Point cloud annotation device, point cloud annotation method, and program

Abstract

This point cloud annotation device is provided with: an acquisition unit that acquires a first point cloud that includes a plurality of first three-dimensional points of a ground object present in a measurement region, a second point cloud that includes a plurality of second three-dimensional points of a ground object present in the measurement region and having a higher density than the first point cloud, and label data that represents a label assigned to each of the plurality of second three-dimensional points; and a label propagation unit that propagates a label based on the label assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points to generate annotation data for the first point cloud.

Description

Point cloud annotation device, point cloud annotation method, and program

[0001] The disclosed technology relates to a point cloud annotation device, a point cloud annotation method, and a program.

[0002] In recent years, high-precision 3D maps of urban environments have been created using 3D point clouds obtained by measuring objects using LiDAR (Light Detection and Ranging) measurement devices mounted on automobiles and other vehicles. These maps are expected to be useful for urban planning and disaster prevention simulations.

[0003] In particular, as a technology for recognizing features such as houses, buildings, and utility poles from a 3D point cloud, Non-Patent Document 1 describes a technology called PointNet that enables highly accurate point cloud segmentation using deep learning technology. To enable deep learning like the technology described in Non-Patent Document 1, it is necessary to create learning data in which a worker annotates and labels a large amount of 3D point cloud data obtained by measuring the object.

[0004] “PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation” Charles R. Qi, Hao Su, Kaichun Mo, Leonidas J. Guibas; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, pp. 652-660

[0005] Because measuring high-density 3D point clouds requires expensive measurement equipment, it is desirable to achieve segmentation using relatively low-density point cloud data. To achieve this, training data annotated from relatively low-density point cloud data is required. However, it can be difficult to obtain annotation data that labels relatively low-density point cloud data. For example, when manually annotating 3D point clouds, the shape of each point in the 3D point cloud is checked using a point cloud viewer to determine which feature each point belongs to. While a high-density 3D point cloud makes it easy to visually confirm its shape, it is difficult to manually recognize features from a low-density 3D point cloud.

[0006] The technology disclosed herein has been developed in consideration of the above points, and provides a point cloud annotation device, a point cloud annotation method, and a program that can obtain annotation data that has been labeled with relatively low-density point cloud data.

[0007] The point cloud annotation device of the present disclosure includes an acquisition unit that acquires a first point cloud including a plurality of first three-dimensional points of features present in a measurement area, a second point cloud including a plurality of second three-dimensional points of features present in the measurement area and having a higher density than the first point cloud, and label data representing labels assigned to each of the plurality of second three-dimensional points, and a label propagation unit that propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0008] In addition, the point cloud annotation method disclosed herein acquires a first point cloud including a plurality of first three-dimensional points of features present in a measurement area, a second point cloud including a plurality of second three-dimensional points of features present in the measurement area and having a higher density than the first point cloud, and label data representing labels assigned to each of the plurality of second three-dimensional points, and propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0009] Furthermore, the program of the present disclosure is for causing a computer to function as each part of the point cloud annotation device of the present disclosure.

[0010] According to the technology of the present disclosure, annotation data can be obtained by adding labels to point cloud data with a relatively low density.

[0011] It is a diagram for explaining a point cloud segmentation model. It is a block diagram showing an example of the functional configuration of the point cloud annotation device of the embodiment. It is a configuration diagram showing an example of the hardware configuration of the point cloud annotation device of the embodiment. It is a flowchart showing an example of point cloud annotation processing in the point cloud annotation device of the embodiment.

[0012] Hereinafter, an example of an embodiment of the disclosed technology will be described with reference to the drawings. Note that the same or equivalent components and parts in each drawing are given the same reference numerals.

[0013] First, an example of the configuration of a point cloud annotation device of this embodiment will be described. FIG. 1 shows an example of a point cloud segmentation model M trained using annotation data generated by the point cloud annotation device of this embodiment. The point cloud segmentation model M is a trained machine learning model that, when a low-density point cloud, which is a group of multiple 3D points obtained by measuring features present in a measurement area using a measurement device such as LiDAR, is input, outputs a segmentation result of the low-density point cloud. An example of the point cloud segmentation model M is a model using PointNet. The point cloud annotation device of this embodiment has a function of generating annotation data, as training data for the point cloud segmentation model M, in which a label representing a feature is assigned to each of multiple 3D points included in the low-density point cloud.

[0014] 2 is a block diagram showing an example of the functional configuration of the point cloud annotation device 10 of this embodiment. As shown in FIG. 2, the point cloud annotation device 10 of this embodiment includes an acquisition unit 20, a change point detection unit 22, an isolated point detection unit 24, an input conversion unit 26, and a label propagation unit 28.

[0015] The acquisition unit 20 acquires the low-density point cloud A, the high-density point cloud B, and the label data L Ｂ The low-density point cloud A is a point cloud containing a plurality of 3D points with a relatively low density, obtained by measuring features existing in a measurement area using a measurement device. The high-density point cloud B is a point cloud containing a plurality of 3D points with a relatively high density, obtained by measuring features existing in the same measurement area as the low-density point cloud A using a measurement device. The specific densities defined as "low density" and "high density" are not particularly limited, and it is sufficient that the density of the high-density point cloud B is at least higher than the density of the low-density point cloud A. For example, the density of a 3D point cloud that allows for relatively easy manual annotation of the 3D point cloud may be experimentally obtained and used as a density threshold, and if the density is lower than the density threshold, it may be called a low-density point cloud A, and if the density is higher than the density threshold, it may be called a high-density point cloud B. Label data L Ｂ is label data representing labels corresponding to features assigned to each of the three-dimensional point clouds included in the high-density point cloud B. The low-density point cloud A of this embodiment is an example of a first point cloud of the present disclosure, and the multiple three-dimensional points included in the low-density point cloud A of this embodiment are an example of the first three-dimensional points of the present disclosure. Furthermore, the high-density point cloud B of this embodiment is an example of a second point cloud of the present disclosure, and the multiple three-dimensional points included in the high-density point cloud B of this embodiment are an example of the second three-dimensional points of the present disclosure.

[0016] The acquisition unit 20 acquires the low-density point group A, the high-density point group B, and the label data L Ｂ The source from which the low-density point cloud A, the high-density point cloud B, and the label data L are acquired is not limited. For example, the low-density point cloud A, the high-density point cloud B, and the label data L may be acquired from an external device via the communication I / F 48. Ｂ The form may be such that the above is acquired.

[0017] The acquisition unit 20 outputs the acquired low-density point cloud A and high-density point cloud B to the change point detection unit 22. The acquisition unit 20 also outputs the acquired low-density point cloud A and high-density point cloud B to the isolated point detection unit 24. The acquisition unit 20 also outputs the acquired low-density point cloud A to the input conversion unit 26. The acquisition unit 20 also outputs the acquired high-density point cloud B and label data L Ｂto the label propagation unit 28.

[0018] The change point detection unit 22 detects change points from the low-density point group A, and extracts a low-density point group A from which the change points have been removed. Ｓ If there is a gap between the measurement of the low-density point cloud A and the measurement of the high-density point cloud B, the shape of the feature may have changed due to the replacement of a building, etc. In this embodiment, a 3D point that exists in an area where the shape of the feature has changed is called a change point.

[0019] The method by which the change point detection unit 22 of this embodiment detects change points is not particularly limited. As an example, the change point detection unit 22 of this embodiment first arranges the low-density point group A and the high-density point group B in the same coordinate system of three-dimensional space. Then, the change point detection unit 22 calculates a distance threshold rh from the reference for a certain three-dimensional point included in the low-density point group A. ｉ The change point detection unit 22 derives the number of three-dimensional points of the high-density point cloud B included within the distance threshold rh ｉ If the number of 3D points included in the low-density point cloud A is equal to or less than a predetermined number, the reference 3D point is detected as a change point. The change point detection unit 22 detects change points for all 3D points included in the low-density point cloud A in the above manner, and the set of detected change points is set as change points S.

[0020] Next, the change point detection unit 22 detects change areas in which the shape of features has changed due to the replacement of buildings, etc., as described above, from the low-density point cloud A. To detect change areas, a clustering algorithm, a geometric shape detection algorithm, or a segmentation model that has been trained in advance may be used.

[0021] Furthermore, the change point detection unit 22 may detect a change area by comparing three-dimensional points between neighboring areas in the low-density point group A and the high-density point group B, and may use the number of three-dimensional points constituting the area, the score of the same object determination, etc. For example, the change point detection unit 22 may use a density parameter e_A and a cluster formation number m, which are parameters of DBSCAN clustering, which is one of the clustering algorithms. ＡWhen some of the change points S are included in a cluster, the change point detection unit 22 adds the three-dimensional points of the low-density point cloud A included in the corresponding cluster to the change points S.

[0022] The change point detection unit 22 of this embodiment extracts a low-density point group A by excluding the change point S from the low-density point group A. Ｓ is output to the input conversion unit 26.

[0023] The isolated point detection unit 24 has a function of detecting 3D points that exist in lower density areas of the low-density point cloud A as isolated points. In the low-density point cloud A, 3D points that are farther away from the measurement point have a lower density. In other words, the point density becomes even sparser, and the number of 3D points that make up a feature may be as few as a few. If such 3D points are labeled and used as annotation data, they may become noise that reduces the learning accuracy of the point cloud segmentation model M. Therefore, in the point cloud annotation device 10 of this embodiment, 3D points that exist in lower density areas are excluded from the low-density point cloud A.

[0024] The method by which the isolated point detection unit 24 of this embodiment detects isolated points is not particularly limited. As an example, the isolated point detection unit 24 of this embodiment detects isolated points by using a certain three-dimensional point included in the low-density point group A as a reference and calculating a distance threshold rh Ｔ The isolated point detection unit 24 calculates the number of three-dimensional points of the low-density point group A included within the distance threshold rh Ｔ If the number of 3D points included in the low-density point group A is less than a predetermined number, the reference 3D point is detected as an isolated point. The isolated point detection unit 24 detects isolated points as described above for all 3D points included in the low-density point group A, and the set of detected isolated points is designated as isolated points T.

[0025] The change point detection unit 22 of this embodiment extracts the low-density point group A by excluding the isolated point T from the low-density point group A. Ｔ is output to the input conversion unit 26.

[0026] The input conversion unit 26 has a function of converting the low-density point group A into a low-density point group A_a. The low-density point group A_a is a low-density three-dimensional point group used as the annotation data 30. The method by which the input conversion unit 26 converts the low-density point group A into the low-density point group A_a is not limited. As an example, the input conversion unit 26 of this embodiment converts the low-density point group A into the low-density point group A_a. Ｓ and low-density point group A Ｔ The low-density point group A_a is derived by extracting three-dimensional points that are common to all of the above. Note that the present invention is not limited to this embodiment, and for example, the isolated point detection unit 24 may derive the low-density point group A_a by excluding the change points S and the isolated points T from the low-density point group A.

[0027] The input conversion unit 26 outputs the low-density point group A_a to the label propagation unit 28 .

[0028] The label propagation unit 28 has a function of propagating labels based on the labels assigned to each of the three-dimensional points included in the high-density point cloud B to each of the three-dimensional points included in the low-density point cloud A.

[0029] The label propagation unit 28 arranges the three-dimensional points included in the low-density point group A_a and the three-dimensional points included in the high-density point group B in the same coordinate system of three-dimensional space. The method by which the label propagation unit 28 propagates the labels of the high-density point group B to the low-density point group A_a is not limited. As an example, for each three-dimensional point included in the low-density point group A_a, the label propagation unit 28 of this embodiment uses that point as a reference, derives the three-dimensional point of the high-density point group B that is closest to the reference three-dimensional point from among the three-dimensional points included in the high-density point group B, and generates label data L of the label assigned to the derived three-dimensional point. Ｂ The label propagation unit 28 assigns the label data L to the reference three-dimensional point using, for example, the label and a weighted score according to the distance. Ｂ The label propagation unit 28 assigns labels to all three-dimensional points included in the low-density point group A_a in the manner described above. The set of assigned label data is referred to as label data L Ａ＿ａ Let's say.

[0030] The label propagation unit 28 of this embodiment receives the low-density point group A_a and the label data L Ａ＿ａand output as annotation data 30.

[0031] The above-described point cloud annotation device 10 is configured with hardware shown in Fig. 3. Fig. 3 is a configuration diagram showing an example of the hardware configuration of the point cloud annotation device 10 of this embodiment. As shown in Fig. 3, the point cloud annotation device 10 includes a CPU (Central Processing Unit) 41, a ROM (Read Only Memory) 42, a RAM (Random Access Memory) 43, storage 44, a display unit 46, an input unit 47, and a communication I / F (Interface) 48. Each component is connected to each other so as to be able to communicate with each other via a bus 49 such as a system bus or a control bus.

[0032] The CPU 41 is a central processing unit, and executes various programs such as a point cloud annotation program 45 stored in the storage 44, and controls each part.

[0033] The ROM 42 stores various programs and data executed by the CPU 41. The RAM 43 temporarily stores programs or data as a work area when the CPU 41 executes the various programs. That is, the CPU 41 reads a program from the storage 44 and executes the program using the RAM 43 as a work area.

[0034] The storage 44 of this embodiment stores a point cloud annotation program 45. The point cloud annotation program 45 may be a single program or a group of programs configured from multiple programs or modules. The storage 44 is configured from an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The point cloud annotation program 45 of this embodiment is an example of a program of the present disclosure.

[0035] The display unit 46 displays various information such as information related to annotations of point clouds. The display unit 46 is not particularly limited, and various displays can be used. The input unit 47 is a user interface such as a mouse or keyboard, and accepts various operations by the user. The display unit 46 and the input unit 47 may be integrated into one unit, or may be a touch panel display.

[0036] The communication I / F 48 is an interface for communicating with external devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).

[0037] Next, the operation of the point cloud annotation device 10 of this embodiment will be described.

[0038] 4 shows a flowchart of an example of a point cloud annotation process executed by the point cloud annotation device 10 of this embodiment. In the point cloud annotation device 10, the CPU 41 executes a point cloud annotation program 45 stored in the storage 44, whereby the CPU 41 functions as the acquisition unit 20, the change point detection unit 22, the isolated point detection unit 24, the input conversion unit 26, and the label propagation unit 28. The point cloud annotation process shown in FIG. 4 is executed at a predetermined timing, such as when an execution instruction is received from a user.

[0039] In step S100 of FIG. 4, the acquisition unit 20 acquires the low-density point cloud A, the high-density point cloud B, and the label data L, as described above. Ｂ Get.

[0040] In the next step S102, as described above, the change point detection unit 22 detects change points S, which are three-dimensional points corresponding to changed topographical features, from the low-density point group A. Also, as described above, the change point detection unit 22 of this embodiment detects a low-density point group A obtained by excluding the change points S from the low-density point group A. Ｓ is derived.

[0041] In the next step S104, as described above, the isolated point detection unit 24 detects isolated points T, which are three-dimensional points existing in a lower density area, from the low-density point group A. Furthermore, as described above, the isolated point detection unit 24 of this embodiment detects a low-density point group A obtained by excluding the isolated points T from the low-density point group A. Ｔ is derived.

[0042] The order of the process in step S102 and the process in step S104 may be reversed, or the processes may be performed in parallel.

[0043] In the next step S106, the input conversion unit 26 has the function of converting the low-density point group A into a low-density point group A_a, as described above.

[0044] In the next step S108, the label propagation unit 28 propagates the label from the high density point group B to the low density point group A_a as described above. Ａ＿ａ is generated.

[0045] In the next step S110, the label propagation unit 28 propagates the low-density point group A_a and the label data L Ａ＿ａ The label propagation unit 28 outputs the annotation data 30 as annotation data 30. The output destination of the annotation data 30 is not limited. For example, the annotation data 30 may be output to the storage 44. Furthermore, for example, the label propagation unit 28 may output the annotation data 30 to an external device via the communication I / F 48.

[0046] When the process of step S110 is completed, the point cloud annotation process shown in FIG. 4 is completed.

[0047] As described above, in the point cloud annotation device 10, the acquisition unit 20 acquires a low-density point cloud A including a plurality of 3D points of features present in the measurement area, a high-density point cloud B including a plurality of 3D points of features present in the measurement area and having a higher density than the low-density point cloud A, and label data representing labels assigned to each of the 3D points in the high-density point cloud B. In addition, the label propagation unit 28 propagates labels based on the labels assigned to each of the 3D points included in the high-density point cloud B to the plurality of 3D points included in the low-density point cloud A, thereby generating annotation data 30 for the low-density point cloud A.

[0048] Therefore, according to the point cloud annotation device 10 of this embodiment, it is possible to obtain annotation data 30 in which labels are added to relatively low-density point cloud data. In this way, by performing model training using the annotation data 30 generated by the point cloud annotation device 10 of this embodiment as training data, it is possible to train the point cloud segmentation model M as shown in FIG.

[0049] In addition, the various processes executed by the CPU after reading the software (program) in each of the above embodiments may be executed by various processors other than the CPU. Examples of processors in this case include dedicated electrical circuits, such as programmable logic devices (PLDs) whose circuit configuration can be changed after manufacture, such as field-programmable gate arrays (FPGAs), and application-specific integrated circuits (ASICs), which are processors having a circuit configuration designed specifically for executing specific processes. Furthermore, each of the learning process and the estimation process may be executed by one of these various processors, or may be executed 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.). Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.

[0050] In addition, in each of the above embodiments, the point cloud annotation program 45 is pre-stored (installed) in the storage 44, but this is not limiting. The point cloud annotation program 45 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 point cloud annotation program 45 may also be downloaded from an external device via a network.

[0051] 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.

[0052] The following additional clauses are disclosed in relation to the above-described embodiment.

[0053] (Supplementary Item 1) A point cloud annotation device comprising: an acquisition unit that acquires a first point cloud including a plurality of first three-dimensional points of features present in a measurement area; a second point cloud that includes a plurality of second three-dimensional points of features present in the measurement area and has a higher density than the first point cloud; and label data that represents labels assigned to each of the plurality of second three-dimensional points; and a label propagation unit that propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0054] (Supplementary Item 2) The point cloud annotation device according to Supplementary Item 1, wherein the label propagation unit assigns, to each of the plurality of first three-dimensional points, a label based on a label assigned to the second three-dimensional point located nearby in three-dimensional space.

[0055] (Supplementary Item 3) The point cloud annotation device according to Supplementary Item 1 or Supplementary Item 2 further comprises a change point detection unit that detects, based on the first point cloud and the second point cloud, a change area where a feature has changed between the time when the first point cloud was measured and the time when the second point cloud was measured, and detects the first three-dimensional points that exist in the change area as change points, and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to the first point cloud excluding the change points.

[0056] (Supplementary Item 4) The point cloud annotation device according to any one of Supplementary Items 1 to 3, further comprising a change point detection unit that detects, from the first point cloud, three-dimensional points at which the number of second three-dimensional points existing in three-dimensional space within a predetermined distance from each first three-dimensional point is equal to or less than a predetermined number, and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to the first point cloud excluding the change points.

[0057] (Supplementary Item 5) The point cloud annotation device according to any one of Supplementary Items 1 to 3, further comprising an isolated point detection unit that detects, from the first point cloud, three-dimensional points such that the number of first three-dimensional points existing in three-dimensional space within a predetermined distance from each first three-dimensional point is equal to or less than a predetermined number, as isolated points; and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to the first point cloud excluding the isolated points.

[0058] (Supplementary Item 6) A point cloud annotation method comprising: acquiring a first point cloud including a plurality of first three-dimensional points of a feature present in a measurement area; a second point cloud including a plurality of second three-dimensional points of the feature present in the measurement area and having a higher density than the first point cloud; and label data representing labels assigned to each of the plurality of second three-dimensional points; and propagating labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0059] (Supplementary Item 7) A program for causing a computer to function as each unit of the point cloud annotation device according to any one of Supplementary Items 1 to 5.

[0060] (Supplementary Item 8) A point cloud annotation device comprising at least one processor, wherein the processor acquires a first point cloud including a plurality of first three-dimensional points of a feature present in a measurement area, a second point cloud including a plurality of second three-dimensional points of the feature present in the measurement area and having a higher density than the first point cloud, and label data representing labels assigned to each of the plurality of second three-dimensional points, and propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0061] (Supplementary Item 9) A non-transitory storage medium storing a program executable by a computer to perform point cloud annotation processing, wherein the point cloud annotation processing is configured to: acquire a first point cloud including a plurality of first three-dimensional points of features present in a measurement area; a second point cloud including a plurality of second three-dimensional points of features present in the measurement area and having a higher density than the first point cloud; and label data representing labels assigned to each of the plurality of second three-dimensional points; and propagate labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

[0062] REFERENCE SIGNS LIST 10 Point cloud annotation device 20 Acquisition unit 22 Change point detection unit 24 Isolated point detection unit 28 Label propagation unit 30 Annotation data 45 Point cloud annotation program M Point cloud segmentation model

Claims

1. A point cloud annotation device comprising: an acquisition unit that acquires a first point cloud including a plurality of first three-dimensional points of features present in a measurement area; a second point cloud including a plurality of second three-dimensional points of features present in the measurement area and having a higher density than the first point cloud; and label data representing labels assigned to each of the plurality of second three-dimensional points; and a label propagation unit that propagates labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

2. The point cloud annotation device according to claim 1, wherein the label propagation unit assigns a label to each of the plurality of first three-dimensional points based on a label assigned to the second three-dimensional point located nearby in three-dimensional space.

3. The point cloud annotation device of claim 1, further comprising a change point detection unit that detects, based on the first point cloud and the second point cloud, a change area in which a feature has changed between the time when the first point cloud was measured and the time when the second point cloud was measured, and detects the first 3D points present in the change area as change points, and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second 3D points to the first point cloud excluding the change points.

4. The point cloud annotation device of claim 1, further comprising a change point detection unit that detects, from the first point cloud, 3D points at which the number of the second 3D points existing in the 3D space within a predetermined distance from each first 3D point is equal to or less than a predetermined number as change points, and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second 3D points to the first point cloud excluding the change points.

5. The point cloud annotation device of claim 1, further comprising an isolated point detection unit that detects, from the first point cloud, 3D points where the number of first 3D points existing in 3D space within a predetermined distance from each first 3D point is a predetermined number or less as isolated points, and the label propagation unit propagates labels based on the labels assigned to each of the plurality of second 3D points to the first point cloud excluding the isolated points.

6. A point cloud annotation method comprising: acquiring a first point cloud including a plurality of first three-dimensional points of a feature present in a measurement area; a second point cloud including a plurality of second three-dimensional points of a feature present in the measurement area and having a higher density than the first point cloud; and label data representing labels assigned to each of the plurality of second three-dimensional points; and propagating labels based on the labels assigned to each of the plurality of second three-dimensional points to each of the plurality of first three-dimensional points, thereby generating annotation data for the first point cloud.

7. A program for causing a computer to function as each part of the point cloud annotation device according to any one of claims 1 to 5.

Images