Pipeline estimation system, pipeline estimation method, and pipeline estimation program

Abstract

The location of buried pipes in geospatial space can be easily and accurately estimated from ground-penetrating radar data. [Solution] A pipeline estimation system for estimating the location of pipelines laid underground, comprising: a vertex detection unit that detects the vertex position of buried object reactions in ground-penetrating radar data; a shape parameter calculation unit that calculates shape parameters representing the shape of the buried object reactions; a buried object reaction extraction unit that outputs an extracted image including the buried object reactions extracted for each shape parameter; a feature quantity extraction unit that calculates the feature quantities of the extracted image; a geospatial position calculation unit that calculates the geospatial position of the vertex position; a clustering unit that calculates detection points having the feature quantities in geospace and calculates clustered detection points to which pipeline labels are assigned according to the results of clustering using the similarity of the feature quantities of the detection points; and a pipeline conversion unit that calculates an estimated pipeline from the clustered detection points for each pipeline label.

Description

Technical Field

[0001] The present invention relates to a pipeline estimation system that estimates the position of a pipeline buried underground by analyzing ground penetrating radar data obtained by scanning an underground exploration apparatus along a plurality of survey lines. Background Art

[0002] Buried pipes such as water pipes, gas pipes, and pipelines accommodating power cables are laid underground. In order to perform maintenance and inspection of existing pipelines and lay new pipelines, it is necessary to excavate the target area after grasping the accurate laying status (current situation) of underground buried objects in geospatial space. Grasping the current situation is also necessary when drawing up a construction plan.

[0003] The current situation of existing pipelines can be estimated by referring to drawings such as construction drawings and equipment ledgers owned by each business entity, etc. However, when the latest drawings are not available, or there is a discrepancy between the information described in the drawings and the actual situation, if excavation work is performed based on inaccurate current situation information, the existing pipelines may be damaged. In addition, when an unexpected buried object is found, or there is no buried object at the location described in the drawing, additional work such as removal of the buried object, relocation of existing pipelines, and review of the construction plan may be required, which causes problems such as construction delay and increased construction costs.

[0004] In order to prevent these problems caused by inaccurate grasping of the current situation, buried objects in the relevant area are sometimes investigated in advance during construction or planning. Ground penetrating radar exploration is one of the investigation methods for buried objects. In ground penetrating radar exploration, an underground exploration apparatus that transmits and receives electromagnetic waves is scanned along a survey line on the ground, and the position of an underground buried object in geospatial space is specified from the obtained ground penetrating radar data. Conventionally, analysis of ground penetrating radar data requires a skilled worker to check the reaction of buried objects in the ground penetrating radar data, which requires a great deal of labor.

[0005] In recent years, automated technologies for detecting buried object reactions in ground penetrating radar data and creating current situation maps in geospatial space have been developed.

[0006] The following prior art exists as background technology for this field. Patent Document 1 (International Publication No. 2022 / 264341) discloses a position estimation device that includes: a candidate point extraction unit that extracts candidate points for the location of buried objects from each of the two-dimensional ground-penetrating radar data obtained by measuring buried objects underground; a two-dimensional processing unit that inputs each of the ground-penetrating radar data into a recognition unit that has been pre-trained to output a recognition score indicating the likelihood of a buried object, and outputs a recognition score indicating the likelihood of a buried object corresponding to the candidate point; and a three-dimensional integration unit that integrates each of the ground-penetrating radar data using a graph structure based on the recognition score and relative position information of the ground-penetrating radar data to estimate the three-dimensional position of the buried object.

[0007] Furthermore, Patent Document 2 (JP 2023-128876 A) discloses a buried pipe location estimation system comprising: a radar exploration device that moves on the ground, irradiates radio waves into the ground and receives radio waves reflected from underground buried objects; an exploration position measuring device that measures the position of the radar exploration device; and a buried pipe location estimation device that estimates the buried location of buried pipes. The buried pipe location estimation device includes a buried location estimation unit that estimates the position and depth of the underground buried object based on the position measured by the exploration position measuring device and the intensity pattern of the radio waves received by the radar exploration device; and a buried pipe determination unit that determines whether the underground buried object is linearly continuous based on the position and depth of the underground buried object calculated by the buried location estimation unit, and determines that underground buried objects determined to be linearly continuous are buried pipes. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] International Publication No. 2022 / 264341 [Patent Document 2] Japanese Patent Publication No. 2023-128876 [Overview of the Initiative] [Problems that the invention aims to solve]

[0009] Patent Document 1 discloses a technique for detecting candidate points of buried objects from a set of two-dimensional ground-penetrating radar data, and estimating the location of the buried object using a graph structure based on a recognition score indicating the likelihood of being a buried object calculated by a machine learning model and the three-dimensional position of the candidate points. However, when multiple buried pipes are intersected, multiple candidate points all have high recognition scores, making it difficult to distinguish individual buried pipes and resulting in the challenge of accurate location estimation.

[0010] Patent Document 2 discloses a technique for estimating the three-dimensional position of buried objects from ground-penetrating radar data, estimating the type of buried object based on the pattern of the ground-penetrating radar data using a learning model, and determining whether the buried object is a buried pipe based on its position and type. However, ground-penetrating radar data includes reactions from underground structures such as geological layers and cavities in addition to the buried object whose type is to be estimated, and there is a problem in that it is difficult to identify the type of buried object when reactions originating from the buried object to be estimated and reactions originating from other underground structures are detected close together.

[0011] One of the objectives of the present invention is to provide a pipeline estimation system that can easily and accurately estimate the location of buried pipes in geospatial space from ground-penetrating radar data. [Means for solving the problem]

[0012] A representative example of the invention disclosed in this application is as follows: A pipeline estimation system for estimating the location of pipelines laid underground, comprising a computer having a computing unit that performs predetermined processing and a storage unit connected to the computing unit, the system includes: a vertex detection unit that takes as input ground-penetrating radar data obtained by scanning a ground-penetrating exploration device that irradiates the ground with electromagnetic waves along a plurality of survey lines and recording reflected waves from buried objects, detects the vertex position of buried object reactions in the ground-penetrating radar data and outputs the detected vertex position; a shape parameter calculation unit that takes as input the ground-penetrating radar data and the vertex position, calculates a shape parameter representing the shape of the buried object reaction for each vertex position, and outputs the calculated shape parameter; a buried object reaction extraction unit that takes as input the ground-penetrating radar data and the shape parameter, extracts the buried object reaction for each shape parameter, and outputs an extracted image including the extracted buried object reaction; and buried objects in the extracted image. The system is characterized by comprising: a feature extraction unit that calculates the feature quantities of the extracted image using a model for calculating feature quantities relating to the appearance of the object reaction and outputs the calculated feature quantities; a geospatial position calculation unit that calculates the geospatial position of the vertex position using the vertex position and the positioning information of the ground-penetrating radar data as input and outputs the calculated geospatial position; a clustering unit that calculates detection points having the feature quantities in geospace using the feature quantities as input, performs clustering using the similarity of the feature quantities of the calculated detection points, calculates clustered detection points to which pipeline labels are assigned according to the results of the clustering, and outputs the calculated clustered detection points; and a pipeline conversion unit that calculates an estimated pipeline from the clustered detection points for each pipeline label using the clustered detection points as input and outputs the calculated estimated pipeline. [Effects of the Invention]

[0013] According to one aspect of the present invention, the location of buried pipes in geospatial space can be easily and accurately estimated from ground-penetrating radar data. Problems, configurations, and effects other than those described above will be clarified by the following description of embodiments. [Brief explanation of the drawing]

[0014] [Figure 1] It is a functional block diagram of the pipe estimation system of Embodiment 1. [Figure 2] It is a functional block diagram of the pipe estimation system of Embodiment 2. [Figure 3] It is a functional block diagram of the pipe estimation system of Embodiment 3. [Figure 4] It is a hardware configuration diagram of the pipe estimation system of Embodiment 1. [Figure 5A] It is a diagram showing an example of a buried object response. [Figure 5B] It is a diagram showing an example of a buried object response. [Figure 6A] It is a diagram showing calculation of an extracted image by the buried object response extraction unit of Embodiment 1. [Figure 6B] It is a diagram showing calculation of an extracted image by the buried object response extraction unit of Embodiment 1. [Figure 6C] It is a diagram showing calculation of an extracted image by the buried object response extraction unit of Embodiment 1. [Figure 6D] It is a diagram showing calculation of an extracted image by the buried object response extraction unit of Embodiment 1. [Figure 6E] It is a diagram showing calculation of an extracted image by the buried object response extraction unit of Embodiment 1. [Figure 7] It is a diagram showing an example of feature quantities of Embodiment 1. [Figure 8] It is a diagram showing an example of detection points of Embodiment 1. [Figure 9] It is a diagram showing an example of clustered detection points of Embodiment 1. [Figure 10] It is a diagram showing an example of survey lines and detection points in geospatial space. [Figure 11] It is a diagram showing clustering processing of Embodiment 3. [Figure 12A] It is a flowchart of clustering processing of Embodiment 3. [Figure 12B] It is a flowchart of clustering processing of Embodiment 3. [Figure 13] It is a diagram showing an example of a cost matrix of Embodiment 3. [Figure 14] This figure shows the clustering results for Example 3. [Figure 15A] This figure shows an example of a piping method using the piping section of Example 1. [Figure 15B] This figure shows an example of a piping method using the piping section of Example 1. [Figure 15C] This figure shows an example of a piping method using the piping section of Example 1. [Figure 16A] This figure shows an example of a piping method using the piping section of Example 1. [Figure 16B] This figure shows an example of a piping method using the piping section of Example 1. [Modes for carrying out the invention]

[0015] Embodiments of the present invention will be described below with reference to the drawings. The following description and drawings are illustrative for illustrating the present invention, and have been omitted and simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms.

[0016] The positions, sizes, shapes, and ranges of the components shown in the drawings may not represent their actual positions, sizes, shapes, and ranges in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the positions, sizes, shapes, and ranges disclosed in the drawings.

[0017] When there are multiple components with the same or similar function, they may be described using the same symbol but with different subscripts. However, if it is not necessary to distinguish between these multiple components, the subscripts may be omitted in the description.

[0018] Furthermore, while the following explanation may describe the processes performed by executing a program, the processor (e.g., CPU (Central Processing Unit), GPU (Graphics Processing Unit)) executes the defined processes, using memory resources (e.g., memory) and / or interface devices (e.g., communication ports) as appropriate; therefore, the processor may be the primary driver of the processing. Similarly, the primary driver of the processing performed by executing a program may be a controller, device, system, computer, or node that has a processor. The primary driver of the processing performed by executing a program may be any arithmetic unit, and may include dedicated circuits that perform specific processing (e.g., FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)).

[0019] A program may be installed from its program source into a device such as a computer. The program source may be, for example, a program distribution server or a computer-readable storage medium. If the program source is a program distribution server, the program distribution server includes a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to other computers. Furthermore, in the following description, two or more programs may be implemented as a single program, or a single program may be implemented as two or more programs.

[0020] Furthermore, while the following explanation may describe a system composed of a single computer, similar functions may be achieved by distributing all or part of the computer's functions across one or more computers, such as a cloud, and communicating with each other via a network.

[0021] <Example 1> Figure 1 is a block diagram of the pipeline estimation system S010 of Example 1, and Figure 4 is a hardware configuration diagram of the pipeline estimation system S010 of Example 1.

[0022] As shown in Figure 1, the pipeline estimation system S010 of Embodiment 1 includes a vertex detection unit P010, a shape parameter calculation unit P020, a buried object reaction extraction unit P030, a feature quantity extraction unit P040, a geospatial position calculation unit P050, a clustering unit P060, and a pipeline generation unit P070.

[0023] The pipeline estimation system S010 may run on a server H010 with the hardware configuration shown in Figure 4. Server H010 consists of a general-purpose computer having information processing resources such as a processor H020 (CPU), memory H030, storage H040, and a communication interface H050.

[0024] Processor H020 is a computing device that executes programs stored in memory H030. By executing various programs, processor H020 enables the functionality of each functional unit of the pipeline estimation system S010 (vertex detection unit P010, shape parameter calculation unit P020, buried object reaction extraction unit P030, feature extraction unit P040, geospatial position calculation unit P050, clustering unit P060, pipeline generation unit P070, etc.). Note that some of the processing performed by processor H020 by executing programs may be executed by other computing devices (e.g., hardware such as ASICs or FPGAs).

[0025] Memory H030 includes ROM, a non-volatile memory element, and RAM, a volatile memory element. ROM stores immutable programs (e.g., BIOS). RAM is a high-speed, volatile memory element such as DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by processor H020 and data used during program execution.

[0026] Storage H040 is a high-capacity, non-volatile storage device such as a magnetic storage device (HDD) or flash memory (SSD). Storage H040 also stores data used by processor H020 during program execution, as well as the programs executed by processor H020. For example, storage H040 stores the pipeline estimation program M010, which causes the computer to execute the pipeline estimation method, and allows the information processing resource to execute it. That is, the program is read from storage H040, loaded into memory H030, and executed by processor H020, thereby realizing the various functions of the pipeline estimation system S010.

[0027] The H050 communication interface is a network interface device that controls communication with other devices according to a predetermined protocol.

[0028] The pipeline estimation system S010 may have an input / output interface. The input / output interface is an interface to which input devices such as a keyboard and mouse are connected, which receive input from the operator, and to which output devices such as a display device and a printer are connected, which output the program execution results in a format that the operator can see. In addition, terminal devices connected to the pipeline estimation system S010 may provide the input and output devices. In this case, the pipeline estimation system S010 may have the functionality of a web server, and the terminal device may access the pipeline estimation system S010 using a predetermined protocol (e.g., http).

[0029] The program executed by processor H020 is provided to the pipeline estimation system S010 via removable media (such as CD-ROM or flash memory) or a network, and stored in the non-volatile storage H040, which is a non-temporary storage medium. For this reason, the pipeline estimation system S010 should have an interface for reading data from the removable media.

[0030] The pipeline estimation system S010 is a computer system that operates on a single physical computer, or on multiple computers configured logically or physically, and may also operate on a virtual computer built on multiple physical computing resources. For example, the multiple programs that implement the functions of the pipeline estimation system S010 may each operate on separate physical or logical computers, or multiple programs may be combined and operate on a single physical or logical computer.

[0031] The pipeline estimation system S010 takes ground-penetrating radar data D010 and positioning information D060 as inputs. Ground-penetrating radar data D010 is obtained by scanning a ground-penetrating survey device along multiple survey lines on the ground. It is desirable that the ground-penetrating radar data D010 is data obtained from surveys along multiple survey lines that cover the entire survey area to be understood. Furthermore, it is desirable that the multiple survey lines be parallel in order to efficiently cover the entire survey area and to simplify the analysis. Ground-penetrating radar data D010 may also be obtained by measuring the entire survey area with a single continuous survey line and then dividing the survey data into multiple parts according to the position and direction of the survey line.

[0032] Ground-penetrating radar data D010 can be treated as image data representing the intensity of reflected waves by plotting the delay time from when the ground-penetrating device transmits electromagnetic waves on the vertical axis, the scanning distance of the ground-penetrating device on the horizontal axis, and associating the intensity of the received signal at each measurement point on the survey line with brightness. Electromagnetic waves emitted from the ground-penetrating device are reflected at the boundaries of underground structures and media, and travel underground while attenuating, so received signals with large delay times generally have low intensity. To make it easier to detect buried object reactions, ground-penetrating radar data D010 may be subjected to gain restoration processing to increase the signal intensity in areas with large delay times, or processing to calculate the difference with surrounding received signals.

[0033] In ground-penetrating radar data D010, with scanning distance x, delay time t, and speed of light in a vacuum c0, the shape of the buried object reaction of a point reflector located at scanning distance x0 and depth d in soil with relative permittivity εr is expressed by equation (1).

[0034]

number

[0035] The vertex detection unit P010 of the pipeline estimation system S010 takes ground-penetrating radar data D010 as input and detects the vertex position D020 of the buried object response in the ground-penetrating radar data D010. In calculating the vertex position D020, migration processing may be applied to the ground-penetrating radar data D010 and points with a value above a predetermined threshold may be calculated as the vertex position D020; template matching may be performed using predetermined buried object responses and points with a degree of agreement above a predetermined threshold may be calculated as the vertex position D020; or an approximation curve of the buried object response may be calculated using points with high brightness in the ground-penetrating radar data D010 and the point with the smallest delay time among the calculated approximation curves may be calculated as the vertex position D020.

[0036] Alternatively, the vertex position D020 may be calculated using a machine learning model. The machine learning model may be an object detection model that outputs a bounding box near the vertex position, a segmentation model that outputs pixels near the vertex position, or an object recognition model that determines whether each small region obtained by dividing the ground-penetrating radar data D010 is the vertex position D020. When using an object detection model, segmentation model, or object recognition model, only the region near the vertex position may be calculated, and the accurate vertex position D020 may not be output. In this case, the centroid of the output region may be calculated as the vertex position D020, or the point with the strongest received signal within the output region may be calculated as the vertex position D020, or the vertex position D020 may be calculated using the aforementioned migration process, template matching, or approximation curve calculation on the output region.

[0037] The shape parameter calculation unit P020 takes ground-penetrating radar data D010 and vertex position D020 as input and outputs shape parameters D030 that represent the shape of the buried object reaction. In calculating the shape parameters D030, an approximation curve using a polynomial or quadratic curve is calculated using points with high received intensity extracted from the ground-penetrating radar data D010 near the vertex position D020, and the parameters of the approximation curve are calculated as the shape parameters D030. Alternatively, migration processing is performed using multiple candidate relative permittivity values, and the scanning distance x0 and depth d in equation (1) are calculated from the relative permittivity εr at which the buried object reaction is most closely aggregated to a single point and the position of the aggregated point, and the shape parameters D030 are calculated from the relative permittivity εr, scanning distance x0, and depth d. Furthermore, the processing of the vertex detection unit P010 and the shape parameter calculation unit P020 may be executed simultaneously. In other words, instead of calculating the vertex position D020 from the ground-penetrating radar data D010 and then calculating the shape parameter D030 of the buried object reaction near the vertex position D020, a migration process may be performed on the ground-penetrating radar data D010 to calculate the vertex position D020 and the shape parameter D030 simultaneously.

[0038] The buried object reaction extraction unit P030 takes ground-penetrating radar data D010 and shape parameters D030 as input, extracts buried object reactions from the ground-penetrating radar data D010 based on the shape parameters D030, and outputs the extracted portion as an extracted image D040. If electromagnetic waves irradiated from the ground-penetrating exploration device are reflected and received by only one buried object, the ground-penetrating radar data D010 will yield a single buried object reaction as shown in Figure 5A. However, if there are multiple buried objects, geological layers, cavities, pebbles, etc. in the ground, the ground-penetrating radar data D010 will contain multiple buried object reactions as shown in Figure 5B. In order to estimate pipelines, it is necessary to identify individual buried pipes from these reactions, and when detecting a certain buried pipe, reactions caused by other buried objects, etc., become noise. If the ground-penetrating radar data D010 contains a lot of noise, the accuracy of the features extracted in the buried pipe feature calculation described later will decrease. Therefore, it is desirable to extract only the buried object response of interest, and then mask or cut out the other regions before extracting the features.

[0039] The buried object response in ground-penetrating radar data D010 is an elongated curve as shown in Figure 5A. Therefore, it is desirable that the extracted region be a shape that follows the curve of the buried object response, rather than a rectangle like a bounding box. Figures 6A to 6E show an example of calculating the extracted image D040 by cutting out the buried object response according to the shape parameter D030. The buried object response at vertex position D020 is assumed to be in the form t=f(x) in the region x1≦x≦x2 and t1≦t≦t2 using the shape parameter D030, as shown in Figure 6A. The extracted region is calculated as the region enclosed by curves t=f(x)-k1 and t=f(x)+k2, obtained by translating the curve t=f(x) by k1 in the negative direction of the delay time axis and k2 in the positive direction, respectively, in the region x1≦x≦x2. Outside the region to be extracted, a masking process is performed to replace all values ​​of the ground-penetrating radar data D010 with zeros, thereby calculating an extracted image D040 with reduced noise effects, as shown in Figure 6B. Furthermore, as shown in Figure 6C, unnecessary regions are removed by cutting out the region x1≦x≦x2 and t1-k1≦t≦t2+k2 from the extracted image D040 to obtain the extracted image D040, thereby reducing the influence of regions outside the extraction area in the feature extraction described later.

[0040] The shape parameter D030 for buried pipes changes depending on the laying direction, depth, pipe diameter, and relative permittivity of the surrounding soil, but the change due to the material of the buried pipe or the material inside the buried pipe is small. Therefore, when identifying differences in buried object reactions due to differences in the material of the buried pipe or the material inside the buried pipe, as shown in Figure 6D, each observation point of the ground-penetrating radar data D010 can be moved parallel to the delay time axis at each scanning distance x so that the buried object reactions have the same delay time, and as shown in Figure 6E, a rectangular region with a scanning distance width of x2-x1 and a delay time width of k1+k2 containing the buried object reactions can be extracted and the extracted image D040 can be calculated.

[0041] The feature extraction unit P040 takes the extracted image D040 as input and calculates the feature vector D050 of the extracted image D040. For example, the feature vector D050 is calculated as a one-dimensional vector by inputting the extracted image D040 into a neural network that has learned the relationship between radar images and features, and outputting the output of the intermediate or final layer. Figure 7 shows an example of the feature vector D050. For example, the feature vector D050 is constructed as table data in which each row contains the feature vector dimN of each extracted image D040. The feature vector D050 may be reduced in dimensionality by methods such as PCA or t-SNE. In addition, the feature vector D050 may include the value of the shape parameter D030 corresponding to the buried object reaction.

[0042] The machine learning model used to compute the features should preferably be trained to distinguish between different buried pipes. The model could be a recognition model trained through supervised learning that outputs the type of buried pipe (water pipe, gas pipe, power cable, etc.), whether the material of the buried pipe is metal, whether it contains liquid inside, or whether the pipe thickness is greater than a predetermined value; or a representation learning model trained through self-supervised learning that reduces the dimensionality of input data with an encoder and then reconstructs it with a decoder; or a model trained to distinguish whether two images originate from the same buried pipe or from different buried objects, given extracted image D040, extracted image D040 with noise such as rectangular noise or scaling applied, and extracted image D040 of different buried pipes.

[0043] When the output of the intermediate layer of a model is used as feature D050, or when the output of the final layer of a model trained without using teacher labels is used as feature D050, it is difficult to interpret the meaning of the values ​​in each dimension of feature D050. However, when the output of the final layer of a model trained using teacher labels is used as D050, the meaning of each dimension can be interpreted, and explainability can be improved. For example, by using feature D050 as the output of a model that determines what type of buried pipe an extracted image D040 is, whether it is a metal pipe, whether it contains liquid inside, and whether the pipe thickness is greater than a predetermined value, it becomes difficult to distinguish between buried pipes with the same label, but the explainability of feature D050 can be improved.

[0044] The geospatial position calculation unit P050 takes the vertex position D020 and the positioning information D060 from the ground-penetrating radar data D010 as input and calculates the geospatial position D070. The geospatial position D070 is the coordinate of the vertex position D020 in three-dimensional space and can be expressed as (x,y,z) in the xyz coordinate system. The positioning information D060 represents the position information of the survey line of the ground-penetrating exploration device and may be position information output from a positioning system such as GNSS mounted on the ground-penetrating exploration device, or it may be the position of the ground-penetrating exploration device measured by surveying equipment such as a total station.

[0045] The electromagnetic wave propagation speed v in the ground is expressed by equation (2), using the relative permittivity εr of the ground and the speed of light c0 in a vacuum. The depth z at the apex position D020 can be calculated from the electromagnetic wave propagation speed v, which is calculated from equation (2) assuming the relative permittivity εr of the ground, and the delay time at the apex position D020.

[0046]

number

[0047] The clustering unit P060 takes the feature quantity D050 and geospatial location D070 as input and outputs clustered detection points D090.

[0048] A point containing geospatial location D070 and feature quantity D050 is defined as a detection point. Figure 8 shows an example of a detection point. For example, detection point D090 is constructed as table data where each row contains the geospatial location x, y, z and feature quantity dimN of the detection point.

[0049] Clustering that assigns pipeline labels indicating which pipeline a detected point belongs to can be calculated using the similarity of the feature vector D050 of the detected points. Clustering may also be performed using the positional relationship of the geospatial position D070 of the detected points and the similarity of the feature vector D050 of the detected points. The similarity of the feature vector D050 can be calculated, for example, as cosine similarity using equation (3) for two feature vectors a and b. Alternatively, the cosine distance can be calculated using equation (4) with respect to the cosine similarity, and clustering may be performed using the calculated cosine distance. In clustering using the positional relationship of the geospatial position D070 and the similarity of the feature vector D050 of the detected points, for example, the same pipeline label may be assigned to two detected points if the cosine similarity is above a predetermined threshold and the distance calculated from the geospatial position D070 is below a predetermined threshold, or the same pipeline label may be assigned to two detected points if the sum of the cosine distance and the distance calculated from the geospatial position D070 is below a predetermined threshold. By utilizing the similarity of feature D050, it is possible to identify detection points originating from different buried objects that are coincidentally located close together.

[0050]

number

number

[0051] Figure 9 shows an example of clustered detection point D090. In clustered detection point D090 shown in Figure 9, a "class" column has been added to the detection point shown in Figure 8. The "class" column records the pipeline label (class) assigned to each detection point as a result of clustering.

[0052] The pipeline generation unit P070 takes clustered detection points D090 as input and outputs an estimated pipeline D110 represented as a line segment. Figures 15A to 15C show an example of the pipeline generation method. First, as shown in Figure 15A, point clouds belonging to the same class are extracted from the clustered detection points D090. Next, as shown in Figure 15B, an approximate line A060 of the extracted clustered detection points D090 is calculated. Subsequently, a perpendicular line is set from the extracted clustered detection points D090 to the approximate line, and the foot of the perpendicular line A070 is set. As shown in Figure 15C, the estimated pipeline D110 is calculated as a line segment whose endpoints are the two outermost points A080 of the feet of the perpendicular line on the approximate line.

[0053] If the pipeline includes bends, the pipeline modeling unit P070 will represent the estimated pipeline D110 as a single line segment, resulting in a large discrepancy between the actual pipeline and the estimated pipeline D110. Figure 16A shows an example where the estimated pipeline D110 of a clustered detection point D090 of a pipeline including bends is represented as a single line segment. Therefore, if the coefficient of determination of the approximation line is below a predetermined threshold, the pipeline modeling unit P070 may further divide the extracted clustered detection point D090 into multiple classes and represent the estimated pipeline D110 by a combination of multiple approximation line segments calculated based on the divided classes. For example, one division point is selected from the extracted clustered detection point D090, and the extracted clustered detection point D090 is divided into two by a plane perpendicular to the approximation line and passing through the division point. A class is assigned to each, and a line is fitted to the clustered detection point D090 assigned the same class. The average value of the coefficient of determination during the fitting of the two lines is calculated. This calculation is repeated by changing the division point, and the division point that maximizes the mean of the coefficient of determination is selected. Then, the estimated pipeline D110 is calculated as a line segment whose endpoints are the two outermost points among the feet of the perpendiculars drawn from the division point and the clustered detection point D090 to the fitted line. Figure 16B shows an example in which the estimated pipeline D110 of the clustered detection point D090 of a pipeline including a bend is represented as two line segments.

[0054] According to this embodiment, the user can easily and accurately estimate the location of buried pipes in geospatial space from ground-penetrating radar data D010.

[0055] <Example 2> Next, Embodiment 2 of the present invention will be described. In Embodiment 2, the differences from Embodiment 1 described above will be mainly explained, and the same components and processes as in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted.

[0056] As shown in Figure 2, the pipeline estimation system S011 of Embodiment 2 includes a vertex detection unit P010, a shape parameter calculation unit P020, a buried object reaction extraction unit P030, a feature quantity extraction unit P040, a geospatial position calculation unit P051, a clustering unit P060, a pipeline visualization unit P070, and a relative permittivity estimation unit P080.

[0057] The relative permittivity estimation unit P080 estimates the relative permittivity of the medium in the ground (e.g., soil). For example, if the shape parameter D030 is expressed by equation (1) which includes the relative permittivity εr, the relative permittivity εr may be extracted from equation (1) and used as the relative permittivity D100. If the shape parameter D030 is expressed by a polynomial or a parabolic parameter, the shape parameter D030 may be recalculated in the form of equation (1) and the relative permittivity D100 may be calculated. Alternatively, the relative permittivity D100 may be calculated from the shape represented by the shape parameter D030 using a machine learning model that has learned the relationship between the shape represented by the shape parameter D030 and the relative permittivity εr. Furthermore, the relative permittivity D100 may be calculated as a statistical quantity such as the median of the relative permittivity εr calculated for each of the multiple vertex positions D020, thereby reducing the influence of outliers.

[0058] The geospatial position calculation unit P051 takes the vertex position D020, the positioning information D060 from the ground-penetrating radar data D010, and the relative permittivity D100 as input and calculates the geospatial position D070. For example, using equation (2), it calculates the electromagnetic wave propagation speed v from the relative permittivity D100 calculated by the relative permittivity estimation unit P080, and calculates the depth z from the delay time of the vertex position D020.

[0059] According to this embodiment, by using the relative permittivity of the soil derived from ground-penetrating radar data D010, rather than a predetermined value, the geospatial position D070 (especially the depth) of the buried object can be calculated more accurately.

[0060] <Example 3> Next, Embodiment 3 of the present invention will be described. In Embodiment 3, the differences from Embodiment 1 described above will be mainly explained, and the same components and processes as in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted.

[0061] As shown in Figure 3, the pipeline estimation system S012 of Embodiment 3 includes a vertex detection unit P010, a shape parameter calculation unit P020, a buried object reaction extraction unit P030, a feature quantity extraction unit P040, a geospatial position calculation unit P050, a clustering unit P061, and a pipeline conversion unit P070. The clustering unit P061 takes feature quantities D050, geospatial positions D070, and positioning information D060 as inputs and outputs clustered detection points D090. By further inputting positioning information D060 in addition to feature quantities D050 and geospatial positions D070, the clustering unit P061 can calculate clustered detection points D090 with higher accuracy than the clustering unit P060 of Embodiment 1.

[0062] Figure 10 shows an example of a survey line A010 and detection point A020 ​​in geospatial space. For example, when a ground-penetrating survey device is scanned along a straight survey line to cross a straight pipeline, the vertex position D020 of the pipeline will be detected at most once along the survey line, and there will not be multiple detection points with the same pipeline label on the same survey line. By imposing this constraint, a more reasonable clustered detection point D090 can be obtained.

[0063] Figure 11 shows an example of clustering by moving the survey line of interest one by one. Based on positioning information D060, adjacent survey lines in geospace are sequentially considered and clustered, allowing for the calculation of the optimal combination of buried pipe A030 estimated from a previously considered detection point A021 on the previously considered survey line A011 and the target detection point A022 on the survey line A012. The combination of the estimated buried pipe A030 and the target detection point A022 can be calculated, for example, by using a first cost matrix based on the positions of the estimated buried pipe A030 and the target detection point A022, and a second cost matrix which is the cosine distance between a representative feature D050 of the estimated buried pipe and a feature D050 of the target detection point A022, and then calculating the combination that minimizes the cost matrix calculated by adding the first and second cost matrices. For the representative feature D050, you can use the feature D050 of the most recent detection point that constitutes the estimated buried pipe A030, or the average value of feature D050.

[0064] The first cost matrix is, for example, a matrix in which each column is assigned an estimated buried pipe A030 and each row is assigned a point of interest A022. The estimated buried pipe A030 is extended to include the survey line A012 of interest, and the intersection point with the plane perpendicular to the ground plane is determined as the estimated detection point A040. The distance between the point of interest A022 and the estimated detection point A040 is calculated to be an element of the matrix.

[0065] The second cost matrix is, for example, a matrix in which each column is assigned an estimated buried pipe A030 and each row is assigned an A022 to focus on. The matrix is ​​calculated such that the cosine distance between the feature quantity D050 of the last focused detection point A023 corresponding to the estimated buried pipe A030 and the feature quantity D050 of the detection point A022 to focus on becomes an element of the matrix.

[0066] Figure 13 shows an example of a cost matrix. In Figure 13, the detection points of interest A022 are denoted as point1, point2, and point3, and the estimated buried pipes A030 are denoted as pipe1, pipe2, and pipe3. Among the combinations of the detection points of interest A022 and the estimated buried pipes A030, the combinations that minimize the sum of costs can be calculated as pipe1 and point3, pipe2 and point2, and pipe3 and point1.

[0067] In calculating the cost matrix, the first and second cost matrices may be added together after multiplying each element by a weight. The weights applied to the first and second cost matrices may be adjusted according to the measurement conditions. For example, if the error in positioning information D060 is large, the weight of the cosine distance may be increased, or if there are obstructions on the ground and it is difficult to distinguish buried pipes using feature D050, the weight of the first cost matrix based on position may be increased.

[0068] Furthermore, for combinations where the value of the first cost matrix is ​​greater than or equal to a predetermined threshold, the cost may be set to a sufficiently large constant so as not to affect the cost minimization calculation, and a process may be added to discard the combination if it is selected. This makes it possible to calculate a reasonable clustered detection point D090 without having to associate the estimated buried pipe A030 with the detection point A022 of interest which has a large distance or cosine distance.

[0069] Furthermore, the clustering process has an initial value dependency, meaning the clustering results change depending on which survey line is initially focused on. Therefore, to obtain robust results, it may be possible to calculate the clustered detection points D090 multiple times by changing the survey line initially focused on, record the number of times detection points classified into the same class have been classified into the same class, initialize the pipeline labels, and assign new pipeline labels to detection points that have been classified into the same class more than a predetermined threshold. Figure 14 shows an example of the number of times detection points have been classified into the same class after calculating the clustered detection points multiple times. When there are N detection points, the result can be represented as an N x N matrix. In Figure 14, if clustering is performed to classify detection points that have been classified into the same class 5 or more times into the same cluster, they are classified as (point1, point3, point4) and (point2, point5, point6). Here, the threshold may be a predetermined value, determined based on the number of clusterings, or calculated based on the density of detection points.

[0070] Figures 12A and 12B are flowcharts of the clustering process in Example 3. In the clustering process of Example 3, the clustering unit P061 first initializes the next pipeline label to be detected by setting class=1 (F020) and initializes the buried pipe list that stores the buried pipes to be calculated (F030). Subsequently, the clustering unit P061 selects one of the sequentially arranged survey lines (F040) and calculates the cost matrix of the detected points on the selected survey line and the buried pipes stored in the buried pipe list (F050). If there are no detected points on the selected survey line or if the buried pipe list is empty, the cost matrix is ​​set to an empty matrix. As mentioned above, the cost matrix is ​​calculated by the sum of the first cost matrix and the second cost matrix, and if any combination of cost matrices has a cost greater than or equal to a threshold, a sufficiently large value (for example, a predetermined value) is set for that combination. Using the calculated cost matrix, the combination that minimizes the cost of the detected points and buried pipes is calculated (F060). If a mapping is established with a cost exceeding the threshold, the mapping is deleted.

[0071] Next, one pair of a detection point and a buried pipe that has been successfully matched is selected (F070), and the pipeline label of the buried pipe is assigned to the detection point (F080). The lifespan of the buried pipe is also initialized (F090). A predetermined lifespan is set for each newly registered buried pipe, and the lifespan is reduced if a combination with a detection point is not found even when focusing on a new survey line. By excluding buried pipes that have not been successfully matched for a predetermined number of consecutive times from the combination candidates, it is possible to suppress unreasonable correspondences between distant buried pipes and detection points. The location of the buried pipe is re-estimated with the newly associated detection point added, and the value in the buried pipe list is updated (F100). If the processing from F070 to F100 has been performed for all successful matching pairs, the process proceeds to F120; if processing has not been performed for some successful matching pairs, the process is repeated from F070 (F110).

[0072] After processing is complete for all valid combinations, select any buried pipes that do not have a valid combination (F120) and decrease their lifespan by 1 (F130). Determine if the lifespan of a buried pipe is 0 (F140), and remove any buried pipes with a lifespan of 0 from the buried pipe list (F150). If steps F110 to F150 have been performed for all buried pipes with invalid combinations, proceed to F170; if steps F120 to F160 have not been performed for some buried pipes with invalid combinations, repeat the process from F120 (F160).

[0073] After processing all unmatched buried pipe combinations, select the detection point for the unmatched combination (F170), assign a pipe label class to the detection point (F180), and then add 1 to the pipe label class to update the pipe number to be registered next (F190). Add the buried pipes calculated up to F190 to the buried pipe list (F200). Here, buried pipes calculated from only one detection point may be registered as small buried pipes with a direction perpendicular to the survey line for convenience. If processing F170 to F200 has been performed for all unmatched detection points, proceed to F220; if processing has not been performed for some unmatched detection points, repeat the process from F170.

[0074] In F220, it is determined whether processing has been completed for all survey lines. If processing is incomplete for some survey lines, a nearby survey line is selected, and the process is repeated from F040. Once processing is complete for all survey lines, the clustering process is terminated, and the clustered detection points are output (F230).

[0075] According to this embodiment, by using the positional information of the survey line, the user can obtain more reasonable clustered detection points D090 and obtain a more accurate estimated pipeline D110.

[0076] <Example 4> Next, Embodiment 4 of the present invention will be described. In Embodiment 4, the differences from Embodiment 1 described above will be mainly explained, and the same components and processes as in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted.

[0077] The estimated pipeline D110 obtained by analyzing ground-penetrating radar data D010 may produce output that differs from drawings such as construction drawings and equipment ledgers. In other words, pipelines shown in the drawings may not be estimated, their positions may be incorrect, or pipelines not shown in the drawings may be estimated. In such cases, it is desirable to know the basis for the output of the pipeline estimation system S010 in order to decide whether to trust the output of the pipeline estimation system S010 or the drawings to understand the current situation.

[0078] In the pipeline estimation system S010, intermediate outputs such as vertex positions D020, geospatial positions D070, clustered detection points D090, and feature quantities D050 from the ground-penetrating radar data may be presented via the user interface.

[0079] The feature vector D050 may be displayed in a radar chart-like format, representing the dimensionality of the representative feature vectors D050 of the selected extracted image D040 and the estimated pipeline D110, in addition to the estimated pipeline D110. The similarity between the representative feature vectors of the extracted image D040 and the estimated pipeline D110 may also be displayed.

[0080] According to this embodiment, the user can find out the basis for the estimated pipeline D110 output from the pipeline estimation system S010.

[0081] <Example 5> Next, Embodiment 5 of the present invention will be described. In Embodiment 5, the differences from Embodiment 1 described above will be mainly explained, and the same components and processes as in Embodiment 1 will be denoted by the same reference numerals, and their descriptions will be omitted.

[0082] In pipeline estimation, accuracy can sometimes be improved by utilizing information obtained in addition to ground-penetrating radar data D010. For example, the existence, location, and laying direction of buried pipes can be confirmed by test excavation in a portion of the exploration area. Surface structures such as manholes also provide useful information for pipeline estimation. To enable analysis based on such information, it would be beneficial to provide an interface that allows users to directly add and edit intermediate outputs of the pipeline estimation system S010, such as vertex positions D020 and clustered detection points D090.

[0083] According to this embodiment, users can estimate pipelines with high accuracy and with minimal effort.

[0084] It should be noted that the present invention is not limited to the embodiments described above, but includes various modifications and equivalent configurations within the spirit of the attached claims. For example, the embodiments described above are described in detail for the purpose of clearly illustrating the present invention, and the present invention is not necessarily limited to having all the described configurations. Furthermore, some of the configurations of one embodiment may be replaced with those of another embodiment. Furthermore, configurations of other embodiments may be added to the configuration of one embodiment. Furthermore, some of the configurations of each embodiment may be added, deleted, or replaced with those of other embodiments.

[0085] Furthermore, each of the aforementioned configurations, functions, processing units, and processing means may be implemented in hardware, for example, by designing them as integrated circuits, or they may be implemented in software by having a processor interpret and execute programs that realize each function.

[0086] Information such as programs, tables, and files that implement each function can be stored in memory, hard disks, SSDs (Solid State Drives), or other storage media such as IC cards, SD cards, and DVDs.

[0087] Furthermore, the drawings show control lines and information lines that are considered necessary to explain the embodiments, and do not necessarily show all control lines and information lines included in actual products to which the present invention is applied. In practice, it can be assumed that almost all components are interconnected. [Explanation of Symbols]

[0088] S010, S011, S012 Pipeline Estimation System P010 Vertex detection unit P020 Shape Parameter Calculation Unit P030 Buried Object Reaction Extraction Unit P040 Feature Extraction Unit P050 Geospatial position calculation section P051 Geospatial position calculation section P060 Clustering Unit P061 Clustering Unit P070 Piping section P080 Relative Permittivity Estimation Unit D010 Ground-penetrating radar data D020 Vertex Position D030 Shape Parameters D040 Extracted image D050 Feature D060 Positioning Information D070 Geospatial location D090 Detection point D100 Relative permittivity D110 Estimated pipeline A010 Survey line A011 Survey lines of interest A012 Survey line of focus A020 Detection point A021 Detection points of interest A022 Detection points to focus on A023 Unnoticed detection points A030 Estimated buried pipes A040 Estimated detection point A060 Approximate straight line A070 Foot of the perpendicular A080 The outermost point H010 Server H020 Processor H030 Memory H040 Storage H050 Communication Interface M010 Pipeline Estimation Program

Claims

1. A pipeline estimation system for estimating the location of pipelines laid underground, It is composed of a computer having an arithmetic unit that performs predetermined processing and a storage unit connected to the arithmetic unit, A ground-penetrating radar device that irradiates electromagnetic waves into the ground scans along multiple survey lines and records reflected waves from buried objects. This ground-penetrating radar data is taken as input, and a vertex detection unit detects the peak position of buried object reactions in the ground-penetrating radar data and outputs the detected vertex position. A shape parameter calculation unit takes the ground-penetrating radar data and the vertex position as inputs, calculates shape parameters representing the shape of the buried object reaction for each vertex position, and outputs the calculated shape parameters. A buried object reaction extraction unit takes the ground-penetrating radar data and the shape parameters as input, extracts the buried object reaction from the ground-penetrating radar data using the shape parameters, and outputs an extracted image including the extracted buried object reaction. A feature extraction unit calculates the feature quantities of the extracted image using a model that calculates the feature quantities related to the appearance of buried object reactions in the extracted image, and outputs the calculated feature quantities. A geospatial position calculation unit takes the vertex position and the positioning information of the ground-penetrating radar data as input, calculates the geospatial position of the vertex position, and outputs the calculated geospatial position. A clustering unit that takes the aforementioned feature quantities as input, calculates detection points having the aforementioned feature quantities in geospatial space, performs clustering using the similarity of the feature quantities of the calculated detection points, calculates clustered detection points to which pipeline labels are assigned according to the results of the clustering, and outputs the calculated clustered detection points. A pipeline estimation system characterized by comprising a pipeline conversion unit that takes the clustered detection points as input, calculates an estimated pipeline from the clustered detection points for each pipeline label, and outputs the calculated estimated pipeline.

2. A pipeline estimation system according to claim 1, The system includes a relative permittivity calculation unit that takes the shape parameters as input, calculates the relative permittivity of the medium in the ground from the shape parameters, and outputs the calculated relative permittivity. The pipeline estimation system is characterized in that the geospatial position calculation unit takes the relative permittivity, the vertex position, and the positioning information of the ground-penetrating radar data as input and calculates the depth in geospatial space from the vertex position and the relative permittivity.

3. A pipeline estimation system according to claim 1, The pipeline estimation system is characterized in that the buried object reaction extraction unit moves each observation point of the ground-penetrating radar data parallel to the delay time axis so that buried object reactions having the shape parameters have the same delay time, and outputs an extracted image including the buried object reactions shown in the radar data where the observation points have moved.

4. A pipeline estimation system according to claim 1, The pipeline estimation system is characterized in that the feature extraction unit uses a machine learning model that has learned the relationship between ground-penetrating radar data and pipeline labels to take the extracted image as input and output pipeline labels as features.

5. A pipeline estimation system according to claim 1, The pipeline estimation system is characterized in that the clustering unit takes the feature quantities and the geospatial location as input, calculates detection points having the feature quantities and the geospatial location, and performs clustering using the similarity of the feature quantities and positional relationships of the calculated detection points.

6. A pipeline estimation system according to claim 5, The pipeline estimation system is characterized in that the clustering unit sequentially focuses on adjacent survey lines in geospace based on the positioning information, calculates the buried pipes estimated from the clustered detection points, calculates the optimal combination of the detection points on the survey line of interest and the estimated buried pipes based on a cost matrix obtained by adding the cost based on the geospace location and the cost based on the similarity of the feature quantities, and performs clustering of the detection points on the survey line of interest.

7. A pipeline estimation system according to claim 6, The pipeline estimation system is characterized in that the clustering unit performs clustering of the detection points such that the combination of the detection point and the estimated buried pipe is not included when the cost in the geospatial area is greater than or equal to a predetermined threshold or when the cost based on similarity is greater than or equal to a predetermined threshold.

8. A pipeline estimation system according to claim 6, The pipeline estimation system is characterized in that the clustering unit calculates the clustered detection points multiple times by changing the survey line initially focused on, and classifies clustered detection points that have been classified into the same class more than a predetermined threshold into the same cluster.

9. A pipeline estimation system according to claim 1, The pipeline estimation system is characterized by the pipeline estimation unit calculating an approximate straight line for detection points classified into the same cluster from the clustered detection points, calculating a line segment whose endpoints are the two outermost points among the feet of the perpendiculars from the detection points to the approximate straight line, and outputting the calculated line segment as an estimated pipeline.

10. A pipeline estimation system according to claim 9, The pipeline estimation system is characterized in that, when the coefficient of determination when calculating the approximate straight line is below a predetermined threshold, the detection points classified into the same cluster are divided into multiple clusters, a line segment is calculated for each cluster of divided detection points, and the calculated line segment is output as an estimated pipeline.

11. A pipeline estimation method in which a pipeline estimation system estimates the location of pipelines laid underground, The pipeline estimation system is comprised of a computer having a processing unit that performs predetermined processing and a storage device connected to the processing unit. The aforementioned pipeline estimation method is: A computing device takes as input ground-penetrating radar data, which is obtained by scanning a ground-penetrating exploration device that irradiates electromagnetic waves into the ground along multiple survey lines and recording reflected waves from buried objects, detects the peak position of the buried object reaction in the ground-penetrating radar data, and outputs the detected peak position, in a peak detection step. A calculation device takes the ground-penetrating radar data and the vertex position as input, calculates shape parameters representing the shape of the buried object reaction for each vertex position, and outputs the calculated shape parameters in a shape parameter calculation step. A arithmetic unit takes the ground-penetrating radar data and the shape parameters as input, extracts the buried object response for each of the shape parameters, and outputs an extracted image including the extracted buried object response, in a buried object response extraction step. A feature extraction step in which a computing device calculates the feature quantities of the extracted image using a model that calculates the feature quantities of the appearance of buried object reactions in the extracted image, and outputs the calculated feature quantities, A arithmetic unit calculates the geospatial position of the vertex position using the vertex position and the positioning information of the ground-penetrating radar data as input, and outputs the calculated geospatial position in a geospatial position calculation step. A clustering step in which a computing device takes the feature quantities as input, calculates detection points having the feature quantities in geospatial space, performs clustering using the similarity of the feature quantities of the calculated detection points, calculates clustered detection points to which pipeline labels are assigned according to the results of the clustering, and outputs the calculated clustered detection points. A pipeline estimation method characterized by comprising a arithmetic unit that takes the clustered detection points as input, calculates an estimated pipeline from the clustered detection points for each pipeline label, and outputs the calculated estimated pipeline in a pipeline estimation step.

12. A pipeline estimation program for estimating the location of pipelines laid underground, The pipeline estimation program is executed on a computer having a processing unit that performs predetermined processing and a storage device connected to the processing unit. The aforementioned pipeline estimation program, A vertex detection procedure involves scanning a ground-penetrating exploration device that irradiates electromagnetic waves into the ground along multiple survey lines to record reflected waves from buried objects, taking ground-penetrating radar data as input, detecting the peak position of buried object reactions in the ground-penetrating radar data, and outputting the detected peak position. A shape parameter calculation procedure that takes the ground-penetrating radar data and the vertex position as inputs, calculates shape parameters representing the shape of the buried object reaction for each vertex position, and outputs the calculated shape parameters, A buried object reaction extraction procedure that takes the ground-penetrating radar data and the shape parameters as input, extracts the buried object reaction for each of the shape parameters, and outputs an extracted image including the extracted buried object reaction, A feature extraction procedure that calculates the features of the extracted image using a model that calculates features related to the appearance of buried object reactions in the extracted image, and outputs the calculated features, A geospatial position calculation procedure that takes the vertex position and the positioning information of the ground-penetrating radar data as input, calculates the geospatial position of the vertex position, and outputs the calculated geospatial position, A clustering procedure that takes the aforementioned feature quantities as input, calculates detection points having the aforementioned feature quantities in geospatial space, performs clustering using the similarity of the feature quantities of the calculated detection points, calculates clustered detection points to which pipeline labels are assigned according to the results of the clustering, and outputs the calculated clustered detection points, A pipeline estimation program that causes the computing device to execute a pipeline creation procedure that takes the clustered detection points as input, calculates an estimated pipeline from the clustered detection points for each pipeline label, and outputs the calculated estimated pipeline.

Images