 may be configured to be set as the reliability. When the number of points is the same, the larger the surface area, the more accurate the surface orientation, so in this case too, the reliability can be expressed by the area. The area of the approximation surface 16 is, for example, the area of the convex hull of the projection of the point cloud data onto the approximation surface 16. Alternatively, it is the area of the range limited by the intersection lines between multiple obtained approximation surfaces 16. In addition, the reliability may be defined from the average nearest neighbor distance value of the nearest neighbor distance method or the K function value of the K function method as a value representing the distribution state of the point cloud data on the approximation surface 16.

In addition, in the above-described second embodiment of the present disclosure, the surface model Ωe is obtained by integrating the surface model Ω1 and the surface model Ω2, but the point cloud correction unit 34 used in the third embodiment of the present disclosure may be provided to correct the coordinate values of the point cloud. The points of the first point cloud data 5 belonging to the surface model Ω1 are corrected by a conversion that converts Ω1 to Ωe, and the points of the second point cloud data 26 belonging to the surface model Ω2 are corrected by a conversion that converts Ω2 to Ωe.

As described above, the device includes a data input/output means for inputting point cloud data and outputting data of a surface model, a surface approximation means for determining a surface shape that approximates the point cloud data, a reliability calculation means for calculating the reliability of the surface shape, and a reliability assignment means for assigning reliability to the surface model. In other words, the device is a shape modeling method for restoring the shape of an object from point cloud data obtained by measuring the object, and includes an input process for inputting point cloud data, a surface approximation process for determining a surface shape that approximates the point cloud data, a reliability calculation process for calculating the reliability of the surface shape, a reliability assignment process for assigning reliability to the surface model, and a data output process for outputting data of the surface model. This makes it possible to obtain a shape modeling device and a shape modeling method using an approximate surface model to which a reliability indicating the likelihood has been assigned.

The data input/output means inputs different first and second point cloud data for substantially the same structure, the surface approximation means obtains first and second surface model sets from the first and second point cloud data, the reliability calculation means calculates reliability for each of the first and second surface models, and includes a conversion calculation means for calculating a weighted average from the reliability for each second surface model corresponding to the first surface model and converting the second surface model set to match the position of the first surface model set. In other words, the data input/output means inputs a first surface model set and a second surface model set having reliability modeled from the first and second point cloud data including a common surface,
The apparatus includes a transformation calculation means for calculating a transformation of the second surface model set to align the position of the second surface model set with the first surface model set, and a model transformation means for transforming the second surface model set, the transformation calculation means determining weights from the reliability of each of the first surface model and the corresponding second surface model, and calculating a transformation for the second surface model set by a weighted average of transformations that transform the corresponding second surface model into the first surface model according to the weights.Therefore, by aligning the first surface model set and the second surface model set using the reliability that indicates the reliability, it is possible to transform and align the surface models most likely.

Furthermore, a point cloud coordinate conversion means is provided for converting the coordinate values of the point cloud data used to model the second surface model set. This makes it possible to align the first and second point cloud data by converting the coordinate values of the point cloud data so that they match the alignment performed using the reliability indicating the degree of error for the first and second surface model sets.

1 Shape modeling device, 2 Data input/output unit, 3 Computer network, 4 Data server, 5 Point cloud data, 6 Point cloud accuracy data, 7 Surface model data, 8 Surface approximation unit, 9 Reliability calculation unit, 10 Reliability assignment unit, 11 Drawing unit, 12 Display device, 13 Input device, 14 Operation input unit, 15 Structure, 16 Approximation surface, 17 Reference point, 18 Measurement point, 19 Normal vector, 23 Image, 24 Integration unit, 25 Reliability synthesis unit, 32 Conversion calculation unit, 33 Model conversion unit, 34 Point cloud correction unit, 40 Normal vector.

Claims (8)

A data input/output means for inputting point cloud data and outputting surface model data;
a surface approximation means for obtaining a surface shape approximating the point cloud data;
A reliability calculation means for calculating a reliability of the surface shape;
a reliability assigning means for assigning the reliability to the surface model;
and an integration means for integrating different surface models representing substantially the same surface into one surface model based on a shape obtained by a weighted average using the reliability for each parameter representing the shape of each surface. A data input/output means for inputting point cloud data and outputting surface model data;
a surface approximation means for obtaining a surface shape approximating the point cloud data;
a reliability calculation means for calculating a reliability of the surface shape by using the first and second point cloud data, which are different and which represent substantially the same surface, as an error matrix in surface approximation by a least squares method;
a reliability synthesis means for determining an inverse matrix, which is a sum of a first inverse matrix of an error matrix based on the first point cloud data and a second inverse matrix of an error matrix based on the second point cloud data, as the reliability of the surface model;
and a reliability assigning means for assigning the reliability to the surface model. a data input/output means for inputting data of first and second surface models which are different for substantially the same structure and have respective reliability levels assigned thereto, and outputting data of the converted second surface model ;
and a transformation calculation means for calculating a weighted average from the reliability of each second surface model corresponding to the first surface model, and for transforming the second surface model set so as to align the position of the first surface model set with the position of the first surface model set. 4. The geometric modeling device according to claim 3, further comprising point cloud coordinate conversion means for converting coordinate values of the second point cloud data used in modeling the second set of surface models. 1. A shape modeling method for restoring a shape of an object from point cloud data obtained by measuring the object, comprising:
a data input step of inputting the point cloud data;
a surface approximation step of obtaining a surface shape that approximates the point cloud data;
a reliability calculation step of calculating a reliability of the surface shape;
a reliability assignment step of assigning the reliability to the surface model;
a merging step of merging different surface models representing substantially the same surface into one surface model based on a shape obtained by a weighted average using the reliability for each parameter representing the shape of each surface;
and outputting data of the surface model. 1. A shape modeling method for restoring a shape of an object from point cloud data obtained by measuring the object, comprising:
a data input step of inputting the point cloud data;
a surface approximation step of obtaining a surface shape that approximates the point cloud data;
a reliability calculation step of calculating the reliability of the surface shape by using the first and second point cloud data, which are different and represent substantially the same surface, as an error matrix in surface approximation by a least squares method;
a reliability synthesis step of determining an inverse matrix, which is a sum of a first inverse matrix of an error matrix based on the first point cloud data and a second inverse matrix of an error matrix based on the second point cloud data, as the reliability of the surface model;
a reliability assignment step of assigning the reliability to the surface model;
and outputting data of the surface model. A shape modeling method for restoring a shape of an object , comprising the steps of:
a data input step of inputting data of first and second surface models different from each other for substantially the same structure, the data of the first and second surface models being assigned respective reliability levels ;
a conversion calculation step of calculating a weighted average of the reliability of each of the second surface models corresponding to the first surface model, and converting the second surface model set so as to align the position of the first surface model set with the position of the first surface model set;
and a data output step of outputting data of the converted second surface model. 8. The geometric modeling device according to claim 7, further comprising a point cloud coordinate conversion step of converting coordinate values of second point cloud data used in modeling the second set of surface models.

Images