JP7344488B2 - Three-dimensional model data generation device, computer program, and three-dimensional model data generation method - Google Patents

Description

The present disclosure relates to a three-dimensional model data generation device and a three-dimensional model data generation method.

Three-dimensional point cloud map data that represents road surfaces and other objects on the ground using three-dimensional point clouds is known (for example, see Non-Patent Document 1). Objects on the ground include natural objects as well as artificial objects such as buildings. The three-dimensional point cloud map data is used after being converted into three-dimensional model data including surface information, for example.

The three-dimensional model data is used, for example, in a simulator for verifying automatic driving of a vehicle, particularly in a LiDAR (Light Detection and Ranging) simulation that includes tracing of light emitted from a laser radar.

When generating three-dimensional model data, a three-dimensional point group is converted into, for example, a group of cubic voxels. Through this conversion, three-dimensional model data having surface information regarding road surfaces and other objects on the ground is generated from the three-dimensional point cloud map data.

Alternatively, the 3D point cloud map data can be converted into surface information by meshing the 3D point cloud using surface reconstruction techniques such as Poisson Surface Reconstruction and Ball Pivoting Algorithm (BPA). is converted to 3D model data with added.

Yuki Kikkawa and 4 others, "Autonomous Driving Demonstration Experiment: Verification of Position Estimation Accuracy", Journal of the International Society of Traffic Safety, vol. 42, No. 2, pp. 48-53, [online], October 2017, [searched on August 10, 2021], Internet <URL: http://www.iatss.or.jp/common/pdf/publication/iatss- review/42-2-06.pdf＞

According to the three-dimensional model data generation technique by meshing using the surface reconstruction technique described above, it is possible to generate three-dimensional model data that smoothly expresses the surface of a three-dimensional shape. On the other hand, according to this technique, the processing load associated with the generation of three-dimensional model data is large, and the calculation time of the processor required for data generation is long.

According to the generation technology of 3D model data by voxelization of a 3D point group, the processing load related to the generation of a 3D model can be reduced compared to the generation technology of 3D model data by meshing. , it is possible to shorten the calculation time of the processor required for data generation. On the other hand, according to this technique, since points are converted into voxels with 8 vertices and 6 sides, the amount of three-dimensional model data is large and the amount of memory required is large.

Therefore, according to one aspect of the present disclosure, it is desirable to be able to provide a technique that can generate three-dimensional model data including surface information from three-dimensional point group data at high speed and with a small amount of data.

A three-dimensional model data generation device according to one aspect of the present disclosure includes an acquisition section, a separation section, and a generation section. The acquisition unit acquires point cloud data representing a three-dimensional shape on the ground surface including a road surface using a three-dimensional point cloud. Point cloud data describes the three-dimensional coordinates of each point in a three-dimensional point cloud.

The separation unit separates the three-dimensional point group into a first point group corresponding to the road surface and a second point group other than the first point group based on the point group data. The generation unit converts the first point group into a group of planar elements and converts the second point group into a group of three-dimensional elements, thereby representing the shape of the road surface with the group of planar elements and expressing the shape of the road surface with the group of three-dimensional elements. Generate three-dimensional model data representing shapes other than road surfaces.

When converting the first point group corresponding to the road surface into a group of planar elements to generate three-dimensional model data with surface information, the entire point group including the first point group and the second point group The amount of 3D model data can be suppressed, and the amount of memory required to store the 3D model data can be reduced, compared to uniformly converting 3D elements into 3D elements with 8 vertices and 6 faces.

According to this conversion technique, the processing load on the three-dimensional model data generation device can be further reduced compared to the three-dimensional model data generation technique using meshing. Therefore, according to one aspect of the present disclosure, it is possible to provide a three-dimensional model data generation device that can generate three-dimensional model data including surface information from three-dimensional point cloud data at high speed and with a small amount of data. .

According to one aspect of the present disclosure, each of the planar elements may be a quadrangular planar element. The three-dimensional model data may include data defining each of the four vertices and one face of the planar element.

According to one aspect of the present disclosure, each of the three-dimensional elements may be a triangular prism three-dimensional element. The three-dimensional model data may include data defining six vertices and three faces of each three-dimensional element. When a triangular prism three-dimensional element is used, the amount of three-dimensional model data can be further reduced compared to a case where a cubic or quadrangular prism three-dimensional element is used.

According to one aspect of the present disclosure, a computer program may be provided that causes a computer to at least partially realize the functions of an acquisition unit, a separation unit, and a generation unit included in a three-dimensional model data generation device. A computer program may be recorded on a computer-readable non-transitory storage medium.

According to one aspect of the present disclosure, a three-dimensional model data generation method may be provided. The three-dimensional model data generation method can be executed by a computer.

The 3D model data generation method represents the 3D shape on the ground surface including the road surface using a 3D point cloud, and acquires point cloud data that describes the 3D coordinates of each point in the 3D point cloud. Based on the data, the three-dimensional point group is separated into a first point group corresponding to the road surface and a second point group other than the first point group, and the first point group is separated into a group of planar elements. By converting the second point group into a group of three-dimensional elements, three-dimensional model data is generated in which the shape of the road surface is represented by a group of planar elements, and the shape other than the road surface is represented by a group of three-dimensional elements. may include.

According to this three-dimensional model data generation method, like the three-dimensional model data generation device, three-dimensional model data including surface information can be generated from three-dimensional point group data at high speed and with a small amount of data.

FIG. 2 is a block diagram showing the configuration of a three-dimensional model data generation device. 3 is a flowchart showing data generation processing executed by a processor. FIG. 3 is a diagram showing a simple example of a three-dimensional point group. FIG. 3 is a diagram showing a simple example of three-dimensional model data.

Exemplary embodiments of the present disclosure will be described below with reference to the drawings.
The three-dimensional model data generation device 1 of the present embodiment shown in FIG. The generation program PR) is installed and configured.

As shown in FIG. 1, the three-dimensional model data generation device 1 includes a processor 11, a memory 13, a storage 15, a display section 17, an operation section 19, and a data input/output section 21. The processor 11 executes processing according to a computer program stored in the storage 15. The memory 13 includes a RAM and is used as a working memory when the processor 11 executes processing.

The storage 15 is configured with a hard disk drive or a solid state drive, and stores computer programs and various data used when executing processes according to the computer programs. The storage 15 stores a three-dimensional model data generation program PR.

The display unit 17 is controlled by the processor 11 and is configured to display various information to a user operating the three-dimensional model data generation device 1. The display unit 17 includes, for example, a liquid crystal display.

The operation unit 19 is configured to input an operation signal from a user to the three-dimensional model data generation device 1 to the processor 11. The operation unit 19 includes, for example, a keyboard and a pointing device.

The data input/output unit 21 is used for inputting data from the outside and outputting data to the outside. The data input/output unit 21 can include a communication interface that communicates with an external device, for example, by wire or wirelessly. The data input/output unit 21 can include a USB interface that can read and write data to and from a USB memory. The data input/output unit 21 can include a media reader/writer capable of reading and writing data to and from a card-type recording medium.

When a three-dimensional model data generation command is input from the user via the operation unit 19, the processor 11 executes data generation processing (see FIG. 2) according to the three-dimensional model data generation program PR stored in the storage 15.

When starting the data generation process shown in FIG. 2, the processor 11 acquires three-dimensional point cloud map data, which is a target for generating three-dimensional model data, from a specified location (S110). The three-dimensional point cloud map data is obtained from an external device through communication, for example. Alternatively, the three-dimensional point cloud map data is read from a USB memory or other recording media.

The three-dimensional point cloud map data is point cloud data that represents the three-dimensional shape of a predetermined area, specifically, the three-dimensional shape (particularly the external shape) of a road surface and an object on the ground surface other than the road surface, using a three-dimensional point cloud. The objects include artificial objects such as buildings and natural objects such as plants.

The three-dimensional point cloud map data describes the three-dimensional coordinates of each point constituting the three-dimensional point cloud. The three-dimensional point cloud map data is created using a measuring device such as a laser radar mounted on a vehicle, for example. In this case, the three-dimensional point cloud map data may be configured as data that represents the three-dimensional shape of an object on the road surface corresponding to the road on which the vehicle is traveling and the surrounding ground surface as a three-dimensional point group.

The three-dimensional point cloud map data is prepared, for example, as point cloud data in PCD (Point Cloud Data) format. PCD format point cloud data describes the position of each point constituting a three-dimensional point group using three-dimensional coordinates (x, y, z).

The PCD format point cloud data may further include information on the normal vector of each point. That is, the three-dimensional point cloud map data may be point cloud data that describes the three-dimensional coordinates and normal vector of each point in the three-dimensional point cloud.

After the processing in S110, the processor 11 processes the three-dimensional point cloud map data by filtering the acquired three-dimensional point cloud map data so as to reduce the number of points in the three-dimensional point cloud (S120).

Specifically, the processor 11 divides the three-dimensional coordinate space including the three-dimensional point group into grids of a predetermined size, places one representative point in each grid cell, and deletes the remaining points. The original point cloud map data can be processed (S120). The representative point may correspond to the centroid of the point cloud within the corresponding grid cell.

Therefore, the three-dimensional point cloud map data after the filtering process can be configured as point cloud data in which each point constituting the point group is arranged at intervals corresponding to grid cells in the three-dimensional coordinate space. The spacing may be, for example, a spacing corresponding to 20 cm in real space.

For example, in the 3D point cloud map data after filtering processing, when the 3D coordinate space is divided into 20cm grids equivalent to real space, a representative point is determined for each grid cell of a 20cm cube, and other points are It can be point cloud data with the clouds removed.

In subsequent S130, the processor 11 performs segmentation of the three-dimensional point cloud indicated by the three-dimensional point cloud map data after the filtering process. That is, the processor 11 separates the three-dimensional point group into a first point group PG1 showing features corresponding to the road surface and a second point group PG2 other than the features. This separation can be achieved using, for example, a point cloud library (PCL).

Among the entire three-dimensional point group, a relatively flat point group located at the bottom in the vertical direction exhibits features corresponding to the road surface. The second point group PG2 includes a point group representing objects on the ground surface other than the road surface.

Through the process of S130, first point cloud data describing the three-dimensional coordinates of each point P1 constituting the first point group PG1 and a second point group PG2 are constructed from the three-dimensional point cloud map data. Second point group data describing the three-dimensional coordinates of each point P2 is generated. Both the first point cloud data and the second point cloud data may be point cloud data in the PCD format, similar to the three-dimensional point cloud map data described above.

In subsequent S140, the processor 11 converts each point P1 constituting the first point group PG1 into a plane element E1 based on the first point group data, thereby generating a first point group having surface information representing the shape of the road surface. Generate shape model data.

In S140, the processor 11 generates first shape model data such that one plane element E1 is assigned to one point P1 in the first point group PG1. For example, the first shape model data is generated so as to convert each point P1 of the first point group PG1 corresponding to the road surface out of the three-dimensional point group shown in FIG. 3 into a plane element E1 shown in FIG. do.

Specifically, the planar element E1 may be a quadrilateral, particularly a square planar element having four vertices and one face. For example, the processor 11 converts each of the points P1 constituting the first point group PG1 into a plane element E1 including a plane having a normal line corresponding to the normal vector of this point, thereby converting the first shape into a plane element E1. Generate model data. The plane element E1 may be a plane element having a size corresponding to the grid cell used in the filtering process (S120), for example.

The first shape model data includes, for each plane element E1, three-dimensional coordinates of four vertices and a description defining that the four vertices constitute a surface, as data defining the corresponding plane element E1. The shape model data is configured as, for example, OBJ format data.

In subsequent S150, the processor 11 represents the shape of the object on the ground other than the road surface by converting each point P2 forming the second point group PG2 into a three-dimensional element E2 based on the second point group data. Generate second shape model data having surface information.

In S150, the processor 11 generates second shape model data such that one three-dimensional element E2 is assigned to one point in the second point group PG2. For example, out of the three-dimensional point group shown in FIG. 3, each point P2 in the second point group PG2 corresponding to an object on the ground surface other than the road surface is converted into the three-dimensional element E2 shown in FIG. Generate shape model data.

The three-dimensional element E2 is specifically a triangular prism three-dimensional element, particularly a triangular prism three-dimensional element having six vertices and three faces. The three faces may be the three sides of the triangular prism that have normals parallel to the horizontal plane (ie, the xy plane). That is, the three-dimensional element E2 may be a triangular prism three-dimensional element that includes three side surfaces, and may be a three-dimensional element that does not include upper and lower surfaces that have normals parallel to the vertical direction (ie, the z direction).

For example, the processor 11 converts each of the points P2 constituting the second point group PG2 into a three-dimensional element E2 having a side surface facing in a direction corresponding to the normal vector of this point P2, and converts the point P2 into a three-dimensional element E2 having a side surface facing in a direction corresponding to the normal vector of this point P2, generate. The three-dimensional element E2 may be, for example, a three-dimensional element E2 having a size corresponding to the grid cell used in the filtering process (S120).

The second shape model data is defined for each three-dimensional element E2 by a combination of the three-dimensional coordinates of six vertices included in the three-dimensional element E2 and four vertices for each of the three sides, as data that defines the corresponding three-dimensional element E2. includes a description of Like the first shape model data, the second shape model data is also configured as, for example, OBJ format data.

In subsequent S160, the processor 11 outputs the combination of the first and second shape model data as three-dimensional model data corresponding to three-dimensional point cloud map data. This three-dimensional model data corresponds to data obtained by adding surface information to a three-dimensional shape corresponding to a three-dimensional point group.

In S160, the processor 11 stores the three-dimensional model data in the storage 15, for example. Alternatively, the processor 11 can output the three-dimensional model data to a simulator.

Examples of simulators include multi-robot simulators for autonomous driving. The LGSVL simulator is known as a corresponding simulator. This type of simulator is used, for example, to verify autonomous driving algorithms by driving multiple vehicles in a virtual three-dimensional coordinate space that corresponds to the vehicle driving environment defined by three-dimensional model data. be done.

According to the three-dimensional model data generation device 1 of the present embodiment described above, three-dimensional model data in which surface information is added to a three-dimensional shape corresponding to a three-dimensional point group can be generated with a small processing load on the processor 11. , can be generated efficiently with a small amount of data.

For example, each point constituting the three-dimensional point group is not separated into a first point group PG1 corresponding to the road surface and a second point group PG2 other than the first point group PG1. Consider a first comparative example in which the elements are uniformly converted into cubic three-dimensional elements (so-called voxels) with eight vertices and six faces.

According to the first comparative example, the amount of 3D model data tends to become unnecessarily large because the 3D elements contain a lot of surface information that is not important for verifying autonomous driving algorithms using LiDAR technology. .

On the other hand, consider a second comparative example in which three-dimensional model data with surface information added is generated using surface reconstruction techniques such as Poisson surface reconstruction and ball pivot algorithm. According to the second comparative example, although it is possible to generate three-dimensional model data that expresses the surface of a three-dimensional shape smoothly and with high precision, the processing load on the processor 11 is large, making it difficult to generate the three-dimensional model data. It takes time.

When considering the use of 3D model data in the above-mentioned simulator, the 3D model data can represent the surfaces of road surfaces and objects on the ground in the corresponding space with the accuracy necessary for verification of autonomous driving using LiDAR. It's fine if you can express it.

From the viewpoint of accuracy, the first comparative example contains a lot of information on aspects that have little influence on verification accuracy, and the amount of three-dimensional model data is wasted. According to the second comparative example, the disadvantage that it takes time to generate the three-dimensional model data may be greater for the user than the advantage regarding the accuracy of the surface information in the three-dimensional model data. In particular, when the three-dimensional point cloud map data is point cloud data relating to a wide area and the amount of data is huge, the benefits of reducing processing load and data generation time are very large for the user.

According to this embodiment, the first point group PG1 is converted into a group of planar elements E1 with 4 vertices and 1 side, and the second point group PG2 is converted into a group of triangular prism elements with 6 vertices and 3 sides. To suppress the amount of three-dimensional model data compared to the first comparative example in which each point is converted into a three-dimensional element with eight vertices and six faces without dividing it into a first point group PG1 and a second point group PG2. This reduces the amount of memory required to store 3D model data.

Furthermore, in order to convert objects on the ground into a group of triangular prism 3D elements E2 that have fewer faces than cube or quadrangular prism 3D elements, even if we focus only on the second point group PG2, the three-dimensional model The amount of memory required to store data can be reduced.

Furthermore, in this embodiment, since the surface information is added using the planar element E1 and the three-dimensional element E2 having a prescribed shape, the surface information is added compared to surface reconstruction techniques using Poisson surface reconstruction and ball pivot algorithms. , it is possible to generate three-dimensional model data with surface information at high speed with a small processing load.

Therefore, according to this embodiment, three-dimensional model data including surface information can be generated from three-dimensional point cloud map data at high speed and with a small amount of data.

It goes without saying that the present disclosure is not limited to the above-described embodiments, and may take various forms. The planar element E1 does not have to be a quadrilateral, and may have a polygonal shape such as a triangle or a pentagon, or a non-polygonal shape. The three-dimensional element E2 also does not have to be a triangular prism, and may have a quadrangular prism, a cube, or other three-dimensional shape.

The technology of the present disclosure is not limited to a three-dimensional model data generation device for a simulator for autonomous driving algorithm verification, but can be applied to various devices that generate three-dimensional model data including surface information based on point cloud data. Applicable to the system.

The function that one component in the above embodiment has may be distributed and provided to multiple components. Functions possessed by multiple components may be integrated into one component. A part of the configuration of the above embodiment may be omitted. At least a part of the configuration of the embodiment described above may be added to or replaced with the configuration of other embodiments described above. All aspects included in the technical idea specified from the words described in the claims are embodiments of the present disclosure.

DESCRIPTION OF SYMBOLS 1...Three-dimensional model data generation device, 11...Processor, 13...Memory, 15...Storage, 17...Display section, 19...Operation section, 21...Data input/output section, E1...Planar element, E2...Three-dimensional element, PG1... First point group, PG2...second point group, PR...three-dimensional model data generation program.

Claims (5)

an acquisition unit that represents a three-dimensional shape on the ground surface including a road surface by a three-dimensional point group, and acquires point cloud data that describes three-dimensional coordinates of each point in the three-dimensional point group;
a separation unit that separates the three-dimensional point group into a first point group corresponding to the road surface and a second point group other than the first point group based on the point group data;
By converting the first point group into a group of planar elements and converting the second point group into a group of three-dimensional elements, the shape of the road surface is represented by the group of planar elements, and the group of three-dimensional elements is expressed. a generation unit that generates three-dimensional model data representing a shape other than a road surface;
Equipped with
Each of the three-dimensional elements is a triangular prism three-dimensional element,
The three-dimensional model data is a three-dimensional model data generation device including data defining six vertices and three faces of each of the three-dimensional elements . Each of the planar elements is a rectangular planar element,
The three-dimensional model data generation device according to claim 1, wherein the three-dimensional model data includes data defining four vertices and one face of each of the planar elements. A computer program for causing a computer to realize the functions of the acquisition unit, the separation unit, and the generation unit included in the three-dimensional model data generation device according to claim 1 or 2 . A three-dimensional model data generation method executed by a computer, the method comprising:
Representing a three-dimensional shape on the ground surface including a road surface by a three-dimensional point group, and obtaining point cloud data describing three-dimensional coordinates of each point in the three-dimensional point group;
Separating the three-dimensional point group into a first point group corresponding to the road surface and a second point group other than the first point group based on the point group data;
By converting the first point group into a group of planar elements and converting the second point group into a group of three-dimensional elements, the shape of the road surface is represented by the group of planar elements, and the group of three-dimensional elements is expressed. generating three-dimensional model data representing a shape other than the road surface;
including;
Each of the three-dimensional elements is a triangular prism three-dimensional element,
The three-dimensional model data generation method includes data defining six vertices and three faces of each of the three-dimensional elements . Each of the planar elements is a rectangular planar element,
5. The three-dimensional model data generation method according to claim 4, wherein the three-dimensional model data includes data defining four vertices and one face of each of the planar elements.

Images