Three-dimensional reconstruction device

Abstract

A three-dimensional reconstruction device (10) according to the present disclosure is provided with: an estimation unit (11) that estimates, in a panoramic image of an internal space of a structure, the positions of boundaries between the wall surfaces of the internal space and each of the ceiling and the floor surfaces, and the probabilities of existence of corners in the internal space; a calculation unit (12) that calculates a weight for each pair (pi, qi) of point pi and point qi on the basis of the distribution shape of the probability of existence of a corner at point pi constituting a source point cloud obtained by converting the coordinates of the estimated boundaries into three-dimensional coordinates, and the distribution shape of the probability of existence of a corner at point qi corresponding to point pi and constituting a target point cloud indicating boundaries that are created from a drawing of the internal space of the structure; and a reconstruction unit (13) that calculates a transformation matrix for converting the source point cloud into the target point cloud on the basis of the weighted distance of each pair (pi, qi), and then converts the coordinates of pi constituting the source point cloud using the calculated transformation matrix to perform three-dimensional reconstruction of the internal space.

Description

3D reconstruction device

[0001] The present disclosure relates to a three-dimensional reconstruction device.

[0002] Layout estimation is an image processing task that estimates the three-dimensional configuration of a structure's interior space (three-dimensional reconstruction) from a panoramic image of the interior space. A typical layout estimation method involves detecting the walls of the structure's interior space and estimating the shape of the interior space from the wall detection results (see Non-Patent Documents 1-3). Recently, among these methods, a method that detects walls by estimating the boundaries between the walls and the ceiling and floor, as well as the positions of corners where the walls intersect, from a panoramic image has become widely used (see Non-Patent Documents 2 and 3). After detecting the walls, three-dimensional data of the interior space can be obtained (the three-dimensional configuration of the interior space can be reconstructed) by performing post-processing.

[0003] As a post-processing step, if the shape of the internal space of the structure is known from drawings or the like, the panoramic image can be texture-mapped onto the three-dimensional data created from the drawings by registering (aligning) the three-dimensional data obtained by layout estimation with the three-dimensional data created from the drawings.

[0004] One registration method is to assign weights to pairs of corresponding points between point groups that contain noise or occlusions (see Non-Patent Document 4). One such method assigns uniform weights to pairs of corresponding points. Another method approximates the distribution of neighboring points with an ellipsoid for each pair of corresponding points, and assigns weights based on the volume difference of the ellipsoid.

[0005] H. Izadinia et al., IM2CAD, CVPR, 2017.Sun, C., et al. HorizonNet: Learning Room Layout With 1D Representation and Pano Stretch Data Augmentation, CVPR, 2019.FE. Wang et al., LED2Net: Monocular 360 Layout Estimation via Differentiable Depth Rendering, CVPR, 2021.A. Gressin et al. al., Towards 3D lidar point cloud registration improvement using optimal neighborhood knowledge. ISPRS Journal of Photogrammetry and Remote Sensing(79):240-251, 2013.

[0006] The method of assigning uniform weights to pairs of corresponding points has the problem of lacking accuracy, while the method of approximating the distribution of neighboring points with an ellipsoid for each pair of corresponding points and assigning weights based on the volume difference of the ellipsoid has the problem of high computational costs.

[0007] The present disclosure has been made in consideration of the above-described problems, and aims to provide a three-dimensional reconstruction device that can reduce the calculation load while ensuring the accuracy of three-dimensional reconstruction.

[0008] In order to solve the above-described problems, a three-dimensional reconstruction device according to the present disclosure is a three-dimensional reconstruction device that performs three-dimensional reconstruction of an internal space of a structure from a panoramic image of the internal space captured by the image capture, the three-dimensional reconstruction device including an estimation unit that estimates, in the panoramic image, positions of boundaries between wall surfaces of the internal space and each of a ceiling and a floor surface, and a probability of existence of a corner where wall surfaces of the internal space intersect with each other, and a point cloud obtained by converting coordinates of the estimated boundaries into three-dimensional coordinates is defined as a Source point cloud, and a three-dimensional point cloud that indicates the boundaries created from a drawing of the internal space of the structure is defined as a Target point cloud, and iThe distribution shape of the existence probability of the corner at the point p i The point q that constitutes the target point group corresponding to i Based on the distribution shape of the existence probability of the corner at the point p i and said point q i Pairs of (p i ,q i ) and a calculation unit for calculating the weight of each of the pairs (p i ,q i ) Based on the respective weighted distances, a transformation matrix is ​​calculated to transform the source point group into the target point group, and points p constituting the source point group are calculated using the calculated transformation matrix. i and a reconstruction unit that converts the coordinates of the internal space and performs three-dimensional reconstruction of the internal space.

[0009] According to the three-dimensional reconstruction device according to the present disclosure, it is possible to reduce the calculation load while ensuring the accuracy of three-dimensional reconstruction.

[0010] Fig. 1 is a diagram illustrating an example of the configuration of a three-dimensional reconstruction device according to an embodiment of the present disclosure. Fig. 2 is a diagram illustrating an example of a panoramic image. Fig. 3 is a diagram illustrating an example of an estimation result of an estimation unit shown in Fig. 1. Fig. 4 is a diagram for explaining the operation of a calculation unit shown in Fig. 1. Fig. 5 is a flowchart illustrating an example of the operation of the three-dimensional reconstruction device shown in Fig. 1. Fig. 6 is a diagram illustrating an example of the configuration of a computer that functions as the three-dimensional reconstruction device according to the present disclosure.

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0012] 1 is a diagram illustrating an example configuration of a 3D reconstruction device 10 according to an embodiment of the present disclosure. The 3D reconstruction device 10 according to the present disclosure performs 3D reconstruction of an internal space (indoor space) of a structure from a panoramic image captured of the internal space, as shown in FIG. 2A . A panoramic image is, for example, an image obtained by capturing the internal space of a structure using a camera or the like at angles ranging from −90° to +90° in the up-down direction along the vertical direction and at angles ranging from 0° to 360° in the left-right direction along the horizontal direction.

[0013] As shown in FIG. 1, a three-dimensional reconstruction device 10 according to this embodiment includes an estimation unit 11, a calculation unit 12, and a reconstruction unit 13.

[0014] A panoramic image of the interior space of a structure is input to the estimation unit 11. The estimation unit 11 estimates, in the input panoramic image, the positions of boundaries between the wall surfaces of the interior space and the ceiling and floor, and the positions of corners where the wall surfaces of the interior space intersect with each other.

[0015] Methods for estimating the positions of wall boundaries and corners include methods using neural networks, such as HorizonNet, described in Non-Patent Document 2, or LED2-Net, described in Non-Patent Document 3. FIG. 2B illustrates an example of the results of estimation of the positions of wall boundaries and corners using HorizonNet. As shown in FIG. 2B , HorizonNet estimates the positions of wall boundaries as the coordinates of the wall boundaries in the height direction (the coordinates of the boundaries between the wall and ceiling and the coordinates of the boundaries between the wall and floor) at each pixel position in the width direction of the panoramic image. Also, as shown in FIG. 2B , HorizonNet estimates the positions of corners as the probability of corner existence at each pixel position in the width direction of the panoramic image. Therefore, if the number of pixels in the width direction of the panoramic image is 1024, HorizonNet outputs three values ​​for each pixel (the image coordinate in the height direction of the boundary between the wall and ceiling, the image coordinate in the height direction of the boundary between the wall and floor, and the probability of corner existence). The estimation unit 11 estimates the positions of the boundaries and corners of the wall surface (probability of the corners existing) using, for example, a neural network such as the above-mentioned HorizonNet or LED2-Net.

[0016] The estimation unit 11 outputs the estimation result of the boundary position of the wall surface to the calculation unit 12 and the reconstruction unit 13. The estimation unit 11 also outputs the estimation result of the corner position (probability of corner existence) to the calculation unit 12.

[0017] The calculation unit 12 receives the estimated results of the boundary positions of the wall surfaces and the estimated results of the corner positions from the estimation unit 11. The calculation unit 12 also receives a drawing of the interior space of the structure. The drawing may be, for example, a completion drawing of the target structure.

[0018] The calculation unit 12 generates a point group (source point group) by converting the coordinates of the boundary estimated by the estimation unit 11 into three-dimensional coordinates. The generation of the source point group will be described below.

[0019] The long side direction (width direction) of the panoramic image is the u-axis direction, and the short side direction (height direction) of the panoramic image is the v-axis direction. The direction along the wall surface of the interior space of the structure is the Y-direction, the direction perpendicular to the Y-direction is the X-direction, and the direction perpendicular to the X-direction and Y-direction is the Z-direction. In other words, the X-direction and Z-direction are directions along the floor and ceiling of the interior space of the structure.

[0020] The image coordinates of a point constituting the boundary between the wall surface and the floor surface are (u, v), and the image coordinates of a point constituting the boundary between the ceiling and the wall surface are (u, u). The height from the floor surface to the position of the camera capturing the panoramic image is h. Note that h is assumed to be known. The calculation unit 12 calculates the three-dimensional coordinates (x, y, z) of the point constituting the boundary between the wall surface and the floor surface and the three-dimensional coordinates (x, y, z) of the coordinate point constituting the boundary between the ceiling and the wall surface based on the following equations. The calculation unit 12 calculates the three-dimensional coordinates for each pixel constituting the boundary extending in the width direction of the panoramic image, for example. floor_d=|h / sin(vf)| yf=floor_d*sin(vf) xf=floor_d*cos(vf)*sin(uf) zf=-floor_d*cos(vf)*cos(uf) ceil_h=|h / tan(vf)*tan(vc)| ceil_d=|ceil_h / sin(vc)|yc=ceil_d*sin(vc) xc=ceil_d*cos(vc)*sin(uc) zc=-ceil_d*cos(vc)*sin(uc)

[0021] The calculation unit 12 also generates a three-dimensional point cloud (target point cloud) indicating the boundary of the wall surface from the input drawing of the internal space of the structure. Then, the calculation unit 12 calculates the three-dimensional point cloud p i The distribution shape of the probability of corner existence at point p i The points q that make up the target point group corresponding to i Based on the distribution shape of the probability of corner existence at point p i and point q i Pairs of (p i ,q i ) Each weight w i Hereinafter, the calculation unit 12 calculates the pair (p i ,q i ) weight w i The calculation of is explained below.

[0022] The calculation unit 12 calculates the points p i The closest point among the points that make up the target point group is the point p i The point q corresponding to i The calculation unit 12 sets the point p i The distribution shape of the probability of corner existence at point q i The difference between the distribution shape of the corner existence probability in i ,q i ) weight w i Calculate.

[0023] In the above-mentioned HorizonNet, the probability of corner existence is C ｜Ｘ’－Ｘ｜ where x' is the pixel coordinate in the width direction of the point of interest. X is the true value or estimated value of the pixel coordinate of the corner closest to x'. C is a constant that satisfies 0<C<1, and is, for example, a value of about 0.90-0.99. The calculation unit 12 calculates the pixel coordinate of the point p in accordance with the above definition. i and point q i The distribution shape of the probability of corner existence is calculated.

[0024] Next, the calculation unit 12 calculates the point p i The distribution shape (estimated value) of the probability of corner existence at point q iSpecifically, the calculation unit 12 normalizes the corner existence probability in an arbitrary region around the corner, and calculates the difference between the distribution shape (true value) of the corner existence probability at point p using KL (Kullback-Leibler) divergence, JS (Jensen-Shannon) divergence, or Wasserstein distance. i The distribution shape (estimated value) of the probability of corner existence at point q i The distribution shape (true value) of the probability of corner existence is compared with the distribution shape (true value) of the probability of corner existence in the following. An example of calculation using KL divergence will be described below.

[0025] The calculation unit 12 calculates the point p i The distribution shape of the probability of corner existence in and point q i The distribution shape of the probability of corner existence in p (x),f q In equation (1), ΔX is an arbitrary constant.

[0026]

[0027] Next, the calculation unit 12 calculates the normalized point p i The distribution shape of the probability of corner existence in p (x) and the normalized point q i The distribution shape of the probability of corner existence in q (x) and evaluates the difference with (x). Specifically, the calculation unit 12 calculates the KL divergence D using the following equation (2).

[0028]

[0029] The calculation unit 12 calculates the inverse of the calculated KL divergence D, i.e., 1 / D, as a pair (p i ,q i ) weight w i It is decided that:

[0030] Pair (p i ,q i ) weight w iThe calculation of only takes into account the uncertainty of the probability of the corner existing near the corner. Therefore, the calculation unit 12 calculates, for example, the number of pairs (p i ,q i ) weight w i For the pair at the corner (p i ,q i ) weight w i can be determined by linear interpolation of

[0031] In this way, the calculation unit 12 defines a point cloud obtained by converting the coordinates of the estimated boundary into three-dimensional coordinates as a Source point cloud, and a three-dimensional point cloud indicating the boundary created from a drawing of the internal space of the structure as a Target point cloud. i The distribution shape of the probability of corner existence at point p i The points q that make up the target point group corresponding to i Based on the distribution shape of the probability of corner existence at point p i and point q i Pairs of (p i ,q i ) Each weight w i Calculate.

[0032] The point p as mentioned above i , point q i Based on the distribution shape of the probability of corner existence in i ,q i ) weight w i The method for calculating requires less computational effort than, for example, the method for approximating the distribution of neighboring points with an ellipsoid for each pair of corresponding points described above and assigning a weight based on the volume difference of the ellipsoids, etc. Therefore, the 3D reconstruction device 10 according to this embodiment can reduce the computational load.

[0033] Referring back to FIG. 1, the calculation unit 12 calculates the points p i For each point p i For each pair containing (p i ,q i ) weight w i Calculate the pair (p i ,q i ) and pairs (pi ,q i ) weight w i is output to the reconstruction unit 13.

[0034] The reconstruction unit 13 reconstructs the pair (p i ,q i ) based on the weighted distances of the source point group and the target point group, a transformation matrix T(R, t) is calculated. R is a rotation matrix, and t is a translation vector. The reconstruction unit 13 calculates the transformation matrix T(R, t) of the pair (p i ,q i ) calculates a transformation matrix T that minimizes the sum of the squares of the weighted distances. The reconstruction unit 13 calculates the transformation matrix T(R, t) by solving equation (3), for example, using singular value decomposition.

[0035]

[0036] After calculating the transformation matrix T(R, t), the reconstruction unit 13 calculates the transformation matrix T(R, t) by using the calculated transformation matrix T(R, t) to reconstruct the points p i The coordinates of the object are transformed to obtain three-dimensional data of the internal space (reconstruct the internal space).

[0037] Points p that compose the Source point group i The distribution shape of the probability of corner existence at point p i The points q that make up the target point group corresponding to i The weight w according to the difference between the distribution shape of the probability of corner existence in i By calculating the weight w i Then, the calculated pair (p i ,q i ) Each weight w i Calculate the transformation matrix T(R, t) using the above, and use the calculated transformation matrix T(R, t) to transform the points p iBy converting the coordinates of the vectors and outputting the three-dimensional data of the internal space, the accuracy of the three-dimensional reconstruction can be ensured. Furthermore, as described above, the calculation of the weights wi based on the differences in the distribution shapes of the corner existence probabilities requires a relatively small amount of calculation. Therefore, the three-dimensional reconstruction device 10 according to this embodiment can reduce the calculation load while ensuring the accuracy of the three-dimensional reconstruction.

[0038] Next, the operation of the three-dimensional reconstruction apparatus 10 according to this embodiment will be described with reference to Fig. 4. Fig. 4 is a flowchart showing an example of the operation of the three-dimensional reconstruction apparatus 10 according to this embodiment, and is a diagram for explaining a three-dimensional reconstruction method executed by the three-dimensional reconstruction apparatus 10.

[0039] The estimation unit 11 estimates the positions of the boundaries between the walls of the interior space of the structure and the ceiling and floor in the input panoramic image, as well as the probability of the existence of corners where the walls of the interior space intersect (step S11).

[0040] The calculation unit 12 converts the coordinates of the estimated boundary into three-dimensional coordinates to obtain points p constituting the source point group. i The distribution shape of the existence probability of the corner at point p i The target points q that form the boundary of the internal space of the structure are created from the drawing. i Based on the distribution shape of the probability of corner existence at point p i and point q i Pairs of (p i ,q i ) is calculated (step S12).

[0041] The reconstruction unit 13 reconstructs the pair (p i ,q i ) Based on the weighted distances, a transformation matrix T(R, t) is calculated to transform the source point cloud into the target point cloud, and the points p that make up the source point cloud are calculated using the calculated transformation matrix T(R, t). i The coordinates of the object are transformed to reconstruct the three-dimensional interior space of the structure (step S13).

[0042] As described above, the 3D reconstruction device 10 according to this embodiment includes an estimation unit 11, a calculation unit 12, and a reconstruction unit 13. The estimation unit 11 estimates the positions of boundaries between the wall surfaces of the internal space and the ceiling and floor surfaces, respectively, in a panoramic image of the internal space of a structure, and the probability of existence of corners where the wall surfaces of the internal space intersect with each other. The calculation unit 12 converts the coordinates of the estimated boundaries into three-dimensional coordinates to obtain points p constituting the Source point cloud. i The distribution shape of the probability of corner existence at point p i The points q that make up the three-dimensional target point cloud that indicates the boundary created from the drawing of the internal space of the structure correspond to i Based on the distribution shape of the probability of corner existence at point p i and point q i Pairs of (p i ,q i ) and calculates the weight of each pair (p i ,q i ) Based on the weighted distances, a transformation matrix T(R, t) is calculated to transform the source point cloud into the target point cloud, and the points p that make up the source point cloud are calculated using the calculated transformation matrix T(R, t). i The coordinates are transformed to perform a three-dimensional reconstruction of the internal space.

[0043] Points p that compose the Source point group i The distribution shape of the probability of corner existence at point p i The points q that make up the target point group corresponding to i The weight w according to the difference between the distribution shape of the probability of corner existence in i By calculating the weight w i Furthermore, the weight wi based on the difference in the distribution shape of the corner existence probability can be calculated with a relatively small amount of calculation. Therefore, the 3D reconstruction device 10 according to this embodiment can reduce the calculation load while ensuring the accuracy of 3D reconstruction.

[0044] The reduction in the computational load of 3D reconstruction enables faster texture mapping, which maps panoramic images onto the interior space of actual structures. As a result, it is possible to improve the processing speed of modeling civil engineering structures after identifying areas of deterioration, as well as the processing speed of modeling furniture arrangements and wallpapering rooms.

[0045] The above-described three-dimensional reconstruction apparatus 10 can be realized by a computer 20 shown in FIG. 5. A program for causing the computer 20 to function as the three-dimensional reconstruction apparatus 10 may be provided. The program may be stored in a storage medium or provided via a network. FIG. 5 is a block diagram showing a schematic configuration of the computer 20 functioning as the three-dimensional reconstruction apparatus 10. The computer 20 may be a general-purpose computer, a dedicated computer, a workstation, a PC (Personal Computer), an electronic notepad, or the like. The program instructions may be program code, code segments, or the like for executing necessary tasks.

[0046] 5, the computer 20 includes a processor 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, a storage 24, an input unit 25, a display unit 26, and a communication interface (I / F) 27. Each component is communicably connected to one another via a bus 29. The processor 21 is specifically a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), a SoC (System on a Chip), or the like, and may be configured with multiple processors of the same or different types.

[0047] The processor 21 is a control unit that controls each component and performs various arithmetic processing. That is, the processor 21 reads a program from the ROM 22 or the storage 24 and executes the program using the RAM 23 as a work area. The processor 21 controls each component and performs various arithmetic processing in accordance with the program stored in the ROM 22 or the storage 24. In this embodiment, the ROM 22 or the storage 24 stores a program for operating the computer 20 as the three-dimensional reconstruction device 10 according to the present disclosure. The processor 21 reads and executes the program to realize the estimation unit 11, calculation unit 12, and reconstruction unit 13 of the three-dimensional reconstruction device 10.

[0048] The program may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), a USB (Universal Serial Bus) memory, etc. The program may also be provided in a form downloaded from an external device via a network.

[0049] The ROM 22 stores various programs and various data. The RAM 23 temporarily stores programs or data as a working area. The storage 24 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including the operating system and various data.

[0050] The input unit 25 includes a pointing device such as a mouse and a keyboard, and is used to input various types of information.

[0051] The display unit 26 is, for example, a liquid crystal display, and displays various types of information. The display unit 26 may employ a touch panel system and function as the input unit 25. The display unit 26 displays, for example, the results of three-dimensional reconstruction of the internal space of the structure by the reconstruction unit 13.

[0052] The communication interface 27 is an interface for communicating with other devices, for example, an interface for a LAN.

[0053] The following additional notes are provided regarding the above-described embodiments.

[0054] [Supplementary Item 1] A three-dimensional reconstruction device that performs three-dimensional reconstruction of an internal space of a structure from a panoramic image of the internal space, comprising a control unit, wherein the control unit estimates, in the panoramic image, positions of boundaries between wall surfaces of the internal space and each of a ceiling and a floor surface, and a probability of existence of a corner where wall surfaces of the internal space intersect, and where a point cloud obtained by converting coordinates of the estimated boundaries into three-dimensional coordinates is defined as a Source point cloud, and a three-dimensional point cloud indicating the boundaries created from a drawing of the internal space of the structure is defined as a Target point cloud, and points p constituting the Source point cloud are i The distribution shape of the existence probability of the corner at the point p i The point q that constitutes the target point group corresponding to i Based on the distribution shape of the existence probability of the corner at the point p i and said point q i Pairs of (p i ,q i ) and calculate the weight of each of the pairs (p i ,q i ) Based on the respective weighted distances, a transformation matrix is ​​calculated to transform the source point group into the target point group, and points p constituting the source point group are calculated using the calculated transformation matrix. i and converting the coordinates of the object to perform a three-dimensional reconstruction of the internal space.

[0055] [Supplementary Item 2] In the three-dimensional reconstruction device according to Supplementary Item 1, the control unit i ,q i[Supplementary Item 1] A three-dimensional reconstruction method executed by a three-dimensional reconstruction device that performs three-dimensional reconstruction of an internal space of a structure from a panoramic image of the internal space, comprising: estimating, in the panoramic image, positions of boundaries between walls of the internal space and each of a ceiling and a floor, and a probability of existence of a corner where walls of the internal space intersect; estimating, as a source point cloud, a point cloud obtained by converting coordinates of the estimated boundaries into three-dimensional coordinates; and estimating, as a target point cloud, a three-dimensional point cloud that indicates the boundaries created from a drawing of the internal space of the structure; i The distribution shape of the existence probability of the corner at the point p i The point q that constitutes the target point group corresponding to i Based on the distribution shape of the existence probability of the corner at the point p i and said point q i Pairs of (p i ,q i ) A three-dimensional reconstruction method that calculates the weights for each.

[0056] [Supplementary Item 4] A non-transitory storage medium storing a program executable by a computer, the non-transitory storage medium storing a program that causes the computer to operate as the three-dimensional reconstruction device described in Supplementary Item 1 or 2.

[0057] Although the above-described embodiments have been described as typical examples, it will be apparent to those skilled in the art that many modifications and substitutions can be made within the spirit and scope of the present disclosure. Therefore, the present invention should not be interpreted as being limited by the above-described embodiments, and various modifications and alterations are possible without departing from the scope of the claims. For example, multiple building blocks shown in the block diagrams of the embodiments can be combined into one, or one building block can be divided.

[0058] 10 Three-dimensional reconstruction device 11 Estimation section 12 Calculation section 13 Reconstruction section

Claims

1. A three-dimensional reconstruction device for three-dimensionally reconstructing an internal space of a structure from a panoramic image of the internal space, the device comprising: an estimation unit that estimates, in the panoramic image, the positions of boundaries between the walls of the internal space and each of the ceiling and floor, and the probability of existence of a corner where the walls of the internal space intersect; and a point cloud obtained by converting the coordinates of the estimated boundaries into three-dimensional coordinates that is a source point cloud, and a three-dimensional point cloud that indicates the boundaries created from a drawing of the internal space of the structure that is a target point cloud, where points p that constitute the source point cloud are i The distribution shape of the existence probability of the corner at the point p i The point q that constitutes the target point group corresponding to i Based on the distribution shape of the existence probability of the corner at the point p i and said point q i Pairs of (p i ,q i a calculation unit for calculating a weight for each of the pairs (p i ,q i ) Based on the respective weighted distances, a transformation matrix is ​​calculated to transform the source point group into the target point group, and points p constituting the source point group are calculated using the calculated transformation matrix. i and a reconstruction unit that converts the coordinates of the object and performs three-dimensional reconstruction of the internal space.

2. The three-dimensional reconstruction device according to claim 1, wherein the reconstruction unit i ,q i ) A three-dimensional reconstruction device that calculates the transformation matrix that minimizes the sum of the squares of the respective weighted distances.

Images