JP7774458B2 - Road edge heightening device - Google Patents

Description

This invention relates to technology for assigning height information such as elevation to features, and more specifically to a road edge height assignment device that determines the height of a road edge (hereinafter referred to as "road edge height") by performing statistical processing of height information near the road edge.

Recently, there has been a growing demand for topographical information (spatial information). For example, an increasing number of managers are requesting spatial information about facilities, such as their shape and location, in order to more effectively manage facilities installed on or along roads. At the same time, advanced maintenance and management of social infrastructure (hereinafter simply referred to as "social infrastructure") is considered an important issue in realizing Society 5.0, which is currently being promoted jointly by the public and private sectors. Furthermore, as autonomous driving technology becomes more widely used, there is a strong demand from many quarters for various spatial information about roads, including road edges (road boundary lines).

Traditionally, two-dimensional (2D) planar drawings (floor plans), such as topographical maps, have been the norm for showing spatial information. Although plan views sometimes show "height information" such as contour lines and endpoint elevations, their primary focus is on showing planar positions, making it difficult to grasp the target area as a three-dimensional (3D) space. However, in recent years, advances in measurement technology have made it easy to acquire large amounts of three-dimensional measurement points (hereafter referred to as "3D point clouds"), and advances in information technology have also made it easier to handle these three-dimensional point clouds.

For example, to obtain a 3D point cloud of terrain including roads, measurement methods such as aerial photogrammetry, airborne laser measurement, terrestrial laser measurement, and MMS (Mobile Mapping System) are used. Among these, MMS is a mobile vehicle equipped with sensors such as a laser scanner, a camera, a satellite positioning system (GNSS: Global Navigation Satellite System) for acquiring its own position, an IMU (Inertial Measurement Unit), and odometry, allowing the laser scanner to obtain a 3D point cloud while moving along the roadway.

The 3D point cloud obtained by MMS or similar is typically used as a 3D model of the terrain being measured (hereafter referred to as a "3D terrain model"). This 3D terrain model represents the target terrain using 3D coordinates, and is a terrain model typified by DSM (Digital Surface Model) and DEM (Digital Elevation Model).

Typically, a 3D terrain model is composed of multiple small regions obtained by dividing the target planar area. These small regions, also known as meshes, are formed by dividing the area into, for example, orthogonal grids, and each small region has a representative point. Because the 3D point clouds obtained by measurement are often random data (data with an irregular arrangement on a plane), geometric calculations are often used to assign heights to the representative points of the small regions. This calculation method includes the TIN (Triangulated Irregular Network) method, which calculates height using an irregular triangulation network formed from random data, the nearest neighbor method, which uses the nearest laser measurement point, as well as inverse distance weighting (IDW), the Kriging method, and the averaging method.

3D terrain models allow users to grasp the target terrain in both two and three dimensions, making them more versatile than flat maps. However, 3D terrain models based on measurement results only present spatial information based on three-dimensional coordinates, and are unable to show the attributes of features. In other words, simply by looking at a 3D terrain model, it is not possible to understand where the edges of roads are or where the outer edges (so-called edges) of office buildings are.

Adding feature attribute information to a 3D terrain model requires a survey of the features. This means extracting feature attributes by visually examining aerial photographs, or having workers personally visit the site and record the visual information on a map; in either case, human judgment is required. However, since roads, for example, generally have a considerable length, the amount of work required for the survey is enormous, and considering the effort and time involved, it requires significant costs.

As mentioned above, traditionally, plan views have been the primary method of data analysis. In some cases, these plan views are used as raster or vector data (i.e., digitized), and in other cases, features are converted into shapes (polylines or polygons) and then given attribute information. Furthermore, recent advances in machine learning technology have made it possible to mechanically (automatically) extract features from aerial photographs and plan views, and then extract their shapes and attribute information. In this way, it is conceivable that a separate "2D terrain model" is available that does not contain elevation or other height information, but does contain feature attribute information. Using such a 2D terrain model can eliminate (or significantly reduce) the need for manual surveying of features when adding feature attribute information to a 3D terrain model.

Patent Document 1 therefore proposes a technology for generating three-dimensional feature shape lines by assigning elevations to feature shape lines represented in two-dimensional map data using point clouds obtained by measurement.

Japanese Patent Application Laid-Open No. 2020-013351

As mentioned above, 3D terrain models merely present spatial information based on three-dimensional coordinates, and it is not possible to determine where the road edge is simply by looking at the 3D terrain model. As a result, assigning road edge heights requires manual work, such as visually processing aerial photographs, which results in significant costs. Therefore, we suggest the possibility of reducing manual work by using "2D terrain models" that digitize plan views such as raster data and vector data.

For example, a two-dimensional terrain model (hereinafter referred to as a "road edge model") having attribute information for "road edge" and its planar position (coordinate information) could be placed on a 3D terrain model, and then component points located around the road edge model could be extracted from the three-dimensional points (hereinafter referred to as "component points") that make up the 3D terrain model and used as the road edge height. However, there are often buildings other than roads, such as houses and office buildings, located around road edges, and there are cases where a component point cannot immediately be used as the road edge height simply because it is located around the road edge model.

The objective of the present invention is to solve the problems inherent in conventional technology, namely, to provide a road edge height assigning device that can assign road edge heights after extracting road surface height information.

This invention is based on the unprecedented idea that, when a region of interest is set based on a road edge model, the pattern of height information for multiple component points differs for each region of interest, and height information equivalent to the target statistical value set for each region of interest is assigned as the road edge height.

The road edge height assignment device of the present invention comprises a three-dimensional terrain model storage means, a road edge model storage means, a region of interest setting means, a target statistical value setting means, and a road edge height calculation means. The three-dimensional terrain model storage means stores a three-dimensional terrain model based on multiple component points having three-dimensional coordinates, and the road edge model storage means stores a two-dimensional road edge model representing the road edge. The region of interest setting means generates multiple reference points of interest on the road edge model and sets a two-dimensional region of interest having a predetermined size and shape based on the reference points of interest. The target statistical value setting means sets a target statistical value. The road edge height calculation means extracts multiple component points belonging to the region of interest when the region of interest is placed on the three-dimensional terrain model as reference points of interest and calculates, as the road edge height, height information for the reference points of interest that correspond to the target statistical value. The road edge height is then assigned to the reference points of interest corresponding to the region of interest.

In the road edge height estimation device of the present invention, the target statistical value setting means may include a learning means and a target statistical value output means. The learning means generates a target statistical value setting model for outputting target statistical values by machine learning multiple types of learning datasets, and the target statistical value output means outputs target statistical values for the input dataset based on the input input dataset and the target statistical value setting model. The learning dataset includes appropriate statistical values set for the learning region of interest and elevation information for multiple constituent points included in the learning region of interest, and the input dataset includes elevation information for multiple constituent points included in the input region of interest. The learning region of interest is a two-dimensional region having the same size and shape as the region of interest, and the input region of interest is a two-dimensional region based on a target reference point and having the same size and shape as the region of interest. The learning means generates a target statistical value setting model by machine learning the appropriate statistical values included in the learning dataset and the elevation information for the multiple constituent points, and the target statistical value output means outputs target statistical values using the elevation information for the multiple constituent points included in the input dataset as input values.

The road edge height assignment device of the present invention can also be configured such that the training dataset includes elevation information for component points corresponding to a plurality of predetermined selection statistical values among the plurality of component points, and the input dataset includes elevation information for component points corresponding to the selection statistical values among the plurality of component points. The selection statistical values can be percentile values calculated based on the elevation information for the plurality of component points.

The road edge height assignment device of the present invention may further include a region of interest division means and a road side region setting means. The region of interest division means divides the region of interest into a plurality of divided regions using the road edge model as a boundary, and the road side region setting means calculates statistical values of height information for a plurality of component points belonging to each divided region as a divided region representative height, and sets the divided region with the lowest divided region representative height as the road side region. In this case, the road edge height calculation means extracts a plurality of component points belonging to the road side region as target component points and then calculates the road edge height.

The road edge height providing device of the present invention has the following effects.
(1) Because the road edge model is used, road edge height can be automatically assigned, which reduces the burden on the operator (worker), reduces the cost of data creation work, and contributes to reducing so-called human error.
(2) Since it is possible to avoid erroneously extracting the height of roofs and the like around the road edge, it is possible to assign a road edge height that is appropriate to the current situation.
(3) The target statistical value setting model can select the optimum target statistical value for the target area, that is, can assign a road edge height that is more appropriate for the target area.
(4) Road edges that are appropriate to the current situation are assigned, meaning that a 3D terrain model that is appropriate to the current situation is obtained, which allows for more advanced management of road facilities, etc. and provides map information that is more useful for automated driving.

1 is a block diagram showing the main configuration of a road edge height providing device according to the present invention; (a) is a planar model diagram that schematically shows focus reference points generated at regular intervals on a road edge model by a focus area setting means, and (b) is a planar model diagram that schematically shows a focus area set as a circle centered on the focus reference point. FIG. 10 is a graph showing the distribution of elevations of constituent points arranged around a point on the road edge. A model diagram that schematically shows a training dataset, training reference points, training areas, and training construction points. FIG. 1 is a step diagram showing a schematic procedure for generating a training dataset. 1 is a model diagram showing a schematic diagram of an input dataset, input reference points, input area, and input configuration points. FIG. 4 is a planar model diagram schematically showing a first divided area and a second divided area set by dividing the region of interest by the region of interest dividing means; FIG. 4 is a planar model diagram schematically showing a first divided region, a second divided region, a third divided region, and a fourth divided region set by the region-of-interest dividing means; FIG. 4 is a flowchart showing an example of a main process flow until the road edge height assignment device outputs a target statistical value. FIG. 4 is a flowchart showing an example of a main process flow until the road edge height calculation device calculates the road edge height.

An example of the road edge height adjustment device of the present invention is explained below with reference to the drawings.

One of the technical features of the road edge height assignment device of the present invention is that it sets multiple reference points (hereinafter referred to as "target reference points") at appropriate intervals on the road edge, sets a predetermined area (hereinafter referred to as "target area") based on the target reference points, and then selects, from among the points (hereinafter referred to as "target points") included in this target area, height information relating to a target point (hereinafter referred to as "specific target point") that satisfies specific conditions as the road edge height. The present invention employs a specific statistical value (hereinafter referred to as "target statistical value") as a condition for selecting a specific target point from among multiple target points, and furthermore, sets this target statistical value uniquely for each target area. Note that height information can refer to various concepts that represent relative height differences from a certain reference height, but for convenience, the height information will be described here as "elevation."

Figure 1 is a block diagram showing the main components of the road edge height assignment device 100 of the present invention. As shown in this figure, the road edge height assignment device 100 is configured to include a region of interest setting means 101, a target statistical value setting means 102, a road edge height calculation means 103, a 3D terrain model storage means 108, and a road edge model storage means 109, and can also be configured to include a learning means 104, a target statistical value output means 105, a region of interest division means 106, a road side region setting means 107, a learning dataset storage means 110, and a target statistical value setting model storage means 111.

The road edge height estimation device 100 comprises the focus area setting means 101, target statistical value setting means 102, road edge height calculation means 103, learning means 104, target statistical value output means 105, focus area division means 106, and road side area setting means 107. These can be manufactured as dedicated components, or general-purpose computers can be used. These computers include a processor such as a CPU, memory such as ROM and RAM, and may also include input means such as a mouse and keyboard, and display means such as a display. They can be configured, for example, as a personal computer (PC) or server.

The 3D terrain model storage means 108, road edge model storage means 109, learning dataset storage means 110, and target statistical value setting model storage means 111 can use the storage device of a general-purpose computer (e.g., a personal computer), or can be built on a database server. If built on a database server, it can be placed on a local network (LAN: Local Area Network), or it can be a cloud server that stores data via the Internet.

The following provides a detailed explanation of each of the main elements that make up the road edge height adjustment device 100 of the present invention.

1. 3D terrain model storage means and road edge model storage means The 3D terrain model storage means 108 is a means for storing 3D terrain models, and the road edge model storage means 109 is a means for storing "road edge models." Here, the "road edge model" is, as described above, two-dimensional data that includes attribute information of "road edges" and their planar positions (coordinate information), and is digitized road edge data such as polylines.

2. Area of Interest Setting Means The area of interest setting means 101 is a means for generating a plurality of reference points of interest on a road edge model. The area of interest setting means 101 can automatically generate reference points of interest SP at regular intervals (or irregular intervals) on the road edge model EL as shown in Figure 2(a), or can set reference points of interest SP at arbitrary positions by an operator's operation. Furthermore, the positions of reference points of interest SP that have been generated can also be changed by an operator's operation.

The region of interest setting means 101 also serves to set a "region of interest RG." This region of interest RG is a region based on the reference point of interest SP, and is a two-dimensional region (planar region) with a predetermined size (dimensions) and shape. For example, as shown in Figure 2(b), the region of interest RG can be a circle with a predetermined radius (predetermined radius) centered on the reference point of interest SP, or it can be an ellipse or polygon with predetermined dimensions centered on the reference point of interest SP.

3. Target Statistical Value Setting Means As described above, the road edge height estimation device 100 of the present invention selects a target component point SP corresponding to a target statistical value from among the component points included in the target region RG (i.e., the target component points SP) as a specific target component point, and sets the height information (here, altitude) possessed by this specific target component point as the road edge height. The target statistical value setting means 102 is a means for setting a target statistical value for selecting the specific target component point. Here, for example, a percentile value when the altitudes are sorted from smallest (lowest) (hereinafter referred to as a "lowest altitude percentile value") can be used as the target statistical value. That is, the target statistical value is set to N (a positive real number, not limited to an integer), such as the 5th percentile or the 10th percentile. Of course, the target statistical value is not limited to percentile values, and various statistical values can be used, such as simply setting the Mth (natural number) lowest elevation as the target statistical value, or setting the elevation closest to the average elevation minus a certain height as the target statistical value.

The inventors discovered that the elevation combination (pattern) of multiple constituent points located around a certain point on a road edge (e.g., a target reference point SP) tends to vary depending on the surrounding environment of the target reference point SP. For example, flat road edges, road edges on viaducts, and road edges with nearby street trees each produce different elevation patterns of the surrounding constituent points, while the elevation patterns of flat road edges show a consistent trend. Figure 3 is a graph showing the distribution of elevations of constituent points located around points on a road edge, with the horizontal axis representing the lowest elevation percentile and the vertical axis representing the elevation. As can be seen from this graph, for flat road edges without surrounding street trees or buildings, a relatively large percentile value (e.g., 5th to 10th percentile) can be used as the target statistical value. Conversely, for road edges with surrounding street trees or buildings, a relatively small percentile value (e.g., 2nd to 3rd percentile) is required. Therefore, in the present invention, a unique target statistical value is set for each region of interest.

The target statistical value can be input for each area of interest by the operator operating the target statistical value setting means 102. Alternatively, the target statistical value setting means 102 can automatically (mechanically) set the target statistical value according to the area of interest. In this case, the road curb height estimation device 100 of the present invention can be configured to include a learning means 104, a target statistical value output means 105, a learning dataset storage means 110, and a target statistical value setting model storage means 111. Below, the specification for the target statistical value setting means 102 automatically (mechanically) setting the target statistical value is described in detail.

(Learning Data Set Storage Means and Target Statistical Value Setting Model Storage Means)
The learning dataset storage means 110 is means for storing a "learning dataset," and the target statistical value setting model storage means 111 is means for storing a "target statistical value setting model." This learning dataset is so-called teacher data for the learning means 104 to perform machine learning, as will be described later, and naturally, a large number of learning datasets are prepared. On the other hand, the target statistical value setting model is a model obtained as a result of machine learning performed by the learning means 104, as will also be described later, and when an "input dataset" is input to this target statistical value setting model, it outputs (returns) a target statistical value.

(Learning tools)
The learning means 104 is a means for generating a "target statistical value setting model" by performing machine learning (e.g., deep learning, etc.) using a training dataset as training data. Here, the training dataset includes the elevations and appropriate statistical values of the component points NP, as shown in FIG. 4 . However, as shown in this figure, the component points NP targeted by the training dataset are component points NP (hereinafter, specifically referred to as "training component points LCP") that belong to (are included in) the training area LRG. Note that "belonging to the training area LRG" here means that when the component points NP are planarly projected onto the target area RG, the component points NP are included within the training area LRG. Furthermore, the training area LRG is set to a predetermined size and shape based on a point on the road edge as a training reference point LSP. For example, in FIG. 4 , the training area LRG is set to be a circle with a predetermined radius (a predetermined radius) centered on the training reference point LSP on the road edge. Of course, the learning area LRG can be set in various shapes, not just circular, such as a polygon or ellipse, or the same shape as the divided area described below (the area obtained by dividing the target area RG by the road edge) (semicircle in the example of Figure 7), and it may also be set in two or more types of shapes.

Learning regions LRG are prepared for various environments (flat road edges, road edges with nearby viaducts and buildings, road edges on viaducts with nearby buildings, etc.), and many are prepared for each environment. Therefore, it is a good idea to specify many learning reference points LSP for each characteristic environment and create many learning regions LRG for each characteristic environment.

Figure 5 is a step diagram that schematically illustrates the procedure for generating a training dataset. To generate a training dataset, first, training reference points LSP are designated, as shown in Figure 5(a). At this time, the operator can designate the desired training reference points LSP using a pointing device (mouse, touch panel, pen tablet, touchpad, trackpad, trackball, etc.) or keyboard while visually viewing the 3D terrain model displayed on a display means (such as a monitor), or the training reference points LSP can be set automatically (mechanically) using image recognition technology, etc. Once the training reference points LSP are designated, a training area LRG is set based on the training reference points LSP, as shown in Figure 5(b). Furthermore, training composition points LCP included in the training area LRG are extracted, and their elevations are set as training data, as shown in Figure 5(c).

The appropriate statistical value included in the training data set is a statistical value that corresponds to an elevation value that is appropriate for the training region LRG, or in other words, a correct answer. As with the target statistical value, the appropriate statistical value can be a lowest elevation percentile value, or various other statistical values. However, the same statistical value is used for the appropriate statistical value and the target statistical value. For example, if the lowest elevation percentile value is used as the target statistical value, the lowest elevation percentile value is also used as the appropriate statistical value. The appropriate statistical value can be a statistical value (e.g., 5th percentile) related to the elevation of one training point LCP selected from multiple training points LCP included in the training region LRG, or a statistical value related to a value (e.g., mean, median, mode, etc.) obtained as a result of statistical processing of the elevations of these training points LCP. The obtained appropriate statistical value is set as training data, for example, by operator operation. By performing machine learning on multiple training data sets prepared in this way, the learning means 104 generates a target statistical value setting model.

(Target statistical value output means)
The target statistical value output means 105 is a means for outputting target statistical values related to an input dataset based on the input dataset and the target statistical value setting model generated by the learning means 104. The input dataset includes the elevations of the composition points NP as shown in FIG. 6. However, as shown in this figure, the composition points NP targeted by the input dataset are composition points NP that belong to (are included in) the region of interest RG (i.e., the composition points CP of interest). Note that the region of interest RG is based on the reference point of interest SP generated by the region of interest setting means 101 as described above. However, it is preferable that the size and shape of the region of interest RG (hereinafter referred to as "region specifications") be the same as the region specifications for setting the learning region LRG. 4, the region of interest RG may also be set as a circle having the same diameter as the learning region LRG and centered on the reference point of interest SP, or, if the learning region LRG is set as a semicircle like the divided regions described below, the region of interest RG may also be set as a semicircle having the same diameter as the learning region LRG and centered on the reference point of interest SP. In either case, the input data set is generated by extracting the constituent points of interest CP included in the region of interest RG set by the region of interest setting means 101 and setting the elevations associated with these constituent points of interest CP.

When an input dataset is input, the target statistical value output means 105 outputs target statistical values related to the input dataset using a target statistical value setting model. Specifically, it compares the elevation combinations (patterns) of the target configuration points CP included in the input dataset with the learned elevation patterns, selects an approximate (or matching) elevation pattern, and outputs target statistical values associated with that elevation pattern.

(Elevation of the dataset)
Up to this point, it has been explained that the training dataset includes the elevations of multiple training composition points LCP, and the input dataset includes the elevations of multiple focus composition points CP. Of course, it is also possible to include the elevations of all training composition points LCP in the training dataset, and the elevations of all focus composition points CP in the input dataset. Alternatively, it is also possible to extract training composition points LCP and focus composition points CP based on predetermined requirements and include them in the respective datasets. More specifically, training composition points LCP and focus composition points CP are extracted according to predetermined statistical values (hereinafter referred to as "selection statistical values"), and only the elevations of the extracted training composition points LCP are included in the training dataset, and only the elevation of the extracted focus composition point CP is included in the input dataset. For example, percentile values sorted in descending order of elevation are used, and percentile values in 5% increments, such as the 0th percentile (the minimum elevation in this case), 5th percentile, 10th percentile, ..., and 100th percentile, are set as selection statistics (i.e., 21 levels of selection statistics), and only the elevations of the learning composing points LCPs and the target composing points CP that correspond to these selection statistical values (0th percentile, 5th percentile, ...) from among all the learning composing points LCPs and the target composing points CP are included in each dataset. Note that the selection statistical values for extracting the learning composing points LCPs and the selection statistical values for extracting the target composing points CP should preferably be set to the same values. In this way, by standardizing the multiple elevations included in the learning dataset and the input dataset, it becomes easier to match the learned elevation pattern with the elevation pattern of the input dataset.

4. Region of Interest Dividing Means The region of interest dividing means 106 is means for dividing the region of interest RG using the road edge model EL as a boundary, thereby setting a plurality of regions (hereinafter referred to as "divided regions RS"). For example, in FIG. 7, the region of interest dividing means 106 divides the circular region of interest RG in half to set a first divided region RS1 and a second divided region RS2. Note that the region of interest RG shown in FIG. 7 is a circle centered on the reference point of interest SP, and the road edge model EL naturally passes through the reference point of interest SP, so the first divided region RS1 and the second divided region RS2 are each semicircular (with a central angle of 180°) with the same shape and area. Of course, if the road edge model EL is bent, the semicircular regions will not have the same area, and fan-shaped divided regions with different areas will be set. If the road edge model EL is bent at multiple points within the region of interest RG, the divided regions will not be fan-shaped. Furthermore, if the region of interest RG is not circular, the divided regions will not be fan-shaped either. In either case, the region of interest RG is divided by the road edge model EL.

5. Road Side Area Setting Means As shown in Figure 7, when a region of interest RG is placed on a 3D topographical model, several component points NP (i.e., component points of interest CP) belong to (are included in) that region of interest RG. Then, by using the elevation of these component points of interest CP, the elevation of the reference point of interest SP related to the region of interest RG can be found.

Usually, one side of a road edge is a road, and the other is a non-road area (hereinafter referred to as the "general area"). Therefore, when the focus area RG is divided into two divided areas RS (first divided area RS1 and second divided area RS2) as shown in Figure 7, it is desirable to select one of them as a road and then determine the elevation of the focus reference point SP based on the focus configuration point CP that belongs to the divided area RS designated as the road (hereinafter referred to specifically as the "road side area").

The road side area setting means 107 is a means for setting road side areas from among the multiple divided areas RS. The processing procedure by the road side area setting means 107 for setting road side areas will be explained in detail below. First, the road side area setting means 107 calculates an elevation that represents the divided area RS (hereinafter referred to as the "divided area representative height") based on the elevations of the multiple focus component points CP that belong to each divided area RS. At this time, the divided area representative height can be the median value of the elevations of the multiple focus component points CP, or another statistical value such as the average value, mode, or lowest elevation percentile of the elevations of the focus component points CP, or can simply be the minimum value of the elevations of the focus component points CP.

Once the divided area representative height for each divided area RS is obtained, the road-side area setting means 107 selects one divided area RS from those divided areas RS. Generally, road-side areas have lower elevations than general areas. Therefore, the road-side area setting means 107 extracts the divided area representative height that shows the smallest (lowest) value, selects the divided area RS associated with that extracted divided area representative height, and sets it as the road-side area.

In the example of Figure 7, two divided regions RS (first divided region RS1 and second divided region RS2) are set by the region of interest dividing means 106, but the region of interest dividing means 106 may also set three or more divided regions RS. For example, Figure 8 shows a so-called T-junction intersection, and the region of interest RG is set by a circle with a relatively large radius. In this case, the region of interest dividing means 106 divides the region of interest RG using three road edge models EL, thereby setting four divided regions RS (first divided region RS1, second divided region RS2, third divided region RS3, and fourth divided region RS4).

When three or more divided areas RS are set, the road-side area setting means 107 can calculate the divided area representative height for each divided area RS, extract the divided area representative height with the smallest (lowest) value, and set the divided area RS associated with the extracted divided area representative height as the road-side area. Alternatively, the road-side area setting means 107 can extract a divided area RS that borders the focus reference point SP (hereinafter referred to as a "candidate divided area") from the three or more divided areas RS, and then set the road-side area. As can be seen from Figure 8, it is clear that the divided area RS that borders the focus reference point SP is a road-side area. Therefore, when three or more divided areas RS are set, the divided area RS that borders the focus reference point SP is extracted as a "candidate divided area," and the road-side area is set by selecting from these candidate divided areas. In the example of Figure 8, four divided areas RS (first divided area RS1 to fourth divided area RS4) have been set, and the road-side area setting means 107 extracts the first divided area RS1 and second divided area RS2, which are adjacent to the target reference point SP, as candidate divided areas, calculates the divided area representative heights for these candidate divided areas (i.e., the first divided area RS1 and the second divided area RS2), extracts the divided area representative height that indicates the smallest elevation, and then sets the road-side area (the second divided area RS2 in Figure 8).

6. Road Edge Height Calculation Means The road edge height calculation means 103 calculates the "road edge height" based on the road side area set by the road side area setting means 107. The processing procedure by which the road edge height calculation means 103 calculates the road edge height will be described in detail below. First, the road edge height calculation means 103 extracts a component point NP (i.e., a target component point CP) belonging to the road side area. Then, the road edge height is calculated using the elevation of the extracted target component point CP. More specifically, from the elevations of a plurality of target component points CP, a target component point CP having an elevation corresponding to the target statistical value set by the target statistical value setting means 102 is selected as a specific target component point, and the elevation of this specific target component point is set as the road edge height. The road edge height obtained here is assigned to a target reference point SP related to the corresponding target area RG.

So far, we have explained the processing procedure by which the road edge height calculation means 103 calculates the road edge height based on road side areas selected from multiple divided areas RS. However, the road edge height calculation means 103 can also be configured to target all of the focus points CP belonging to the focus area RG, that is, to select an elevation corresponding to the target statistical value from the elevations of all of the focus points CP belonging to the focus area RG. In other words, in this case, the process of dividing the focus area RG and setting divided areas RS, or the process of setting road side areas, is not performed, and the road edge height is simply calculated using the focus points CP belonging to the focus area RG. Therefore, in this case, the focus area division means 106 and road side area setting means 107 can be omitted.

7. Processing Flow The main processing of the road edge height estimation device 100 will be described in detail below with reference to Figures 9 and 10. Figure 9 is a flow diagram showing an example of the main processing flow until the road edge height estimation device 100 outputs the target statistical value, and Figure 10 is a flow diagram showing an example of the main processing flow until the road edge height estimation device 100 calculates the road edge height. In these flow diagrams, the processing to be performed is shown in the center column, the items necessary for that processing are shown in the left column, and the items resulting from that processing are shown in the right column.

As shown in Figure 9, the process leading up to outputting the target statistical values can be broadly divided into a "learning step" leading up to generating a target statistical value setting model, and an "output step" leading up to actually setting the target statistical values for the input dataset. In the learning step, a learning reference point LSP is first specified to create a learning dataset (Step 211 in Figure 9). At this time, the learning reference point LSP can be specified by an operator, or the learning reference point LSP can be set automatically (mechanically) using image recognition technology, etc.

Once the learning reference point LSP is set, a learning region of interest LRG is generated based on the learning reference point LSP and in accordance with predetermined region specifications (Step 212 in Figure 9). Next, the elevations of the learning configuration points LCP included in this learning region of interest LRG are set, and appropriate statistical values for the learning region of interest LRG are set (Step 213 in Figure 9), thereby creating a learning dataset (Step 214 in Figure 9). By repeating this series of processes (Steps 211 to 214 in Figure 9), numerous learning datasets are prepared for each of various roadside environments. These learning datasets are then trained by the learning means 104 to generate a target statistical value setting model (Step 215 in Figure 8).

Once the target statistical value setting model has been generated, the road edge model EL is read from the road edge model storage means 109 as shown in FIG. 10, and the focus area setting means 101 generates multiple focus reference points SP on the road edge model EL (Step 231 in FIG. 10). Furthermore, the focus area setting means 101 sets a focus area RG based on the focus reference points SP (Step 232 in FIG. 10).

Once the focus reference point SP has been generated and the focus region RG has been set, the target statistical values for that focus region RG are output in the output step shown in FIG. 9. More specifically, first, the focus composing points CP included in the focus region RG are extracted, and the elevations for these focus composing points CP are set to generate an input dataset (Step 221 in FIG. 9). Then, the target statistical value output means 105 outputs the target statistical values for the input dataset based on the input input dataset and the target statistical value setting model (Step 222 in FIG. 9).

Furthermore, once the region of interest RG is set, the region of interest division means 106 sets multiple divided regions RS (Step 233 in Figure 10). Then, the road side region setting means 107 calculates a divided region representative height for each divided region RS (Step 234 in Figure 10) and sets the road side region based on this divided region representative height (Step 235 in Figure 10).

Once the road side area is set, the road edge height calculation means 103 extracts a focus point CP belonging to this road side area (Step 236 in FIG. 10). The target statistical value setting means 102 then sets the value output by the target statistical value output means 105 as the target statistical value (Step 237 in FIG. 10). The road edge height calculation means 103 then selects a focus point CP having an elevation equivalent to the target statistical value from among the elevations of multiple focus points CP belonging to the road side area as a specific focus point, and sets the elevation of this specific focus point as the road edge height (Step 238 in FIG. 10). Note that the series of processes from setting the focus area RA (Step 232) to calculating the road edge height (Step 238) can be performed for all focus reference points PA, or can be performed for a specific section (e.g., intersection to intersection) as a unit. In this case, this series of processes (Steps 232 to 238) can be performed on only one representative reference point PA of interest within a specific section to set a target statistical value, and that target statistical value can then be applied to other reference points PA of interest within that specific section. Alternatively, target statistical values can be set by performing this process on all or some of the reference points PA of interest within a specific section, and the target statistical value with the greatest number (for example, the 5th percentile) can be set as the overall target statistical value.

The road curb height adjustment device of the present invention is particularly suitable for use in managing various facilities, including road infrastructure, and as map information for autonomous driving. Furthermore, the present invention can provide highly accurate step information that is useful for elderly people and wheelchair users, and can also be effectively used in disaster prevention planning. Therefore, the road curb height adjustment device of the present invention is not only applicable to industry, but is also expected to make a significant contribution to society.

100 Road edge height assignment device of the present invention 101 Area of interest setting means (of road edge height assignment device) 102 Target statistical value setting means (of road edge height assignment device) 103 Road edge height calculation means (of road edge height assignment device) 104 Learning means (of road edge height assignment device) 105 Target statistical value output means (of road edge height assignment device) 106 Area of interest division means (of road edge height assignment device) 107 Road side area setting means (of road edge height assignment device) 108 3D terrain model storage means (of road edge height assignment device) 109 Road edge model storage means (of road edge height assignment device) 110 Learning dataset storage means (of road edge height assignment device) 111 Target statistical value setting model storage means (of road edge height assignment device) CP Point of interest configuration point EL Road edge model LCP Learning configuration point LRG Learning area of interest LSP Learning reference point NP Composition point RG Region of interest RS Split region SP Reference point of interest

Claims (3)

a three-dimensional terrain model storage means for storing a three-dimensional terrain model based on a plurality of constituent points having three-dimensional coordinates;
road edge model storage means for storing a two-dimensional road edge model representing a road edge;
an area-of-interest setting means for generating a plurality of reference points of interest on the road edge model and setting a two-dimensional area of interest having a predetermined size and shape based on the reference points of interest;
a target statistical value setting means for setting a target statistical value;
a road edge height calculation means for extracting a plurality of said constituent points belonging to said region of interest as constituent points of interest when said region of interest is arranged on said three-dimensional topographical model, and for calculating height information relating to said constituent points of interest that correspond to said target statistical value among said constituent points of interest as a road edge height,
The target statistical value setting means
a learning means for generating a target statistical value setting model for outputting the target statistical value by machine learning a plurality of learning data sets;
a target statistical value output means for outputting the target statistical value related to the input data set based on the input input data set and the target statistical value setting model,
the learning data set includes an appropriate statistical value set for a learning region of interest and height information of a plurality of the constituent points included in the learning region of interest,
the input data set includes height information of the plurality of constituent points included in the input region of interest,
the learning region of interest is a two-dimensional region that is based on the reference point of interest and has the same size and shape as the region of interest,
the input region of interest is a two-dimensional region that is based on the reference point of interest and has the same size and shape as the region of interest,
assigning the road edge height to the target reference point corresponding to the target area;
A road edge height providing device. the learning data set includes height information of the constituent points corresponding to a plurality of predetermined selection statistical values among the plurality of constituent points,
the input data set includes height information of the constituent points corresponding to the selection statistical value among the plurality of constituent points,
the selection statistical value is a percentile value calculated based on height information of a plurality of the constituent points;
2. The road edge height providing device according to claim 1 . a three-dimensional terrain model storage means for storing a three-dimensional terrain model based on a plurality of constituent points having three-dimensional coordinates;
road edge model storage means for storing a two-dimensional road edge model representing a road edge;
an area-of-interest setting means for generating a plurality of reference points of interest on the road edge model and setting a two-dimensional area of interest having a predetermined size and shape based on the reference points of interest;
a target statistical value setting means for setting a target statistical value;
a region-of-interest dividing means for dividing the region of interest into a plurality of divided regions using the road edge model as a boundary;
a road-side area setting means for calculating a statistical value of height information relating to the plurality of component points belonging to the divided area as a divided area representative height, and setting the divided area relating to the lowest divided area representative height as a road-side area;
a road edge height calculation means for extracting a plurality of said constituent points belonging to said road side area as constituent points of interest when said road side area is arranged on said three-dimensional terrain model, and for calculating height information relating to said constituent points of interest that correspond to said target statistical value among said constituent points of interest as a road edge height,
assigning the road edge height to the target reference point corresponding to the target area;
A road edge height providing device.