Three-dimensional model production device

Abstract

This three-dimensional model production device comprises: an acquisition unit that acquires point group data of a plurality of point groups representing three-dimensional coordinates for points on the surface of a structure; a candidate point group extraction unit that extracts, from the plurality of point groups and on the basis of the reflection intensity, a candidate point group that is a candidate for a point group constituting a linear object; a linear object point group extraction unit that, for each of a plurality of groups obtained by grouping a plurality of clusters generated by clustering the extracted candidate point groups, detects a straight line on the basis of the coordinates of the candidate point group, generates a catenary curve using the candidate point group constituting the detected straight line, and extracts a linear object point group constituting the same linear object, from the catenary curve candidate point groups generated for each group; and a production unit that produces a model of the linear object using the linear object point group.

Description

3D model creation device

[0001] The disclosed technology relates to a three-dimensional model creation device.

[0002] Conventionally, a technology (Mobile Mapping System: MMS) has been developed that uses an on-board 3D laser scanner to create a 3D model of an outdoor structure. For example, Patent Document 1 discloses a technology that creates a point cloud and a scan line in a space where no 3D point cloud exists, and then creates a 3D model, thereby achieving good results in recall and precision even when the point cloud is sparse (when the vehicle speed is high).

[0003] The cable model, which is a linear object model, is created using the point density of a three-dimensional point cloud and training data.

[0004] Furthermore, for example, Patent Document 2 discloses a technology in which the spacing between scan lines is determined by the running speed and scan frequency, and the measurement time is also known, making it possible to estimate where the next point will be, and then connecting those points to create a cable model.

[0005] JP 2017-156179 A JP 2015-078849 A

[0006] By the way, we would like to model thin linear objects such as cables and thin rods in real time using point clouds acquired by a real-time 3D laser scanner. Since the MMS can acquire point clouds while moving along the object, it can acquire point clouds evenly and at fairly regular intervals within the measurement range, so by connecting points with narrow intervals, a straight line model can be created. We also implemented machine learning using training data to create highly accurate models.

[0007] However, conventional technologies require post-processing of point clouds for calculations and require enormous calculation times, making real-time calculations difficult. Furthermore, real-time 3D laser scanners have a sparser point density than MMSs, so simply connecting densely spaced points does not result in a straight line. Therefore, even if the 3D point cloud has a sparse point density, it is desirable to extract target point clouds from point cloud data and create a 3D model of a thin linear object in real time.

[0008] The disclosed technology has been made in consideration of the above points, and aims to provide a three-dimensional model creation device that can create three-dimensional models of thin linear objects with high precision and in real time, even if the three-dimensional point cloud representing the three-dimensional coordinates is sparse.

[0009] A first aspect of the present disclosure is a three-dimensional model creation device comprising: an acquisition unit that acquires point cloud data of a plurality of point clouds representing the three-dimensional coordinates of points on the surface of a structure; a candidate point cloud extraction unit that extracts candidate point clouds from the plurality of point clouds based on reflection intensity as candidates for a point cloud constituting a linear object; a linear object point cloud extraction unit that detects straight lines based on the coordinates of the candidate point clouds for each of a plurality of groups into which a plurality of clusters are generated by clustering the extracted candidate point clouds, creates catenary curves using the candidate point clouds constituting the detected straight lines, and extracts linear object point clouds constituting the same linear object from the candidate point clouds of the catenary curves created for each group; and a creation unit that creates a model of the linear object using the linear object point clouds.

[0010] According to the disclosed technology, even if the three-dimensional point cloud representing the three-dimensional coordinates is sparse, a three-dimensional model of a thin linear object can be created with high accuracy in real time.

[0011] FIG. 1 is a schematic diagram showing an example of the configuration of a three-dimensional model creation system. FIG. 2 is a block diagram showing an example of the hardware configuration of a three-dimensional model creation device. FIG. 3 is a block diagram showing an example of the functional configuration of a three-dimensional model creation device. FIG. 4 is a diagram for explaining processing performed in a linear object point cloud extraction unit. FIG. 5 is a diagram for explaining processing performed in a linear object point cloud extraction unit. FIG. 6 is a diagram for explaining processing performed in a linear object point cloud extraction unit. FIG. 7 is a diagram for explaining processing performed in a linear object point cloud extraction unit. FIG. 8 is a diagram for explaining processing performed in a linear object point cloud extraction unit. FIG. 9 is a flowchart showing an example of a three-dimensional model creation process. FIG. 10 is a flowchart showing details of processing executed in the three-dimensional model creation process.

[0012] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that the same or equivalent components and parts in each drawing are given the same reference numerals. Also, the dimensional proportions in the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.

[0013] First, an example of the configuration of a 3D model creation system 1 according to the technology of this embodiment will be described. As shown in Fig. 1, the 3D model creation system 1 includes a 3D model creation device 10 and a 3D laser scanner 20. Hereinafter, as shown in Fig. 1, with the position of the 3D laser scanner 20 as the reference, the east-west direction is defined as the X-axis, the north-south direction as the Y-axis, and altitude as the Z-axis.

[0014] The 3D laser scanner 20 sequentially measures point cloud data of points on the surface of a structure such as a utility pole, a wall, or a cable, and outputs the point cloud data to the 3D model creation device 10. The point cloud data is data that represents the three-dimensional (X, Y, Z) coordinates of points on the surface of the structure.

[0015] The 3D laser scanner 20 may be, for example, a LiDAR (Light Detection and Ranging) sensor. A LiDAR sensor irradiates an object with laser light and measures the distance between the object based on the time it takes for the reflected light to be received or the phase change between the emitted light and the received light. The LiDAR sensor measures the 3D coordinates of the object by arranging multiple laser light emitters vertically and horizontally scanning (rotating) each emitter. The LiDAR sensor can also measure the reflection intensity at each coordinate on the object surface using the ratio of the amount of light between the emitted light and the received light. Therefore, in this embodiment, the point cloud data output from the 3D laser scanner 20 to the 3D model creation device 10 also includes data on the reflection intensity of the point cloud.

[0016] The three-dimensional model creation device 10 is a device that creates a three-dimensional model of a linear object from point cloud data output from a three-dimensional laser scanner 20 .

[0017] The hardware configuration of the 3D model creation device 10 of this embodiment will be described with reference to Fig. 2. Fig. 2 is a block diagram showing an example of the hardware configuration of the 3D model creation device 10. As shown in Fig. 2, the 3D model creation device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 16, a display unit 17, and a communication I / F (Interface) 18. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0018] The CPU 11 is a central processing unit that executes various programs, such as a 3D model creation program 15, stored in the storage 14, and controls each component. That is, the CPU 11 reads the programs from the ROM 12 or the storage 14, and executes the programs using the RAM 13 as a work area. The CPU 11 controls each of the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or the storage 14.

[0019] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores various programs including an operating system and a 3D model creation program 15, as well as various data. The 3D model creation program 15 may be a single program or a group of programs configured by multiple programs or modules. The storage 14 may also store point cloud data acquired from the 3D laser scanner 20.

[0020] The input unit 16 includes a pointing device such as a mouse and a keyboard, and is used to input various information to the device itself. The display unit 17 is, for example, a liquid crystal display, and displays various information. Note that a touch panel system in which the input unit 16 and the display unit 17 are integrated may also be adopted.

[0021] The communication I / F 18 is an interface for the device to communicate with other external devices such as the 3D laser scanner 20. For this communication, for example, a wired communication standard such as Ethernet (registered trademark) or FDDI (Fiber Distributed Data Interface), or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark) is used.

[0022] The three-dimensional model creation device 10 of this embodiment is implemented by a general-purpose computer such as a server computer or a personal computer (PC).

[0023] Next, the functional configuration of the 3D model creation device 10 of this embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the functional configuration of the 3D model creation device 10. Each functional configuration is realized by the CPU 11 reading out the 3D model creation program 15 stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it.

[0024] 3, the 3D model creation device 10 includes processing units including an acquisition unit 30, a candidate point cloud extraction unit 32, a linear object point cloud extraction unit 34, and a creation unit 36. Each processing unit may be configured as an apparatus.

[0025] The acquisition unit 30 acquires point cloud data of a plurality of point clouds representing the three-dimensional coordinates of points on the surface of a structure. Note that the acquisition method by which the acquisition unit 30 acquires the point cloud data is not limited. For example, the acquisition unit 30 may acquire the point cloud data from the 3D laser scanner 20, or, if the point cloud data is stored in the storage 14, may acquire the point cloud data from the storage 14. Furthermore, the acquisition unit 30 may acquire a three-dimensional point cloud for one frame measured by the 3D laser scanner 20 as the point cloud data for one run of the 3D model creation process, or may acquire a superposition of three-dimensional point clouds for multiple frames as the point cloud data for one run of the 3D model creation process.

[0026] The acquisition unit 30 outputs the point cloud data acquired as described above to the candidate point cloud extraction unit 32 .

[0027] The candidate point cloud extractor 32 extracts candidate points from the plurality of point clouds as candidates for the point cloud constituting the linear object based on the reflection intensity. The reflection intensity is represented by, for example, the ratio of the amount of emitted light to the amount of received light measured by the 3D laser scanner 20 (LiDAR sensor), and is included in the point cloud data as described above.

[0028] For example, if the linear object is a cable, the reflection intensity is relatively low. Considering that the reflection intensity of a cable is thus lower than that of other objects, the candidate point group extraction unit 32 extracts, from among the multiple point groups, those with low reflection intensities as candidate point groups that are candidates for the point group constituting the linear object. As a specific example, the candidate point group extraction unit 32 of this embodiment sets the threshold to "5" and extracts point groups with reflection intensities of 5 or less (reflection intensity≦5) as candidate point groups that are candidates for the point group constituting the linear object. The candidate point group extraction unit 32 outputs the extracted candidate point groups to the linear object point group extraction unit 34.

[0029] The linear object point cloud extraction unit 34 clusters the extracted candidate point cloud to generate multiple clusters. As shown in FIG. 4 , the linear object point cloud extraction unit 34 clusters the candidate point cloud 50 based on voxel size or point density, and divides it into multiple clusters 52. By clustering, several point clusters are created from the extracted candidate point cloud 50. By clustering into multiple clusters 52 in this way, the number of candidate point clouds to be calculated can be reduced. Furthermore, by using the center of gravity of each cluster 52, it is possible to reduce the deviation of points at the acquired locations. This improves the speed and accuracy of subsequent processing.

[0030] The linear object point cloud extraction unit 34 also groups the generated clusters to generate groups. As shown in Fig. 5A, the linear object point cloud extraction unit 34 generates groups of candidate points 50 that are larger than the cluster 52 through grouping. Note that a hierarchical clustering method such as Ward's method or single-link method may be used as the grouping method.

[0031] The clustered candidate point group 50 may be grouped multiple times. In other words, larger groups may be generated by repeatedly grouping the clustered candidate point group 50. FIG. 5A shows a case where grouping is performed twice, generating two large groups, group A and group B. Specifically, the first grouping generates three groups, groups a1 to a3, and two groups, groups b1 and b2. The second grouping generates two groups, group A including groups a1 to a3, and group B including groups b1 and b2. When grouping is performed multiple times in this manner, the same grouping method may be combined, or different grouping methods may be combined. For example, when grouping is performed twice, the Ward method may be used the first time and the single-link method may be used the second time. Alternatively, for example, the single-link method may be used both times.

[0032] The linear object point cloud extraction unit 34 of this embodiment calculates the center of gravity of each cluster 52 and the center of gravity of each group (groups a1 to a3, b1, and b2 in FIG. 5A), and calculates the distance between clusters and groups using this position information. If the distance is within a threshold, the clusters are considered to be in the same group, and larger groups are formed.

[0033] When multiple linear objects exist at positions of different heights (positions in the Z direction), multiple groups are generated at positions of different heights, as shown in Fig. 5B. In the example shown in Fig. 5B, groups C and D are generated at positions of different heights from groups A and B.

[0034] Furthermore, even if the detection of lines to be performed in the subsequent calculations is performed before grouping, the cluster spacing is too large, so only lines for each cluster are detected, and it is not possible to detect lines that follow the desired cable, so this type of grouping is performed.

[0035] Furthermore, the linear object point cloud extraction unit 34 detects straight lines for each of the multiple groups based on the two-dimensional coordinates of the candidate point cloud in the horizontal plane (X-Y plane). As shown in FIG. 6A, the linear object point cloud extraction unit 34 detects straight lines 54a and 54b in the horizontal X-Y plane. According to the algorithm used in this embodiment, only one straight line is detected from one group. For example, when groups A and B are generated as shown in FIG. 5A, straight line 54a is detected from group A, and straight line 54b is detected from group B, as shown in FIG. 6A. For example, when groups A to D are generated as shown in FIG. 5B, straight line 54a is detected from group A, straight line 54b is detected from group B, straight line 54c is detected from group C, and straight line 54d is detected from group D, as shown in FIG. 6B. As shown in FIG. 6B, when multiple linear objects do not overlap in the height direction (Z direction), multiple straight lines (54a and 54b, and 54c and 54d) are detected for each of the multiple linear objects. In addition, if multiple linear objects are located at a position where they overlap in the vertical direction, the lines corresponding to each linear object will overlap in the X-Y plane, but by detecting lines for each group, line detection will be performed for each linear object, so this does not cause a problem.

[0036] 5A and 5B, the candidate point group 50 is arranged in the height direction (Z direction), and therefore when viewed from the XY plane as shown in Figures 6A and 6B, each cluster 52 can be processed as one large block of candidate point group 50. This reduces the amount of processing.

[0037] The method by which the linear object point cloud extraction unit 34 detects the straight line 54 from the plurality of candidate point clouds 50 is not limited, and for example, a known Hough transform or the like may be used.

[0038] The linear object point cloud extraction unit 34 creates a catenary curve for each detected straight line 54 using the candidate point cloud 50 that constitutes the straight line 54. The linear object point cloud extraction unit 34 extracts, as linear object point clouds that constitute the same linear object, the candidate point cloud 50 of the catenary curve generated in one group and the candidate point cloud 50 of the catenary curve generated in another group that is determined to match the curve obtained by extending the catenary curve generated in the first group. In the example shown in FIG. 7A , the curve obtained by extending the catenary curve 60a generated in group A matches the catenary curve 60b generated in group B, so the linear object point cloud extraction unit 34 extracts the candidate point cloud 50 of the catenary curve 60a and the candidate point cloud 50 of the catenary curve 60b as linear object point clouds.

[0039] 7B , the curve obtained by extending catenary curve 60a created in group A matches catenary curve 60b created in group B, so linear object point cloud extraction unit 34 extracts candidate point cloud 50 for catenary curve 60a and candidate point cloud 50 for catenary curve 60b as linear object point clouds for the same linear object (first linear object point cloud). Furthermore, the curve obtained by extending catenary curve 60c created in group C matches catenary curve 60d created in group D, so linear object point cloud extraction unit 34 extracts candidate point cloud 50 for catenary curve 60c and candidate point cloud 50 for catenary curve 60d as linear object point clouds for the same linear object (second linear object point cloud) that are different from the first linear object point cloud.

[0040] The catenary curve may be generated using, for example, RANSAC curve approximation, which is a learning method that estimates the parameters of a mathematical model to be calculated by excluding as much of the influence of outliers as possible from data that includes outliers.

[0041] Furthermore, when creating a catenary curve, using areas with high point density, such as utility poles or wall surfaces, may result in poor accuracy of the resulting 3D model. Therefore, object detection may be performed in advance using a different method to exclude surrounding candidate point groups 50, or the candidate point groups may be manually excluded from the target. Furthermore, if a predetermined number or more of points with reflection intensities equal to or greater than a threshold exist within a specified range of the candidate point group 50, the candidate point group 50 may also be excluded. By excluding the candidate point group 50 using such a method, the accuracy of the 3D model may be ensured.

[0042] The creation unit 36 ​​creates a catenary curve that serves as a model of the linear object using the linear object point cloud extracted by the linear object point cloud extraction unit 34. For example, in the example shown in Fig. 7A, the creation unit 36 ​​creates one catenary curve using the linear object point cloud. Also, for example, in the example shown in Fig. 7B, the creation unit 36 ​​creates a total of two catenary curves, one using the first linear object point cloud and one using the second linear object point cloud.

[0043] In this way, the linear object point cloud extraction unit 34 and the creation unit 36 ​​perform fitting on the multiple catenary curves generated using all of the straight lines 54 to unify the reference model. A specific example will be described below. The linear object point cloud extraction unit 34 of this embodiment first sets, as the reference model, a catenary curve created using the candidate point cloud 50 that constitutes the longest straight line 54 out of the multiple detected straight lines 54.

[0044] Then, it is determined whether the candidate points 56 of the other straight lines 54 are on the reference model, and if so, they are extracted as candidate points constituting the same linear object. This determination is repeated until there are no candidate points on the reference model.

[0045] Note that, for example, the following two methods can be used to determine whether the candidate point group 56 is on the catenary curve of the reference model. The first method is to create a catenary curve for each group, and then determine whether there are other catenary curves on the extension line of the reference model. The second method is to determine whether the distance between the candidate point group 56 and the catenary curve when the catenary curve of the reference model is extended is within a threshold. Note that the length to which the catenary curve is extended may be given a value in advance, or a method may be used in which the catenary curve is extended to a landmark object (such as a utility pole or other cable).

[0046] The creation unit 36 ​​creates a catenary curve again using the extracted linear object point cloud, and sets it as a linear object model. The creation unit 36 ​​outputs the created linear object model. Note that the destination to which the creation unit 36 ​​outputs the linear object model is not limited. For example, the creation unit 36 ​​may output the linear object model to the display unit 17, which displays the linear object model. Alternatively, the creation unit 36 ​​may output the linear object model to the storage 14, which stores the linear object model.

[0047] Next, the operation of the three-dimensional model creating device 10 will be described.

[0048] Fig. 8 shows a flowchart of an example of a 3D model creation process executed by the 3D model creation device 10 of this embodiment. The 3D model creation device 10 executes the 3D model creation process shown in Fig. 8 by executing a 3D model creation program 15 stored in the storage 14. The 3D model creation process shown in Fig. 8 is executed at a predetermined timing, such as when an execution instruction is received from a user.

[0049] In step S100, the acquisition unit 30 acquires point cloud data of the three-dimensional point cloud obtained by the three-dimensional laser scanner 20 as described above.

[0050] In the next step S102, the candidate point group extracting unit 32 extracts the candidate point group 50 based on the reflection intensity of the point group, as described above.

[0051] In the next step S104, the linear object point cloud extraction unit 34 clusters the candidate point cloud 50 extracted in step S102 above using voxels of any size or point density, as described above, and divides it into a plurality of clusters 52.

[0052] In the next step S106, the linear object point cloud extraction unit 34 groups the multiple clusters 52 divided in step S104 based on the distance of the center of gravity of each cluster 52, as described above, to generate multiple groups.

[0053] In the next step S108, the linear object point cloud extraction unit 34 extracts a linear object point cloud from the candidate point cloud, as described above, and in the next step S110, the creation unit 36 ​​creates a catenary curve that serves as a linear object model, using the linear object point cloud extracted in step S108, as described above. Note that in this embodiment, steps S108 and S110 are performed as a series of processes, and will be described in detail later.

[0054] In the next step S112, the creating unit 36 ​​outputs the created linear object model to an arbitrary output destination, as described above. When the process of step S112 ends, the 3D model creating process shown in FIG. 8 ends.

[0055] Further, details of the processing of steps S108 and S110 of the three-dimensional model creation processing shown in Fig. 8 will be described with reference to Fig. 9. That is, in this embodiment, the processing of steps S108 and S110 of the three-dimensional model creation processing is performed as the processing of the flowchart shown in Fig. 9.

[0056] In step S200 of FIG. 9, the linear object point cloud extraction unit 34 detects a plurality of straight lines 54 for each of the plurality of groups based on the two-dimensional coordinates (XY plane coordinates) of the candidate point cloud 50.

[0057] In the next step S202, the linear object point cloud extraction unit 34 creates a catenary curve for each straight line 54 detected in step S200. For example, as shown in Fig. 6, in group A, two straight lines 54 are detected, so two catenary curves are created, and in group B, two straight lines 54 are detected, so two catenary curves are created.

[0058] In the next step S204, the linear object point cloud extraction unit 34 sets the catenary curve corresponding to the longest straight line 54 among the straight lines 54 detected in step S200 as the reference model.

[0059] In the next step S206, the linear object point cloud extraction unit 34 determines whether there are any other catenary curves that fit on the reference model. For example, if the reference model belongs to group A, it determines whether there are any catenary curves that are determined to match the curve obtained by extending the reference model among the catenary curves that belong to group B. Note that "match" here does not necessarily mean a perfect match, but rather it is sufficient if the conditions for being considered to match, including errors, are met.

[0060] If there is another catenary curve that fits on the reference model, the determination in step S206 is affirmative, and the process proceeds to step S208. In step S208, the linear object point cloud extraction unit 34 updates the reference model. For example, the linear object point cloud extraction unit 34 may generate a new catenary curve using the candidate point cloud 50 of the reference model and the candidate point cloud 50 of the catenary curve that fits on the reference model, and use the generated catenary curve as the updated reference model. Also, for example, instead of the current reference model, a catenary curve that fits on the current reference model may be used as the updated reference model. When the process in step S208 is completed, the process returns to step S206.

[0061] On the other hand, if there are no other catenary curves that fit on the reference model, the determination in step S206 is negative, and the process proceeds to step S210. In step S210, the creation unit 36 ​​creates a catenary curve again using the reference model candidate point group 50. The process of this step corresponds to the process of step S110 in the 3D model creation process of FIG.

[0062] In the next step S212, the linear object point cloud extraction unit 34 determines whether or not there are any unused catenary curves among the catenary curves created in step S202. If there are any unused catenary curves, the determination in step S212 becomes positive, and the process proceeds to step S214.

[0063] In step S214, the linear object point group extraction unit 34 excludes the used catenary curves from the multiple catenary curves (catenary curve group) created in step S202, and then returns to step S204 to repeat the processing of steps S204 to S212 for the new catenary curve group.

[0064] On the other hand, if there is no unused catenary curve, the determination in step S212 is negative, and the processing of the flowchart shown in FIG. 9 ends.

[0065] As described above, the 3D model creation device 10 of this embodiment includes an acquisition unit 30, a candidate point cloud extraction unit 32, a linear object point cloud extraction unit 34, and a creation unit 36. The acquisition unit 30 acquires point cloud data of a plurality of point clouds representing the three-dimensional coordinates of points on the surface of a structure. The candidate point cloud extraction unit 32 extracts candidate point clouds 50 that are candidates for point clouds constituting linear objects from the plurality of point clouds based on reflection intensity. The linear object point cloud extraction unit 34 clusters the extracted candidate point clouds 50 to generate a plurality of clusters 52, and for each of the plurality of groups, detects a straight line 54 based on the two-dimensional coordinates in the horizontal plane of the candidate point cloud 50, and creates a catenary curve using the candidate point cloud 50 that constitutes the detected straight line 54. The linear object point cloud extraction unit 34 also extracts linear object point clouds that constitute the same linear object from the candidate point clouds 50 for the catenary curve created for each group. The creating unit 36 ​​creates a model of the linear object using the linear object point cloud.

[0066] If the intervals between clusters are constant, it is possible to easily create a linear object model by setting a threshold for the interval and connecting the points within that threshold. However, if the point density is sparse and the intervals between clusters are not constant, setting the threshold is difficult, and there is the problem that if the threshold is set too small, the clusters will not be connected, while if it is set too large, the clusters will be connected to the point clouds of other objects.

[0067] On the other hand, in the three-dimensional model creation device 10 of this embodiment, a linear object model is created using catenary curves created for each of multiple groups as described above, so that a linear object model can be created even if the distance (interval) between each cluster 52 is not constant, as in the case of the clusters 52 of this embodiment (see Figure 4, etc.).

[0068] Therefore, according to the three-dimensional model creation device 10 of this embodiment, even if the three-dimensional point cloud representing the three-dimensional coordinates is sparse, a three-dimensional model of a thin linear object can be created with high precision and in real time.

[0069] In the above embodiment, the linear object point cloud extraction unit 34 detects straight lines for each of the plurality of groups based on the two-dimensional coordinates of the candidate point cloud in the horizontal plane (X-Y plane). However, this is not limiting. The linear object point cloud extraction unit 34 may detect straight lines for each of the plurality of groups. For example, the linear object point cloud extraction unit 34 may detect straight lines for each of the plurality of groups based on the two-dimensional coordinates of the candidate point cloud in another plane (the Y-Z plane or the X-Z plane) for each of the plurality of groups. Alternatively, for example, the linear object point cloud extraction unit 34 may detect straight lines for each of the plurality of groups based on the three-dimensional coordinates of the candidate point cloud in three-dimensional space. That is, in the process of step S200 in FIG. 9 described above, the linear object point cloud extraction unit 34 may detect straight lines 54 for each of the plurality of groups based on the two-dimensional coordinates of the candidate point cloud 50 in the Y-Z plane, the two-dimensional coordinates in the X-Z plane, or the three-dimensional coordinates in three-dimensional space.

[0070] Furthermore, the 3D model creation process executed by the CPU 11 after reading the 3D model creation program 15 in the above embodiment may be executed by various processors other than the CPU 11. Examples of processors in this case include dedicated electrical circuits, such as programmable logic devices (PLDs) (such as field-programmable gate arrays (FPGAs)) whose circuit configuration can be changed after manufacture, and application-specific integrated circuits (ASICs) that are processors having a circuit configuration specifically designed to execute specific processes. The 3D model creation process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, a combination of a CPU and an FPGA, etc.). The hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.

[0071] In the above embodiment, the three-dimensional model creation program 15 is pre-stored (also referred to as "installed") in the ROM 12 or the storage 14, but this is not limiting. The three-dimensional model creation program 15 may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The three-dimensional model creation program 15 may also be downloaded from an external device via a network.

[0072] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[0073] The following additional notes are provided regarding the above-described embodiments.

[0074] (Supplementary Item 1) A three-dimensional model creation device including: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire point cloud data of a plurality of point clouds representing three-dimensional coordinates of points on the surface of a structure; extract, from the plurality of point clouds, candidate point clouds that are candidates for a point cloud constituting a linear object based on reflection intensity; cluster the extracted candidate point clouds to generate a plurality of clusters, and for each of the plurality of groups, detect a straight line based on the coordinates of the candidate point cloud; create a catenary curve using the candidate point clouds that constitute the detected straight line; extract a linear object point cloud that constitutes the same linear object from the candidate point clouds of the catenary curve generated for each of the groups; and create a model of the linear object using the linear object point clouds.

[0075] (Supplementary Item 2) A non-transitory storage medium storing a program executable by a computer to execute a three-dimensional model creation process, wherein the three-dimensional model creation process: acquires point cloud data of a plurality of point clouds representing three-dimensional coordinates of points on the surface of a structure; extracts, from the plurality of point clouds, candidate point clouds that are candidates for a point cloud constituting a linear object based on reflection intensity; clusters the extracted candidate point clouds to generate a plurality of clusters, and for each of the plurality of groups, detects a straight line based on the coordinates of the candidate point cloud; creates a catenary curve using the candidate point clouds that constitute the detected straight line; extracts a linear object point cloud that constitutes the same linear object from the candidate point clouds of the catenary curve created for each of the groups; and creates a model of the linear object using the linear object point clouds.

[0076] REFERENCE SIGNS LIST 10 3D model creation device 11 CPU 14 Storage 15 3D model creation program 20 3D laser scanner 30 Acquisition unit 32 Candidate point cloud extraction unit 34 Linear object point cloud extraction unit 36 ​​Creation unit

Claims

1. A three-dimensional model creation device comprising: an acquisition unit that acquires point cloud data of multiple point clouds that represent the three-dimensional coordinates of points on the surface of a structure; a candidate point cloud extraction unit that extracts candidate point clouds from the multiple point clouds based on reflection intensity as candidates for point clouds that constitute linear objects; a linear object point cloud extraction unit that detects straight lines based on the coordinates of the candidate point clouds for each of multiple groups into which multiple clusters are generated by clustering the extracted candidate point clouds, creates catenary curves using the candidate point clouds that constitute the detected straight lines, and extracts linear object point clouds that constitute the same linear object from the candidate point clouds of the catenary curves created for each group; and a creation unit that creates a model of the linear object using the linear object point clouds.

2. A three-dimensional model creation device as described in claim 1, wherein the linear object point cloud extraction unit extracts a candidate point cloud of a catenary curve generated in one group and a candidate point cloud of a catenary curve generated in another group that is determined to match a curve obtained by extending the catenary curve generated in the first group, as linear object point clouds that constitute the same linear object.

3. The three-dimensional model creation device according to claim 2, wherein the linear object point cloud extraction unit uses the longest catenary curve among the catenary curves generated for each group as a reference model, extracts candidate points for the reference model and candidate points for other catenary curves that match the curve obtained by extending the reference model as the linear object point cloud, generates a catenary curve using the linear object point cloud to use as the reference model, and extracts candidate points for the reference model and candidate points for other catenary curves that match the curve obtained by extending the reference model as the linear object point cloud.

4. The three-dimensional model creation device according to claim 1, wherein the linear object point cloud extraction unit generates a plurality of groups by clustering the extracted candidate point cloud and, based on the center of gravity of each of the clusters, grouping clusters whose center of gravity is within a threshold distance to each other to generate the plurality of groups.

Images