JP7164652B2 - Model generation device, model generation method, and model generation program - Google Patents

Description

Article 30, Paragraph 2 of the Patent Act applies September 17, 2020 The 2020 IEICE Society Conference will be held online (URL: https://www.ieice-taikai.jp/2020society/jpn/index.html) published in

The present invention relates to a model generation device, a model generation method, and a model generation program for creating an indoor 3D model from indoor point cloud data.

Techniques for creating indoor 3D models from indoor point cloud data are known, and there is a demand for using such 3D models for radio wave propagation simulators.
There is a technique for creating a mesh 3D model that can be read into a radio propagation simulator, but the model created by meshing may have holes or fine surfaces, making it unsuitable for radio propagation simulation.
As a 3D model creation technology that does not rely on meshing, for example, the device of Patent Document 1 reads 3D indoor point cloud data acquired about the inside of a building, extracts a plane from the read indoor point cloud data, and extracts The planes are classified into floor, ceiling, and wall surfaces. Then, the classified floor surface, ceiling surface, and wall surface are polygonized to generate a three-dimensional model.
Further, Patent Literature 2 discloses an information processing apparatus that generates three-dimensional model information of a room having a floor surface, a ceiling surface, and wall surfaces based on measurement point cloud information obtained by 3D measurement.
The number of measurement points in the measurement point group information is counted in the height direction, and the distribution of the number of measurement points in the height direction is specified.
A plane having the largest number of measurement points in the measurement point group information on the upper side in the height direction is specified as the ceiling plane. A plane having the largest number of measurement points in the measurement point cloud information on the lower side in the height direction is specified as the floor surface.
After specifying the ceiling surface and the floor surface, the vertical surface (wall surface) from the ceiling surface and the floor surface is specified, and the three-dimensional model information of the room having the floor surface, the ceiling surface and the wall surface is generated.

Japanese Patent Application Laid-Open No. 2019-168976 Japanese Patent Application Laid-Open No. 2018-151835

By applying the techniques disclosed in Cited Documents 1 and 2, a 3D model for a radio wave propagation simulator can be created.
However, the conventional method has many parameters to be set by the user, and there is a problem that an inexperienced user cannot easily create a 3D model.
The present invention has been made in view of the above problems, and makes it possible to easily create a 3D model that can also be used for a radio wave propagation simulator from indoor point cloud data.

In order to solve the above problems, the present invention provides a data input unit for receiving input of indoor point cloud data, a parameter setting unit for receiving input parameter settings by a user, and generating internal parameters based on the input parameters. A parameter generation unit, a histogram generation unit that generates a height histogram based on the indoor point cloud data that has been input, and a ceiling data and floor data that are detected from the indoor point cloud data based on the height histogram. a room region separation unit that applies region growth to the ceiling data to separate one or more room regions; and a contour detection unit that detects the contour of the separated ceiling data of each room. and a room model generation unit that generates a plane model of each room from the detected contour, the floor height in the floor data, and the ceiling height in the ceiling data, wherein the input parameter is the room When the number of room regions separated by the room region separation unit is different from the number of rooms set as the input parameters, the parameter generation unit regenerates the internal parameters, and regenerates the internal parameters. and a model generation device that generates the planar model based on the internal parameters .

Since it is configured as described above, according to the present invention, it is possible to easily create a 3D model that can also be used for a radio wave propagation simulator from indoor point cloud data.

FIG. 4 is a diagram showing an example of filtered indoor point cloud data; FIG. 4 is a diagram showing a height histogram created from indoor point cloud data; FIG. 4 is a diagram showing indoor point cloud data before separation into ceiling data and floor data; FIG. 10 is a diagram showing indoor point cloud data after separating ceiling data and floor data; FIG. 10 is a diagram showing ceiling data separated into room regions by the region growing method; FIG. 10 is a diagram showing a projection image in which the outline of a room is detected; It is a figure which shows a simple object model. FIG. 10 is a diagram showing a mesh object model; It is a figure which shows the indoor 3D model which the model generation apparatus of this embodiment produced|generated. It is a figure explaining the functional composition of the model generation device of this embodiment. It is a figure explaining automatic adjustment of a smoothing threshold value and a curvature threshold value. 10 is a flowchart (part 1) for explaining the flow of model generation processing executed by the CPU of the model generation device; 2 is a flowchart (part 2) for explaining the flow of model generation processing executed by the CPU of the model generation device;

BEST MODE FOR CARRYING OUT THE INVENTION Below, embodiments of the present invention will be described in detail with reference to the drawings.
The present embodiment relates to a model generation algorithm for generating an indoor 3D model for radio wave propagation simulation (ray trace analysis) from indoor point cloud data, and a model generation device for executing this.
The model generating apparatus of this embodiment is characterized by automating part of the parameter setting for the model generating algorithm.
The model generation apparatus of this embodiment generates internal parameters for a model generation algorithm from parameters that can be easily set by the user, and executes the model generation algorithm based on the generated internal parameters. Further, necessary parameters are automatically adjusted based on the execution result of the algorithm (model generation result) and the parameters set by the user.
According to this embodiment, even a user who is not familiar with the algorithm for generating an indoor 3D model from indoor point cloud data can easily generate an indoor 3D model suitable for radio wave propagation simulation in a short time. can do

In this embodiment, an existing algorithm such as the region growing method is utilized to separate the indoor point cloud data into one or more room regions. After that, by using image processing technology, an indoor 3D model suitable for radio wave propagation simulation and a 3D model of an object existing indoors (object model) are created.
It took several days to several weeks to completely manually create an indoor 3D model for radio wave propagation simulation from indoor point cloud data. can create indoor models and object models.
Moreover, the known method for creating a 3D model for radio wave propagation simulation has many parameters that the user should set by himself, and it is difficult to set appropriate values. The model generation device of this embodiment can create a model using parameters that can be easily set by the user. In addition, it is difficult to model indoor objects with known methods, but the model generation device of this embodiment can also model indoor objects, enabling radio wave propagation simulations that take into account the effects of objects. Become.

An outline of the operation of the model generation device of this embodiment will be described. After the model generator starts up, the user inputs the separately created filtered indoor point cloud data to the model generator, and uses user parameters such as point intervals during filtering and the number of assumed indoor rooms. Enter
The model generation device automatically generates and adjusts internal parameters from user parameters, executes a model generation flow, and outputs model generation results to a display device.
The user can determine if the output results match the user parameters (number of rooms). If the output results match the user parameters, the model generator outputs a 3D model. If not, the model generator readjusts the internal parameters and executes the model creation flow again.

The configuration of the model generation device of this embodiment and the model generation process will be described in detail below.
FIG. 1 is a diagram showing an example of filtered indoor point cloud data.
The indoor point cloud data shown in FIG. 1 is a set of points obtained by scanning indoors using a 3D laser scanner. The indoor point cloud data in FIG. 1 has undergone filtering in advance. Filter processing is processing for thinning out points and reducing the density of the point cloud in order to speed up the processing of the point cloud data. For filtering, for example, a method called Voxel Grid filter can be used. The Voxel Grid filter arranges a three-dimensional cubic grid in the point cloud, regards the center of gravity of the point cloud existing in each grid as an approximation point of each point, and converts the original indoor point cloud data to one point of the center of gravity. It is a method to replace with
As a result, the indoor point cloud data can be thinned out at regular intervals to reduce the density of the point cloud. In the indoor point cloud data shown in FIG. 1, the points are reduced at intervals of 1 cm, for example.
The filtered indoor point cloud data 1 may be input from an external device or via a storage medium in a filtered state. It may be obtained by filtering. In the following description, indoor point cloud data 1 refers to filtered indoor point cloud data unless otherwise specified.

[Ceiling/floor separation]
In the model generation method of this embodiment, the model generation device first performs ceiling/floor separation processing. In this ceiling/floor separation processing, the model generation device creates a height histogram from the filtered indoor point cloud data 1 shown in FIG.
FIG. 2 shows a height histogram. Each point included in the indoor point cloud data 1 obtained by 3D scanning has height information. A height histogram 2 shows the distribution of the number of points having the same height in the indoor point cloud data 1 . In the indoor point cloud data 1, it is considered that many points are gathered at the height of the ceiling and the floor. Using this point, the model generation device 10 separates and extracts the ceiling data and the floor data from the indoor point cloud data 1 based on the most numerous peak points in the histogram 2 .

That is, the model generation device detects the ceiling surface and the floor surface by regarding the two peak points in the histogram 2 in FIG. 2 as the height of the ceiling and the floor. The model generation device detects the point group belonging to the higher peak point among the peak points in the histogram 2 as the ceiling data 2A corresponding to the ceiling surface. In addition, the model generation device detects a point group belonging to the lower peak point among the peak points of the histogram 2 as the floor data 2B corresponding to the floor surface. Then, the model generation device separates and extracts the detected ceiling data 2A and floor data 2B from the indoor point cloud data 1 .

FIG. 3 is a diagram showing indoor point cloud data before separation of ceiling data and floor data, and FIG. 4 is a diagram showing indoor point cloud data 1A after separation of ceiling data and floor data.
In the indoor point cloud data 1A in the state of FIG. 3, an object such as a table existing indoors cannot be visually recognized, but in the indoor point cloud data 1A in the state of FIG. 4, the ceiling data and the floor data are separated. , the object point cloud data 2C of objects existing indoors can be visually recognized. The object point cloud data 2C included in the indoor point cloud data 1 can be used for 3D modeling of objects existing indoors.
By separating the ceiling data 2A and the floor data 2B from the indoor point cloud data 1, objects existing indoors can be easily separated by point cloud of each object.
When separating the point cloud of the object into individual objects in the [object modeling] process described below, cluster classification is performed based on distance. end up Therefore, the indoor point cloud data 1A is created by separating the floor data 2B as well as the ceiling data 2A.

[Segmentation processing]
Next, the model generation device performs segmentation processing. Segmentation processing is processing for separating and dividing ceiling data into one or more room regions.
In the segmentation process, the model generation device applies Region Growing, which is a method of point group processing, to the ceiling data 2A separated and extracted by the above-described ceiling/floor separation process to obtain the ceiling data 2A. into one or more room regions. The region growing method is an example of the point group processing method, and other methods may be adopted as long as the ceiling data 2A can be separated.

A region growing method will be described. In general, the region growing method is a method of recognizing spatially continuous pixels satisfying conditions such as pixel values (luminance) as one region. Any pixel that satisfies the above conditions is manually determined as a "seed point" and labeled. Then, the process of assigning the same label to pixels satisfying the conditions near the seed point is repeated.
On the other hand, the region growing method of this embodiment is applied particularly to a point group. is used to determine seed points and conditions.
That is, in the processing of the region growing method of this embodiment, the seed points are determined using the curvature values of the points included in the point group and the following "curvature threshold", and the normal line of the seed points and the adjacent points of the seed points Adjacent points that belong to the same region as the seed point are determined based on the angle formed by the normal to and the "smoothing threshold" described below. By repeatedly classifying points based on the seed points, the model generator grows regions based on the seed points. A point adjacent to a seed point is a point that exists within a predetermined range (number of points) from the seed point. The range defining adjacent points is set by the parameters described below.

Processing of the region growing method in this embodiment will be described in detail.
First, the model generation device performs a sorting process for sorting points included in a point group to be subjected to the region growing method by curvature values. Then, the model generation device selects the point with the smallest curvature value from the sorted point group as the minimum curvature point. This minimum curvature point becomes the seed point. The threshold used as the criterion for selecting the minimum curvature point is the "curvature threshold", and the model generation device selects a point having a curvature value smaller than the curvature threshold as the minimum curvature point, and uses this minimum curvature point as a seed point to divide the region. growth. This is because the point of minimum curvature is located in a flat area, and by growing the area from a flatter area, the number of segments can be reduced (prevented from becoming too many). The "curvature threshold" is set by parameters described below.

When the model generator grows regions from the determined seed points, it classifies points that are sufficiently close with respect to the "smoothness constraint" into one region (segment). Since the points within the same segment form a smooth surface, the angles of the normals to each other do not differ greatly (the angles formed by the normals of the points are small). Therefore, the model generating device regards points that are smaller than the angle formed by the normals to each other as points within the same segment. A "smoothing threshold" is a threshold for an angle between normals for regarding points as the same segment.
The model generation device determines that, among adjacent points to the seed point, points at which the angle between the normal and the normal to the seed point is smaller than the smoothing threshold value belongs to the same segment as the seed point. The "smoothing threshold" is set by parameters described below.

The output result of the region growing process is a set of segments, a set of points that are considered part of the same smooth surface. By setting the smoothness condition in defining the segments, it is ensured that the smooth surface ends (the region stops growing) at the boundary of the wall.
As for the present embodiment, it is assumed that in the point cloud data of the ceiling surface (ceiling data 2A), there will be a gap due to a door or the like between the rooms. At points near the gap, the angle of the normal to the seed point greatly exceeds the "smoothing threshold", and expansion (growth) of the region (segment) is considered to stop. Therefore, it is expected that the ceiling surface (ceiling data 2A) will be suitably separated into one or more room areas by the area growing method. Therefore, it can be said that the region growing method is a more preferable method for the application of separating the room regions.

Processing of the region growing method of the present embodiment is performed as follows.
The model generation device performs the sorting process described above, labels the minimum curvature points whose curvature is smaller than the curvature threshold value as seed points, and adds them to the "seed group". The model generator proceeds to start growing regions from each seed point until there are no more unlabeled points in the point cloud.

After searching for adjacent points for each seed point, the model generator performs the following processing.
(1) The model generator adds a neighboring point to the current region (segment) if the angle between the normal of each neighboring point and the normal of the current seed point is less than the above smoothing threshold. label the same as the seed point. This process is performed for all neighboring points of the current seed point.
(2) The model generator examines all adjacent points for curvature and adds adjacent points whose curvature value is less than the curvature threshold to the 'seed group'.
(3) The model generator removes the current seed point from the "seed group". The processes (1) and (2) are executed for the next seed point, and the processed seed point is deleted from the "seed group". When the "seed group" becomes empty, it means that the area has been grown, and the process ends.

FIG. 5 is a diagram showing a state in which the ceiling data is separated into room areas by the area growing method.
In the case of this embodiment, as a result of using the region growing method, the point cloud of the ceiling data 2A could be separated into three segmented room regions (room region 3a, room region 3b, and room region 3c).
As is clear from FIG. 5, the contours of the room regions and the boundaries between the room regions separated by the region growing method are unclear, and it is difficult to use them for creating an indoor 3D model. Therefore, processing for detecting and clarifying the outline of the room area is performed as follows.

[Image processing/contour approximation]
FIG. 6 is a diagram showing a projected image in which the outline of the room is detected.
The model generation device creates projected images 4a to 4c by projecting the ceiling data of the room areas 3a to 3c separated in FIG. Thus, the contours of the room regions 3a-3c are detected as polygons 5a-5c.
If the detected contour includes many points, a model that is not suitable for radio wave propagation simulation is created. is approximated with fewer points.

[Room modeling]
The model generator uses the detected polygons 5a-5c and height information of the ceiling data 2A and floor data 2B to create room 3D models 6a-6c (see FIG. 9) of each room.
The above is the procedure for creating an indoor 3D model.

[Object modeling]
Next, a procedure for creating a 3D model of an object existing in a room will be described.
The model generating device separates the point cloud 2C (FIG. 4) of the object into each room region (segment) using the outline of the room (polygons 5a to 5c) shown in FIG. Then, the model generation device performs clustering processing on the point groups of the object 2C separated for each room region (segment), and separates individual object point groups.
The clustering processing performed here is Euclidian Clustering. Euclidean clustering is a process of regarding points within a certain distance as the same cluster.
Furthermore, the model generation device obtains an object model by performing either of the following two types of modeling processing (simple object modeling, mesh object modeling) on point groups of individual objects separated by clustering processing. .

FIG. 7 is a diagram showing a simple object model. In simple object modeling, an image is created by projecting the point group of the object two-dimensionally (xy plane), and the contour of the object is detected by the same processing as that for the room area. Polygonal approximation is then performed using the detected contour and the maximum and minimum heights of the object, and the object is modeled as a polygonal prism. As a result, the simple object model shown in FIG. 7 is obtained.

FIG. 8 is a diagram showing a mesh object model. In mesh object modeling, for example, meshing (GST: Greedy Surface Triangulation), which is one of point group processing techniques, is performed to obtain a mesh object model shown in FIG.
A model generated by either simple object modeling or mesh object modeling can be read into a commercially available radio wave propagation simulator.

FIG. 9 is a diagram showing an indoor 3D model generated by the model generation device of this embodiment.
The indoor 3D model 6 shown in FIG. 9 includes a room 3D model 6a based on polygons 5a, a room 3D model 6b based on polygons 5b, and a room 3D model 6c based on polygons 5c. Furthermore, object models 7 are included in the room 3D models 6a-6c.

10A and 10B are diagrams for explaining the functional configuration of the model generation device of this embodiment, where FIG. 10A is a hardware functional block diagram and FIG. 10B is a software functional block diagram.
As shown in FIG. 10(a), the model generation device 10 executes a general-purpose operating system that controls the entire device, and a CPU (Central Processing Unit) that executes a program that implements the functions of the model generation device 10. 11, RAM (Random Access Memory) 12 in which various programs, temporary data, and variables are developed for processing by CPU 11, HDD (Hard Disk Drive) 13 in which programs and data are stored, and ROM (not shown). Read Only Memory), an I/O interface 14, a reading device 15, a network I/F 16, a display device 17, and an input device 18 such as a mouse or keyboard.

The CPU 11 is an example of a control circuit that is provided in the model generation device 10 and implements a control unit. can be applied.
The HDD 13 is a drive device that drives an internal hard disk. The HDD 13 reads programs and data stored in the hard disk as a storage medium, and writes data to the hard disk.

In addition, the model generation device 10 can also be used to store optical discs such as FDs (Floppy Disks), CDs (Compact Discs), DVDs (Digital Versatile Discs), and BDs (Blu-ray (registered trademark) Disks), as well as detachable discs such as flash memories. A reader/writer 15 that reads and writes programs and data to/from the storage medium 200 may be provided.
The reading device 15 is FDD (Floppy Disk Drive), CDD (Compact Disc Drive), DVDD (Digital Versatile Disc Drive), BDD (Blu-ray (registered trademark) Disk Drive), USB (Universal Serial Bus), or the like.
The CPU 11, RAM 12, HDD 13, I/O interface 14, and reading device 15 are connected via an internal bus, for example.
The model generating device 10 can input filtered or unfiltered indoor point cloud data from another device via the network I/F 16 . The model generation device 10 may obtain filtered or pre-filtered indoor point cloud data by the reading device 15 from a storage medium storing indoor point cloud data.
The model generator 10 may receive indoor point cloud data from a 3D scanner connected via the I/O interface 14 . In that case, it is particularly desirable that the model generation device 10 has a function of filtering the indoor point cloud data input from the 3D scanner.

As shown in FIG. 10B, the CPU 11 includes, as a control unit 10A, a data input unit 31, a parameter setting unit 32, a parameter generation unit 33, a histogram generation unit 34, a floor/ceiling detection unit 35, A room region separation unit 36, an outline detection unit 37, a room model generation unit 38, an object separation unit 39, and an object model generation unit 40 are executed.
These processing units are programs that implement the functions of the control unit 10A of the model generating device 10, and the programs can be stored in the hard disk included in the HDD 13 or the storage medium 200. FIG. The program is read from the hard disk or storage medium 200 to the RAM 12 by the HDD 13 or reading device 15 and executed by the CPU 11 .
The RAM 12 and HDD 13 constitute a storage unit 10B of the model generating device 10. FIG.
The storage unit 10</b>B mainly includes a point cloud data storage unit 51 and a parameter storage unit 52 .

The data input unit 31 receives the input of the filtered indoor point cloud data 1 and stores it in the point cloud data storage unit 51 .
The parameter setting unit 32 accepts a setting input of parameters (user parameters) by the user using the input device 18, for example.
The parameter generator 33 automatically sets or adjusts the internal parameters based on the user parameters, and stores them in the parameter storage 52 .
The histogram generator 34 generates a height histogram 2 based on the height information for each point included in the indoor point cloud data 1 .

The floor/ceiling detector 35 detects ceiling data 2A and floor data 2B from the indoor point cloud data based on the height histogram 2, and separates the ceiling data 2A and floor data 2B from the indoor point cloud data 1. FIG.
The room area separation unit 36 applies the area growing method to the ceiling data 2A to separate the room areas.
The contour detection unit 37 creates an image by two-dimensionally projecting the ceiling data of each separated room, and detects the contour of the room using image processing.
The room model generator 38 creates an indoor 3D model of each room from the detected outline, the floor height in the ceiling data 2A, and the ceiling height in the floor data 2B.
The object separation unit 39 separates point groups of objects inside the room for each room using the contour of the room.
The object model generator 40 generates an object model from the point group of the object.
When indoor point cloud data before filtering is input from an external device, the point cloud filter unit 41 performs filtering on the input indoor point cloud data.

The model generation device 10 automatically generates and automatically adjusts internal parameters for the model generation algorithm by the following processing.
The user of the model generation device 10 provides the model generation device 10 with a point interval (filter interval) in filtering, an assumed room area, an object surface area, an assumed detection number of rooms, a maximum room area Sets the number of points, the maximum number of points for the contour of the object.
At this time, the model generating device 10 receives setting inputs such as filter_spacing, room_min_area, room_max_area, obj_min_area, obj_max_area, room_num, room_max_vertices, obj_max_vertices, etc. as user parameters specifying these pieces of information.

filter_spacing is a parameter that specifies the filter spacing in filtering.
room_min_area and room_max_area are parameters that specify the minimum area and maximum area of the detected room, respectively.
obj_min_area and obj_max_area are parameters that specify the minimum surface area and maximum surface area of the detected object, respectively.
room_num is a parameter that specifies the number of rooms to be detected or the number of rooms to be detected.
room_max_vertices is a parameter that specifies the maximum number of vertices of the outline of the room area.
obj_max_vertices is a parameter that specifies the maximum number of vertices of the contour of the object.

The model generator 10 uses the user parameters described above to generate internal parameters as follows.
The model generator 10 generates internal parameters rc_minsize and rc_maxsize based on the minimum and maximum room areas specified by the user parameters room_min_area and room_max_area and the filter spacing specified by the user parameter filter_spacing.
rc_minsize is a parameter that specifies the minimum number of points in an area to be detected as a room.
rc_maxsize is a parameter that specifies the maximum number of points in an area detected as a room.
Specifically, rc_minsize and rc_maxsize are
rc_minsize=round(room_min_area/μ ² )
rc_maxsize=round(room_max_area/μ ² )
However, μ = filter_spacing (filter spacing)
is obtained by computing Here, round means an operation of rounding a value by rounding off. The same applies to the following.

The model generator 10 generates internal parameters os_minizie and os_maxsize based on the minimum surface area and maximum surface area of the object specified by the user parameters obj_min_area and obj_max_area and the filter spacing specified by the user parameter filter_spacing.
os_minizie is a parameter that specifies the minimum number of points in an area to be detected as an object.
os_maxsize is a parameter that specifies the maximum number of points in an area to be detected as an object.
Specifically, the above os_minizie and os_maxsize are
os_minsize=round(obj_min_area/μ ² )
os_maxsize=round(obj_max_area/μ ² )
However, μ = filter_spacing (filter spacing)
is obtained by computing
In addition to these, an internal parameter ksearch_neighbors may be set. This is the parameter used in the normal calculation algorithm.
You may set the internal parameter pixel_coeff. This is a parameter indicating how many pixels 1m corresponds to.
An internal parameter rc_flag may be set. Normally it is set to 0, but if the value of rc_flag is 1, the entire ceiling is regarded as one room.

The model generation device 10 also automatically generates and automatically adjusts the internal parameters rc_curv_thresh, rc_smooth_thresh, and rc_neighbors of the region growing method so that the number of rooms designated by room_num and the number of rooms separated by the region growing method match.
rc_curv_thresh is the curvature threshold for selecting the "minimum curvature point" to be adopted as the seed point from the point group targeted by the region growing method, as described above. is selected as the seed point.
rc_smooth_thresh is a parameter for determining whether or not one point in the point group targeted by the region growing method belongs to a predetermined segment. That is, the above smoothing method determines whether or not a point adjacent to a seed point belongs to the same region (segment) as the seed point based on the normal angle between this adjacent point and the seed point. is the threshold. Neighboring points with normals whose angle with the normal of the seed point is equal to or less than the "smooth threshold" are assumed to belong to the same region (segment) as the seed point.
rc_neighbors is a parameter that specifies the number of neighboring points (search range) searched from the seed point when expanding the region from the seed point. That is, the model generation device 10 determines whether or not the normals of the number of adjacent points determined by rc_neighbours with the seed point as a base point satisfy rc_smooth_thresh. It is possible to prevent the processing load from becoming excessively high by making decisions within an appropriate range without repeating the decision until there are no more points that satisfy rc_smooth_thresh.

The model generation device 10 classifies the point group into segments using a smooth threshold value and a curvature threshold value in the region growing method. That is, the model generation device 10 determines points that satisfy the curvature threshold as seed points, and classifies points that have normals whose angles with respect to the normals of the seed points satisfy the smoothing threshold into the same segment as the seed value.
Both the smoothing threshold and the curvature threshold are values of 0 or more. In either case, the larger the threshold, the easier it is for the points to be regarded as the same area, and as a result, the number of detected areas (number of segments) decreases.
The model generating device 10 adjusts the optimum combination of the smoothing threshold and the curvature threshold in a binary search manner using general set values (smoothing threshold=3.0 degrees, curvature threshold=1.0) as base points.

Specifically, the model generation device 10 automatically sets and adjusts rc_smooth_thresh (smooth threshold value) and rc_curv_thresh (curvature threshold value) like a binary search as follows.
FIG. 11 is a diagram illustrating automatic adjustment of the smoothing threshold and the curvature threshold.
In the following description, the unit degree of the smoothing threshold is omitted.
Let lhs=(0,0) and rhs(6.0,2.0) so as to satisfy (lhs+rhs)/2=(3.0,1.0). The model generation device 10 performs a process of searching for a combination of optimal setting values (smooth threshold, curvature threshold) between lhs=(0, 0) and rhs=(6.0, 2.0).
(1) (lhs+rhs)/2=((0,0)+(6.0,2.0))/2 yields the above general setting A(3.0,1.0).
The model generation device 10 executes the region growing method with this set value A to separate the ceiling data 2A into room regions. If the number of separated room areas matches the number of rooms specified by room_num, the process ends.
(2-1) If the number of separated room regions is greater than the number of rooms specified by room_num, then the set value A is set to a new lhs, and (lhs+rhs)/2=((3.0, 1.0) +(6.0, 2.0))/2 to obtain a new set value B1 (4.5, 1.5).
(2-2) If the number of separated room regions is smaller than the number of rooms specified by room_num, (lhs+rhs)/2=((0, 0)+(3. 0, 1.0))/2 to obtain a new set value B2 (1.5, 0.5).

When the set value B1 is obtained, the ceiling data 2A is separated into room regions by executing the region growing method using the set value B1. If the number of separated room areas matches the number of rooms specified by room_num, the process ends.
(3-1) If the number of separated room regions is greater than the number of rooms specified by room_num, then (lhs+rhs)/2=((4.5, 1.5) with the set value B1 as the new lhs +(6.0, 2.0))/2 to obtain a new set value C1.
(3-1) If the number of separated room areas is smaller than the number of rooms specified by room_num, (lhs+rhs)/2=((3.0, 1.0) with the set value B1 as the new rhs +((4.5, 1.5))/2 to obtain new set value C2

When the set value B2 is obtained as described above, the ceiling data 2A is separated into room regions by executing the region growing method using the set value B2. If the number of separated room areas matches the number of rooms specified by room_num, the process ends.
(3-3) If the number of separated room regions is greater than the number of rooms specified by room_num, ((1.5, 0.5)) is set as a new lhs, and (lhs+rhs) /2=((1.5, 0.5)+(3.0, 1.0))/2 to obtain a new set value C3.
(3-4) If the number of separated room areas is smaller than the number of rooms specified by room_num, ((1.5, 0.5)) is set as a new rhs, and (lhs+rhs) /2=((0.0,0.0)+((1.5,0.5))/2 is calculated to obtain a new set value C4.

Furthermore, the ceiling data 2A is separated by executing the area growing method using the new set values C1, C2, C3, and C4, and the separation result is compared with the number of rooms specified by room_num.
The process described above is repeated in the same way, but if the number of detected room areas does not match the number of rooms specified by room_num even after repeating the process a certain number of times, the process ends when the numbers are close to each other. good too.

The model generation device 10 automatically sets contour approximation parameters (wd_epsilon, os_epsilon) so as not to exceed the maximum number of vertices of the contour of the room region specified by room_max_vertices and the maximum number of vertices of the contour of the object specified by obj_max_vertices. adjust.
In the Douglas Peucker algorithm, the model generator 10 uses the value of epsilon to simplify the contour. Larger values of epsilon result in more simplified contours.
The method of automatic adjustment is to set epsilon=0.001 for both wd_epsilon and os_epsilon and implement the Douglas Peucker algorithm to simplify the contour.

If the number of vertices of the room after simplification > the maximum number of vertices specified by room_max_vertices or the number of vertices of the object after simplification > the maximum number of vertices specified by obj_max_vertices, the model generator 10 doubles the epsilon. and run the Douglas Peucker algorithm again.
If the number of vertices of the room after simplification>the maximum number of vertices specified by room_max_vertices or the number of vertices of the object after simplification>the maximum number of vertices specified by obj_max_vertices, the model generator 10 completes the adjustment.

12 and 13 are flowcharts for explaining the flow of model generation processing executed by the CPU of the model generation device.
In step S101, the CPU 11 determines whether indoor point cloud data has been input.
When it is determined that indoor point cloud data has been input (Yes in step S101), the CPU 11 determines in step 102 whether user parameters have been input.
As described above, the input user parameters are filter_spacing that specifies the filter spacing in the filtering process, room_min_area and room_max_area that specify the minimum area and maximum area of the room to be detected, the maximum surface area of the object to be detected, and the minimum surface area. obj_min_area and obj_max_area that specify respectively, room_num that specifies the number of rooms expected to be detected, room_max_vertices that specifies the maximum number of vertices of the outline of the room area, and obj_max_vertices that specifies the maximum number of vertices of the outline of the object.

When determining that a user parameter has been input (Yes in step S102), the CPU 11 automatically sets internal parameters based on the user parameters and stores them in the parameter storage unit 52. FIG.
In particular, the CPU 11 uses internal parameters rc_minsize, maximum Generate an internal parameter rc_maxsize that specifies the number of points.
The model generation device 10 also calculates the minimum number of points (os_minizie) and the maximum number of points (os_minizie) and the maximum number of points ( os_maxsize).

In step S104, the CPU 11 creates a height histogram based on the indoor point cloud data.
In step S105, the CPU 11 detects indoor point cloud data of the floor surface (floor data) and indoor point cloud data of the ceiling surface (ceiling data) in the input indoor point cloud data from the created histogram.
The CPU 11 detects a bin having a value equal to or greater than [maximum value of histogram]×fc_cutoff_coeff as a peak. The internal parameter fc_cutoff_coeff is the ratio that determines the threshold of bins to be detected as peaks relative to the largest bin in the histogram. Note that fc_cutoff_coeff is a fixed parameter that is not subject to automatic setting or automatic adjustment.
Among the bin clusters detected as peaks, the CPU 11 regards the place with the lowest height as the floor, and the rest as the ceiling group.
If peak detection is not successful, use the minimum and maximum ceiling heights specified by the internal parameters fc_min_thresh and fc_max_thresh, and consider all points whose height is greater than or equal to fc_min_thresh and less than or equal to fc_max_thresh as ceiling data. may fc_min_thresh and fc_max_thresh are fixed parameters that are not automatically set or adjusted.

In step S106, the CPU 11 performs a region growing method on the ceiling data using the internal parameters rc_neighbors, rc_smooth_thresh, and rc_curv_thresh, and separates the point cloud of the ceiling data into one or more segments corresponding to room regions. For the values of rc_smooth_thresh and rc_curv_thresh, general values are used as initial values as described above.
After separating the ceiling data point group into segments, the CPU 11 adopts the segment when the number of points included in the segment is equal to or greater than rc_minsize generated in step S103 and equal to or less than rc_maxsize, and otherwise does not adopt.

In step S107, the CPU 11 creates a planar image of each room by projecting the separated room areas two-dimensionally. The CPU 11 regards each segmented ceiling data as a room area, and creates image data (planar image) by two-dimensionally projecting point groups of each room area.
By calculating point_size=1.414×100×μ (filter interval), the radius (unit: pixel) when projecting the point group onto the two-dimensional image may be set. This point_size can be used both when projecting the point cloud of the room area and when projecting the point cloud of the object.

In step S108, the CPU 11 performs image processing on the planar image to detect the outline of each room, and reduces the number of vertices of the outline by the Douglas Peucker algorithm. At this time, an internal parameter wd_epsilon, which is an index indicating the degree of simplification, is used. The CPU 11 sets wd_epsilon=0.001 as an initial value and implements the Douglas Peucker algorithm to simplify the outline of the room area.
In S109, the CPU 11 determines whether or not the maximum number of vertices of the room after simplification is greater than the maximum number of vertices specified by room_max_vertices. If the number of vertices of the room after simplification is greater than the maximum number of vertices specified by room_max_vertices (Yes in S109), the CPU 11 doubles wd_epsilon in S110, returns to step S108, and performs the Douglas Peucker algorithm. to run. If the number of vertices of the room after simplification is equal to or smaller than the maximum number of vertices specified by room_max_vertices (No in S109), the CPU 11 completes adjustment of wd_epsilon and proceeds to the next step.

In step S111, the CPU 11 determines whether or not the number of room regions matches the number of rooms designated by the user as room_num. If it is determined that they do not match (No in step S111), the CPU 11 resets the parameters rc_smooth_thresh and rc_curv_thresh used in the region growing method in step S112 as described above. Then, the process is returned to step S106, and the process is redone from the separation of the room area. If the number of rooms separated in step S111 does not match the number specified by the user even after adjustment, the closest number of rooms may be used.

If it is determined that the number of room regions matches the number of rooms specified by the user in room_num (Yes in step S111), the CPU 11 separates point groups of objects inside each room using the outline of the room in step S113. .
In step S113, the CPU 11 separates indoor objects using the outline of the room simplified in step S108.
When separating objects, the CPU 11 uses internal parameters dist_from_wall and wd_max_vertices.
dist_from_wall is a threshold value that determines how far inside the wall a point must be to be considered as a point inside the room. dist_from_wall is a fixed parameter that is not subject to automatic setting or automatic adjustment.
wd_max_vertices is a fixed parameter that is not automatically set or adjusted.

Although isolated objects are modeled, as described above, the model generator 10 is capable of simple object modeling and mesh object modeling. This flow chart mainly describes simple object modeling. Simplified object modeling is a method of detecting the contour of an object using a method similar to that for detecting the contour of a room, and modeling the detected object.
Simplified object modeling uses an internal parameter os_tolerance, which is a distance threshold for separating objects, and os_minsize and os_maxsize generated in step S103. os_tolerance is a fixed parameter that is not subject to automatic setting or automatic adjustment.
The CPU 11 regards points that are not separated by a distance specified by os_tolerance as belonging to the same object, and points that are separated by this distance or more as belonging to different objects, thereby dividing the objects separated in step S113. Separate into individual objects.
Furthermore, the CPU 11 adopts as an object the number of points greater than the number specified by os_minsize and less than the number specified by os_maxsize among the separated individual objects, and does not employ other objects.

In step S115, the CPU 11 creates image data (planar image) by two-dimensionally projecting the point groups of the individual objects separated in step S114.
The CPU 11 performs image processing on the planar image to detect the contour of each object, and at this time, reduces the number of vertices by the Douglas Peucker algorithm. At this time, an internal parameter os_epsilon, which is an index indicating the degree of simplification when reducing the number of contour vertices, is used. The CPU 11 first sets os_epsilon=0.001 as an initial value and executes the Douglas Peucker algorithm to simplify the outline of the object.

In S117, the CPU 11 determines whether or not the number of vertices of the object after simplification>the maximum number of vertices specified by obj_max_vertices. If the number of vertices of the object after simplification>the maximum number of vertices specified by obj_max_vertices (Yes in S117), the CPU 11 doubles os_epsilon in S118, returns to step S108, and executes the Douglas Peucker algorithm. Run. If the number of vertices of the object after simplification<the maximum number of vertices specified by obj_max_vertices (No in S117), the CPU 11 completes the adjustment of wd_epsilon and proceeds to the next step.

In step S119, the CPU 11 creates a 3D model of the room from the outline of the room and the heights of the floor and ceiling based on the histogram.
In step S120, the CPU 11 creates a 3D model of the object based on the contour, minimum height, and maximum height of the object.
The CPU 11 generates an indoor 3D model as shown in FIG. 9 from the 3D model of the room and the 3D model of the object respectively created in steps S119 and S120.

Note that mesh object modeling is performed as follows. The CPU 11 meshes the object using a Greedy Surface Triangulation (GST) algorithm. In mesh object modeling, the CPU 11 uses the internal parameters om_search_radius, om_mu, om_neighbors, om_max_surface_angle, om_min_angle, om_max_angle.
om_search_radius is a GST parameter that indicates the maximum allowable distance of meshing triangles.
om_mu is a parameter that controls the size of the neighborhood to search.
om_neighbours is a parameter that controls the size of the neighborhood to search.
om_max_surface_angle is a parameter to handle cases where there are sharp edges and two triangles are very close to each other.
om_min_angle is a parameter that indicates the minimum angle of triangles to be meshed.
om_max_angle is a parameter that indicates the maximum angle of triangles to be meshed.
The CPU 11 meshes the object with the above parameters and reduces the number of meshes using the Quadratic Decimation (QD) algorithm. At that time, an internal parameter om_target_reduction that indicates the ratio of the target number of reduction surfaces is used.

When only a room model was generated without objects from the indoor point cloud data of FIG. 1, the number of planes was 54, and the calculation time was 5 minutes and 16 seconds.
When the simple object model was generated in addition to the room model, the total number of surfaces was 2447, and the calculation time was 12 minutes and 01 seconds.
When the mesh object model was generated in addition to the room model, the number of surfaces was 331264 and the calculation time was 5 hours, 36 minutes and 18 seconds.
It can be seen that the simple object model has fewer faces than the mesh object model, and the calculation time can be greatly reduced.

Note that the processing shown in FIGS. 12 and 13 is an example and may be performed in a different order. For example, after separating the room area in step S106, the determination in step S111 and the update of rc_smooth_thresh and rc_curv_thresh in step S112 may be performed before performing steps S107 to S110 related to the contour detection of the room area.

With the above configuration, according to the present embodiment, the user can easily create an indoor 3D model from indoor point cloud data simply by setting and inputting simple parameters such as the number of rooms. . In addition, objects such as tables and chairs existing in a room, which were difficult in the past, can also be 3D modeled together with the room. As a result, an indoor 3D model more suitable for radio wave propagation simulation (ray tracing analysis) can be created from indoor point cloud data.

10 model generation device, 11 CPU, 12 RAM, 13 HDD, 14 I/O interface, 15 reading device, 16 network I/F, 17 display device, 18 input device, 30A control unit, 30B storage unit, 31 data input unit , 32 parameter setting unit, 33 parameter generation unit, 34 histogram generation unit, 35 floor/ceiling detection unit, 36 room region separation unit, 37 contour detection unit, 38 room model generation unit, 39 object separation unit, 40 object model generation unit , 200 storage medium, 41 point cloud filter unit

Claims (5)

a data input unit for receiving input of indoor point cloud data;
a parameter setting unit that accepts input parameter settings by a user;
a parameter generator that generates internal parameters based on the input parameters;
a histogram generation unit that generates a height histogram based on the input indoor point cloud data;
a floor/ceiling detection unit that detects ceiling data and floor data from the indoor point cloud data based on the height histogram;
a room region separation unit that applies region growth to the ceiling data and separates one or more room regions;
a contour detection unit that detects the contour of the ceiling data of each separated room;
a room model generation unit that generates a plane model of each room from the detected outline, the floor height in the floor data, and the ceiling height in the ceiling data;
with
the input parameters include the number of rooms;
When the number of room regions separated by the room region separation unit is different from the number of rooms set as the input parameter, the parameter generation unit regenerates the internal parameters,
generating the planar model based on the regenerated internal parameters;
A model generation device characterized by: In the model generation device according to claim 1,
The indoor point cloud data is pre-filtered filtered indoor point cloud data,
the input parameters further include point cloud spacing in the filtered indoor point cloud data;
A model generation device characterized by: In the model generation device according to claim 1 or 2,
an object separation unit that separates the point cloud of the object inside the room for each room using the contour of the room;
an object model generation unit that generates an object model using the separated point cloud;
A model generating device comprising: A model generation method executed by a processor, comprising:
Accepts input of indoor point cloud data by the user,
accepts input parameters set by the user,
generating internal parameters based on said input parameters;
generating a height histogram based on the input indoor point cloud data;
Detecting ceiling data and floor data from indoor point cloud data based on the height histogram;
subjecting the ceiling data to a region growing method to separate room regions;
Detect the contours of the ceiling data for each separated room,
creating a plane model of each room from the detected outline, the height of the floor in the floor data, and the height of the ceiling in the ceiling data;
the input parameters include the number of rooms;
if the number of the separated room regions is different from the number of rooms set as the input parameters, regenerate the internal parameters;
generating the planar model based on the regenerated internal parameters;
A model generation method characterized by: A model generation program to be executed by a processor,
Accepts input of indoor point cloud data by the user,
accepts input parameters set by the user,
generating internal parameters based on said input parameters;
generating a height histogram based on the input indoor point cloud data;
Detecting ceiling data and floor data from indoor point cloud data based on the height histogram;
subjecting the ceiling data to a region growing method to separate room regions;
Detect the contours of the ceiling data for each separated room,
creating a plane model of each room from the detected outline, the height of the floor in the floor data, and the height of the ceiling in the ceiling data;
the input parameters include the number of rooms;
if the number of the separated room regions is different from the number of rooms set as the input parameters, regenerate the internal parameters;
generating the planar model based on the regenerated internal parameters;
A model generation program characterized by:

Images