JP6785751B2 - Model generator, generator, and program - Google Patents

Description

The present disclosure relates to a model generator, a generation method, and a program.

With the widespread use of long-distance laser range finders, environmental measurement in a wide range of outdoor spaces has been realized, and technological development has been carried out to detect useful structures from the measured three-dimensional point cloud. In particular, power lines, communication lines used for telephones, the Internet, etc., which are important as infrastructure equipment, are managed by companies, local governments, etc., and it is necessary to regularly inspect whether there are any abnormalities. However, since there are a huge number of thin wire-like objects such as cables (hereinafter referred to as "wires") such as power lines and communication lines, it takes a lot of labor and time to visually check them. It is work. Therefore, it automatically grasps the location of the wire, and whether the ground clearance is sufficient, whether it interferes with the surrounding structures in the case of power lines, and whether it is sufficiently far away (hereinafter). , Called "separation distance") is required for technological development. In addition, at present, when constructing a new cable, the exact length of the existing cable is not known, so a field survey has been started. It is required to automatically detect and 3D model the cables having such a huge number in order to suppress the on-site operating time.

As a technique for automatically detecting such a wire and modeling it in 3D, for example, there is a technique described in Non-Patent Document 1. The technique described in Non-Patent Document 1 assumes that the wire looks like a straight line when viewed from directly above, projects a three-dimensional point cloud directly below (XY plane), and then Houghs the shape of the point cloud after projection. The transform is used to detect the position on the plane where the straight line exists. Next, since the wires having different heights exist on the same straight line, as a process of separating the wires, the point clouds existing on the straight line detected by the Hough transform are hierarchically clustered in the three-dimensional space. By doing so, it is cut out in units of objects. Finally, the parameters of the electric wire are estimated by applying the catenary model to the detected clusters. Since the technique described in Non-Patent Document 1 uses a global model fitting such as the Hough transform, it is robust against the influence of objects other than electric wires, such as the influence of buildings and trees, and noise. In addition, it has the feature that it can be detected even when the density of the measurement point cloud is sparse.

However, in the technique described in Non-Patent Document 1, since a process of finding a straight line from a plane by using a Hough transform is performed, it is suitable for detecting a wire having a long horizontal distance of several tens of meters, but a utility pole. It is difficult to detect the wire that connects the wire to other structures. For example, as shown in FIGS. 22A to 22B, lead-in lines such as power lines and communication lines extending from between utility poles to electric lights or from utility poles to private homes tend to have a short horizontal length and are oriented in the vertical direction. It is difficult to detect because the length of the straight line after projection is short. In order to detect it, it is necessary to lower the threshold value of the Hough transform, but lowering the threshold value causes erroneous detection at a position where an object other than the wire, such as a tree, exists.

In the technique described in Non-Patent Document 2, a three-dimensional space is divided at intervals of several meters, and short linear elements are detected in each divided space. A catenary model is applied to the detected short straight line elements, and clustering is performed on nearby straight line elements on the catenary model. This is a process in which the catenary model is fitted and clustered alternately, and the process in which the fluctuation of the catenary model parameters is sufficiently small is repeated. When clustering short straight elements, the linear elements to be clustered are determined using a catenary model, so that even if the measurement point cloud is sparse, a decrease in clustering accuracy can be suppressed.

However, in the technique described in Non-Patent Document 2, although linear elements are clustered using a catenary model, a wire in which a plurality of wires are connected in an urban area, a tension adjustment and a separation distance as shown in FIG. 22C, are used. For cables with adjustment parts (called accessories), the shape is different from the catenary model shape, and the clustering accuracy is reduced at that time.

For example, when the wire has a metal fitting or a reinforcing material for adjusting the tension, when a plurality of wires are connected as shown in FIGS. 22A to 22C, or when the tension and the separation distance are adjusted by accessories. If this is the case, the load applied to the wire is not uniform, so that the wire shape is distorted three-dimensionally and the estimation accuracy of the fitted catenary model is lowered. As a result, there is a technical problem that the clustering accuracy is lowered and originally different wires are detected as the same wire.

On the other hand, in the technique described in Patent Document 1, it is optimal to detect locally continuous line segments (straight line elements) divided in a section of several meters in a three-dimensional space and smoothly connect the straight line elements. By introducing an evaluation index (connectivity index), it is possible to detect not only a specific wire shape but also a wire that causes three-dimensional distortion. The shape (position, orientation) of each straight line element can be fine-tuned by changing the parameters of the connectivity index so that it can be smoothly connected to nearby straight line elements, improving robustness to measurement noise.

Japanese Patent No. 5981886

Melzer, T. and Briese, C., "Extraction and Modeling of Power Lines from ALS Point Clouds," Proceedings of Austrian Association for Pattern Recognition (OAGM), Hagenberg, 06-17-2004 --06-18-2004, 47- 54 (2004). G. Sohn, Y. Jwa, and H. B. Kim, AUTOMATIC POWERLINE SCENE CLASSIFICATION AND RECONSTRUCTION USING AIRBORNE LIDAR DATA, SPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., I-3, 167-172, 2012 Hitoe Arakaki, Ken Tsutsuguchi, Tetsuya Fuchibuchi, "Detection of Tree Trunks from Mobile Mapping System Measurement Point Clouds-Feature Extraction for Sweep Shape Detection", 1st Annual Meeting of the Society of Imaging Electronics, (2017)

According to Patent Document 1, it is possible to detect a wire having an arbitrary shape by connecting linear elements, but there is a problem that the detection accuracy is lowered when a large point cloud is missing. The point cloud measured by MMS (Mobile Mapping System) is prone to loss of the point cloud due to the influence of occlusion, and if the shape of the wire model of the missing part cannot be estimated correctly, the separation distance of that part can be calculated accurately. Can not.

Occlusion in point cloud measurement means that the subject is not measured when an obstacle exists between the laser sensor of the MMS and the subject. For example, when a tree exists between the laser sensor and the wire, the wire behind the tree is not measured with respect to the sensor, and the wire shape may not be measured by several meters, or close to 10 meters under bad conditions.

At this time, in the connectivity index of Patent Document 1, it is difficult to accurately define the smoothness of linear elements between distant ones, so that wires that are not originally the same are connected, and as a result, the detection accuracy tends to decrease. The reason why it cannot be defined accurately is that the connectivity index of Patent Document 1 does not assume that the attitudes (direction vectors of straight lines) of the linear model are similar in the vicinity.

In addition, the wire model can be approximated by a straight line from a micro viewpoint (small scale), but the approximation accuracy by a straight line tends to deteriorate from a macro viewpoint (wide scale). That is, as shown in FIG. 23, since the straight line element at a certain position physically means the tangential direction of the wire, it is difficult to estimate the position of the wire at a distance, and the wire is approximated by a straight line. If the scale of the local region is increased, there is also a problem that the shape error from the original wire shape becomes large and the estimation accuracy of the separation distance decreases. That is, the inability to estimate the correct shape in the missing region reduces the accuracy of estimating the separation distance.

The present disclosure has been made in view of the above points, and provides a model generator, a method, and a program capable of robustly and accurately detecting a wire from a three-dimensional point cloud and calculating a separation distance. With the goal.

In order to achieve the above object, the model generation device of the first aspect of the present disclosure is a model generation device that generates a wire model from the input three-dimensional point group information of the subject, and is the three-dimensional point group. From the information, a piecewise continuous line detection unit that detects a group of wire constituent points constituting the wire and detects a plurality of piecewise continuous lines that are continuous curves existing in a local region from the detected wire constituent points group. Of the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit, it is determined whether or not to connect the two piecewise continuous lines to be connected, and the two piecewise continuous lines determined to be connected are connected. A piecewise continuous line connecting portion that generates a wire model by connecting the continuous lines is provided.

In order to achieve the above object, the model generation device of the second aspect of the present disclosure is the model generation device of the first aspect, in which the wire model generated by the piecewise continuous line connecting portion and the periphery of the wire model. A peripheral distance estimation unit for calculating the distance to the structure of the above is further provided.

The model generation device of the third aspect of the present disclosure is the model generation device of the second aspect, in which the peripheral distance estimation unit has a wire configuration detected by the sectional continuous line detection unit from the three-dimensional point cloud information. The point cloud excluding the point cloud is clustered based on the similarity of the local shape to obtain the peripheral structure point cloud, and the distance between the peripheral structure point cloud and the wire model is calculated.

In the model generator of the fourth aspect of the present disclosure, in the model generator of the second aspect, the peripheral distance estimation unit performs noise removal processing based on the density of the point cloud and clustering processing with neighboring points. Do.

The model generator according to the fifth aspect of the present disclosure is the model generator according to any one of the first to fourth aspects, wherein the piecewise continuous line connecting portion is the two objects to be connected and determined. Whether or not to connect the two piecewise continuous lines to be connected is determined by using the error when the piecewise continuous line is approximated as one continuous line and the length of the continuous line as an index.

The model generator according to the sixth aspect of the present disclosure is the model generator according to any one of the first to fourth aspects, wherein the piecewise continuous line connecting portion is two pieces to be determined for connection. The error when the target continuous line is approximated as one continuous line, the length of the continuous line, and the three-dimensional features extracted from the two piecewise continuous lines to be connected and the periphery of the two piecewise continuous lines. Using a classifier based on the quantity, it is determined whether or not to connect the two piecewise continuous lines to be connected.

In order to achieve the above object, the generation method of the seventh aspect of the present disclosure is a generation method of generating a wire model from the input three-dimensional point group information of the subject by the piecewise continuous line detection unit. , The wire constituent points that make up the wire are detected from the three-dimensional point group information, and a plurality of piecewise continuous lines that are continuous curves existing in the local region are detected from the detected wire constituent points and divided. When it is determined whether or not to connect two piecewise continuous lines to be connected, among the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit, the target continuous line connecting unit is connected. A wire model is generated by connecting the two determined piecewise continuous lines.

In order to achieve the above object, the program of the eighth aspect of the present disclosure is a program for causing a computer to function as each part of the model generation device according to any one of the first to sixth aspects. Is.

According to the present disclosure, it is possible to obtain the effect that the wire can be detected robustly and accurately from the three-dimensional point cloud. Further, since the wire shape in the occlusion region can be estimated, the separation distance can be calculated accurately.

It is a block diagram which shows the structure of an example of the model generation apparatus of 1st Embodiment. It is a block diagram which shows the structure of an example of the piecewise continuous line connecting part of 1st Embodiment. It is a flowchart which shows an example of the operation of the wire model generation part of 1st Embodiment. It is a flowchart of the estimation example of the quadratic curve by RANSAC. It is a conceptual diagram for demonstrating the distance between a quadratic curve Cn and a point cloud pi. It is a flowchart which shows an example of the piecewise continuous line connection processing of 1st Embodiment. It is a conceptual diagram for demonstrating a continuous candidate line. It is a block diagram which shows the structure of an example of the connectivity index calculation part in the method (A) of 1st Embodiment. It is a flowchart which shows an example of the operation of the geometric index calculation part in the method (A) of 1st Embodiment. It is a conceptual diagram for demonstrating the integration curve candidate in the method (A) of 1st Embodiment. It is a conceptual diagram for demonstrating the wire constituent points used for evaluation of an integrated curve model. It is a block diagram which shows the structure of an example of the connectivity index calculation part in the method (B) of 1st Embodiment. It is a flowchart which shows an example of the operation of the geometric index calculation part in the method (B) of 1st Embodiment. It is a conceptual diagram for demonstrating the feature amount which is the histogram generated from the attention point and the peripheral point in the method (B) of 1st Embodiment. It is a conceptual diagram for demonstrating the method (B) of 1st Embodiment. It is a conceptual diagram for demonstrating the method (B) of 1st Embodiment. It is a conceptual diagram for demonstrating the integration method of the feature amount in the method (B) of 1st Embodiment. It is a conceptual diagram for demonstrating the connection process executed by the connection determination part of 1st Embodiment. It is a block diagram which shows the structure of an example of the model generation apparatus of 2nd Embodiment. It is a flowchart which shows an example of the operation of the wire model generation part of 2nd Embodiment. It is a flowchart which shows an example of the peripheral distance estimation processing of 2nd Embodiment. It is a figure for demonstrating the wiring example of a wire. It is a figure for demonstrating the wiring example of a wire. It is an image diagram of the accessory (adjustment metal fitting) shown in FIG. 22B. It is a conceptual diagram for demonstrating the connection of the straight line element by an occlusion area.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The present embodiment does not limit the present invention.

[Outline of Invention]
First, an outline of the embodiment of the present invention will be described.

The present disclosure is a technique relating to a model generator, a wire model generation method, and a program for robustly detecting a wire model of a wire such as a communication line such as a power line or a telephone line from a three-dimensional point cloud. In particular, it is an effective technique even in a situation where a large defect occurs in a three-dimensional point cloud measured due to the influence of occlusion or a situation in which a plurality of wires are distorted in shape. In the following, "three-dimensional" may be expressed as "3D".

In the present disclosure, clustering (concatenating curved elements into the same group) is performed on locally continuous curved elements detected in advance by an index that defines the ease of connecting the curved elements and neighboring curved elements. .. By approximating the local region with a curve, it is possible to estimate the position and shape (physical parameters) of the curve element at a distant position.

However, if a connectivity index that considers only the smoothness at the time of connection is defined as in the conventional technique, the estimation accuracy tends to decrease when the point cloud has a large defect, and there is a problem that incorrect connection is performed. It will occur. In the present disclosure, by defining an index using the approximation accuracy when two curve elements that are candidates for connection are integrated, it is possible to suppress a decrease in accuracy in the case of a large defect and to make a robust connection to measurement noise. To do.

Here, there are information criteria such as AIC (Akaike's Information Criterion) and BIC (Schwarz's Bayesian Information Criterion) as indicators that have been conventionally used when integrating a plurality of geometric models. This is an index considering the approximation accuracy of the observation data (three-dimensional point cloud in this embodiment) and the complexity of the model parameters when a certain geometric model is selected. In the information criterion, a geometric model is selected that sufficiently expresses the approximation accuracy of the observed data but has low complexity. In the connectivity index in the present disclosure, the approximation accuracy used in the information criterion is used in order to make a robust estimate of noise.

Regarding infrastructure equipment, the shape of the lead-in wire and cable (wire) is not necessarily a catenary due to the influence of accessories, but it is installed as a regular shape according to the in-house inspection work manual. Regularity often follows some standard, such as maintaining contrast or wiring cables so that they are shaped to maintain a certain distance from the surrounding structure. Therefore, it means that there is a high probability that a wire shape similar to the wire shape installed in one utility pole section also exists in another utility pole section.

In other words, from the viewpoint of maintenance work and safety, the cable wiring method that satisfies the installation conditions as infrastructure equipment, such as not hanging too much to maintain the height above the ground, or the cables are routed so as to maintain a certain interval. It is considered that learning regularity such as safety is effective for connecting wires accurately.

In the present disclosure, the shape of the wires to be connected is learned by using the 3D feature amount, and whether or not the wires can be connected to each other is determined by the classifier obtained by the framework of machine learning.

[Outline of Embodiment]
In the first embodiment, in the present disclosure, similarly to this information criterion, the increase / decrease in the approximation error and the increase / decrease in the size of the model when a plurality of models are represented by one are used as indexes. Specifically, the observation three-dimensional point cloud on the curve element to be connected is used as a judgment index of whether or not it should be represented by two curve elements or one curve element. However, when a plurality of curve elements are distorted, it becomes difficult to connect the two curves, and if the threshold value for connection possibility is lowered, the possibility that erroneous wires are connected tends to increase. Therefore, in the present disclosure, the influence of distortion is suppressed to a low value by calculating the approximation error by focusing on the 3D point cloud in the local region from the connection end point position for the curve element to be connected. As a result, even when a plurality of wires exist in parallel as in an urban area, the wires can be connected with high accuracy.

In the calculation method (B) of the connectivity index, an evaluation function for determining the possibility of connectivity is defined in consideration of not only the curve element to be determined for connection but also the surrounding conditions. For example, wires arranged three by three can be connected by learning "regularity" such that three wires are likely to be connected while maintaining a certain distance even if there is a large defect. Suppresses a decrease in accuracy. Specifically, by recording the shape information of the curved element and the surrounding object as a three-dimensional feature quantity, and recognizing the pattern that integrates the approximation error and the features of the two curved elements to be connected, the connection is possible or not. To judge.

In the second embodiment, the separation distance is calculated by examining the distance between the obtained wire model and the surrounding structure. At this time, by applying a noise removal filter and further removing accessories by point cloud clustering, it is possible to accurately calculate the distance to a large structure such as a leaf of a tree or a building.

The present embodiment is a technique for detecting individual wires from a three-dimensional point cloud, and is a technique for estimating the distance (separation distance) between the 3D model of the detected wire and the surrounding structure. First, a point cloud determined to form a wire is detected, then a curved element is detected from the wire constituent point cloud, and a wire model is generated by connecting the curved elements. The separation distance can be output by measuring the distance between this wire and the surrounding structure.

Here, the wire model means that a wire existing in a three-dimensional space is parametrically represented by a 3D geometric model. In the present embodiment, in order to express a curve element using a quadratic curve existing in a three-dimensional space, each curve element has a local coordinate system for expressing a quadratic curve (origin position and coordinates of the coordinate system). The orientation of the system) is a parameter. Each wire model holds the minimum ground clearance (position and distance closest to the ground) and separation distance information (position and distance closest to the surrounding structure) as 3D shape information as parameter information and peripheral information for multiple curved elements. To do.

Further, in the present embodiment, the curve element means a continuous line segment in a local region, and is hereinafter referred to as a piecewise continuous line. The piecewise continuous line can be applied not only to a quadratic curve but also to a linear straight line or, in principle, a quartic curve. It means that it has two endpoints in a certain section and the interval between the endpoints is a continuous line.

The purpose of the present disclosure is to detect wires such as power lines, telephones, and Internet communication lines and to obtain information on their surroundings, but it is also a technique capable of detecting and modeling other fine line shapes. .. For example, it also means an elongated structure such as a white line on a road. In the following embodiment, as a specific example, a wire detection method using a point cloud having position information (three-dimensional coordinates) acquired by a laser range finder will be described.

Here, the three-dimensional coordinates in the measured point group may be latitude, longitude, and sea level (height) information, or may be a three-dimensional Euclidean coordinate system or a polar coordinate system with a specific position set by the user as the origin. .. In the following example, a three-dimensional Euclidean coordinate system (each direction is defined as X, Y, Z coordinates) at the origin set by the user is assumed, and the unit of each coordinate is meters. The three-dimensional point is a point to which the above-mentioned three-dimensional coordinates are given measurement information such as the time when the point cloud was photographed, the reflection intensity of the laser, and the attribute information of the point. There is no limit to the information given to the 3D points, but at least the position information (X, Y, Z coordinates) is given, and the 3D point cloud is a collection of a plurality of the 3D points. ..

[First Embodiment]
FIG. 1 is a block diagram showing an example of the configuration of the model generation device 100 of the present embodiment. As shown in FIG. 1, the model generation device 100 of this embodiment includes a storage unit 103 and a calculation unit 104.

The subject measuring unit 101 measures a three-dimensional point cloud, and is a device capable of measuring the distance between the subject and the sensor, such as a laser range finder, an infrared sensor, or an ultrasonic sensor, and is a three-dimensional point of the subject. The group information is output to the model generator 100.

For example, by mounting a laser range finder on a car equipped with GPS (Global Positioning System) or on an airplane equipped with GPS and measuring while moving, artificial objects in an outdoor environment, such as wires and buildings, can be measured. , A system that measures the three-dimensional position of an unspecified number of subjects such as trees, roads, and the ground other than roads.

In the present embodiment, an MMS (Mobile Mapping System) in which a GPS and a laser range finder are mounted on a vehicle is assumed as the subject measurement unit 101. The subject measuring means 101 is not limited to the present embodiment, and may be, for example, an MMS mounted on an airplane or the like, or is fixed to measure from a specific position such as an intersection. It may be a laser scan sensor device or the like.

The input unit 102 is a user interface such as a mouse or a keyboard, and is used when changing the value of the parameter storage unit 111 for arithmetic processing. It means a hardware device in which a user inputs necessary values using a keyboard or the like. Alternatively, it may be a device that copies the parameters stored in the USB HDD device or the like to the parameter storage unit 111 for arithmetic processing.

As shown in FIG. 1, the storage unit 103 of the present embodiment includes a three-dimensional point cloud storage unit 110, a parameter storage unit 111 for arithmetic processing, and an equipment information storage unit 112.

The 3D point cloud storage unit 110 is input with the 3D point cloud information from the subject measurement unit 101, and the input 3D point cloud information is stored. The three-dimensional point cloud information stored in the three-dimensional point cloud storage unit 110 is output to the calculation unit 104.

The calculation processing parameter storage unit 111 is input with the calculation processing parameters used for the calculation of the calculation unit 104 from the input unit 102, and the input calculation processing parameters are stored. The arithmetic processing parameters stored in the arithmetic processing parameter storage unit 111 are output to the arithmetic unit 104.

The wire model generated by the calculation unit 104, which will be described in detail later, is input to the equipment information storage unit 112, and the input wire model is stored.

The three-dimensional point cloud storage unit 110, the arithmetic processing parameter storage unit 111, and the equipment information storage unit 112 are hardware storage devices such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive), for example.

The calculation unit 104 includes a piecewise continuous line detection unit 121 and a piecewise continuous line connecting unit 122.

The piecewise continuous line detection unit 121 acquires three-dimensional point cloud information from the three-dimensional point cloud storage unit 110, and detects a curve element (piecewise continuous line) based on the three-dimensional point cloud information. The piecewise continuous line detection unit 121 outputs the detected curve element (piecewise continuous line) to the piecewise continuous line connecting unit 122.

A piecewise continuous line is input from the piecewise continuous line detection unit 121 to the piecewise continuous line connecting unit 122. The piecewise continuous line connecting unit 122 connects the curve elements (piecewise continuous lines) detected by the piecewise continuous line detecting unit 121 to generate a wire model. The wire model generated by the piecewise continuous line connecting unit 122 is output to the equipment information storage unit 112 of the storage unit 103.

As shown in FIG. 2, the piecewise continuous line connecting unit 122 of the present embodiment includes a connection candidate line detection unit 131, a connectivity index calculation unit 132, and a connection determination unit 133, which will be described in detail later.

The model generation device 100 of this embodiment includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory) that stores programs and various data for executing processes described later. Can be configured with a computer that includes. When the CPU of the present embodiment executes the program, it functions as each unit of the wire model generation unit 120.

(Processing flow of the wire model generation unit 120 of the calculation unit 104)
FIG. 3 is a flowchart showing an example of the operation of the calculation unit 104 of the present embodiment. First, the entire operation (action) of the wire model generation unit 120 of the calculation unit 104 in the model generation device 100 of the present embodiment will be described. It should be noted that steps S1 to S3 correspond to the piecewise continuous line detection unit 121, and step S4 corresponds to the piecewise continuous line connecting unit 122. Further, "model fitting" means a process of estimating a geometric model defined parametrically. In this embodiment, RANSAC (RANdom SAmple Consensus), which is a general algorithm for estimating the parameters of the shape that maximizes the evaluation score, is used for the input point cloud.

As shown in FIG. 3, in step S1, the piecewise continuous line detection unit 121 performs a local shape analysis from the three-dimensional point cloud input from the three-dimensional point cloud storage unit 110.

In the next step S2, the piecewise continuous line detection unit 121 detects a three-dimensional point cloud constituting the wire, and sets the detected three-dimensional point cloud as the wire constituent point.

In the next step S3, the piecewise continuous line detection unit 121 applies a model to the detected wire constituent points, detects the piecewise continuous line, and outputs the piecewise continuous line to the piecewise continuous line connecting unit 122.

In the next step S4, the piecewise continuous line connecting unit 122 performs a connecting process on the piecewise continuous line input from the piecewise continuous line detecting unit 121, detects a wire model, and sets the parameters of the detected wire model as equipment information. Output to the storage unit 112 to end this operation.

Hereinafter, the detailed operation of the algorithms of steps 1 to 4 in the wire model generation unit 120 described in the flowchart shown in FIG. 3 will be described.

(Step S1: Local shape analysis)
In step S1, the connection candidate line detection unit 131 acquires the three-dimensional point cloud from the three-dimensional point cloud storage unit 110, and acquires the arithmetic processing parameters used by the wire model generation unit 120 from the arithmetic processing parameter storage unit 111. .. Then, the connection candidate line detection unit 131 calculates the normal and the minimum curvature direction (extrusion direction) at the position of interest as a local shape analysis for the three-dimensional point cloud. Assuming that the number that distinguishes the input point cloud is i and the total number of point clouds is Ni, the three-dimensional position of each point is expressed by the following equation.

However, [x _i , y _i , z _i ] means the coordinate components of the X, Y, and Z axes, and the subscript symbol "T" in the upper right means transpose.

The existing method may be used to estimate the normal direction, but an example is shown below. For target points p _i, the distance within a radius r is defined as the local region, the local region inside the point of _{p j (j∈1,2,3, ..., N} j), the point group in the local region Total number of

Notated as. The points included in the distance within the radius r are obtained as points satisfying the following equation (1).
... (1)

However, in equation (1), the symbol "|| ||" means the 2 norms of the vector, and the symbol "||" means the absolute value.

In this case, the normal vector n _i is Motomari as the third eigenvalue vector when the covariance matrix P of the point group of the internal local region and eigenvalue decomposition (vector corresponding to the smallest eigenvalue), the minimum curvature direction m _i, Obtained as the first eigenvalue vector. In the present embodiment, the size of r in the local region when the normal is obtained is set to 0.2 [m].

As in Non-Patent Document 3, the extrusion direction may be estimated robustly based on the measurement noise and used in the subsequent feature extraction processing. In this embodiment, a method with a small amount of calculation is shown. In the following, the meaning of "attention" in the words "attention point" and "attention curve" means "is" or "certain", and does not mean that it is selected by some index, but all processing candidates. Is processed in the same way. In addition, since the number of letters in the alphabet is limited, the same letters may be used as different meanings in different paragraphs (sections).

(Step S2: Wire component point detection)
In step 2, the piecewise continuous line detection unit 121 detects the wire constituent points. The piecewise continuous line detection unit 121 performs identification processing for identifying which point (wire constituent point) is a wire from the input three-dimensional point cloud. This process uses existing technology and can be realized by, for example, the technology described in Patent Document 1. Further, for example, when it is possible to prepare teacher data in advance, it is possible to realize the detection of a point cloud on an elongated object by using an existing method such as Non-Patent Document 3.

(Step S3: Piecewise continuous line detection)
In step S3, the piecewise continuous line detection unit 121 detects the piecewise continuous line. The piecewise continuous line detection unit 121 detects a continuous curve existing in a local region (a region within a certain scale) for this process from the wire constituent point group detected in step S2. Specifically, the piecewise continuous line detection unit 121 performs model fitting on a group of wire constituent points in a certain region, and determines a curve element having an evaluation score equal to or higher than a threshold value.

In this embodiment, the parameters of the quadratic curve (arc) are estimated using RANSAC, which is widely used in robust estimation. Parameters of the quadratic curve, the two axes Xc of the coordinate system where the presence of the arc is in three-dimensional space, Yc direction axis perpendicular Zc _i direction in that direction, the center position Vc and its radius Rc of the circle determined. The end point is obtained as the position of the arc in the three-dimensional point cloud (wire point cloud) existing on the arc.

FIG. 4 shows a flowchart of an example of estimating a quadratic curve by RANSAC executed by the piecewise continuous line detection unit 121. At the start of step S3, a three-dimensional point cloud determined to be a wire constituent point is input to the piecewise continuous line detection unit 121.

In step S3-1, the piecewise continuous line detection unit 121 adds “1” to the management variable Loop for repeating the predetermined predetermined number of times of processing. The initial value of the repeating variable Loop when the execution of this process is started is 0. When the end condition is satisfied, Loop becomes the initial value 0.

In the next step S3-2, the piecewise continuous line detection unit 121 calculates the parameters of the piecewise continuous line. In the present embodiment, since the continuous line is a quadratic curve (arc), the piecewise continuous line detection unit 121 can obtain the parameter from three or more three-dimensional points. Hereinafter, the "direction" is defined as a unit vector having a length of 1.0.

For the Cnth compartmentalized continuous line, the center of gravity positions of three randomly selected points from the wire constituent points are VV _Cn , and the direction of the first point randomly selected from the center of gravity position VV _Cn is XX _Cn and the center of gravity position VV _{Cn. Let} YY _Cn be the direction of the first point selected at random. Further, regarding the two-dimensional plane stretched by XX _Cn and YY _Cn, the projection of the vector in the Z-axis direction is referred to as Y _Cn , and the one obtained by _removing the components of the direction Y _Cn from the direction XX _{Cn is} referred to as the direction X _Cn .

However, the symbol "・" means multiplication of numerical values or the inner product between vectors, and the symbol "→" means assigning a value to a variable.

The direction obtained here is the local coordinate system of the Cnth piecewise continuous line. A random number generator such as Mersenne Twister can be used to randomly select a point cloud. For example, the generated random number is divided by the number of input points and the quotient (rounded down to the nearest whole number). You can choose the point with the same number as the value of. In rare cases, when three randomly selected points are lined up in a straight line, three points may be selected again, and when the same number is selected by a random number, they may be selected again.

Next, the equations of the circle are obtained from the above three randomly selected points, and the center position Tn (Tx, Ty) of the circle and the radius Rn of the arc in the local coordinate system are obtained. Since the solution of the circle equations is known, the description thereof is omitted here. However, Tx and Ty are scalar values and are components of the XCn axis and the YCn axis of the local coordinate system.

The center position (three-dimensional coordinates) of the circle of the Cnth piecewise continuous line can be obtained by the following equation.

In the next step S3-3, the piecewise continuous line detection unit 121 evaluates the tentative parameters of the piecewise continuous line calculated in step S3-2. The piecewise continuous line detection unit 121 obtains an evaluation score for the parameter of the Cnth piecewise continuous line using the wire constituent points within a distance R from the center of gravity position VV _Cn .

Here, the function D (Cn ^loop , pi) is a function that outputs the distance between the quadratic curve and the wire constituent point pi. As shown in FIG. 5, the curve (arc) defined by the wire constituent point pi and the nth parameter can be obtained by the following equation D from the geometrical relationship.

The threshold value εc is an experimentally determined parameter, and in this embodiment, εc = 0.05 [m].

The length L _Cn of the _Cnth piecewise continuous line can be obtained by the following equation.

However, θmax and θmax are the maximum and minimum angles formed from the X _Cn axis, and can be obtained by the following equation.

In the next step S3-4, the piecewise continuous line detection unit 121 determines whether or not the end condition 1 is satisfied.

(End condition 1)
The number of iterations Loop is equal to or greater than the maximum number of iterations LoopMax.

If the end condition 1 is not satisfied, the determination in step S3-4 becomes a negative determination, the process returns to step S3-1, and each process of steps S3-1 to S3-3 is executed until the end condition 1 is satisfied. On the other hand, when the end condition 1 is satisfied, the determination in step S3-4 becomes an affirmative determination, and the process proceeds to step S3-5.

In step S3-5, the piecewise continuous line detection unit 121 determines whether or not the end condition 2 is satisfied.

(End condition 2)
The highest score among the candidate parameters is less than the threshold TH_score_curve, or the length of the quadratic curve is less than the threshold length TH_length_curve.

If the end condition 2 is not satisfied, the determination in step S3-5 is a negative determination, the parameter with the highest score is determined to be the Cnth piecewise continuous line, and the process proceeds to step S3-6. On the other hand, when the end condition 2 is satisfied, the determination in step S3-5 becomes an affirmative determination, and this process ends.

In step S3-6, the piecewise continuous line detection unit 121 determines that the wire model constituent point on the Cnth piecewise continuous line belongs to the Cnth piecewise continuous line and removes it. However, "removal" in step S3-6 means to remove from the RANSAC process in step S3. Further, the point cloud belonging to the Cnth continuous line corresponds to the point cloud within the range of θmin and θmax for the arc and the distance εc or less from the arc when the evaluation score is calculated. In the subsequent connection processing, each continuous line and a point cloud belonging to that line are used. After the process of step S3-6, the process returns to step S3-1 and the process of steps S3-1 to S3-5 is repeated.

Here, the maximum number of repetitions LoopMax, the score threshold TH_score_curve, and the length threshold are parameters that are experimentally determined. In this embodiment, LoopMaxx = 1000 [times], TH_score_curve = 50 [points], TH_score_curve = 1.0 [m]. And said. The size of the constant scale Rk that defines the local region in the evaluation of the parameter of the quadratic curve is an experimental parameter, and in this embodiment, the scale Rk = 5 [m].

(Step S4: Piecewise continuous line connection)
In step S4, the piecewise continuous line connecting unit 122 determines whether or not to connect the piecewise continuous line of interest and the piecewise continuous line around it, and connects the piecewise continuous line based on the determination result. FIG. 6 shows a flowchart of an example of the piecewise continuous line connecting process executed by the piecewise continuous line connecting unit 122 of the present embodiment. In FIG. 6, the continuous curves input from steps S4-1 to S4-5 are described as sequential processing, but the processing may be performed in parallel.

As shown in FIG. 6, in step S4-1, the connection candidate line detection unit 131 pays attention to one of the piecewise continuous lines input from the piecewise continuous line detection unit 121. The connection candidate line detection unit 131 describes the continuous line of interest as the number c (c = 1,2,3, ..., Nc). The initial value of c when the execution of this process is started is set to 0, and the connection candidate line detection unit 131 adds "1" to c in this step. However, Nc is the total number of input piecewise continuous lines.

In the next step S4-2, the connection candidate line detection unit 131 detects the connection candidate line group. The connection candidate line detection unit 131 first detects a piecewise continuous line having an end point within a radius Rt [m] from the end point of the piecewise continuous line of interest as a connection candidate line, and detects the detection result as a connectivity index calculation unit. Output to 132.

The two end point positions of the piecewise continuous line of the attention number c can be obtained by the following equation.

Using the peripheral search scale Rt [m], the connection candidate line number k is obtained as a piecewise continuous line satisfying the following equation (2) or equation (3).

A piecewise continuous line having an end point V _term ^(K) satisfying the above equation (2) or (3) is a connection candidate line.

However, V ^(k) means a piecewise continuous line of the number k. The peripheral search scale is the permissible length (upper limit value) of the region where occlusion occurs, and is a value experimentally obtained from the measured data. In this embodiment, Rt = 10 [m].

Further, the connection candidate line detection unit 131 determines whether or not the end point of the connection candidate line exists at the end of the extension of the piecewise continuous line of interest. As shown in FIG. 7, the connection candidate line detection unit 131 determines that the divisional continuous line having an end point within a distance H [m] from the line segment extending the divisional continuous line of interest is a connection candidate line. Here, at least one of the two endpoints may be a distance H or less. The distance H is an experimentally determined parameter, and in this embodiment, H = 2 [m].

In the next step S4-3, the connectivity index calculation unit 132 pays attention to the piecewise continuous line and each connection candidate line based on the detection result (continuous line of connection candidate) input from the connection candidate line detection unit 131. Calculate the connectivity index with and. Regarding the process of step S4-3 executed by the connectivity index calculation unit 132, the following two methods (A) and (B) will be described in detail.

In the method (A), the connectivity index calculation unit 132 calculates the geometric index. The connectivity index calculation unit 132 in this case has a geometric index calculation unit 141 as in the example shown in FIG.

The geometric index calculation unit 141 uses as an index whether or not two piecewise continuous lines can be approximated as one continuous line, an approximation error, and the length of an approximate curve. In the present embodiment, an approximation of two piecewise continuous lines as one curve is called an integrated curve model.

Since this integrated curve model cannot be obtained analytically, it is obtained by RANSAC in the same manner as in the process of step S3. That is, the curve parameter having the smallest approximation error is estimated from the hypothetical parameters. Then, the geometric index calculation unit 141 obtains it as a weighted sum of the variation amount of the approximation error of the estimated integrated curve model and the variation amount of the line segment length.

The details of the above-mentioned operation of the geometric index calculation unit 141 will be described. FIG. 9 shows a flowchart of an example of the operation of the geometric index calculation unit 141.

As shown in FIG. 9, in step S4-3-1A, the geometric index calculation unit 141 adds “1” to the management variable Loop for repeating the predetermined predetermined number of times of processing. The initial value of the repeating variable Loop when the execution of this process is started is 0. In this embodiment, the specified number of repetitions is 2000 times.

As shown on the left side of FIG. 10, two continuous lines that are easily connected are easily approximated by one quadratic curve, and the value of the connectivity index is output small. On the other hand, as shown in the figure on the right side of FIG. 10, two continuous lines that are difficult to be connected are difficult to approximate with one quadratic curve (the approximation error becomes large), so that the connectivity index tends to be large.

In the next step S4-3-2A, the geometric index calculation unit 141 derives an integrated curve model candidate. In deriving the integrated curve model candidate, the parameters of the quadratic curve may be obtained by the same process as in step S3-2. However, it is different from the above step S3 that the point cloud used here is processed by using the wire constituent points belonging to the piecewise continuous or connection candidate line of interest.

In the next step S4-3-3A, the geometric index calculation unit 141 calculates the geometric index value which is the evaluation value of the integrated curve model. Similar to step S3-3, the evaluation score is calculated using the wire constituent points having the distance threshold value εc or less from the quadratic curve. However, the point cloud used for the evaluation is different from step S3 in that the processing is performed using the wire constituent points belonging to the piecewise continuous or connection candidate line of interest.

The geometric index can be obtained by the following equation (4).

Geometric index value = ψ + λ × ξ ・ ・ ・ (4)
However, in equation (4), ψ is the approximation error increase rate ((approximate error of the sectioned continuous line of interest + approximation error of the connection candidate) / approximation error of the integrated curve model), λ is the weighting coefficient, and ξ is the line segment. The rate of increase in the length of the minute (the length of the continuous line segmented by the line of interest + the length of the connection candidate line) / the length of the integrated curve model).

The approximation error is the distance of a group of points included in the quadratic curve (distance of εc or less from the arc), and can be calculated as the average value of the distances from the arc. The approximation error in the integrated curve model may be evaluated using the wire component points belonging to the piecewise continuous line of interest and the wire model component points belonging to the connection candidates.

For the derivation of the approximation error, the average error may be obtained. The error E ^(Cn) of the ^Cnth piecewise continuous line of interest can be obtained by the following equation using the distance function D.

However, N _Cn denotes the number of wires constituting points belonging to piecewise continuous line, p _i denotes the wire configuration points belonging to piecewise continuous line. Similarly, the average error of the connection candidate curve and the integrated curve model can be calculated.

The length of the integrated curve model can also be obtained by using the wire constituent points having the distance threshold value εc or less in the same manner as in step S3. In this example, λ = 0.1.

However, since the wire model may have three-dimensional distortion, the wire constituent points within the distance Rcon [m] from the end points to be connected are used for the piecewise continuous line and the candidate line, respectively. To evaluate. As shown in FIG. 11, a point cloud within Rcon [m] from an end point close to the line segment to be connected is targeted. In this embodiment, Rcon = 2 [m].

In the next step S4-3-4A, the geometric index calculation unit 141 determines whether or not the specified number of times has been processed. If the variable Loop has not reached the specified number of times, the process of step S4-3-4A is determined to be negative, the process returns to step S4-3-1A, and the above process is repeated. On the other hand, when the variable Loop reaches the specified number of times, the process of step S4-3-4A becomes an affirmative determination, and the calculation of the connectivity index shown in FIG. 9 ends.

In the method (B), the connectivity index calculation unit 132 uses a three-dimensional feature quantity in order to consider not only the piecewise continuous line and the connection candidate line of interest but also the shape information of the surrounding object. Feature quantities such as PFH, FPFH, and SHOT may be used as the three-dimensional feature quantity, but in order to improve the accuracy, the two line segments to be connected and the shape information around the two line segments are used. The technique of Non-Patent Document 3 is used in order to clearly separate and characterize.

Using the method of Non-Patent Document 3, the shape feature amount of the piecewise continuous line of interest itself and the shape feature amount of the piecewise continuous line existing around the piecewise continuous line of interest are calculated, and ( The feature amount that integrates the geometric index amount and the shape feature amount calculated in the method of A) is calculated. In (A), threshold processing is performed on a scalar value to which a geometric index amount is added as a weighted sum, but in (B), which feature is important is automatically determined by machine learning.

In this case, the connectivity index calculation unit 132 includes a geometric index calculation unit 141, a 3D feature amount calculation unit 142, and a feature vector derivation unit 143, as in the example shown in FIG. Details of the operation of each of the above units of the connectivity index calculation unit 132 will be described. FIG. 13 shows a flowchart of an example of the operation of the 3D feature amount calculation unit 142 from the piecewise continuous line and the connection candidate line of interest.

In step S4-3-1B, the 3D feature amount calculation unit 142 manages by the number p (p = 1,2,3, ...) Of the wire constituent points belonging to the piecewise continuous line (curve) of interest. The initial value of p when the execution of this process is started is set to 0, and the 3D feature amount calculation unit 142 adds "1" to p in this step.

In the next step S4-3-2B, the 3D feature calculation unit 142 sets a local region at the position of the piecewise continuous line of interest, and in the next step S4-3-3B, the 3D feature calculation unit 142 sets the periphery. Calculate the features of the histogram generated between the points.

As shown in FIG. 14, for the wire constituent points belonging to the piecewise continuous line of interest, the points of interest that exist in the region within the radius R centered on the point of interest p _i ^(c) (query point).

Generates a relative angle histogram (see also FIG. 16) with. For the point of interest and the point of interest, the local coordinate system vectors u, v, w and the relative position vector t _k are defined by the two relative vectors, the normal vector, and the minimum curvature direction (extrusion direction). Regarding the local coordinate system vector u, a normal is used as an example in this embodiment.

In FIG. 14, the filled point cloud is a piecewise continuous line of interest, and the unfilled point cloud is a connection candidate line from the peripheral continuous line (the connection candidate line of FIG. 14). 1 and connection candidate line 2) are excluded.

As shown in FIG. 16, a histogram bin of the histogram feature (relative angle feature) is determined by the magnitudes of the relative angles α, ψ, θ of the defined local coordinate system vectors and the magnitude of the angle τ formed in the vertical direction. The position of is found.


However, the symbol "・" means multiplication of numerical values or the inner product between vectors, and the symbol "→" means assigning a value to a variable. However, u ⁽ q _k ⁾ means the normal direction at the point of interest q _k ^(c) .

The feature amount is described by the frequency distribution of the obtained bin positions. In the present embodiment, if the number N _in ^{(c) of the} point cloud of interest is, for example, N _in ^(c) = 600 points, the total value of the frequency distribution extracted from the area of interest is also 600. When features are extracted from the normal vector and the minimum curvature direction under the same conditions as in the non-patent documents, the feature vector becomes a 1500-dimensional vector.

Similarly, regarding the point of interest, the histogram feature may be extracted from the point cloud existing in the area around the object of interest as shown in FIGS. 19 and 16.

In the next step S4-3-4B, the feature vector derivation unit 143 determines whether or not a predetermined termination condition is satisfied. If the predetermined end condition is not satisfied, the process of step S4-3-4B is determined to be negative, the process returns to step S4-3-1B, and the above process is repeated. On the other hand, when the end condition is satisfied, such as when the variable P reaches the specified number of times, the process of step S4-3-4B becomes an affirmative determination, and the calculation of the connectivity index shown in FIG. 13 ends. The feature vector deriving unit 143 has the geometric index value calculated by the geometric index calculation unit 141 and the 3D feature when the above end condition is satisfied, that is, when the determination in step S4-3-4B is affirmative. A feature vector integrated with the 3D feature calculated by the quantity calculation unit 142 is output.

The feature amount μ obtained in (B) is the value of the geometric index (approximate error increase / decrease rate E, curve length increase / decrease rate r), and the 3D feature amount extracted from each of the piecewise continuous line and the connection candidate line of interest. It is a thing. As shown in FIG. 11, as a method of integrating 3D feature quantities, feature vectors may be added or connected.

When the process of step S4-3 shown in FIG. 6 is completed in this way, the connection determination unit 133 performs the connection determination process for each connection candidate line in the next step S4-4.

When the connectivity index is calculated by the method (A), the connectivity determination unit 133 determines the threshold value in which the approximate error increase / decrease value and the curve length increase / decrease value are predetermined for the integrated curve model obtained by RANSAC. A piecewise continuous line, which is a connection candidate line smaller than THERR, is determined to be a connection determination target. In the present embodiment, the threshold value Therr = 0.2.

On the other hand, when the connectivity index is calculated by the method (B) above, the connection determination unit 133 determines the connection determination target by the classifier derived in advance by machine learning. The features and geometric index quantities extracted from each piecewise continuous line are integrated by concatenating them.

As shown in FIG. 17, two 3D feature quantity vectors (1500-dimensional vector) are connected to the geometric index vector (two-dimensional vector) in which the approximate error amount obtained by the geometric index and the curve model length increase / decrease amount are used as vectors. As a result, it becomes a 3002 dimensional vector.

There are several methods for integrating 3D features. For example, a 1500-dimensional vector may be obtained by adding vector elements instead of concatenating them.

In this embodiment, the AdaBoost classifier was used as in Non-Patent Document 3. In this case, it can be solved as a two-class problem in which the case of connecting is positive and the case of not connecting is negative.

When the output of the classifier H when the feature vector μ is input is positive (0 or more), it is determined to be connected. Here, the number T of the weak classifiers h is T = 1000.

In the present embodiment, the boosting classifier is shown, but a classifier such as SVM (support vector machine) or DNN (deep neural network) may be used. Further, although a two-value discriminator of connection and non-connection is used, a classifier that ranks the ease of connection as in rank learning may be used. In the case of rank learning, if the output value of ease of connection is equal to or greater than the threshold value, it may be determined that the connection is performed.

In the next step S4-5, the connection determination unit 133 determines whether or not a predetermined termination condition is satisfied. In the present embodiment, it is determined that the predetermined end condition is satisfied when the number c is Nc, that is, when all the piecewise continuous lines are processed. If the end condition is not satisfied, the determination in step S4-5 becomes a negative determination, the process returns to step S4-1, and the above process is repeated. On the other hand, when the end condition is satisfied, the process of step S4-5 becomes an affirmative determination, and the process proceeds to step S4-6.

In step S4-6, the connection determination unit 133 performs the connection process for connecting the piecewise continuous lines, and then ends the piecewise continuous line connection process shown in FIG.

Here, with respect to the piecewise continuous line and the connection candidate line of interest, those existing on the same wire tend to have a small value of the connectivity index derived in step S4-3. As shown in FIG. 18, when it is determined that the connection candidate line 1 and the connection candidate line 2 are connected to the piecewise continuous line of interest, the connection determination unit 133 selects the one having the closest center position distance in this step. Judge as a connection judgment target. In the case shown in FIG. 18, the connection determination unit 133 determines the piecewise continuous line of interest and the connection candidate line 1 as the connection determination target, and connects the piecewise continuous line of interest and the connection candidate line 1.

When the piecewise continuous line connection process in step S4 is completed in this way, the wire model generation unit 120 ends the operation as shown in FIG. At the end of this operation, the wire model generation unit 120 outputs the wire model obtained by connecting the obtained piecewise continuous lines to the equipment information storage unit 112 and stores it.

[Second Embodiment]
As described above, the model generation device 100 of the second embodiment can estimate the distance between the connected wire model and the surrounding structure, and obtain the separation distance in each wire model. Since the model generation device 100 of the present embodiment includes the same configuration and operation as the model generation device 100 of the first embodiment, detailed description of the same configuration and operation will be omitted.

FIG. 19 is a block diagram showing an example of the configuration of the model generation device 100 of the present embodiment. As shown in FIG. 19, the model generation device 100 of the present embodiment is the model generation device 100 of the first embodiment in that the wire model generation unit 120 of the calculation unit 104 further includes the peripheral distance estimation unit 123 (see FIG. 1). ) Is different.

The wire model generated by the piecewise continuous line connecting unit 122 is input to the peripheral distance estimation unit 123, and the point cloud attribute information (details will be described later) is input. The peripheral distance estimation unit 123 calculates the distance between the wire model and the surrounding structure. The distance between the wire model generated by the peripheral distance estimation unit 123 and the peripheral structure is output to the equipment information storage unit 112 of the storage unit 103 and stored in the equipment information storage unit 112.

(Processing flow of the wire model generation unit 120 of the calculation unit 104)
FIG. 20 is a flowchart showing an overall example of the operation of the wire model generation unit 120 of the present embodiment. As shown in FIG. 20, the operation of the wire model generation unit 120 of the present embodiment is different from the operation of the wire model generation unit 120 of the first embodiment (see FIG. 3) in that the process of step S5 is included. Step S5 corresponds to the peripheral distance estimation unit 123.

As shown in FIG. 20, the peripheral distance estimation unit 123 in step S5 calculates the distance between the wire model and the peripheral structure detected by the process of step S4 described above. Specifically, the peripheral distance estimation unit 123 executes the peripheral distance estimation process shown in FIG. 21 as an example, and the detected wire model and the three-dimensional point cloud (wire constituent points) acquired from the three-dimensional point cloud storage unit 110. And other point clouds) and the point cloud attribute information are used to estimate the separation distance of each wire model. The point cloud attribute information is the result of identifying what each point is, and in the present embodiment, it is referred to as an infrastructure facility "wire constituent point" and "telephone pole". The separation distance is particularly important information about the power cable, and the magnitude of the voltage often defines the minimum required distance from the surrounding structure. Communication cables are also often specified from the viewpoint of maintenance work, such as the minimum ground clearance to avoid contact with automobiles.

As shown in FIG. 21, the peripheral distance estimation unit 123 removes the point cloud determined to be the infrastructure equipment from the acquired three-dimensional point cloud in step S5-1, and then performs a filtering process for noise removal. Since dust in the air and metal fittings for adjusting tension are present around the cable, when measuring the distance between the wire model and the surrounding point cloud, a very short distance of several cm or less is measured. Therefore, first, dust and measurement noise in the air are removed by noise filtering, and then small objects (point clouds) are removed in order to ignore the distance from fine metal fittings and cable accessories.

Noise filtering is to remove sparse outliers using a commonly used sphere region. When paying attention to a certain point, if the point cloud density in the sphere in a certain region from that point is small, it is determined as noise. In the present embodiment, the radius of the sphere that defines a certain area is set to r = 0.2 [m].

In the next step S5-2, the peripheral distance estimation unit 123 clusters the three-dimensional point cloud after filtering by the process of step S5-1 with the k-nearest neighbor points. In this embodiment, the number of neighboring points k = 8 is set. However, in order to suppress the influence of the isolated point cloud of k points or less, clustering is not performed when the neighboring point distances are separated by an allowable distance of 1 [m] or more.

In the next step S5-3, the peripheral distance estimation unit 123 refers to the three-dimensional point cloud clustered by the process of the above step S5-2, and accessories having a predetermined size or less are present around the wire (for example, a utility pole). Remove it by regarding it as a hardware, a device that holds the tension of the wire, etc.). The peripheral distance estimation unit 123 determines that a three-dimensional point cloud having a predetermined size or larger is a peripheral structure point cloud. In the present embodiment, when the size of the point cloud cluster is 1 [m] or more in any of the X-axis direction, the Y-axis direction, and the Z-axis direction, it is determined to be a structure. If all are below the threshold value, it is judged as an accessory and removed.

In the next step S5-4, the peripheral distance estimation unit 123 calculates the ground clearance by calculating the distance between each wire model input from the piecewise continuous line connecting unit 122 and the point cloud of the ground attribute. Keep a short distance and the corresponding position as the minimum ground clearance. Next, the distance to the peripheral structure point cloud is calculated, the closest distance (nearest distance) and the corresponding position on the wire model are derived, and the peripheral distance estimation process in step S5 is completed. Here, the corresponding position means the position of the wire model at the closest distance to the point cloud of the peripheral structure or the ground attribute.

When the peripheral distance estimation process in step S5 is completed in this way, the wire model generation unit 120 ends the operation as shown in FIG. 20. At the end of this operation, the peripheral distance estimation unit 123 (wire model generation unit 120) outputs the detected wire model parameter and the distance between the wire parameter and the peripheral structure to the equipment information storage unit 112. Remember.

As described above, the model generation device 100 of each of the above embodiments is a model generation device 100 that generates a wire model from the input three-dimensional point cloud information of the subject. The model generator 100 detects the wire constituent points that make up the wire from the three-dimensional point cloud information, and detects a plurality of piecewise continuous lines that are continuous curves existing in the local region from the detected wire constituent points. Of the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit 121 and the piecewise continuous line detection unit 121, it is determined whether or not to connect the two piecewise continuous lines to be connected. It includes a piecewise continuous line connecting portion 122 that generates a wire model by connecting two piecewise continuous lines determined to be connected. Further, according to the second embodiment, the wire model maintains the separation distance from the surrounding structure and the ground clearance. As the definition of the wording of the embodiment of the present invention, "separation distance" and "ground clearance" are described as different physical quantities, but in the sense of inclusion, the ground clearance is also considered to be a structure. May be included.

As described above, according to the model generation device 100 of each of the above embodiments, the wire can be robustly and accurately detected from the three-dimensional point cloud.

That is, according to the model generation device 100 of each of the above embodiments, even if the wire has a three-dimensional distortion in its shape, even if a large defect occurs in the measurement point due to the influence of occlusion or the like. The wire can be detected robustly. Further, since the shape of the wire can be estimated accurately even in a region where the point cloud is missing, such as an occlusion region, the separation distance estimation accuracy can be improved even if the point cloud of the wire is missing.

As a result, according to the model generator 100 of each of the above embodiments, it is possible to grasp the position and shape of the wire of the infrastructure equipment and the separation distance from the surrounding structures, and support the work efficiency of the equipment maintenance work. it can.

As described above, each of the above embodiments is an example of the present disclosure, and the specific configuration is not limited to the above-mentioned case embodiment, and it goes without saying that various changes can be made without departing from the gist thereof.

The model generator 100 of the present embodiment has a computer system inside, but if the "computer system" is using a WWW (World Wide Web) system, the homepage providing environment ( Alternatively, the display environment) is also included.

Further, in the present embodiment, the mode in which the program is pre-installed has been described, but the program can be stored in a computer-readable recording medium and provided, or provided via a network. Is also possible.

100 Model generation device 101 Subject measurement unit 102 Input unit 103 Storage unit 104 Calculation unit 110 Three-dimensional point cloud storage unit 111 Calculation processing parameter storage unit 112 Equipment information storage unit 120 Wire model generation unit 121 Piecewise continuous line detection unit 122 Target continuous line connection unit 131 Connection candidate line detection unit 132 Connectivity index calculation unit 133 Connection judgment unit 141 Geometric index calculation unit 142 3D feature quantity calculation unit 143 Feature vector calculation unit

Claims (8)

It is a model generator that generates a wire model from the input 3D point cloud information of the subject.
From the three-dimensional point cloud information, a wire constituent point cloud constituting a wire is detected, and from the detected wire constituent point cloud, a plurality of piecewise continuous lines which are continuous curves existing in a local region are detected. Line detector and
Of the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit, it is determined whether or not to connect the two piecewise continuous lines to be connected, and the two piecewise continuous lines determined to be connected are connected. A piecewise continuous line connecting part that generates a wire model by connecting continuous lines,
A wire model generated by the piecewise continuous line connecting portion and a peripheral distance estimation unit for calculating the distance between the peripheral structures of the wire model are provided.
The peripheral distance estimation unit
From the three-dimensional point cloud information, the point cloud excluding the wire constituent point cloud detected by the sectional continuous line detection unit is clustered based on the similarity of the local shape to obtain the peripheral structure point cloud, and the peripheral structure point cloud is obtained. Calculate the distance between the structure point cloud and the wire model,
Model generator. It is a model generator that generates a wire model from the input 3D point cloud information of the subject.
From the three-dimensional point cloud information, a wire constituent point cloud constituting a wire is detected, and from the detected wire constituent point cloud, a plurality of piecewise continuous lines which are continuous curves existing in a local region are detected. Line detector and
Of the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit, it is determined whether or not to connect the two piecewise continuous lines to be connected, and the two piecewise continuous lines determined to be connected are connected. A piecewise continuous line connecting part that generates a wire model by connecting continuous lines,
A wire model generated by the piecewise continuous line connecting portion and a peripheral distance estimation unit for calculating the distance between the peripheral structures of the wire model are provided.
The peripheral distance estimation unit performs noise removal processing based on the density of the point cloud and clustering processing with neighboring points.
Model generator. The piecewise continuous line connecting portion is subject to the connection determination target 2 using the error when the two piecewise continuous lines to be the connection determination target are approximated as one continuous line and the length of the continuous line as an index. The model generator according to claim 1 or 2 , which determines whether or not to connect two piecewise continuous lines. It is a model generator that generates a wire model from the input 3D point cloud information of the subject.
From the three-dimensional point cloud information, a wire constituent point cloud constituting a wire is detected, and from the detected wire constituent point cloud, a plurality of piecewise continuous lines which are continuous curves existing in a local region are detected. Line detector and
Of the plurality of piecewise continuous lines detected by the piecewise continuous line detection unit, it is determined whether or not to connect the two piecewise continuous lines to be connected, and the two piecewise continuous lines determined to be connected are connected. A piecewise continuous line connecting part that generates a wire model by connecting continuous lines,
With
The piecewise continuous line connecting portion includes an error when two piecewise continuous lines to be connected and determined are approximated as one piecewise continuous line, the length of the piecewise line, and the two piecewise continuous lines to be judged to be connected. Using a classifier based on the line and the three-dimensional feature quantity extracted from the periphery of the two piecewise continuous lines, it is determined whether or not the two piecewise continuous lines to be connected are connected.
Model generator. It is a generation method that generates a wire model from the input 3D point cloud information of the subject.
The piecewise continuous line detection unit detects the wire point cloud that constitutes the wire from the three-dimensional point cloud information, and from the detected wire point cloud, a plurality of piecewise curves that are continuous curves existing in the local region. Detects continuous lines and
The sectioned continuous line connecting unit determines whether or not to connect the two piecewise continuous lines to be connected, among the plurality of piecewise continuous lines detected by the piecewise continuous line detecting unit, and connects them. then by connecting the two piecewise continuous line which is determined to generate a wire model,
The peripheral distance estimation unit includes calculating the distance between the wire model generated by the piecewise continuous line connecting unit and the structure around the wire model.
According to the calculation by the peripheral distance estimation unit,
From the three-dimensional point cloud information, the point cloud excluding the wire constituent point cloud detected by the sectional continuous line detection unit is clustered based on the similarity of the local shape to obtain the peripheral structure point cloud, and the peripheral structure point cloud is obtained. Calculate the distance between the structure point cloud and the wire model,
How to generate a wire model. It is a generation method that generates a wire model from the input 3D point cloud information of the subject.
The piecewise continuous line detection unit detects the wire point cloud that constitutes the wire from the three-dimensional point cloud information, and from the detected wire point cloud, a plurality of piecewise curves that are continuous curves existing in the local region. Detects continuous lines and
The sectioned continuous line connecting unit determines whether or not to connect the two piecewise continuous lines to be connected, among the plurality of piecewise continuous lines detected by the piecewise continuous line detecting unit, and connects them. then by connecting the two piecewise continuous line which is determined to generate a wire model,
The peripheral distance estimation unit includes calculating the distance between the wire model generated by the piecewise continuous line connecting unit and the structure around the wire model.
In the calculation by the peripheral distance estimation unit, noise removal processing based on the density of the point cloud and clustering processing with neighboring points are performed.
How to generate a wire model. It is a generation method that generates a wire model from the input 3D point cloud information of the subject.
The piecewise continuous line detection unit detects the wire point cloud that constitutes the wire from the three-dimensional point cloud information, and from the detected wire point cloud, a plurality of piecewise curves that are continuous curves existing in the local region. Detects continuous lines and
The sectioned continuous line connecting unit determines whether or not to connect the two piecewise continuous lines to be connected, among the plurality of piecewise continuous lines detected by the piecewise continuous line detecting unit, and connects them. Including generating a wire model by connecting two piecewise continuous lines determined to be
When the piecewise continuous line connecting unit determines, the error when the two piecewise continuous lines to be connected are approximated as one continuous line, the length of the continuous line, and the above 2 to be connected. Whether or not to connect the two piecewise continuous lines to be connected is determined by using a discriminator based on the two piecewise continuous lines and the three-dimensional feature quantity extracted from the periphery of the two piecewise continuous lines. judge,
How to generate a wire model. A program for causing a computer to function as each part of the model generator according to any one of claims 1 to 4 .