Measurement point classification device

Abstract

To provide a measurement point classification device which can quickly classify measurement points into classes compared to the prior art.SOLUTION: A measurement point classification device is a device which classifies measurement points, and comprises region of interest setting means, grouping means, representative point setting means, search region setting means, and feature amount calculation means. The grouping means is means which sets measurement points included in a region of interest as "a measurement point collection". The representative point setting means is the means which sets "a representative point" of the measurement point collection. The feature amount calculation means is the means which calculates a feature amount of an interest representative point on the basis of an attribute which the representative point included in a search region possesses.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technology for estimating, for example, the type of a feature using a set of measurement points obtained by measurement (hereinafter referred to as a "measurement point cloud"), and more specifically to a measurement point classification device that sets a measurement point that represents a plurality of measurement points and classifies the measurement points. [Background technology]

[0002] In recent years, with the advancement of measurement technology, the demand for topographical information (spatial information) has increased, and an increasing number of managers are requesting spatial information on facilities, such as their shape and location, in order to more effectively manage facilities installed on and along roads. Furthermore, the realization of Society 5.0, which is currently being promoted jointly by the public and private sectors, also sees advanced maintenance and management of social infrastructure (hereinafter simply referred to as "social infrastructure") as a key issue, and for this reason spatial information on facilities, etc., has become extremely important and indispensable information.

[0003] Previously, to obtain spatial information on road facilities, direct surveying of the target feature was carried out using a TS (Total Station) or similar, but recently measurement methods that obtain comprehensive spatial information without targeting specific features are often adopted. For example, SfM (Structure from Motion) uses multiple photographs, and MMS (Mobile Mapping System) measures while moving using a vehicle equipped with a laser scanner, etc., making it possible to obtain a large number of measurement points in a relatively short period of time.

[0004] Each measurement point obtained through such measurements may contain various "attributes," such as RGB (photographic measurement) and reflectance intensity (laser measurement) in addition to 3D coordinates. However, the "feature type" corresponding to the measurement point is not included as an attribute, meaning it is not possible to directly determine what the measurement point measured. Therefore, efforts are being made to estimate the feature type for each measurement point (or a group of nearby measurement points) by using the attributes possessed by the measurement point. For example, feature types can be estimated from the attributes of measurement points by performing machine learning using a combination of the attributes of the measurement points and the feature types, or feature types can be estimated from the attributes of measurement points by setting predetermined rules based on the relationship between the attributes of the measurement points and the feature types.

[0005] Various techniques have been proposed for estimating feature types based on the attributes of measurement points. For example, Patent Document 1 discloses an invention that uses "missing measurement points" that were not measured by laser measurement to estimate "water areas" or "objects with significant unevenness" that do not reflect lasers well. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2022-030559 Summary of the Invention [Problem to be solved by the invention]

[0007] As mentioned above, feature types can be estimated directly from the attributes of measurement points, but feature types can also be estimated after determining feature quantities related to the measurement points. Specifically, feature quantities such as linearity, planarity, and point density are determined using the attributes of multiple nearby measurement points, and the feature type is estimated based on these feature quantities using machine learning technology, rule-based methods, or the like. When selecting nearby measurement points, a spherical (or circular) region of a specified radius is often set around the measurement point whose feature quantities are to be determined, and feature quantities (multi-scale feature quantities) can also be determined while setting spherical regions of multiple radii.

[0008] Conventionally, when estimating feature types using feature quantities at measurement points, the feature quantities are calculated for all measurement points within the target area. However, when using laser measurement, for example, a huge number of measurement points are acquired, meaning that feature quantities must be calculated for a large number of measurement points, which unavoidably requires a huge amount of calculation. In particular, when calculating multi-scale feature quantities, an even greater amount of calculation is required. As a result, it takes a long time to obtain results, which leads to high calculation costs, and various problems arise, such as the need to store large amounts of data.

[0009] An object of the present invention is to solve the problems of the prior art, that is, to provide a measurement point classification device that can classify measurement points more quickly than the prior art. [Means for solving the problem]

[0010] The present invention is based on a novel idea of ​​setting a "representative point" that represents multiple nearby measurement points, using that representative point to determine feature values, and then classifying the measurement points.

[0011] The measurement point classification device of the present invention is a device for classifying measurement points and includes a region-of-interest setting means, a grouping means, a representative point setting means, a search region setting means, and a feature calculation means. Among these, the region-of-interest setting means is a means for selecting a "measurement point of interest" from a group of measurement points and setting a "region of interest (region based on the measurement point of interest)." The grouping means is a means for setting measurement points (including the measurement point of interest) included in the region of interest as a "measurement point set." The representative point setting means is a means for setting a "representative point" of the measurement point set. Furthermore, the search region setting means is a means for selecting a "representative point of interest" and setting a "search region (region based on the representative point of interest)." The feature calculation means is a means for calculating the feature of the representative point of interest based on attributes of the representative point included in the search region. The class setting means is a means for assigning a class to the representative point of interest according to the feature. The search area is larger than the area of ​​interest, and the area of ​​interest setting means repeatedly selects the measurement point of interest from the "unprocessed measurement points (measurement points excluding those included in the measurement point set that has already been set)." The grouping means repeatedly sets the measurement point set, targeting only the unprocessed measurement points.

[0012] The measurement point classification device of the present invention can also be configured to perform special processing on a representative point when the representative point is near the boundary of the search area. In other words, in this case, when the representative point is near the boundary of the search area, the feature calculation means calculates the feature taking into account the attributes of measurement points that are inside the search area among the measurement points included in the measurement point set related to the representative point.

[0013] The measurement point classification device of the present invention may also be configured such that the representative point setting means sets the target measurement point as the representative point.

[0014] The measurement point classification device of the present invention can also set a newly determined measurement point as a representative point. In this case, the representative point setting means determines the new measurement point by statistically processing the attributes of the measurement points included in the measurement point set.

[0015] The measurement point classification device of the present invention may also use the type of feature as a class. [Effects of the Invention]

[0016] The measurement point classification device of the present invention has the following effects. (1) The measurement point classification device of the present invention can classify measurement points into classes (types of feature) without the need for human judgment such as visual inspection, thereby avoiding human error and other human mistakes. (2) Compared to conventional techniques, the measurement points can be classified into classes (funny feature types) in an extremely short time, thereby reducing the calculation cost. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a block diagram showing the main configuration of a measurement point classification device according to the present invention; [Figure 2] FIG. 1 is a model diagram showing a measurement point and an area of ​​interest. [Figure 3] (a) is a model diagram that shows a schematic diagram of a situation in which measurement points of interest are repeatedly selected from left to right to set a region of interest, and (b) is a model diagram that shows a schematic diagram of a process in which a set of measurement points is set and a representative point is set. [Figure 4] 1 is a model diagram showing a representative point of interest and a search area. [Figure 5] FIG. 10 is a table showing multiple types of feature amounts calculated using three-dimensional coordinates, which are one of the attributes of a representative point of interest. [Figure 6] (a) is a model diagram that shows a schematic representation of a "buffer zone" set near the boundary of the search area, and (b) is a model diagram that shows a schematic representation of measurement points within the area of ​​interest related to the buffer zone that are included in the search area and measurement points that are not included in the search area. [Figure 7] FIG. 2 is a flowchart showing the main processing flow of the measurement point classification device of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0018] An example of an embodiment of the measurement point classification device of the present invention will be described with reference to the drawings. The present invention is capable of classifying measurement points by assigning them various classes. For convenience, the following example will be described using an example in which the classes are "feature types," i.e., an example in which measurement points are classified by feature type.

[0019] 1 is a block diagram showing the main components of a measurement point classification device 100 of the present invention. As shown in this figure, the measurement point classification device 100 of the present invention comprises an area of ​​interest setting means 101, a grouping means 102, a representative point setting means 103, a search area setting means 104, a feature calculation means 105, and a class setting means 106, and can also comprise a measurement point cloud storage means 107, a feature type storage means 108, etc.

[0020] Each of the means constituting the measurement point classification device 100 (particularly, the area of ​​interest setting means 101 to the class setting means 106) can be manufactured as a dedicated means, or a general-purpose computer device can be used. That is, the computer device executes arithmetic processing according to a predetermined program, thereby performing processing specific to each means. This computer device is equipped with a processor such as a CPU, memories such as ROM and RAM, and some also include input means such as a mouse and keyboard, and a display, and can be configured, for example, as a personal computer (PC) or a server.

[0021] The measurement point cloud storage means 107 and feature type storage means 108 can be configured using a storage device of a general-purpose computer (for example, a personal computer), or can be configured on a database server. When configured on a database server, they can be placed on a local network (LAN: Local Area Network), or can be configured as a cloud server that stores data via the Internet.

[0022] Below, each of the main elements that make up the measurement point classification device 100 of the present invention will be described in detail.

[0023] (Means for storing measurement points and means for storing feature types) The measurement point cloud storage means 107 is a means for storing multiple measurement points (i.e., a measurement point cloud) obtained by photogrammetry (e.g., SfM), laser measurement (e.g., MMS), or measurement using TS. The measurement points are represented by so-called three-dimensional coordinates consisting of plane coordinates and height information. These coordinates are expressed as "latitude and longitude" or "X coordinate and Y coordinate," and height information from a predetermined reference horizontal plane, such as elevation. The feature type storage means 108 is a means for storing multiple pre-defined feature types, such as structures (e.g., electric wires, utility poles, roads, bridges), slopes, buildings (e.g., office buildings), and road facilities. As will be described later, the present invention estimates feature types based on the feature amounts of measurement points. The feature types can be estimated from the feature amounts by machine learning using a combination of the feature amounts and feature types of the measurement points as training data, or by rules established based on the relationship between the feature amounts of the measurement points and the feature types. Therefore, in cases where feature types are estimated by setting this rule, the feature type storage means 108 should store a table in which feature amounts correspond to feature types (hereinafter referred to as a "class table").

[0024] (Method for setting an area of ​​interest and grouping) The region of interest setting means 101 is a means for selecting a "measurement point of interest PMf" and further setting a "region of interest RF." Below, the processing up to the point where the region of interest setting means 101 sets the region of interest RF will be described in detail with reference to Fig. 2. Fig. 2 is a model diagram that schematically shows the measurement point of interest PMf and the region of interest RF.

[0025] The region of interest setting means 101 first reads out the measurement point cloud from the measurement point cloud storage means 107, and selects one measurement point PM from the measurement point cloud as a "measurement point of interest PMf." As will be described later, the region of interest setting means 101 repeatedly selects measurement points of interest PMf, and the selection order can be determined according to the attributes of the measurement points PM. For example, the selection can be in order of earliest measurement time or ascending order of X coordinate.

[0026] Once the measurement point of interest PMf has been selected, a predetermined region based on the measurement point of interest PMf, i.e., the "region of interest RF," is set. This region of interest RF is set based on predetermined requirements, and can be set, for example, as a sphere of a predetermined radius (e.g., 10 cm) centered on the measurement point of interest PMf. Alternatively, the region of interest RF can be set as a cube, or as a circle centered on the measurement point of interest PMf after projecting the measurement point PM onto a plane (e.g., a horizontal or vertical plane). As long as the measurement point of interest PMf is used as the base, the region of interest RF can be set in various shapes and sizes.

[0027] When the region of interest setting means 101 sets the region of interest RF, the grouping means 102 sets a "measurement point set" by grouping the measurement points PM included in the region of interest RF. In other words, the grouping means 102 sets the measurement point of interest PMf and the measurement points PM in the vicinity of the measurement point PMf as a measurement point set. For convenience, the measurement points PM included in the measurement point set set by the grouping means 102, that is, the measurement points PM selected as the measurement point set, will be referred to as "processed measurement points PMt (black points in Figure 2)," and the measurement points PM excluding the processed measurement points PMt will be referred to as "unprocessed measurement points PMu (white points in Figure 2)."

[0028] Once the grouping means 102 has set the measurement point set, the region of interest setting means 101 further selects a new measurement point of interest PMf and sets a region of interest RF based on the measurement point of interest PMf. However, the region of interest setting means 101 selects the measurement point of interest PMf from the measurement points PM that have not yet been selected as part of the measurement point set, i.e., from unprocessed measurement points PMu, among the measurement point group. Then, once the region of interest setting means 101 has set the region of interest RF, the grouping means 102 again sets the measurement points PM included in the region of interest RF as a new measurement point set. However, the grouping means 102 sets the measurement point set using only unprocessed measurement points PMu.

[0029] In this way, the region of interest setting means 101 selects measurement points of interest PMf one after another to set the region of interest RF, and the grouping means 102 sequentially sets measurement point sets accordingly. For example, in Figure 3(a), by repeatedly setting the region of interest RF and measurement point sets from left to right, many measurement points PM are set as processed measurement points PMt (black points). As shown in this figure, the measurement point of interest PMf for the region of interest RF being set this time (the rightmost region of interest RF in the figure) is selected from the unprocessed measurement points PMu (white points).

[0030] 3(b), when the currently set region of interest RF includes a previously processed measurement point PMt (in the figure, a previously processed measurement point PMt of the measurement point set set previously), the measurement point set is set using the measurement points PM excluding the previously processed measurement point PMt, i.e., only the unprocessed measurement points PMu (white point 9 in the figure). The series of processes consisting of "selection of the target measurement point PMf and setting of the region of interest RF" by the region of interest setting means 101 to "setting of the measurement point set" by the grouping means 102 should be repeated until all measurement points PM become previously processed measurement points PMt (in other words, until there are no unprocessed measurement points PMu among the measurement points PM).

[0031] (Representative point setting means) The representative point setting means 103 is a means for setting a "representative point PR" for each set of measurement points (i.e., for each region of interest RF). When the representative point setting means 103 sets the representative point PR, it can directly set any one of the measurement points PM (pre-processed measurement points PMt) that make up the set of measurement points as the representative point PR. For example, as shown in FIG. 3(b), it is possible to directly set the target measurement point PMf as the representative point PR, or to set the measurement point PM with the earliest measurement time as the representative point PR, or to set the measurement point PM with the lowest altitude as the representative point PR.

[0032] Alternatively, new measurement points PM can be generated by statistically processing the attributes of multiple measurement points PM that make up the measurement point set, and these measurement points PM can be set as representative points PR. For example, measurement points PM related to new coordinates can be generated using the average or median of each coordinate that makes up the three-dimensional coordinates, and these measurement points PM can be set as representative points PR. In this case, attributes other than coordinates (such as RGB and reflection intensity) can also be set to statistically processed values ​​such as averages and medians.

[0033] (Search area setting means and feature calculation means) The search area setting means 104 is a means for selecting a "representative point of interest PRf" and setting a "search area RS." The processing up to the point where the search area setting means 104 sets the search area RS will be described in detail below with reference to Fig. 4. Fig. 4 is a model diagram that schematically shows the representative point of interest PRf and the search area RS.

[0034] The region-of-interest setting means 101 first selects one representative point PR as a "representative point of interest PRf" from among the representative points PR set by the representative point setting means 103. After selecting the representative point of interest PRf, the region-of-interest setting means 101 sets a predetermined region based on the representative point of interest PRf, i.e., a "search region RS." This search region RS is set based on predetermined requirements, and can be set, for example, as a sphere of a predetermined radius (e.g., 1.0 m) centered on the representative point of interest PRf. Alternatively, the search region RS can be set as a cube, or as a circle centered on the representative point of interest PRf after projecting the representative point PR onto a plane (e.g., a horizontal or vertical plane). As long as the representative point of interest PRf is used as the base, the search region RS can be set in various shapes and sizes. However, the search region RS is set to be larger than the region of interest RF. In other words, the search region RS is set to include multiple regions of interest RF, as shown in FIG. 4.

[0035] Once the search area setting means 104 sets the search area RS, the feature amount calculation means 105 calculates the "feature amount" of the target representative point PRf. Specifically, the means focuses on the search area RS related to the target representative point PRf for which the feature amount is to be found, extracts a representative point PR (representative point PR number 8 in FIG. 4) included in the search area RS, and calculates the feature amount of the target representative point PRf based on the attributes of the extracted representative point PR. In other words, the feature amount calculation means 105 calculates the feature amount for the target representative point PRf and representative points PR in the vicinity of the target representative point PRf.

[0036] As the feature, various values ​​can be adopted as long as they can be calculated based on the attributes (coordinates, RGB, reflection intensity, etc.) of the representative point PR. For example, FIG. 5 shows multiple types of feature calculated using three-dimensional coordinates, which are one of the attributes of the target representative point PRf. That is, the feature shown includes "three-dimensional eigenvalues" obtained by performing principal component analysis on the target representative point PRf, and feature values ​​such as "linearity," "planarity," and "point cloud density" calculated based on these eigenvalues. Note that the feature calculation means 105 can be configured to calculate one type of feature (e.g., linearity only), or can be configured to calculate two or more types of feature (e.g., planarity and perpendicularity).

[0037] By setting a search region RS, representative points PR near the target representative point PRf are extracted, and feature values ​​are calculated based on these representative points PR. However, because the representative point PR originates from the target region RF, as shown in FIG. 6, even if the representative point PR itself is included in the search region RS, a portion of the target region RF may be outside the search region RS. Alternatively, even if the representative point PR is located outside the search region RS, a portion of the target region RF may be included in the search region RS. In this case, it is recommended to set a "buffer zone BZ" near the boundary of the search region RS, as shown in FIG. 6(a). The process of feature value calculation by the feature value calculation means 105 after setting the buffer zone BZ will be described in detail below.

[0038] First, a buffer zone BZ is set near the boundary of the search region RS. For example, if the search region RS is set as a sphere of radius R centered on the representative point of interest PRf, the space between a sphere of radius Rr, which is slightly shorter than the radius of the search region RS, and a sphere of radius R+r, which is slightly longer than the radius of the search region RS, can be set as the buffer zone BZ. In this case, the "length r," which increases or decreases with respect to the radius of the search region RS, should be set to the same value as the default radius (e.g., 10 cm) when the region of interest RF was set.

[0039] When a buffer zone BZ is set, representative points PR within the buffer zone BZ are extracted, and attention is paid to measurement points PM within the region of interest RF associated with the representative points PR. Then, as shown in Figure 6(b), among the measurement points PM, those within the search region RS are extracted, and feature quantities are calculated based on the attributes of the extracted measurement points PM. In other words, when a buffer zone BZ is set, feature quantities are calculated based on representative points PR within the search region RS that are located inside the buffer zone BZ (e.g., representative points PR within "radius Rr"), and on measurement points PM (i.e., a set of measurement points) within the region of interest RF associated with the buffer zone BZ (more specifically, the region of interest RF associated with the representative points PR included in the buffer zone BZ) that are within the search region RS.

[0040] A series of processes from "selection of target representative point PRf and setting of search region RS" by the search region setting means 104 to "calculation of feature amount" by the feature amount calculation means 105 may be repeatedly executed for all representative points PR.

[0041] (Class setting method) The class setting means 106 is a means for assigning a class (here, a feature type) to the representative point of interest PRf according to the feature amount. When the class setting means 106 assigns a feature type to the representative point of interest PRf according to the feature amount, it can use, for example, machine learning or a rule base. When using machine learning, first, machine learning is performed using training data consisting of a combination of feature amounts and feature types to generate a "trained model," and the feature amount is input into the trained model, and the resulting feature type can be assigned to the representative point of interest PRf. On the other hand, when using a rule base, the feature amount can be obtained by querying the "class table (a correspondence table between feature amounts and feature types)" described above for the feature amount, and the resulting feature type can be assigned to the representative point of interest PRf.

[0042] (Processing flow) The main processing of the measurement point classification device 100 of the present invention will be described in detail below with reference to Fig. 7. Fig. 7 is a flow chart showing the flow of the main processing of the measurement point classification device 100 of the present invention. In Fig. 7, the action to be performed is shown in the center column, what is necessary for that action is shown in the left column, and what results from that action is shown in the right column.

[0043] When classifying measurement points using the measurement point classification device 100 of the present invention, first, as shown in Fig. 7, the region of interest setting means 101 reads out the measurement point cloud from the measurement point cloud storage means 107 and selects one measurement point PM from the measurement point cloud as a "measurement point of interest PMf" (Step 201 in Fig. 7). Next, the region of interest setting means 101 sets a "region of interest RF" based on the measurement point of interest PMf (Step 202 in Fig. 7).

[0044] When the region of interest RF is set by the region of interest setting means 101, the grouping means 102 sets the measurement points PM included in the region of interest RF as a "measurement point set" (Step 203 in FIG. 7), and the representative point setting means 103 sets a "representative point PR" for the measurement point set (Step 204 in FIG. 7). At this time, if unprocessed measurement points PMu still remain (Yes in Step 205 in FIG. 7), a series of processes consisting of "selection of target measurement points PMf (Step 201)" to "setting of representative points PR (Step 204)" are repeatedly executed. On the other hand, if unprocessed measurement points PMu do not remain (No in Step 205 in FIG. 7), the process proceeds to the subsequent process.

[0045] When all measurement points PM are set as processed measurement points PMt, the search area setting means 104 selects one representative point PR from the plurality of representative points PR as a "representative point of interest PRf" and sets a "search area RS" based on the representative point of interest PRf (Step 206 in FIG. 7). When the search area RS is set by the search area setting means 104, the feature amount calculation means 105 calculates the "feature amount" of the representative point of interest PRf (Step 207 in FIG. 7). Then, the class setting means 106 assigns a feature type to the representative point of interest PRf according to the feature amount of the representative point of interest PRf. Note that the series of processes from "setting the search area RS (Step 206)" to "assigning a feature type (Step 208)" are repeatedly executed for all representative points PR. [Industrial Applicability]

[0046] The measurement point classification device of the present invention can be used to estimate various types of features, such as structures such as electric wires and utility poles, roads and bridges, slopes, buildings such as office buildings, road facilities, etc. Considering that the present invention provides spatial information about features in an ideal manner and ultimately leads to advanced maintenance and management of social infrastructure, it can be said to be an invention that can be expected to not only be used industrially but also make a great contribution to society. [Explanation of symbols]

[0047] 100 Measurement point classification device of the present invention 101 (measurement point classification device) area of ​​interest setting means 102 Grouping means (of measuring point classification device) 103 (Measuring point classification device) representative point setting means 104 (measurement point classification device) search area setting means 105 (measurement point classification device) feature calculation means 106 Class setting means (for measuring point classification device) 107 (measurement point classification device) measurement point group storage means 108 (measurement point classification device) feature type storage means BZ Buffer Zone PM measurement points PMf Measurement point of interest PMt Processed measurement point PMu Raw measurement point PR representative point PRf Representative point of interest RF Region of Interest RS search area

Claims

1. An apparatus for classifying a plurality of measurement points using a measurement point group, the apparatus comprising: an area-of-interest setting means for selecting one of the measurement points from the group of measurement points as a "measurement point of interest" and setting an "area of ​​interest" that is an area based on the measurement point of interest; a grouping means for setting the measurement point of interest and the measurement points included in the region of interest related to the measurement point of interest as one "measurement point set"; a representative point setting means for setting a "representative point" relating to the set of measurement points based on the set of measurement points; a search area setting means for selecting one representative point from the plurality of representative points as a "representative point of interest" and setting a "search area" that is an area based on the representative point of interest; a feature value calculation means for calculating a feature value of the representative point of interest related to the search area based on an attribute of the representative point included in the search area; a class setting means for assigning a class corresponding to the feature amount from among a plurality of classes set in advance to the representative point of interest, the search area is an area larger than the area of ​​interest, the region-of-interest setting means sets the measurement points excluding the measurement points included in the set of measurement points that have already been set as "unprocessed measurement points," and then repeatedly selects the measurement points of interest from the unprocessed measurement points; the grouping means repeatedly sets the measurement point set by targeting only the unprocessed measurement points. A measurement point classification device characterized by:

2. when the representative point is located near the boundary of the search area, the feature amount calculation means calculates the feature amount while taking into account attributes of the measurement points that are located inside the search area among the measurement points included in the measurement point set related to the representative point.

2. The measurement point classification device according to claim 1.

3. the representative point setting means sets the target measurement point as the representative point; 2. The measurement point classification device according to claim 1.

4. the representative point setting means determines a new measurement point by statistically processing attributes of the measurement points included in the measurement point set, and sets the measurement point as the representative point.

2. The measurement point classification device according to claim 1.

5. The class is a type of feature.

2. The measurement point classification device according to claim 1.

Images