JP6652865B2 - Feature detection device, feature detection method, and program - Google Patents

Description

  The present invention relates to a feature detection device, a feature detection method, and a program for detecting a target feature rising from the ground based on three-dimensional coordinate data of a point cloud extracted from a feature surface in a target space.

  2. Description of the Related Art Conventionally, there has been disclosed a technique for acquiring three-dimensional point cloud data representing a shape of a terrestrial object using a laser scanner. For example, in a mobile mapping system, a laser scanner is mounted on an upper part of an automobile and irradiates a laser to the surroundings. The optical axis of the laser is scanned in the vertical direction and the horizontal direction by changing the elevation angle and the azimuth angle, and laser pulses are emitted at small angles within the scanning range. The distance is measured based on the time from the emission of the laser to the reception of the reflected light, and at that time, the emission direction of the laser, the time, and the position / posture of the vehicle body are measured. From these measurement data, point group data representing the three-dimensional coordinates of the point reflecting the laser pulse is obtained.

  Patent Document 1 discloses a technology for automatically detecting the wall surface of a feature from point cloud data, and Patent Document 2 discloses a technology for automatically detecting a columnar feature such as a telephone pole from point cloud data. Has been disclosed.

JP 2013-053915 A JP 2013-064688 A

  In the case of detecting a plurality of types of features from the point cloud data, processing for the point cloud data is conventionally performed for each type of the features, and the processing efficiency is not necessarily high. In this regard, it is conceivable to improve the processing efficiency by creating a device or a program for directly detecting a plurality of types of predetermined features from point cloud data. However, there is a problem in that adding a type of a feature to be detected complicates changes in necessary devices, program configurations, and the like.

  The present invention has been made to solve the above problems, and can improve the efficiency of processing for detecting a plurality of types of destinations from point cloud data, and can flexibly add / change destinations. An object of the present invention is to provide a feature detection device, a feature detection method, and a program that can cope with the above.

  (1) A feature detection device according to the present invention is a device for detecting a target feature rising from the ground surface based on three-dimensional coordinate data of a point cloud extracted from a feature surface in a target space. Using a plurality of types of element figures having predetermined shape characteristics in the original space, a feature model in which the standard shape of the destination is represented by one or a plurality of the element figures arranged in a height direction in advance. The stored model storage means, the layered space setting means for dividing the target space into a plurality of horizontal layered spaces, and in each of the layered spaces, the point group projected on a horizontal plane and gathered near a predetermined reference or more. Edge detecting means for detecting an edge that is a position of a feature surface, and classifying the edge into ones of the elemental figures that produce a projection corresponding to the shape of the edge, and associating the edge with the classification Object generation means for generating a stratified object, and obtaining a group consisting of a plurality of stratified objects, each edge of which is arranged along the height direction, and determining how to sort the classification along the height direction in the group. A destination feature detecting unit that detects the destination feature by comparing with the feature model.

  (2) In the feature detection device according to (1), the plurality of layered spaces may have the same height dimension.

  (3) In the feature detection device according to (1) or (2), the destination feature detection unit may be configured to determine a position of a first edge and a second edge existing in the different layered spaces in a horizontal plane with each other. The first edge and the second edge may be determined to be aligned in the height direction when an edge vicinity region having a predetermined width with the first edge as a center line and the second edge overlap. it can.

  (4) In the feature detection device according to any one of (1) to (3), the element graphic has a non-circular shape in a horizontal cross section and a first substantially vertical surface having a length equal to or longer than a predetermined threshold length. And a second substantially vertical surface having a non-circular shape in a horizontal cross section and a length less than the threshold length, and a first cylinder having a horizontal cross section having a radius equal to or greater than a predetermined threshold radius. And a second cylinder having a horizontal cross section having a circular shape having a radius smaller than the threshold radius.

  (5) The feature detection method according to the present invention is a method for detecting a target feature rising from the ground based on three-dimensional coordinate data of a point cloud extracted from a feature surface in a target space, A layered space setting step of dividing the target space into a plurality of horizontal layered spaces; and, in each of the layered spaces, an edge which is a position on the surface of the feature, wherein the point cloud projected on a horizontal plane and gathered near a predetermined reference or more is located. Edge detection step, and classifying the edge into a plurality of types of elemental figures having a predetermined shape characteristic in a three-dimensional space that produce a projection corresponding to the shape of the edge, and the edge and the classification And an object generating step of generating a stratified object in which each of the stratums is arranged along the height direction. A grouping step of obtaining a group; and a step of storing the feature model from a model storage unit in which a feature model represented by one or a plurality of the elemental figures arranged in a standard shape of the destination feature along one or more height directions is stored in advance. Reading, and a feature model matching step of detecting the destination feature by matching the arrangement of the classifications along the height direction in the group with the feature model.

  (6) A program according to the present invention is a program for causing a computer to perform a process of detecting a destination object rising from the ground surface based on three-dimensional coordinate data of a point cloud extracted from a surface of a feature in a target space. Wherein the computer uses a plurality of types of element figures having a predetermined shape characteristic in a three-dimensional space, and the standard shape of the destination object is one or a plurality of the element elements arranged in a height direction. Model storage means pre-stored a feature model represented by a graphic, layered space setting means for dividing the target space into a plurality of horizontal layered spaces, and in each of the layered spaces, a level projected on a horizontal plane and higher than a predetermined reference. Edge detecting means for detecting the point group gathering in the vicinity as an edge which is a position on the surface of a feature, and projecting the edge corresponding to the shape of the edge in the element graphic Object generating means for generating a stratified object in which the edge and the classification are associated with each other, and a group including a plurality of stratified objects in which the edges are arranged along the height direction. Then, it is made to function as destination feature detection means for detecting the destination by comparing the arrangement of the classifications along the height direction in the group with the feature model.

  ADVANTAGE OF THE INVENTION According to this invention, the efficiency of the process which detects several types of destinations from point cloud data can be improved, and it can respond flexibly to addition / change of a destination.

1 is a block diagram illustrating a schematic configuration of a feature detection system according to an embodiment of the present invention. It is a general | schematic flowchart of the process of the layered space setting means of the feature detection system which concerns on embodiment of this invention, an edge detection means, and an object generation means. It is a general | schematic flowchart of the process of the layered space setting means of the feature detection system which concerns on embodiment of this invention, an edge detection means, and an object generation means. It is a general | schematic flowchart of the process of the layered space setting means of the feature detection system which concerns on embodiment of this invention, an edge detection means, and an object generation means. FIG. 9 is a schematic diagram illustrating an edge search process in a block space. It is a schematic plan view explaining the processing which checks whether an edge has a possibility of a destination. It is a schematic plan view explaining the processing which checks whether an edge has a possibility of a destination. It is an outline flow figure of a continuous edge search processing. It is an outline flow figure of a continuous edge search processing. It is a schematic flowchart of a continuous tracking process. It is a schematic flowchart of a continuous tracking process. It is a general | schematic flowchart of a cylinder search process. It is a schematic plan view which shows an example of the continuous line segment produced | generated by the continuous tracking process. FIG. 9 is a schematic flowchart of edge fine adjustment processing. It is a schematic plan view explaining the fine adjustment processing of an edge. FIG. 14 is a schematic diagram showing representative points obtained by fine adjustment processing on the continuous line segment shown in FIG. 13. It is a schematic diagram which shows an example of a re-search space. FIG. 4 is a schematic flowchart of a short edge process. It is a schematic plan view explaining the process which determines a thin pole. It is a schematic plan view explaining the overlap of the edge of a different layered space. It is a general | schematic flowchart which shows the example of a process of the destination detection means about the destination including a pole, such as a road sign. It is a general | schematic flowchart of a pole discriminating process. It is a general | schematic flowchart of a pole discriminating process.

  Hereinafter, a feature detection system according to an embodiment of the present invention (hereinafter, referred to as an embodiment) will be described with reference to the drawings. This system is a feature detection device that detects a destination feature rising from the ground surface based on three-dimensional coordinate data of a point cloud extracted from a feature surface in a target space. The data of the point cloud is acquired by, for example, a laser scanner mounted on a vehicle traveling on the ground as in the above-described mobile mapping system. Alternatively, the measurement may be performed by installing a laser scanner on the ground. In order for the point cloud data to represent the three-dimensional shape of the feature surface, laser scanning must be performed at a point density corresponding to the scale of the unevenness of the feature surface. In this regard, a laser scan performed from the ground such as a road using a vehicle or a tripod can achieve a scan density enough to capture shapes of buildings, walls, signboards, signs, telephone poles, and the like that are built near the road, for example. .

  FIG. 1 is a block diagram illustrating a schematic configuration of the feature detection system 2. The present system includes an arithmetic processing unit 4, a storage device 6, an input device 8, and an output device 10. Although it is possible to create dedicated hardware for performing various arithmetic processes of the present system as the arithmetic processing device 4, in the present embodiment, the arithmetic processing device 4 uses a computer and a program executed on the computer. Is built.

  A CPU (Central Processing Unit) of the computer constitutes the arithmetic processing unit 4 and functions as a layered space setting unit 20, an edge detection unit 22, an object generation unit 24, and a destination object detection unit 26, which will be described later.

  The storage device 6 is configured by a hard disk or the like built in the computer. The storage device 6 is a program for causing the arithmetic processing device 4 to function as the layered space setting unit 20, the edge detection unit 22, the object generation unit 24, and the destination object detection unit 26, and other programs, and necessary for processing of the present system. Various data is stored. For example, the storage device 6 stores the point cloud data 30 of the analysis target space as the processing target data. For example, an area along a road is a target space for analysis. The storage device 6 is a model storage unit that stores in advance a feature model 32 that describes a typical shape, a standard shape, or a condition of a destination feature that is a feature to be detected.

  The input device 8 is a keyboard, a mouse, and the like, and is used by a user to operate the system.

  The output device 10 is a display, a printer, or the like, and is used, for example, to display the position of the object generated by the present system or the detected destination in the target space to the user by screen display, printing, or the like.

The layered space setting means 20 divides the target space into a plurality of horizontal layered spaces. When the target space is defined by an XYZ orthogonal coordinate system in which two axes defining the horizontal plane are defined as the X axis and the Y axis with the axis having the upward direction in the height direction as the positive axis defined as the Z axis, the layered space setting means 20 sets a plurality of Z values. The target space is divided using the set horizontal plane (XY plane) as a division plane. It is preferable that the dimensions in the height direction, that is, the thicknesses, of the respective layered spaces be the same, thereby simplifying the processing and ensuring the processing efficiency. For example, the thickness _HL of each layered space can be set to 50 cm [cm]. In the program constituting the layered space setting means 20, _HL is parameterized and can be changed by the user using the input device 8, for example. In addition, as described later, the layered space setting unit 20 further divides each layered space into a mesh in the processing of edge detection.

  In each layered space, the edge detecting means 22 projects a point group existing in the layered space on a horizontal plane, and indicates a line on which the projected point group gathers closer than a predetermined reference or more on the surface of the feature. Detect as an edge.

  The object generation unit 24 classifies each edge into a plurality of types of elemental figures having a predetermined shape characteristic in a three-dimensional space that generate a projection corresponding to the shape of the edge, and classifies the edge and the classification. Generate the associated stratified object.

  The destination feature detection unit 26 obtains a group including a plurality of stratified objects, in which the edges of the stratified objects are arranged along the height direction. Then, the destination feature detecting means 26 reads the feature model 32 from the storage device 6 and detects the destination feature by comparing the arrangement of the classification of stratified objects along the height direction in the group with the feature model 32. I do.

The feature model 32 is, as described above, a standard shape of a destination feature represented by one or a plurality of element figures arranged along the height direction or a condition for defining the shape. The element graphic is a shape having a predetermined characteristic in a three-dimensional space. For example, the element graphic includes a substantially vertical surface having a non-circular shape in a horizontal cross section, and a cylinder having a circular horizontal cross section. Further, the element graphics can be classified according to the size. For example, in the present embodiment, the substantially vertical surface has a first substantially vertical surface having a length equal to or longer than a predetermined threshold length L _TH1 (for example, 1.5 m [m]), and the first vertical surface has a length less than the threshold length L _TH1. And a second substantially vertical surface having a length of Further, the first cylinder is a circular cylinder having a horizontal cross section having a radius equal to or larger than a predetermined threshold radius R _TH (for example, 0.1 m), and the first cylinder is a circle having a horizontal cross section having a radius smaller than the threshold radius R _TH . It is divided into 2 cylinders.

  FIGS. 2 to 4 are schematic flowcharts of the processing of the layered space setting unit 20, the edge detection unit 22, and the object generation unit 24 of the feature detection system 2.

In the feature detection system 2, the layered space setting means 20 divides the target space at intervals _HL in the Z-axis direction to set a layered space, and performs processing for detecting an edge existing in each layered space. In the present embodiment, a block space obtained by further dividing the layered space into a two-dimensional mesh is searched for a block having an edge, and when a block space having an edge is found, the continuity of the edge in the layered space is checked. For example, the target space is divided into meshes along the X-axis direction and the Y-axis direction. Here, assuming that an interval of the mesh in the X-axis direction is width W and an interval of the mesh in the Y-axis direction is depth D, W and D are set as parameters similarly to _HL . In this embodiment, W and D are set to 50 cm. When the target space is divided into meshes, the target space is divided into columnar spaces. The columnar space is referred to as a partial space. The partial space has a structure in which block spaces for each layer of the layered space are vertically stacked. The number of arrays in the X and Y directions of the subspace is determined according to the size of the target space and the values of the width W and the depth D, and the range of the index indicating the position of the subspace in the X and Y directions is determined by the number of arrays. Is determined. The edge detecting means 22 changes indices representing the two-dimensional mesh arrangement of the subspace in a predetermined order, and defines a subspace defined by a position (coordinate range in each of the X and Y directions) according to the set index. Is selected as a processing target (step S2).

  If the partial space of the mesh designated by the index is unprocessed ("No" in step S4), the layered space setting means 20 stores the point group in the partial space in a storage device 6 such as a hard disk. (Step S6). When there is a height difference equal to or greater than the predetermined threshold Th1 in the point group taken into the partial space (in the case of "Yes" in step S8), the layered space setting means 20 determines that the partial space is the target partial space. Then, processing such as searching for an edge in the focused partial space is started (processing proceeds to node B in FIG. 3). For example, the threshold Th1 can be set small enough to exclude a partial space including only a substantially flat portion of the ground such as a road.

  If the height difference of the point group taken from the set partial space is smaller than the threshold value Th1 ("No" in S8), the edge detection processing in the partial space is not performed. In this case, the layered space setting means 20 selects the next partial space (the processing returns to the node A).

  If the processing has been completed for all the meshes ("Yes" in S4), the arithmetic processing unit 4 ends the edge detection processing and the object generation processing (step S10).

  The number of layers in the layered space set in the target space is n, and each layered space is represented as an I-th layer (I = 1, 2,..., N) from the bottom using an index I.

  The layered space setting means 20 sequentially selects the n-stage block spaces included in the selected partial space. For example, the layered space setting means 20 increments the index I by one and selects a block space for examining the presence of an edge (steps S20, S22). If I exceeds n, the process returns to the node A in FIG. 2 and selects the next target partial space (step S24). The edge detecting means 22 reads a point group included in the selected block space of the I-th layer (step S26), and checks whether an edge exists in the block space (step S28). When there is no edge in the selected block space of the I-th layer (in the case of “No” in step S30), the edge detection unit 22 selects the next block space (S22). If the number of point groups belonging to the block space is equal to or smaller than the predetermined threshold value in step S28, the edge detection processing can be omitted, and it can be determined in step S30 that no edge exists.

  FIG. 5 is a schematic diagram illustrating edge search processing S28 in the block space. FIG. 5 is a view of the block space as viewed from above, showing an example of a rectangle 40 defined by a width W and a depth D of the block space and an arrangement of point groups in the block space on the XY plane. I have. The edge detecting means 22 selects any two data points PA and PB from the point group in the block space, sets a line segment L0 in the XY plane having these two points as both ends as a candidate line of the edge, and A band-shaped area EA having a width w (total width 2w) is set on both sides of the line segment L0. The edge detecting means 22 counts the number of data points in the point group whose coordinates on the XY plane are located in the area EA. Whether or not a data point is within the area EA can be determined, for example, based on whether or not the length of a perpendicular from the data point to the line segment L0 on the XY plane is equal to or less than w. In addition, as a condition of the edge, the three-dimensional shape represented by the point group in the area EA is not flat. Specifically, the edge detecting unit 22 determines that the data point is not flat if the height difference (the difference between the maximum elevation and the minimum elevation) ΔZ of the data points included in the area EA is equal to or larger than a preset step threshold γ. I do.

  Here, the width w is a parameter, and absorbs a variation in position in a direction orthogonal to the edge of the data point corresponding to the surface of the destination. For example, w can be about 3 cm. In addition, the height difference threshold γ can be set to, for example, about 2 cm.

The edge detecting means 22 sets the candidate lines for all combinations of two data points in the block space and makes the above-described determination. The most data points are gathered in the area EA, and ΔZ is equal to or larger than the threshold γ. The line segment L0 is an edge found in the block space. If there is an edge ("Yes" in S30), it is determined whether the edge is newly detected (step S32). Specifically, if the edge is not an edge already detected as being continuous with an edge detected in another block space in the layered space of the same layer, it is determined as a new edge. The determination is made based on whether or not the same edge as the edge found in the process S28 exists by searching for the previously detected edge. If the same item does not exist in the list, it is determined that the edge is a new edge (in the case of “No” in S32), and the coordinates (ax, ay) and (bx, by) of both ends PA and PB of the edge are registered in the list. while, determine the length _{L E} of the line segment L0 (line PA-PB) as the length of the edge (step S34).

  On the other hand, if the same edge as the edge found in the processing S28 has already been registered (in the case of “Yes” in S32), the registration for the edge is omitted, and the arithmetic processing device 4 proceeds to the next block. Processing is performed on the space (processing returns to node C).

Edge detection means 22 is newly detected edge, and when _{L E} is the threshold _{L TH2 (L TH2 <L TH1} ) short edge is below (the case of "Yes" in S40), processing relating to short edges to be described later (Short edge processing) is performed (step S42).

On the other hand, if L _E new detection edge (the case of "No" in S40) is greater than the threshold L _TH2, the edge detection means 22 is likely the edge is the object feature, such as a wall, Alternatively, it is checked whether the plant is a plant (leaf of a tree) that is not a destination in the present embodiment (step S44). If it is an artificial structure that is not a plant but a destination (in the case of "No" in step S46), the edge detecting means 22 performs a continuous edge search process described later (step S48).

  When it is determined that the edge is planting (“Yes” in S46), and when the short edge processing S42 and the continuous edge search processing S48 are completed, the arithmetic processing device 4 performs processing for the next block space. The processing shifts to processing (processing returns to node C).

6 and 7 are schematic plan views illustrating the process of step S44 for checking whether or not the edge may be a destination. The edge detecting means 22 sets the space S _A shown in FIG. 6 and the spaces S _B1 and S _B2 shown in FIG. 7 in the same layered space as the block space to be processed, and checks the distribution of point groups near the edge. When the point group is appropriately concentrated at the edge position, it is determined that the point group is an artificial surface. When the point group has a large variation with respect to the edge position, it is determined that the plant is planting. Specifically, the space S _A shown in FIG. 6 is a rectangular parallelepiped is a rectangular 42 horizontal section and the center line of the line segment L0. The height of the cuboid is H _L, the size of the edge direction is L _E, the direction of the size perpendicular to the edge and width respectively on both sides about the line segment L0 w _{A (total} width 2w _A) . Width _{w A} is a parameter, for example, it may be about 20 cm. Space S _B1, S _B2 shown in FIG. 7 are aligned in a direction perpendicular to the line segment L0 is sandwiched between the line L0, a rectangular parallelepiped placed from line segments at a distance g _B. The height of the cuboid is H _L, the size of the edge direction is L _E, the direction of the size perpendicular to the edge and w _B. g _B and w _B are parameters, for example, g _B can be about 3 cm.

In any of the following cases (A1) to (A3), the edge detection unit 22 determines that the edge is due to an artificial surface and is not a plant.
(A1) _{A case where the number} of points in the group of points in the space SA whose perpendicular to the line L0 is within a threshold (for example, 3 cm) is larger than a predetermined number (for example, 10).
(A2) _A case where the average value of the length of the perpendicular from the point group in the space SA to the line segment L0 is within a threshold (for example, 3 cm).
(A3) A case where no point group exists in the spaces S _B1 and S _B2 .

  8 and 9 are schematic flowcharts of the continuous edge search processing S48. The edge detecting means 22 generates a linear list for buffering edge information. In the continuous edge search process, a process of tracking an edge connected to the edge detected in step S28 as a base point is performed. In the edge tracking processing, a rectangular parallelepiped space is set and the edge is searched for in the space, similarly to the case where the edge serving as the base point is searched for in the block space. Hereinafter, the block space from which the edge serving as the base point is extracted is referred to as a base space, and the space of the rectangular solid set in the tracking processing is referred to as a tracking space.

  The tracking processing is performed in opposite directions from both ends PA and PB of the edge, and two linear lists are created in correspondence with each other. Here, tracking from one end PA of the edge is referred to as tracking in the A direction, and tracking from the other end PB of the edge is referred to as tracking in the B direction. The edge detection unit 22 creates a linear list for tracking in the A direction and a linear list for tracking in the B direction (S60). The graphic data of the edge serving as the tracking base point obtained in step S28 can be stored in, for example, a linear list in the A direction (S62). Thereafter, the edge detecting means 22 performs the continuous tracking process S64 in the A direction and the continuous tracking process S66 in the B direction.

  The edge information obtained by the continuous tracking processing in each of the A direction and the B direction is stored in a linear list in the A direction and the B direction. When the two-way tracking of the edge obtained in a certain base space is completed, the edge detecting means 22 connects the end points of the edges stored in the linear list in the A direction and the linear list in the B direction according to the tracking order. To create a continuous line segment (S68). The continuous line segment is not limited to a straight line, but may be a polygonal line.

The process continues to the node E in FIG. 9. If the length L _C of the continuous line segment is equal to or longer than the threshold length L _TH1 (“Yes” in step S80), the object generating unit 24 The object is classified into the first substantially vertical plane, and a stratified object is generated by associating the created continuous line segment with the classification, and registered in the storage device 6 (step S82). In this case, the continuous edge search process ends with “TRUE” indicating that the classification has been set as the return value (step S84), and returns to the calling source that called the continuous edge search process as a subroutine. . The stratified object includes information on the shape and position of the continuous line segment.

  In the present embodiment, the first substantially vertical plane is classified into two types of “building wall surface” and “fence / fence” according to the height of the layered space in which the stratified object exists. Specifically, using a threshold (for example, 3 m from the road surface) set at about the height of the fence / fence, objects that are less than the threshold are classified as “fence / fence”, and objects that are more than the threshold are “fence / fence”. Classified as "Building wall". However, this classification of “building wall surface” or “fence / fence” is for convenience and provisional purposes, and does not necessarily mean that the object is the same feature as the classification. Here, a code representing the classification of the elemental graphic associated with the stratified object is defined as a graphic code, and a “building wall” is defined as a graphic code 201 and a “fence / fence” is defined as a graphic code 202.

If the length L _C of the continuous line segment is less than the threshold length L _TH1 (“No” in step S80), the object generation unit 24 further _checks whether the length L _C is greater than the threshold length L _TH3. Check (step S86). The threshold length L _TH3 can be set, for example, according to the size of the smallest destination object, and is set to 0.2 m in the present embodiment.

If the length L _C is equal to or less than a threshold length L _TH3 (the case of "No" in S86), the object generation unit 24 does not perform the classification with respect to the continuous line segment, the continuous edge search processing not determined classification Is set as the return value, and the process ends (step S88), and the process returns to the calling source.

On the other hand, if the length L _C is larger than the threshold length L _TH3 , that is, if L _TH3 <L _C <L _TH1 (“Yes” in S86), the object generation unit 24 sets a continuous line segment in the layered space. Is performed to determine whether the substantially vertical plane corresponding to is cylindrical (step S90). Specifically, it is checked whether the shape of the continuous line segment is a circle.

  When the object is in the shape of a cylinder (in the case of “Yes” in step S92), the object generating unit 24 classifies the continuous line segment into a first cylinder (figure code 2005), and creates the generated continuous line segment and the classification. Are associated with each other and a stratified object is generated and registered in the storage device 6 (step S94). In this case, the continuous edge search process ends with "TRUE" set as the return value (step S96), and the process returns to the calling source.

  If the object is not a columnar shape ("No" in S92), the object generating means 24 decomposes the continuous line segment into individual edges constituting the continuous line segment, and separates the individual edges into a second substantially vertical plane (graphic code). 2003), a stratified object is generated, and registered in the storage device 6 (step S98). The classification has an attribute of “short line segment and no circle element”. For example, destinations such as road signs and signboards may fall into this category. In this case, the continuous edge search process ends with "TRUE" set as the return value (step S100), and the process returns to the calling source.

  FIG. 10 and FIG. 11 are schematic flowcharts of the continuous tracking processing S64 and S66 by the edge detecting means 22. When an edge is detected in the base space or the previously set tracking space, the edge detection unit 22 determines a tracking space adjacent to the space in which the preceding edge detection was performed (referred to as an analysis unit space). It is set (step S120). The new tracking space is arranged so that the center line of the rectangle, which is a cross section in the cutting plane, matches the direction of the edge detected in the preceding analysis unit space. When an edge is detected in the previously set tracking space, the direction line of the edge is extended in the same direction as the direction in which the tracking space was obtained, and further tracking is performed.

  When the tracking space is set, the edge detecting means 22 fetches the points in the tracking space from the storage device 6 into a work area such as a RAM (step S122), and sets the points along the extension of the edge of the preceding analysis unit space. It is checked whether an edge exists in the tracking space (step S124). If a group of points equal to or more than the reference point are gathered in the tracking space set along the extension line and the height difference is equal to or greater than the threshold γ, it is determined that an edge exists in the tracking space (“Yes” in step S126). in the case of).

  The edge detecting unit 22 checks whether or not the extracted edge (edge candidate) intersects with the edge stored on the linear list (step S128). Here, the edges registered in the linear list originate in the same base space as the edges newly extracted from the tracking space. When a figure having a closed edge such as a cross section of a tunnel is formed, the closed figure may be extracted by continuous tracking processing from a certain base space. Therefore, in step S128, it is determined whether there is an intersection (including coincidence) between the newly extracted edge and the edge on the linear list, and when the intersection is detected (in the case of "Yes" in step S130). In order to avoid the self-loop or the infinite loop, the continuous tracking process ends without adding the newly extracted edge to the linear list (step S132).

  If no intersection is detected ("No" in S130), the edge detection unit 22 further checks whether the newly extracted edge is already registered in the figure (step S134). If the figure is registered (in the case of “Yes” in step S136), the continuous tracking process is terminated without adding the newly extracted edge to the linear list (S138), and the process is called as a subroutine. The process returns to Steps S64 and S66 of FIG.

  On the other hand, if the figure is not registered (in the case of “No” in S136), the edge detection unit 22 adds the edge information about the current tracking space to the linear list (step S140), and the node in FIG. Returning to F, a new tracking space following the current tracking space is set. In this way, the edge detecting means 22 sequentially sets the tracking space and tracks the edge.

  If the edge detecting means 22 cannot detect the edge in the tracking space set along the extension of the detected edge ("No" in S126), the process proceeds to node G in FIG. 11 to extend the detected edge. A search process is performed in a direction deviating from the line. In the search processing, the tracking space is set by changing its center axis in the cutting plane from the position of the extension line in the clockwise direction and the counterclockwise direction. For example, a change range (± θ) of the direction of the central axis from the direction of the extension line and the division number k are set, a plurality of tracking spaces having different angles in the cutting plane by θ / k are set, and each tracking is performed. An edge is searched for in the space, and an edge that collects the most point groups is extracted from the edges obtained in the plurality of tracking spaces set by changing the angle (step S160).

  When an edge is thus extracted (in the case of “Yes” in step S162), the edge detecting unit 22 performs the same processing as in steps S128 to S140. That is, it is checked whether or not the extracted edge intersects with the edge on the linear list (step S166), and it is checked whether the extracted edge is already registered in the figure (step S172). When an intersection is detected (“Yes” in step S168) and when a figure is registered (“Yes” in step S174), the newly extracted edge is added to the linear list. Without this, the continuous tracking process is terminated (steps S170 and S176), and the process returns to steps S64 and S66 in FIG. 8 where the process is called as a subroutine. On the other hand, if no intersection is detected (in the case of “No” in S168) and the figure is not registered (in the case of “No” in S174), the edge detecting means 22 stores the extracted edge information in a linear list. (Step S178), and returns to the node F in FIG. 10 to set a new tracking space following the current tracking space. If no edge has been extracted ("No" in S162), the process returns to steps S64 and S66 in FIG. 8 where the continuous tracking process is called as a subroutine (S164).

  Next, the cylinder search processing S90 in FIG. 9 will be described. FIG. 12 is a schematic flowchart of the cylinder search processing S90. FIG. 13 is a schematic plan view showing an example of the continuous line segment generated in step S68 of FIG. 8, and shows a continuous line segment 52 detected corresponding to the cylinder 50. Naturally, from the surface of the feature in which a plurality of vertical planes are bent and connected, there may occur a polygonal line edge in which a plurality of straight lines (line segments) such as a continuous line segment 52 are connected. Since the process of detecting an edge is basically a straight line approximation, an edge like a continuous line segment 52 can be detected even from a smoothly curved cylindrical side surface. Therefore, the edge detection unit 22 positions the edge detected in the relatively large base space (block space) or the tracking space as the approximate edge of the feature, and in the cylinder search processing S90, the line segment which is the approximate edge , An edge fine adjustment process for obtaining a more detailed fold line representing the edge of the terrestrial feature is performed (step S200).

FIG. 14 is a schematic flowchart of the edge fine adjustment processing S200, and FIG. 15 is a schematic plan view illustrating the edge fine adjustment processing. The fine adjustment processing S200 will be described with reference to FIGS. FIG. 15 shows an example of a detailed edge Lf generated corresponding to one general edge L0 constituting the continuous line segment 52. The edge detecting means 22 divides the line segment L0 indicating the approximate edge by a predetermined length from one end (step S220), searches for a point (representative point) through which the detailed edge passes in each of the generated sections, and Instead of the line segment L0 indicating a general edge connecting PA and PB in the plane, a bent line Lf connecting representative points obtained in each section is set as an edge. Here, the number of sections is N. Processing device 4, in the loop process of S224~S232 edge fine adjusting process of FIG. 14, the representative point _{P K} is obtained from the section K selected, the coordinates in the XY plane of the representative point _{P K (X K} , Y _K ) as representative point information.

In the edge fine adjustment process, the edge detecting means 22 increments the section number K by one, selects a section (steps S222, S224), and searches for a representative point in each section (step S228). The search is performed in a compartment 62 obtained by dividing the edge peripheral space 60 into sections. The edge peripheral space 60 is a region whose shape projected on the XY plane is substantially in the vicinity of the edge, and can be set as a space having the height _HL of the layered space in the Z direction. Specifically, the shape of the edge peripheral space 60 on the XY plane is a rectangle in which both ends PA and PB of the edge are located at the midpoint of two opposing sides. The dimension of the rectangle in the direction orthogonal to the general edge can be set in consideration of the curvature and the size of the unevenness assumed in the cross-sectional shape of the columnar feature to be detected. In the present embodiment, the columnar feature to be detected is a telephone pole, and the dimension of the edge peripheral space 60 in the direction orthogonal to the general edge needs to be large enough to accurately capture the edge of the telephone pole in the cutting plane. . On the other hand, it is preferable that the dimension is set smaller than the width W and the depth D of the partial space, for example, from the viewpoint that a point group not caused by the edge of the telephone pole is hardly erroneously detected as a representative point. . The compartment 62 is generated by partitioning the edge peripheral space 60 by a plane passing through the boundary point of the section that has partitioned the general edge and being orthogonal to the general edge. The edge detecting means 22 takes in the point cloud data in the compartment 62 corresponding to the section for each section, and obtains the position of the compartment 62 where the point cloud is most dense in the XY plane. The dense position is determined by the number of point groups included in a circle centered on an arbitrary point of the point group. For example, the radius of the circle can be 2 cm. If the number of points densely located at the position is equal to or greater than a predetermined number ("Yes" in step S230), the position is added to the edge representative point list as a representative point (step S232), and the next The process proceeds to the section (Step S224). On the other hand, if the number of points densely located at the position is less than the predetermined number (in the case of “No” in S230), it is determined that there is no edge representative point in the section and the point is not added to the edge representative point list. The process proceeds to the next section (S224). When the processing is completed for all the sections (in the case of “Yes” in step S226), the fine adjustment processing of the edge shape ends.

In the example illustrated in FIG. 15, the line segment L <b> 0 indicating an approximate edge is divided into, for example, four sections, and the edge peripheral space 60 represented by a dotted rectangle having L <b> 0 as a center line also has four sections corresponding to the section. It is partitioned into compartments 62a to 62d. For example, the representative points P _k , P _{k + 1} , and P _{k + 2} are extracted in the compartments 62 a, 62 b, and 62 d, and the fold line P _k P _{k + 1} P _{k + 2} connecting these in the arrangement order of the compartments becomes the detailed edge Lf. Incidentally, in the example shown in FIG. 15, it is shown that the division of the edge peripheral space 60 is performed at regular intervals from the PA side, and as a result, the last section can be smaller than other sections.

  The above-described fine adjustment processing is performed on a plurality of general edges constituting the continuous line segment 52 shown in FIG. 13, and upon completion thereof, the edge detecting means 22 returns to the processing of FIG. 12 and estimates whether or not the column is a cylinder ( Step S202). FIG. 16 shows the representative point 70 obtained by the fine adjustment processing on the continuous line segment 52. The processing of step S202 will be described with reference to FIG.

  The edge detecting means 22 determines that all the representative points other than the representative points 70a and 70b are on the same side with respect to a straight line 72 connecting both ends (representative points 70a and 70b) among the representative points 70 arranged along the continuous line segment 52. Check if there is. If not on the same side, it is determined that the continuous line segment 52 is not a column (in the case of “No” in step S204).

When all the representative points other than the representative points 70a and 70b are on the same side of the straight line 72 as shown in FIG. 16, the length of the perpendicular from each representative point 70 to the straight line 72 is obtained. If the length of the maximum perpendicular 74 is equal to or longer than the predetermined reference length, it is determined that the cylinder is a cylinder (in the case of “Yes” in step S204), and the foot of the perpendicular is moved to the temporary center point C of the cylinder. _α is set (step S206). On the other hand, if the maximum perpendicular length is less than the reference length, it is determined that the object is not a cylinder (in the case of “No” in step S204). The predetermined reference length can be set to, for example, a predetermined ratio (for example, about 30 to 40%) of the distance between the representative points 70a and 70b.

If it is determined that the column is a cylinder, a re-search space is set, and data of a point group existing in the space is read from the storage device 6 (step S208). FIG. 17 is a schematic diagram illustrating an example of the re-search space. FIG. 17 is a diagram of the target space as viewed from above, and shows the arrangement of the search space (rectangle 80) and the distribution of the point group 82 obtained from the cylindrical surface of the telephone pole. Re-search space is set the center of the horizontal cross-section to match the center point C _alpha provisional cylindrical. One side of the rectangle 80 can be set larger than the block space. For example, in the present embodiment, W and D of the block space are set to 50 cm, and one side of the rectangle 80 is set to 60 cm, which is 10 cm larger than that. By giving a margin to the size of the re-search space in this way, the point group generated from the cylindrical surface can be taken into the re-search space as a processing target point group without omission.

  The edge detection unit 22 checks whether the spread of the processing target point group in the re-search space is equal to or smaller than the upper limit set based on the size of the destination object (step S210). In the present embodiment, it is checked whether or not the spread of the processing target point group in the X-axis direction and the spread in the Y-axis direction are each 50 cm or less. If it is not equal to or less than the upper limit (in the case of “No” in S210), it is determined that the destination is not a cylinder, specifically a telephone pole, and the process returns to the caller of the subroutine of the cylinder search processing S90.

On the other hand, if it is equal to or less than the upper limit (in the case of “Yes” in S210), the edge detecting means 22 performs processing for obtaining the center and radius of the cylindrical surface based on the processing target point group in the re-search space (step S212). ). Specifically, a process of searching for a circle passing through the four points selected from the processing target point group is performed for all combinations of the four points in the processing target point group. Then, the searched circle having the largest number of processing target points near the circle is selected. For example, among the searched circles (center C, radius R), circles (center C _β , center C _β , where the center is C and the radius is [R−δ, R + δ]) in the donut-shaped region are the largest. Select the radius R _β ). Here, δ is a predetermined value, and can be set according to the columnar feature to be detected. For example, it can be set to 2-3 cm for a utility pole.

Then, using the center C _β and the radius R _β of the circle obtained in the above-described processing, a point existing in a donut-shaped region whose center is C _{β and} whose radius is [R _β −δ, R _β + δ] is extracted. and, the center C _gamma compatible circle to the extracted point cloud, the radius R _gamma is calculated by the least squares method. In the cylinder search processing, the circle (center _Cγ , radius _Rγ ) is set as a search result, and the processing is returned to the calling source. In the continuous edge search processing of FIG. 9, when a circle (center _Cγ , radius _Rγ ) is obtained as a search result in the cylinder search processing S90, the stratified object whose classification is the first cylinder is obtained for the continuous line segment 52. (Graphic code 2005) is generated. Object generating means 24 to the stratified object can be associated with the center C _gamma, radius R _gamma as information representing the shape characteristics of the continuous line segment 52 shown in FIG. 16.

  FIG. 18 is a schematic flowchart of the short edge processing in step S42 in FIG. The object generation unit 24 checks whether the detected short edge is a thin pole (step S240), and if it is determined to be a thin pole ("Yes" in step S242), classifies the second cylinder with respect to the edge. Then, a stratified object set as is generated and registered in the storage device 6 (step S244). In the present embodiment, specifically, the stratified object is classified as a “thin pole” (graphic code 1001), which is one type of the second cylinder.

FIG. 19 is a schematic plan view illustrating a process S240 of determining a thin pole. Set the space S _C having a horizontal cross-section containing the short edges 90, capturing point cloud data of the space S _C as a processing target point group. Horizontal section space S _C may be a rectangle having a side in a direction perpendicular to the sides of the direction along the short edge 90 (the ζ axis) (the η axis). The rectangular respect to the ζ-axis direction, and opposite ends from to the side of the rectangle distance d _C of the edge 90, can be the ζ axial size of the space S _C and (L E ₊ 2d _C). On the other hand, in the η-axis direction, the size can be set to 2w _C with the edge 90 as the center. Further, Z-axis direction of the range of the space S _C may be the same as the layered space edge 90 has been detected. Object generating means 24 checks whether or not more than the upper limit the spread in the horizontal plane of the target point groups in space S _C is set based on the size of the thin pole is an object feature. In the present embodiment, it is checked whether the spread of the processing target point group in the X-axis direction and the spread in the Y-axis direction are each 20 cm or less.

Then (if at step S242 "Yes") when the upper limit or less point cloud in the space S _C is determined to be narrow pole, object generator 24, to edge 90, classification of the second A stratified object that is a column “thin pole” is generated (step S244). The object generating means 24 can associate the stratified object with the center of the distribution of the processing target point group as the center point of the thin pole, and can associate, for example, 0 as the radius of the thin pole. When the stratified object is generated, the process returns to the calling source of the subroutine of the short edge processing S42.

On the other hand, point cloud in the space S _C if the distribution of the target point group is wider than the upper limit is determined not to be a thin pole (if at step S242 "No"). In this case, the edge detecting means 22 performs the same processing as in step S44 in FIG. 4, and checks whether the edge is a planting (leaf of a tree) or the like (step S246). In the case of an artificial structure that is not a plant but a destination (in the case of “No” in step S248), the edge detection unit 22 uses FIGS. 8 and 9 as in step S48 of FIG. The described continuous edge search process is performed (step S250). Thereafter, the process returns to the calling source of the subroutine of the short edge processing S42 in FIG.

  By the processing of the layered space setting unit 20, the edge detecting unit 22, and the object generating unit 24 described above, the feature detection system 2 divides the target space into layered spaces stacked in the Z-axis direction, and calculates a point group in each layered space. Then, an object of a plurality of types of element graphics having predetermined shape characteristics is extracted and registered in the storage device 6.

The destination detecting means 26 combines the stratified objects in the vertical direction, and checks whether or not the object is a destination standing on the ground. Specifically, the destination feature detecting means 26 combines stratified objects whose edges are arranged along the height direction. Here, the cross section of each destination object rising from the ground surface in each layered space may ideally have a common part or intersect in an adjacent layered space, but is generated from a point cloud. Often, the feature surface represented by a sharp edge does not. Therefore, the target object detecting means 26 treats the first edge and the second edge existing in different layered spaces as being arranged in the height direction if the positions on the horizontal plane are somewhat close to each other. FIG. 20 is a schematic plan view for explaining the overlapping of edges in different layered spaces. Object feature detection unit 26 is different for the first edge 100 and second edge 102 which belongs to a layer space, the near edge region 104 having a predetermined width w _M a first edge 100 and the center line second mutually When the edge 102 and the edge 102 overlap, it is determined that the edges 100 and 102 are arranged in the height direction. w _M, for example, it can be set to about 5cm.

  The destination detecting means 26 checks whether or not the vertical arrangement of stratified objects has the characteristics of the destination described in the feature model 32 of FIG. 1. It is determined that the stratified objects arranged are the destination. As a simple example, the destination feature "road sign beside the road" is composed of a cylinder corresponding to a pillar and a vertical plane corresponding to a sign thereon, and the feature model 32 having a height of 0 to 3 m. In the range, a second column such as a "thin pole" is vertically arranged as an element graphic in the range, and a second substantially vertical plane corresponding to the size expected for the sign as the element graphic exists above the second column. Are stored in the storage device 6. Also, as the feature model 32 of the destination feature "guide sign on the road", a first substantially vertical plane is vertically arranged as an element graphic in a plurality of continuous layered spaces at relatively high positions, and the element graphic below the element graphic is shown below. Are stored in the storage device 6. When the feature model 32 is represented by a simple array pattern of element figures as described above, the destination feature detection unit 26 performs a relatively simple process of comparing the vertical arrangement of stratified objects with the array pattern. Can be used to determine the destination. However, the feature model 32 may define more complicated conditions relating to the element figure, and a program for determining whether or not the conditions are satisfied is stored in the storage device 6 as the feature model 32 substantially. It may be something.

An example of a more detailed process of the destination object detecting means 26 will be described. FIG. 21 is a schematic flow chart showing a processing example of the destination object detecting means 26 for a destination object including a pole such as a road sign, and a program describing the processing is stored in the storage device as the object model 32 as described above. 6 can be stored in advance. The stratified objects generated in the target space by the object generating means 24 are assigned a natural number Number from 1 in order. The destination object detecting means 26 increments the Index in order from the initial value 0 (step S280) set at the start of the process and selects a stratified object to be processed (step S282), and the Index indicates the total number N of stratified objects. _{If TOT} is exceeded, the process ends ("No" in step S284).

When the stratified object is selected (in the case of “Yes” in step S284), the destination object detecting means 26 determines that the graphic code Code of the stratified object is “1001” and the stratified object is in the form of a feature. The graphic mark Mark indicating whether or not it is already used is a value “0” indicating that it is not used (if “Yes” in step S286), and is further arranged in the height direction on the stratified object. If the number N _V of other stratified object predetermined reference value N _VTH above (if in step S288 "Yes"), the elevation multiple stratification objects in the positional relationship arranged in a height direction Sorted in ascending order (step S290). _NVTH can be, for example, 2 or 3.

Note that when the code of the selected stratified object is not “1001”, when the Mark of the stratified object is a value “1” indicating “used” (“No” in step S286), and in the height direction If the number N _V of stratification objects arranged in is less than the reference value (step S288 "No"), the process returns to node a, to select the next stratified object (step S282).

  After the sorting process in step S290, the destination object detecting means 26 performs a pole determination process S292 described later. If the return value from the subroutine of the pole discrimination processing is "1", it means that a pole has been detected ("Yes" in step S294), and the destination object detecting means 26 determines that the center of the horizontal section is The radius is calculated (step S296). Then, based on the calculation result or the like, it is determined whether or not the pole is already registered in the storage device 6 as a detected feature. If not already registered (in the case of "No" in step S298), the lower and upper elevations of the pole are obtained from the radius and the center point of the pole (step S300), and the pole is stored as a detected feature. 6 is newly registered (step S302). Further, the stratified object having the index set in step S282 and having Code = “1001” and the stratified object having Code = “1001” aligned in the height direction with the stratified object are marked with Mark = If it is set to “1” and used (step S304), the process returns to the node A, and selects the next stratified object (step S282).

  Also, as a result of the pole discriminating process S292, when it is determined that it is not a pole (in the case of "No" in step S294) and when the detected pole is already registered (in the case of "Yes" in step S298) ) Also returns to the node A, and selects the next stratified object (step S282).

FIG. 22 and FIG. 23 are schematic flowcharts of the pole discrimination processing S292. The pole discriminating process S292 discriminates whether a feature corresponding to an object group composed of stratified objects arranged in the height direction includes a pole. The destination object detecting means 26 reads the input from the input device 8 or the parameters set in the storage device 6 to set the minimum number _CMIN1001 of “thin pole” objects required for judging a pole. (Step S320). Also, an index CI for identifying each stratified object in the object group to be processed, a value PL indicating the probability of being a pole, a count number FC ₁₀₀₁ of stratified objects with Code = “1001”, and Code = “2001”. initialize count _FC 2001 of a stratified object, Code = "2003" count _FC 2003 stratification object is, Code = "2005" the count _{FC 2005} stratification object is to 0, respectively (step S322).

Object feature detection unit 26 selects the stratified object to be processed is incremented sequentially CI from the initial value 0 (step S324), CI is the number N _V of stratification objects constituting the object group to be processed If it exceeds, the process proceeds to the node D (in the case of “No” in step S326).

If the count FC ₁₀₀₁ is equal to or more than C _{MIN 1001} (“Yes” in step S328), it is determined that the object group is a feature including a pole, and the value of PL is set to a value “1” indicating the determination result. "(Step S330), and the process proceeds to node D.

If FC ₁₀₀₁ <C _{MIN 1001} (“No” in step S328), the destination object detecting unit 26 determines that the selected stratified object is Code = “1001” (“Yes” in step S332). ")", The count number FC ₁₀₀₁ is increased by 1 (step S334), and if Code = "2001"("Yes" in step S336), the count number FC ₂₀₀₁ is increased by 1 (step S338), and Code = " If “2003” (“Yes” in step S340), the count FC ₂₀₀₃ is incremented by 1 (step S342), and if Code = “2005” (if “Yes” in step S344), the count FC ₂₀₀₅ is increased by 1 (step S346), and the process returns to the node B.

On the other hand, if FC ₁₀₀₁ <C _{MIN 1001} (“No” in step S328), the selected stratified object is not Code = “1001”, “2001”, “2003”, or “2005”. In this case, it is determined whether or not the selected stratified object is the second or more from the bottom (step S364). If the stratified object is the lowest object in the object group to be processed (“No” in step S364), the process returns to node B.

  If it is the second or more (if “Yes” in step S364), the stratified object is Code = “201” indicating “building wall surface”, and the height Z (CI) of the stratified object is It is checked whether or not the difference between the height and the height Z (CI-1) of the stratified object below it is less than 1.5 m (step S366). When these conditions are satisfied (in the case of “Yes” in step S366), the destination feature detecting unit 26 has a low possibility that the object group is a feature including a “building wall surface” and a feature including a pole. Is determined, the value of PL is set to a value “−1” indicating the determination result (step S368), and the process proceeds to the node D. By the way, in the case where Z (CI) -Z (CI-1) is 1.5 m or more, for example, there is a possibility that a viaduct wall is located above a road sign and exists as a possibility. PL remains at 0.

  If PL = 1 (“Yes” in step S370), the return value is set to “1” (step S372), and the process returns to the call source of the poll determination subroutine.

  If PL = −1 (“Yes” in step S374), the return value is set to “0” (step S376), and the process returns to the call source of the poll determination subroutine.

If FC ₁₀₀₁ is 1, that is, it is not confirmed that the objects of “thin poles” are stacked in the height direction (“No” in step S378), the return value is set to “0” (step S380). The processing returns to the caller of the subroutine of the pole determination processing.

If PL is 0 and FC ₁₀₀₁ is 2 or more (“No” in steps S370 and S374 and “Yes” in step S378), the processing target object group is a combination of a pole and a sign. It is determined whether it is a pole marker that has been set (step S382). Specifically, when a stratified object with Code = “2003” is combined with a “thin pole” with Code = “1001”, it is determined to be a pole marker (in the case of “Yes” in step S382). In this case, the return value is set to "1" (step S384), and the process returns to the caller of the subroutine of the poll determination process. On the other hand, if it is determined that it is not a pole indicator ("No" in step S382), the return value is set to "0" (step S386), and the process returns to the call source of the pole determination subroutine.

  The example of the processing of the destination object detecting means 26 has been described above. In this process, basically, not the point cloud data itself, but a group consisting of a plurality of stratified objects arranged along the height direction is to be processed. The feature detection system 2 describes, as a feature model 32, how to classify the stratified objects in the group along the height direction, for example, by the processing described with reference to FIGS. The destination is detected by matching the condition. In the processing in the destination detecting means 26, the stratified objects are commonly used for detecting various destinations as so-called components constituting the terrestrial feature. In other words, the process of searching and extracting the feature of the shape of the destination object from the point cloud data for each destination object is omitted, and the processing efficiency is improved. In addition, the process of searching and extracting the feature of the shape of the destination based on the combination of the componentized objects requires less man-hour than the process of searching and extracting the feature of the shape of the destination from the point cloud data. And can flexibly respond to the addition / change of destinations.

  2 feature detection system, 4 arithmetic processing device, 6 storage device, 8 input device, 10 output device, 20 layered space setting means, 22 edge detection means, 24 object generation means, 26 destination feature detection means, 30 point group data , 32 feature model.

Claims (6)

Based on three-dimensional coordinate data of a point cloud extracted from the surface of the feature in the target space, a feature detection device that detects a target feature rising from the ground surface,
Using a plurality of types of elemental figures having a predetermined shape characteristic in a three-dimensional space, a feature model representing a standard shape of the destination by one or a plurality of the elemental figures arranged in a height direction. Model storage means stored in advance,
Layered space setting means for dividing the target space into a plurality of horizontal layered spaces,
In each of the layered spaces, an edge detection unit that detects the point group that is projected on a horizontal plane and gathers in the vicinity of a predetermined reference or more as an edge that is a position on the surface of a feature,
Object generation means for classifying the edge into those which produce a projection corresponding to the shape of the edge among the element figures, and generating a stratified object in which the edge is associated with the classification,
Obtain a group consisting of a plurality of the stratified objects, each edge of which is arranged along the height direction, and collate the classification in the group along the height direction with the feature model to determine the destination feature. Destination detecting means for detecting
A feature detection device comprising: The feature detection device according to claim 1,
The feature detecting device, wherein the plurality of layered spaces have the same dimension in the height direction. In the feature detection device according to claim 1 or 2,
The destination feature detecting means is configured to determine a position of a first edge and a second edge existing in the layered space different from each other in a horizontal plane, the edge vicinity area having a predetermined width with the first edge as a center line, and The feature detection device, wherein when the edge has an overlap, the first edge and the second edge are determined to be arranged in a height direction. In the feature detection device according to any one of claims 1 to 3,
The element graphic has a non-circular shape in a horizontal cross section and a first substantially vertical surface having a length equal to or greater than a predetermined threshold length, and a non-circular shape in a horizontal cross section and a length less than the threshold length. A second column having a second substantially vertical surface, a first column having a horizontal cross section having a radius greater than or equal to a predetermined threshold radius, and a second column having a horizontal cross section having a radius smaller than the threshold radius. A feature detection device comprising: a column. A feature detection method for detecting a destination feature rising from the ground based on three-dimensional coordinate data of a point cloud extracted from a feature surface in a target space,
A layered space setting step of dividing the target space into a plurality of horizontal layered spaces,
In each of the layered spaces, an edge detection step of detecting the point cloud projected on a horizontal plane and gathering in the vicinity of a predetermined reference or more as an edge that is a position on the surface of a feature,
The edge is classified into a plurality of types of elemental figures having a predetermined shape characteristic in a three-dimensional space that generate a projection corresponding to the shape of the edge, and a stratified object in which the edge is associated with the classification. An object generation step for generating
A grouping step of obtaining a group consisting of a plurality of the stratified objects, each edge of which is arranged along a height direction;
The feature model is read from a model storage unit in which a feature model represented by one or a plurality of the elemental figures arranged in one or more standard shapes of the destination feature along the height direction is read in the height direction in the group. A feature model matching step of matching the arrangement of the classification along the feature model with the feature model to detect the destination feature;
A feature detection method comprising: Based on the three-dimensional coordinate data of a point cloud extracted from the surface of the feature in the target space, a program for causing a computer to perform a process of detecting a destination feature rising from the surface of the ground, the computer,
Using a plurality of types of elemental figures having a predetermined shape characteristic in a three-dimensional space, a feature model representing a standard shape of the destination by one or a plurality of the elemental figures arranged in a height direction. Model storage means stored in advance,
Layered space setting means for dividing the target space into a plurality of horizontal layered spaces,
In each of the layered spaces, an edge detection unit that detects the point group that is projected on a horizontal plane and gathers in the vicinity of a predetermined reference or more as an edge that is a position of a feature surface,
An object generating unit that classifies the edge into those that generate a projection corresponding to the shape of the edge among the element figures, and generates a stratified object that associates the edge with the classification; and
Obtain a group consisting of a plurality of the stratified objects, each edge of which is arranged along the height direction, and collate the classification in the group along the height direction with the feature model to determine the destination feature. Destination detecting means for detecting
A program characterized by functioning as a program.

Images