Point cloud data reduction device, point cloud data reduction method, and point cloud data reduction program

Abstract

This invention provides a point cloud data reduction device that can compress point cloud data with a high compression ratio while suppressing the loss of necessary information. [Solution] The point cloud discrimination unit 5 compares the target point cloud data with the base point cloud data and classifies the target point cloud into fixed point clouds and non-fixed point clouds. The point cloud discrimination unit 5 sets point clouds of corners or edges in the fixed point cloud as exceptional point clouds, and sets point clouds within clusters that represent thin objects narrower than a predetermined width or objects smaller than a predetermined size, among the clusters obtained by clustering the fixed point cloud, as exceptional point clouds. The downsampling unit 6 compresses the target point cloud data by downsampling the point clouds in the fixed point cloud that are not set as exceptional point clouds.

Description

Technical Field

[0001] The present invention relates to a point cloud data reduction device, a point cloud data reduction method, and a point cloud data reduction program.

Background Art

[0002] Since the point cloud data obtained by measuring a three-dimensional space with a three-dimensional laser scanner has a large capacity, it uses a large capacity of a storage unit for storing the point cloud data and requires a long time when processing the point cloud data. Therefore, the point cloud data may be data-compressed by PCC (Point Cloud Compression) or the like, which is the latest international standard method of the three-point cloud compression technology developed by MPEG (Moving Picture Experts Group).

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Non-Patent Document 2

Non-Patent Document 3

Summary of the Invention

[0004] When point cloud data of approximately 150MB is compressed using the PCC method mpeg-pcc-tmc13 or the lossless compression algorithm Zstandard (see Non-Patent Document 1), the compression ratio is only about 30% to 60%, resulting in limited capacity reduction. There is a need to compress point cloud data at a high compression ratio while minimizing the loss of essential information.

[0005] The present invention aims to provide a point cloud data reduction device, a point cloud data reduction method, and a point cloud data reduction program that can compress point cloud data with a high compression ratio while suppressing the loss of necessary information. [Means for solving the problem]

[0006] A first aspect of the present invention provides a point cloud data reduction device comprising: a point cloud discrimination unit that compares target point cloud data, which is a target point cloud to be compressed, with base point cloud data, which is a base point cloud used as a reference, to classify the target point cloud into fixed point clouds, which are point clouds that are commonly included at the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud; sets point clouds of corners or edges in the fixed point clouds as exceptional point clouds that are not subject to downsampling; sets point clouds in clusters obtained by clustering the fixed point clouds, which are thin objects narrower than a predetermined width or objects smaller than a predetermined size, as exceptional point clouds; and a downsampling unit that compresses the target point cloud data by downsampling point clouds from the fixed point clouds that are not set as exceptional point clouds.

[0007] A second aspect of the present invention provides a point cloud data reduction method that compresses the target point cloud data by comparing target point cloud data, which represents a target point cloud to be compressed, with base point cloud data, which represents a base point cloud as a reference, and classifying the target point cloud into fixed point clouds, which are point clouds that are commonly included in the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud, setting point clouds of corners or edges in the fixed point cloud as exceptional point clouds that are not subject to downsampling, setting point clouds in clusters that represent thin objects narrower than a predetermined width or objects smaller than a predetermined size among clusters obtained by clustering the fixed point cloud as exceptional point clouds, and downsampling point clouds in the fixed point cloud that are not set as exceptional point clouds.

[0008] A third aspect of the present invention provides a point cloud data reduction program that causes a computer to perform the following steps: compare target point cloud data indicating a target point cloud to be compressed with base point cloud data indicating a base point cloud as a reference; classify the target point cloud into fixed point clouds, which are point clouds that are commonly included in the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud, based on the result of the comparison between the target point cloud data and the base point cloud data; set point clouds of corners or edges in the fixed point cloud as exceptional point clouds that are not subject to downsampling; set point clouds in clusters that represent thin objects narrower than a predetermined width or objects smaller than a predetermined size, among clusters obtained by clustering the fixed point cloud, as exceptional point clouds; and compress the target point cloud data by downsampling point clouds from the fixed point cloud that are not set as exceptional point clouds. [Effects of the Invention]

[0009] According to the point cloud data reduction device, point cloud data reduction method, and point cloud data reduction program of the present invention, point cloud data can be compressed with a high compression ratio while suppressing the loss of necessary information. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a block diagram showing a point cloud data reduction device according to one embodiment. [Figure 2A] Figure 2A is a conceptual diagram illustrating an example of a base point cloud. [Figure 2B] Figure 2B is a conceptual diagram illustrating an example of the target point cloud. [Figure 3] Figure 3 is a conceptual diagram showing the exceptional point clouds set by the point cloud discrimination unit of a point cloud data reduction device according to one embodiment, within the target point cloud shown in Figure 2B. [Figure 4A] Figure 4A is a partial flowchart illustrating a point cloud data reduction process according to one embodiment. [Figure 4B] Figure 4B is a partial flowchart following Figure 4A. [Figure 4C] Figure 4C is a partial flowchart that follows Figure 4B. [Figure 5] Figure 5 is a block diagram showing an example of a computer configuration. [Modes for carrying out the invention]

[0011] Hereinafter, a point cloud data reduction device, a point cloud data reduction method, and a point cloud data reduction program according to one embodiment will be described with reference to the attached drawings. Figure 1 shows a point cloud data reduction device 100 according to one embodiment. The point cloud data reduction device 100 includes a point cloud data storage unit 1, a data selection unit 2, a base point cloud data holding unit 3, a target point cloud data holding unit 4, a point cloud discrimination unit 5, a downsampling unit 6, and a processed point cloud data writing unit 7.

[0012] The point cloud data storage unit 1 stores base point cloud data as a reference and target point cloud data. In this embodiment, the point cloud data is assumed to be acquired by measuring the office as a three-dimensional space using a 3D laser scanner (hereinafter referred to as a 3D laser scanner). A 3D laser scanner is sometimes called LiDAR (Light Detection and Ranging). Figure 2A conceptually shows the point cloud acquired by measuring the office with a 3D laser scanner at a certain date and time. The point cloud shown in Figure 2A is referred to as the base point cloud PC1.

[0013] As shown in Figure 2A, the base point cloud PC1 includes point clouds representing columns 11-13, desk 14, chair 15, floor lamp 16, cabinet 17, floor cabinet 18, and potted plant 19. The point cloud data storage unit 1 stores base point cloud data representing the base point cloud PC1 shown in Figure 2A.

[0014] Figure 2B conceptually shows a point cloud acquired by the 3D laser scanner at a date and time later than the date and time the base point cloud PC1 was acquired. The point cloud shown in Figure 2B is referred to as the target point cloud PC2. The target point cloud PC2 is the point cloud that will be subjected to data compression by the point cloud data reduction device 100. As shown in Figure 2B, the target point cloud PC2 includes the point cloud contained in the base point cloud PC1, as well as point clouds representing the trash can 21 and luggage 22. In the target point cloud PC2, the potted plant 19 that was present in the base point cloud PC1 has disappeared. The point cloud data storage unit 1 stores the target point cloud data representing the target point cloud PC2 shown in Figure 2B.

[0015] The point cloud data storage unit 1 may also store base point cloud data that represents a base point cloud different from the base point cloud PC1 shown in Figure 2A. A base point cloud different from the base point cloud PC1 is, for example, a point cloud representing an office without any furniture in place, excluding the point clouds representing the desk 14, chair 15, floor stand 16, cabinet 17, floor cabinet 18, and potted plant 19 in the base point cloud PC1.

[0016] In FIG. 1, the user operates the data selection unit 2 to select the base point cloud PC1 and input it into the base point cloud data holding unit 3, and selects the target point cloud PC2 and inputs it into the target point cloud data holding unit 4. The base point cloud data holding unit 3 and the target point cloud data holding unit 4 respectively hold the base point cloud data indicating the input base point cloud PC1 and the target point cloud data indicating the target point cloud PC2.

[0017] The point cloud discrimination unit 5 discriminates the fixed point cloud included in the target point cloud PC2 by comparing the base point cloud data and the target point cloud data. The fixed point cloud is a point cloud indicating a fixed object that exists fixedly and is commonly included at the same position in both the base point cloud PC1 and the target point cloud PC2. In the target point cloud PC2 shown in FIG. 2B, the point cloud discrimination unit 5 discriminates the point clouds indicating the columns 11 to 13, the desk 14, the chair 15, the floor stand 16, the cabinet 17, and the floor cabinet 18 as the fixed point cloud. The point cloud discrimination unit 5 discriminates the point clouds indicating the trash can 21 and the luggage 22 that exist only in the target point cloud PC2 as non-fixed point clouds that may be moving objects. The point cloud discrimination unit 5 may also discriminate the point cloud indicating the disappeared potted plant 19 as a non-fixed point cloud.

[0018] The point cloud discrimination unit 5 sets the point cloud of the corner or edge in the fixed point cloud as an exception point cloud that is an exception to the object of the downsampling described later. The corner or edge can be detected by using a known method. It is possible to detect both the corner and the edge by one method, or to detect the corner and the edge by separate methods. As an example, it is possible to use the Harris corner detector to detect the corner. The technique described in Non-Patent Document 2 in which the Harris corner detector is applied to a three-dimensional point cloud may be used. In order to extract the edge of the three-dimensional point cloud, the technique described in Non-Patent Document 3 can be used.

[0019] The processed point cloud data, described later, obtained by compressing the target point cloud data representing the target point cloud PC2, may be used, for example, as map data for estimating the self-position of a service robot when it moves around an office, as an example of a mobile machine. Point clouds of corners or edges in a fixed point cloud are important point clouds when a service robot estimates its own position. Therefore, the point cloud discrimination unit 5 sets point clouds of corners or edges in a fixed point cloud as exceptional point clouds.

[0020] An example of a service robot is a cleaning robot. Service robots are equipped with sensors to estimate their own position. These sensors may be a camera, gyroscope, geomagnetic sensor, accelerometer, inertial measurement device, tire pulse sensor, LiDAR, radar sensor (RADAR: Radio Detection and Ranging), or any combination of two or more of these sensors.

[0021] For service robots used outdoors, GNSS sensors are used to determine their position by receiving GNSS signals transmitted from three or more satellites of the Global Navigation Satellite System (GNSS). Typically, GNSS is GPS (Global Positioning System), and GNSS sensors are GPS sensors.

[0022] Furthermore, the point cloud discrimination unit 5 extracts point clouds from the fixed point cloud that represent thin objects with a width narrower than a predetermined width, and point clouds that represent objects smaller than a predetermined size, and sets these extracted point clouds as exceptional point clouds. In the target point cloud PC2 shown in Figure 2B, the point cloud discrimination unit 5 sets the point cloud representing the floor stand 16 as an exceptional point cloud.

[0023] Point clouds representing thin objects can be extracted using known methods such as curvature analysis as an analysis of local geometric features, methods using the variance of normal vectors as another analysis of local geometric features, and skeleton extraction. Curvature analysis is a method that calculates the local curvature of the point cloud and identifies points with high curvature as candidates for thin objects. The method using the variance of normal vectors is a method that calculates the normal vector of each point and evaluates its variance. Skeleton extraction is a method that uses a skeletonization algorithm to identify thin objects by extracting a skeleton (framework) from the point cloud data.

[0024] Point clouds representing small objects can be extracted using known methods such as clustering called DBSCAN (Density-Based Spatial Clustering of Applications with Noise), voxel grid filtering, and region growth algorithms.

[0025] In methods using clustering, point cloud data is clustered, the size of each cluster is evaluated, and smaller clusters are determined to represent point clouds of small objects. In methods using voxel grid filtering, point cloud data is divided into a voxel grid, the number of points in each voxel is evaluated, and voxels with fewer points are determined to represent point clouds of small objects. In methods using region growth algorithms, clusters are formed by adding points adjacent to a specific seed point, the size of the formed clusters is evaluated, and smaller clusters are determined to represent point clouds of small objects.

[0026] The point cloud discrimination unit 5 may extract point clouds representing thin objects by combining multiple methods, or it may extract point clouds representing small objects by combining multiple methods. How the point cloud discrimination unit 5 extracts point clouds representing thin objects or how it extracts point clouds representing small objects is not limited.

[0027] As described above, the point cloud discrimination unit 5 sets point clouds of corners or edges in the fixed point cloud included in the target point cloud PC2 shown in Figure 2B, and point clouds within clusters representing thin objects narrower than a predetermined width or objects smaller than a predetermined size within the fixed point cloud as exceptional point clouds. Point clouds of corners or edges are important point clouds when the service robot estimates its own position. Thin or small objects have fewer points that make them up, so they are point clouds that should not be downsampled in order to ensure they are reliably maintained in the processed point cloud data.

[0028] Figure 3 conceptually shows the exceptional point clouds that the point cloud discrimination unit 5 has set in the target point cloud PC2 shown in Figure 2B. In Figure 3, as indicated by the black circles at the corners and edges of the fixed point cloud, the point cloud discrimination unit 5 sets the point clouds of corners or edges in the fixed point cloud as exceptional point clouds. As shown enclosed by the dashed line, the point cloud discrimination unit 5 sets the point cloud representing the floor stand 16 as an exceptional point cloud among the fixed point clouds. As shown enclosed by the dashed line, the point cloud discrimination unit 5 sets the point clouds representing the trash can 21 and luggage 22 as non-fixed point clouds.

[0029] Returning to Figure 1, the target point cloud data PC2, which is classified into fixed point clouds and non-fixed point clouds, and in which a portion of the fixed point cloud is identified as an exceptional point cloud, is supplied to the downsampling unit 6 along with the identification information. The downsampling unit 6 performs downsampling on the point clouds in the input target point cloud data, excluding the non-fixed point clouds and exceptional point clouds, i.e., the point clouds of the fixed point cloud other than the exceptional point clouds. The downsampling unit 6 supplies the downsampled target point cloud data as processed point cloud data to the processed point cloud data writing unit 7.

[0030] The processed point cloud data writing unit 7 writes the processed point cloud data to the point cloud data storage unit 1. Writing the processed point cloud data to the point cloud data storage unit 1 is not mandatory. The processed point cloud data may also be transmitted to an external device of the point cloud data reduction device 100.

[0031] The proportion of corner or edge points in a fixed point cloud is significantly smaller than the proportion of non-corner and non-edge points in the fixed point cloud. The point cloud data reduction device 100 downsamples the non-corner and non-edge points in the fixed point cloud, which make up a significantly larger proportion, thus enabling high compression of the point cloud data. Furthermore, since the point cloud data reduction device 100 does not downsample non-fixed point clouds and exceptional point clouds, the loss of necessary information can be suppressed. When a service robot uses the processed point cloud data as map data for estimating its own position, the information necessary for estimating its own position is maintained even with high compression, so the movement of the service robot is not hindered.

[0032] If the 3D laser scanner acquires further point cloud data of the office after the date and time when the target point cloud PC2 shown in Figure 2B was acquired, it is advisable to use the target point cloud PC2 shown in Figure 2B as the base point cloud data and the most recent point cloud data as the target point cloud data.

[0033] The operation of the point cloud data reduction device 100 and the point cloud data reduction method performed by the point cloud data reduction device 100 will be explained in more detail using the flowcharts shown in Figures 4A to 4C. In Figure 4A, when processing starts, the base point cloud data holding unit 3 reads and holds the base point cloud data in step S1. The target point cloud data holding unit 4 reads and holds the target point cloud data in step S2. The order of steps S1 and S2 may be reversed or they may be performed simultaneously.

[0034] In step S3, enclosed by a dashed line, the point cloud data reduction device 100 sequentially determines whether all points included in the target point cloud data are fixed points or not. Specifically, in step S31, the point cloud discrimination unit 5 compares the target point cloud with the base point cloud, and in step S32, it determines whether the selected point exists in the base point cloud. If the selected point exists in the base point cloud (YES), the point cloud discrimination unit 5 assigns the fixed point attribute to the selected point in step S33.

[0035] Next, in step S34, the point cloud discrimination unit 5 determines whether the selected point is an important point for self-localization, such as a corner or edge of an object. If the selected point is an important point (YES), the point cloud discrimination unit 5 assigns an exception attribute to the selected point in step S35. If the point selected in step S32 does not exist in the base point cloud (NO), the point cloud discrimination unit 5 assigns an exception attribute to the selected point in step S35. The point cloud discrimination unit 5 repeats the same process until the determination is completed for all points included in the target point cloud data.

[0036] In step S4, the point cloud discrimination unit 5 generates a fixed point cloud consisting only of fixed points, and in step S5, it generates a non-fixed point cloud consisting of points other than fixed points. In step S6, the point cloud discrimination unit 5 clusters the fixed point cloud. By clustering the fixed point cloud, the fixed point cloud is grouped based on the similarity between adjacent points, and the fixed point cloud is divided into multiple clusters.

[0037] In Figure 4B, the point cloud discrimination unit 5, in step S7 enclosed by the dashed line, sequentially selects whether or not to set an exceptional point cloud for all clusters based on the shape determination of the cluster.

[0038] Specifically, in step S71, the point cloud discrimination unit 5 analyzes the shape of the selected cluster, and in step S72, it determines whether the selected cluster is thin or small. If the selected cluster is thin or small (YES), in step S73, the point cloud discrimination unit 5 repeats the process of assigning an exception attribute to the points within the selected cluster until an exception attribute is assigned to all points within the cluster.

[0039] If the cluster selected in step S72 is not a thin or small cluster (NO), the exception attribute is not assigned. The point cloud discrimination unit 5 repeats the same process until processing for all clusters is complete.

[0040] In Figure 4C, the downsampling unit 6 performs downsampling on the fixed point group in step S8, which is enclosed by a dashed line. Specifically, in step S81, the downsampling unit 6 determines whether or not an exception attribute is assigned to the selected point in the fixed point group. If the selected point does not have an exception attribute (NO), the downsampling unit 6 generates a random number in step S82. In step S83, the downsampling unit 6 determines whether or not the random number is above a threshold. If the random number is above a threshold (YES), the downsampling unit 6 deletes the selected point from the fixed point group in step S84.

[0041] If the point selected in step S81 has an exception attribute (YES), or if the random number in step S83 is not greater than or equal to the threshold, the downsampling unit 6 does not delete the selected point from the fixed point group. The downsampling unit 6 repeats the same process for all points in the fixed point group until it has completed the process of deciding whether or not to delete a point based on the comparison result between the random number and the threshold.

[0042] In step S9, the downsampling unit 6 integrates the fixed point cloud and the non-fixed point cloud to generate processed point cloud data. In step S10, the processed point cloud data writing unit 7 reads the processed point cloud data into the point cloud data storage unit 1 and terminates the process.

[0043] The method by which the downsampling unit 6 selects points to be removed from the fixed point cloud is not limited to using random numbers. The downsampling unit 6 may determine whether the distance of the selected point from the centroid of the voxel is greater than or equal to a threshold, and if so, remove the selected point from the fixed point cloud. The downsampling unit 6 may also determine whether the selected point is the centroid of the voxel, and if it is not the centroid of the voxel, remove the selected point from the fixed point cloud.

[0044] When the downsampling unit 6 determines which points to delete from the fixed point cloud using random numbers, it has the advantage that the points to be deleted are not fixed. When the downsampling unit 6 deletes points whose distance from the centroid of a voxel is greater than a threshold, it is possible to downsample the target point cloud while retaining multiple points, including points located at the centroid that are important as points. When the downsampling unit 6 deletes points other than the centroid of a voxel, it is possible to downsample the target point cloud with a higher compression ratio while retaining only points located at the centroid that are important as points.

[0045] The point cloud data reduction device 100 shown in Figure 1 may be composed of hardware, or it may be functionally configured by a computer executing a point cloud data reduction program. Figure 5 shows the configuration of a general-purpose computer. The computer includes a central processing unit (hereinafter referred to as CPU) 31, main memory 32, non-temporary storage medium 33, input device 34, and output device 35. The point cloud data reduction device 100 can be realized by the CPU 31 executing a point cloud data reduction program stored in the non-temporary storage medium 33. The point cloud data reduction program can also be distributed via a network.

[0046] The point cloud data reduction program stored in the non-temporary storage medium 33 is loaded into the main memory 32, and the CPU 31 executes the point cloud data reduction program. The point cloud data storage unit 1 may be the non-temporary storage medium 33 or another storage medium not shown. The data selection unit 2 can be configured by an input device 34 or an input device 34 and an output device 35. The input device 34 is a keyboard or mouse, and the output device 35 is, for example, a display.

[0047] The CPU 31 displays candidate base point clouds and target point clouds on the display, and the user selects the base point cloud and target point cloud using the input device 34. This makes it easy to select the base point cloud data to be stored in the base point cloud data storage unit 3 and the target point cloud data to be stored in the target point cloud data storage unit 4.

[0048] The base point cloud data storage unit 3 and the target point cloud data storage unit 4 may be the main memory 32 or the memory built into the CPU 31. The CPU 31 functions as a point cloud discrimination unit 5, a downsampling unit 6, and a processed point cloud data writing unit 7. The flowcharts shown in Figures 4A to 4C are also processes that the point cloud data reduction program has the CPU 31 execute. In Figure 5, a GPU (Graphics Processing Unit) may be used instead of the CPU 31.

[0049] The computer may be equipped with a communication device that transmits processed point cloud data to external devices such as service robots via a network such as the Internet.

[0050] The processed point cloud data should ideally include information that distinguishes between fixed and non-fixed point clouds. If the processed point cloud data used by the service robot as map data includes information that distinguishes between fixed and non-fixed point clouds, the following secondary effects can be obtained: The service robot can change its behavior depending on whether the object is a fixed object represented by a fixed point cloud or a moving object represented by a non-fixed point cloud.

[0051] Let's take the example of a service robot that is a cleaning robot. If the robot cleans around a moving object and then the object moves, the area where the moving object was located will not be cleaned. Therefore, the cleaning robot should be given information that there are no non-fixed point clouds in the latest target point cloud, and should spend more time cleaning the areas where non-fixed point clouds existed in the previous target point cloud than other areas.

[0052] The present invention is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the invention. [Explanation of symbols]

[0053] 1. Point cloud data storage unit 2. Data Selection Section 3. Base point cloud data storage unit 4. Target point cloud data storage unit 5 Point cloud discriminator 6 Downsampling section 7. Processed point cloud data writing unit 31 Central Processing Unit 32 Main Memory 33 Non-transitory storage medium 34 Input devices 35 Output device 100-point cloud data reduction device

Claims

1. A point cloud discrimination unit compares target point cloud data, which represents the target point cloud to be compressed, with base point cloud data, which represents a base point cloud as a reference, and classifies the target point cloud into fixed point clouds, which are point clouds that are commonly included at the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud, sets point clouds of corners or edges in the fixed point cloud as exceptional point clouds that are not subject to downsampling, and sets point clouds within clusters that represent thin objects narrower than a predetermined width or objects smaller than a predetermined size, among clusters obtained by clustering the fixed point cloud, as exceptional point clouds. A downsampling unit that compresses the target point cloud data by downsampling the point clouds among the fixed point clouds that are not set in the exceptional point cloud, A point cloud data reduction device equipped with the following features.

2. The point cloud data reduction device according to claim 1, wherein the downsampling unit determines whether or not to delete points based on the result of comparing the generated random number with a threshold, and downsamples the fixed point cloud.

3. The point cloud data reduction device according to claim 1, wherein the downsampling unit downsamples the fixed point cloud by deleting points whose distance from the centroid of a voxel is greater than or equal to a threshold.

4. The point cloud data reduction device according to claim 1, wherein the downsampling unit downsamples the fixed point cloud by deleting points that are not located at the centroid of the voxel.

5. The target point cloud data, which represents the point cloud to be compressed, is compared with the base point cloud data, which represents the base point cloud used as a reference. Based on the results of comparing the target point cloud data with the base point cloud data, the target point cloud is classified into fixed point clouds, which are point clouds that are commonly included at the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud. The point clouds of corners or edges in the aforementioned fixed point cloud are set as exceptional point clouds that are not subject to downsampling, Among the clusters obtained by clustering the fixed point clouds, the point clouds within clusters that represent thin objects narrower than a predetermined width or objects smaller than a predetermined size are set as the exceptional point clouds. The target point cloud data is compressed by downsampling the point clouds among the fixed point clouds that are not set in the exceptional point clouds. Methods for reducing point cloud data.

6. On the computer, The steps include comparing the target point cloud data, which represents the point cloud to be compressed, with the base point cloud data, which represents the base point cloud used as a reference, Based on the results of comparing the target point cloud data with the base point cloud data, the target point cloud is classified into fixed point clouds, which are point clouds that are commonly included at the same positions in both the base point cloud and the target point cloud, and non-fixed point clouds, which are point clouds that are not included in the base point cloud. The steps include setting the point clouds of corners or edges in the fixed point cloud as exceptional point clouds that are not subject to downsampling, The steps include setting the point clouds within clusters obtained by clustering the fixed point clouds, which represent thin objects narrower than a predetermined width or objects smaller than a predetermined size, as the exceptional point clouds, The step of compressing the target point cloud data by downsampling the point clouds from the fixed point cloud that are not set as the exceptional point clouds, A point cloud data reduction program that performs this operation.

Images