KR102513683B1 - Point cloud matching method and point cloud matching device - Google Patents

Abstract

In the point cloud matching method and apparatus for matching a source point cloud and a target point cloud according to embodiments of the present invention, a source point cloud obtained at a first point in time, a target point cloud and locus data obtained at a second point after the first point in time is received, duplicate points and noise for each of the source point cloud and the target point cloud are removed, and the driving trajectory from which the source point cloud and the target point cloud are obtained based on the locus data is divided into a plurality of sections to obtain a plurality of source point clouds and target point clouds. is divided into a source point group for each section and a target point group for each section belonging to each section of , and corresponding point pairs of source points included in the source point group for each section and target points included in the target point group for each section are determined, and the distance between the corresponding point pairs is determined. The source point cloud for each section and the target point cloud for each section are matched so that the summation is minimized. Accordingly, the source point cloud and the target point cloud are accurately matched, and more accurate high-precision map data can be generated.

Description

Point cloud matching method and point cloud matching device {POINT CLOUD MATCHING METHOD AND POINT CLOUD MATCHING DEVICE}

The present invention relates to a method and apparatus for constructing map data, and more particularly, to a point cloud matching method and apparatus for constructing vehicle map data.

Recently, in order to provide services such as route guidance and automatic driving in vehicles, etc., more precise and high-precision map data is required. On the other hand, the method of building map data using aerial photographs is not suitable for building high-precision map data that requires centimeter (cm) level precision, and LIDAR (Light Detection And Ranging) is used to build high-precision map data. For example, high-precision map data can be constructed by collecting point clouds several times using LIDAR or the like and matching the point clouds collected several times.

However, in matching the point clouds, since the conventional matching method matches the point clouds for the entire driving trajectory, there is a problem in that the position error of the matched point clouds increases as the distance from the starting position of the driving trajectory increases.

One object of the present invention is to provide a point cloud matching method capable of more accurately matching a source point cloud and a target point cloud.

Another object of the present invention is to provide a point cloud matching device capable of more accurately matching a source point cloud and a target point cloud.

However, the object of the present invention is not limited to the above-mentioned objects, and may be expanded in various ways without departing from the spirit and scope of the present invention.

In order to achieve one object of the present invention, in a point cloud matching method for matching a source point cloud and a target point cloud according to embodiments of the present invention, the source point cloud obtained at a first point in time and a second point cloud obtained after the first point in time The target point cloud and trajectory data acquired at a viewpoint are received, overlapping points and noise for each of the source point cloud and the target point cloud are removed, and a driving trajectory from which the source point cloud and the target point cloud are obtained based on the trajectory data is divided into a plurality of sections so that the source point cloud and the target point cloud are divided into a source point cloud for each section and a target point cloud for each section belonging to each of the plurality of sections, and the source points included in the source point cloud for each section and the Corresponding point pairs of target points included in the target point cloud for each section are determined, and the source point cloud for each section and the target point cloud for each section are matched so that the sum of the distances of the corresponding point pairs is minimized.

In one embodiment, each of the source point cloud and the target point cloud is obtained through a LiDaR device of a Mobile Mapping System (MMS), and the trajectory data is obtained through a trajectory collection device of the mobile mapping system can be obtained

In an embodiment, downsampling may be performed on each of the source point cloud and the target point cloud to remove the overlapping point.

In one embodiment, to perform the downsampling, for each of the source point cloud and the target point cloud, a space in which each point cloud is located is divided into a plurality of voxels having a size corresponding to an error range of a LIDAR device. When two or more points of each point group exist in one voxel among the plurality of voxels, at least one of the two or more points may be removed so that one point among the two or more points remains.

In an embodiment, filtering may be performed on each of the source point cloud and the target point cloud to remove the noise.

In one embodiment, for a plurality of points of each of the source point group and the target point group, a predetermined number of neighboring points adjacent to each point are searched for to perform the filtering, and a distance from each point to the neighboring points is searched. An average distance of the distances is calculated, an average value and a standard deviation value for the average distances for the plurality of points are calculated, a reference distance is determined based on the average value and the standard deviation value, and the plurality of Among the points, a point having the average distance greater than the reference distance may be removed.

In one embodiment, for a plurality of points of each of the source point group and the target point group, neighboring points having a distance less than or equal to a predetermined distance from each point are searched for to perform the filtering, and a minimum distance from the neighboring points is searched. A nearby point plane having is determined, a distance of each point from the nearby point plane is calculated, and each point having a calculated distance equal to or greater than a reference distance may be removed.

In one embodiment, the driving trajectory from which the source point cloud and the target point cloud are obtained to divide the source point cloud and the target point cloud into the source point cloud for each section and the target point cloud for each section belonging to each of the plurality of sections, respectively A plurality of dividing points having predetermined intervals are set in, a plurality of vertical planes perpendicular to the traveling direction of the driving trajectory are determined at the plurality of dividing points, and the spaces between the plurality of vertical planes are It may be determined as a plurality of sections, the source point cloud may be divided into source point clouds for each section belonging to each of the plurality of sections, and the target point cloud may be divided into target point clouds for each section belonging to each of the plurality of sections. there is.

In one embodiment, for each source point among the source points, to determine pairs of corresponding points between the source points included in the source point group for each section and the target points included in the target point group for each section, the source point; Among the target points, if a target point having the minimum distance from the source point is determined as a neighbor point pair, and one of the target points belongs to two or more of the neighbor point pairs, among the two or more neighbor point pairs The neighboring point pairs left by removing at least one of them are set as the corresponding point pairs, and among the corresponding point pairs, a corresponding point pair having a distance greater than or equal to a predetermined correspondence pair reference distance may be removed as an outlier corresponding point pair.

In an embodiment, at least one of the source point cloud for each section and the target point cloud for each section is rotated and moved to match the source point cloud for each section and the target point cloud for each section, and the source point cloud for each section and the target point cloud for each section At least one of them may be moved in parallel.

In order to achieve another object of the present invention, a point cloud matching apparatus for matching a source point cloud and a target point cloud according to embodiments of the present invention receives the source point cloud obtained at a first point in time, and A point cloud preprocessing block for further receiving the target point cloud obtained at a second point in time and removing duplicate points and noise for each of the source point cloud and the target point cloud; receiving locus data obtained at the second point in time; Based on the data, the driving trajectory from which the source point cloud and the target point cloud are obtained is divided into a plurality of sections, and the source point cloud and the target point cloud are divided into a source point cloud for each section and a target point cloud for each section belonging to each of the plurality of sections. A section division block to be divided, and corresponding point pairs of source points included in the source point cloud for each section and target points included in the target point cloud for each section are determined, and the source points for each section are determined such that the sum of the distances of the corresponding point pairs is minimized. A point cloud matching block for each section matching the point cloud with the target point cloud for each section is included.

In one embodiment, each of the source point cloud and the target point cloud is obtained through a LiDaR device of a Mobile Mapping System (MMS), and the trajectory data is obtained through a trajectory collection device of the mobile mapping system can be obtained

In an embodiment, the point cloud preprocessing block may perform downsampling on each of the source point cloud and the target point cloud to remove the overlapping point.

In an embodiment, the point cloud preprocessing block divides a space where each point cloud is located into a plurality of voxels having a size corresponding to an error range of a LIDAR device, for each of the source point cloud and the target point cloud. In addition, when two or more points of each point group exist in one voxel of the plurality of voxels, at least one of the two or more points may be removed so that one point among the two or more points remains.

In an embodiment, the point cloud preprocessing block may perform filtering on each of the source point cloud and the target point cloud to remove the noise.

In an embodiment, the point cloud preprocessing block searches for a predetermined number of neighboring points adjacent to each of a plurality of points of the source point group and the target point group, and from each point to the neighboring points. Calculate an average distance of the distances, calculate an average value and a standard deviation value for the average distances for the plurality of points, determine a reference distance based on the average value and the standard deviation value, and determine the plurality of points. Among the points, points having the average distance greater than the reference distance may be removed.

In an embodiment, the point cloud preprocessing block searches for neighboring points having a distance less than or equal to a predetermined distance from each of the plurality of points of the source point group and the target point group, and a minimum distance from the neighboring points. Determines a nearby point plane having , calculates the distance of each point from the nearby point plane, and removes each point whose calculated distance is equal to or greater than the reference distance.

In one embodiment, the section division block sets a plurality of division points having predetermined intervals in the driving trajectory from which the source point cloud and the target point cloud are obtained, and at the plurality of division points, the driving trajectory A plurality of vertical planes perpendicular to the traveling direction are determined, spaces between the plurality of vertical planes are determined as the plurality of sections, and the source point cloud is a source point cloud for each section belonging to each of the plurality of sections. segmentation, and the target point cloud may be divided into target point clouds for each section belonging to each of the plurality of sections.

In one embodiment, the section-by-section point cloud matching block determines, for each source point among the source points, the source point and a target point having a minimum distance from the source point among the target points as a pair of neighboring points; When one of the target points belongs to two or more of the neighboring point pairs, at least one of the two or more neighboring point pairs is removed, and the remaining neighbor point pairs are set as the corresponding point pairs, and the corresponding point pairs Correspondence point pairs having a distance equal to or greater than a predetermined correspondence pair reference distance may be removed as outlier correspondence point pairs.

In an embodiment, the point cloud matching block for each section rotates at least one of the source point cloud for each section and the target point cloud for each section, and moves at least one of the source point cloud for each section and the target point cloud for each section in parallel. can

In the point cloud matching method and point cloud matching apparatus for matching a source point cloud and a target point cloud according to embodiments of the present invention, the source point cloud obtained at a first point in time and the target point obtained at a second point in time after the first point in time Point cloud and trajectory data are received, overlapping points and noise for each of the source point cloud and the target point cloud are removed, and driving trajectories from which the source point cloud and the target point cloud are obtained based on the trajectory data are divided into a plurality of sections By dividing, the source point cloud and the target point cloud are divided into a source point cloud for each section and a target point cloud for each section belonging to each of the plurality of sections, and the source points included in the source point cloud for each section and the target point cloud for each section are included. Corresponding point pairs of the corresponding target points are determined, and the source point cloud for each section and the target point group for each section may be matched so that the sum of the distances of the corresponding point pairs is minimized. Accordingly, since point cloud matching is performed section by section, position errors of matched point clouds are reduced, and high-precision map data can be generated. In addition, since high-precision map data is generated by matching a plurality of point clouds, road shaded sections can be restored and changes in road facilities can be detected and updated.

However, the effects of the present invention are not limited to the above-described effects, and may be variously extended within a range that does not deviate from the spirit and scope of the present invention.

1 is a flowchart illustrating a point cloud matching method for matching a source point cloud and a target point cloud according to an embodiment of the present invention.
2 is a diagram for explaining an example of downsampling performed on each of a source point cloud and a target point cloud.
3 is a diagram for explaining an example of Statistical Outlier Removal (SOR) filtering performed on each of a source point cloud and a target point cloud.
4 is a diagram for explaining an example of noise filtering performed on each of a source point cloud and a target point cloud.
5 is a diagram for explaining an example in which a source point cloud and a target point cloud are divided into a source point cloud for each section and a target point cloud for each section.
6 is a diagram for explaining an example in which pairs of neighboring points of a source point and a target point are determined.
7 is a diagram for explaining an example of corresponding point pairs of source points and target points.
8 is a diagram for explaining an example in which an outlier corresponding point pair among corresponding point pairs is removed.
9 is a diagram for explaining an example in which rotational movement and parallel movement are performed on at least one of a source point cloud and a target point cloud.
10 is a block diagram illustrating a point cloud matching device for matching a source point cloud and a target point cloud according to an embodiment of the present invention.

For the embodiments of the present invention disclosed in the text, specific structural or functional descriptions are only exemplified for the purpose of explaining the embodiments of the present invention, and the embodiments of the present invention may be implemented in various forms and the text It should not be construed as being limited to the embodiments described above.

Since the present invention can have various changes and various forms, specific embodiments will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to a specific form disclosed, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

Terms such as first and second may be used to describe various components, but the components should not be limited by the terms. The terms may be used for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first element may be termed a second element, and similarly, the second element may also be termed a first element.

It is understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, but other elements may exist in the middle. It should be. On the other hand, when a component is referred to as “directly connected” or “directly connected” to another component, it should be understood that no other component exists in the middle. Other expressions describing the relationship between components, such as "between" and "directly between" or "adjacent to" and "directly adjacent to", etc., should be interpreted similarly.

Terms used in this application are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as "comprise" or "having" are intended to indicate that the described feature, number, step, operation, component, part, or combination thereof exists, but that one or more other features or numbers are present. However, it should be understood that it does not preclude the presence or addition of steps, operations, components, parts, or combinations thereof.

In addition, terms such as “~ unit” described in the text refer to a unit that processes at least one function or operation, which may be implemented by hardware or software or a combination of hardware and software.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in this application, they are not interpreted in an ideal or excessively formal meaning. .

Hereinafter, embodiments of the present invention will be described in more detail with reference to the accompanying drawings. The same or similar reference numerals are used for like elements in the drawings.

1 is a flowchart illustrating a point cloud matching method for matching a source point cloud and a target point cloud according to an embodiment of the present invention, and FIG. 2 is a diagram for explaining an example of downsampling performed on each of the source point cloud and the target point cloud. 3 is a diagram for explaining an example of SOR (Statistical Outlier Removal) filtering performed on each of the source point cloud and the target point group, and FIG. 4 illustrates an example of noise filtering performed on each of the source point group and the target point group. 5 is a diagram for explaining an example in which a source point cloud and a target point cloud are divided into a source point cloud for each section and a target point cloud for each section, and FIG. 7 is a diagram for explaining an example of correspondence point pairs of source points and target points, and FIG. 8 is a diagram for explaining an example in which an outlier correspondence point pair among correspondence point pairs is removed. FIG. 9 is a diagram for explaining an example in which rotational movement and parallel movement are performed on at least one of a source point cloud and a target point cloud.

Referring to FIG. 1 , in the point cloud matching method for matching a source point cloud 110 and a target point cloud 120, a source point cloud 110 obtained at a first time point and a source point cloud 110 obtained at a second time point after the first time point The target point cloud 120 and trajectory data 130 may be received (S140). In one embodiment, each of the source point cloud 110 and the target point cloud 120 is obtained through a LiDaR device of a Mobile Mapping System (MMS), and the trajectory data 130 is the mobile mapping system It can be obtained through the trajectory collection device of For example, the lidar device may be a Velodyne LiDAR, an HDL-32 channel LiDAR, or the like, but is not limited thereto. Also, for example, the trajectory collection device may be a Global Positioning System (GPS) device, and/or a Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) device, but is not limited thereto. Meanwhile, the source point cloud 110 is a point cloud obtained when the vehicle equipped with the mobile mapping system travels along the driving trajectory at the first time point, and the target point cloud 120 and trajectory data 130 are the mobile mapping system The mounted vehicle may be point cloud and trajectory data obtained at the second point in time after the first point in time (eg, a point cloud several days, months, or years after the first point in time).

Redundant points and noise for each of the source point cloud 110 and the target point cloud 120 may be removed (S150). In an embodiment, downsampling is performed on each of the source point cloud 110 and the target point cloud 120 to remove the overlapping points, and filtering is performed on each of the source point cloud 110 and the target point cloud 120 to remove the noise. can be performed

For example, in order to perform the downsampling, as shown in 210 of FIG. 2, the space in which each point cloud is located is divided into a plurality of voxels, and as shown in 220 of FIG. Duplicate points 230 may be removed so that only one point remains. That is, the downsampling is performed for each of the source point cloud 110 and the target point cloud 120, and the space where each point cloud is located has a size corresponding to the error range of the LIDAR device as shown in 210 in FIG. It can be divided into the plurality of voxels having . In addition, as shown in 220 of FIG. 2 , when two or more points 230 of each point group exist in each voxel, the two or more points 230 are determined to be the overlapping point, and the two or more points 230 are determined as overlapping points. At least one of the two or more dots 230 may be removed so that one of the dots 230 remains. For example, among the two or more points 230 , only one point close to the center point of each voxel may remain, and the remaining points may be removed. Meanwhile, the LIDAR device has a certain error range, and two or more points 230 present in each voxel having a size corresponding to the error range may be the erroneously acquired overlapping points, and downsampling using such a voxel Through this, the overlapping point can be removed.

In one embodiment, as shown in FIG. 3 , statistical outlier removal (SOR) filtering may be performed to remove the noise of each of the source point cloud 110 and the target point cloud 120 . For example, as shown in FIG. 3, the SPR filtering is performed on each of the source point cloud 110 and the target point cloud 120, and a plurality of points 310, 320, 330, 340, 350, 360 and 370), a predetermined number of adjacent points 320, 330, 340, 350, 360, and 370 adjacent to each point 310 (six in the example of FIG. 3, but not limited thereto) are Searched, and the average distance of the distances (d0, d1, d2, d3, d4, d5) from each point 310 to the neighboring points 320, 330, 340, 350, 360, 370 may be calculated. . In this way, the average distance is determined as the average distance of the points 310, and this average distance can be determined for all point clouds of each point group. Also, an average value and a standard deviation value of the average distances of all the point groups may be calculated, and a reference distance may be determined based on the average value and the standard deviation value. In one embodiment, the reference distance may be determined as a sum of the average value and N times (N is a real number greater than or equal to 0) of the standard deviation value. For example, the reference distance may be determined as a sum of the average value and the standard deviation value, and in this case, about 68% of the average distances may be less than or equal to the reference distance. In another example, the reference distance may be determined as a sum of twice the average value and the standard deviation value, and in this case, about 95% of the average distances may be less than or equal to the reference distance. Also, among all point clouds of each point group, a point having the average distance greater than the reference distance may be removed. Accordingly, with respect to each of the source point cloud 110 and the target point cloud 120, points far from other points may be removed as erroneously acquired noise.

In another embodiment, as shown in FIG. 4 , noise filtering may be performed to remove the noise of each of the source point cloud 110 and the target point cloud 120 . For example, as shown in FIG. 4, the noise filtering is performed on each of the source point cloud 110 and the target point cloud 120, and a plurality of points 410, 420, 430, 440, 450, 460), the neighboring points 420, 430, 440, 450, and 460 having a distance less than or equal to a predetermined distance (in the example of FIG. 4, but not limited to about 20 cm) from each point 410 are A nearby point plane 470 having a minimum distance from the neighboring points 420 , 430 , 440 , 450 , and 460 may be determined by searching and performing plane fitting. Also, the distance of each point 410 from the nearby point plane 470 is calculated, and each point 410 whose calculated distance is greater than or equal to a reference distance (predetermined according to embodiments) may be removed. Meanwhile, the neighboring point search, plane fitting, and point removal beyond the reference distance may be performed on all point clouds of each point group. Accordingly, with respect to each of the source point cloud 110 and the target point cloud 120, points far from the nearby point plane 470 can be removed as erroneously acquired noise.

For each of the source point cloud 110 and the target point cloud 120, according to embodiments, the SOR filtering shown in FIG. 3 is performed, the noise filtering shown in FIG. 4 is performed, or the noise filtering shown in FIG. 3 is performed. Both the SOR filtering and the noise filtering shown in FIG. 4 may be performed.

Based on the locus data 130, the driving trajectory obtained by the source point cloud 110 and the target point cloud 120 is divided into a plurality of sections, and the source point cloud 110 and the target point cloud 120 are each of the plurality of sections. It can be divided into a source point cloud for each section and a target point cloud for each section belonging to (S160).

For example, as shown in FIG. 5, the driving trajectory in which the source point cloud 110 is obtained at the first time point and the target point cloud 120 obtained at the second time point after the first time point is obtained; That is, a plurality of dividing points (511, 512, 513, 514, 515, 516, 517, 518, 519) are set, and a plurality of vertical planes perpendicular to the traveling direction of the driving trajectory at a plurality of dividing points 511, 512, 513, 514, 515, 516, 517, 518, 519) (521, 522, 523, 514, 525, 526, 527, 528, 529) can be determined. Meanwhile, in FIG. 5 , each vertical plane is represented by a line for convenience of description, but each vertical plane may be a plane. Spaces between the plurality of vertical planes 521, 522, 523, 514, 525, 526, 527, 528, and 529 are determined as the plurality of sections, and the source point cloud 110 is applied to each of the plurality of sections. The target point cloud 120 may be divided into the target point clouds for each section belonging to each of the plurality of sections. In this way, the source point cloud 110 and the target point cloud 120 may be divided into the source point cloud for each section and the target point cloud for each section along the driving trajectory indicated by the target point cloud 120 .

Corresponding point pairs of source points included in the source point cloud for each section and target points included in the target point cloud for each section may be determined for the source point cloud for each section and the target point cloud for each section that correspond to each other belonging to each section ( S170).

For example, as shown in FIG. 6 , for each source point 610 among the source points of the source point group for each section, the source point 610 and the target points 621 and 622 of the target point group for each section , 623, 624, and 625, the target point 621 having the minimum distance from the source point 610 may be determined as a pair of neighboring points. Meanwhile, since the neighbor point pair is determined for each source point 610, a single target point 621 may be included in two or more neighbor point pairs. A neighbor point pair having a target point not included in other neighbor point pairs may be set as the corresponding point pair. Meanwhile, when one of the target points belongs to two or more pairs of neighbor points among the pairs of neighbor points, the pairs of neighbor points left by removing at least one of the two or more pairs of neighbor points may be set as the pairs of corresponding points. . Accordingly, as shown in FIG. 7, the corresponding point pairs include non-overlapping source points 711, 712, 713, 714, and 715 and non-overlapping target points 721, 722, 723, 724, and 725. can do.

In an embodiment, an outlier correspondence point pair among the correspondence point pairs may be removed. For example, as shown in FIG. 8 , a corresponding pair reference distance is predetermined according to embodiments, and the distances between source points 813, 814 and 815 and target points 823, 824 and 825 are Correspondence point pairs that are less than the correspondence pair reference distance are maintained as inlier correspondence point pairs, and correspondence point pairs whose distances between the source points 811 and 812 and the target points 821 and 822 are equal to or greater than the correspondence pair reference distance are outliers ( outliers) can be eliminated as pairs of correspondence points.

The source point cloud for each section and the target point cloud for each section may be matched so that the sum of the distances of the corresponding point pairs is minimized (S180). In an embodiment, rotational movement and/or parallel movement may be performed on at least one of the source point group for each section and the target point group for each section so that the sum of the distances of the corresponding point pairs is minimized.

In the example of FIG. 9 , 910 may indicate the source point cloud for each section, and 930 may indicate the target point cloud for each section. For example, as shown in FIG. 9 , rotation is performed on the target point cloud 930 for each section, so that the rotated target point cloud 950 for each section is derived, and then the rotated target point cloud for each section ( 950), the target point cloud 930 for each section may come closer to the source point cloud 910 for each section. Accordingly, the sum of the distances of pairs of corresponding points of the matched source point cloud 910 for each section and the target point cloud 930 for each section can be minimized. Meanwhile, FIG. 9 shows an example in which the target point cloud 930 for each section is rotated and moved in parallel so as to approach the source point cloud 910 for each section. However, in another embodiment, the source point cloud 910 for each section is It can be rotated and translated to approximate the point cloud 930 .

In this way, the matched point clouds for each section are merged, the source point cloud 110 and the target point cloud 120 are matched with respect to the entire driving route, and high-precision map data can be generated.

As described above, in the point cloud matching method for matching the source point cloud 110 and the target point cloud 120 according to embodiments of the present invention, overlapping points for each of the source point cloud 110 and the target point cloud 120 And since the noise is removed, the accuracy of each of the source point cloud 110 and the target point cloud 120 can be improved. In addition, in the point cloud matching method according to embodiments of the present invention, since point cloud matching is performed section by section, position errors of matched point clouds are reduced, and more accurate high-precision map data can be generated. In addition, in the point cloud matching method according to embodiments of the present invention, rather than a point cloud obtained by a single driving, a plurality of points acquired by a plurality of driving (of a vehicle equipped with a mobile mapping system) at a plurality of viewpoints. Since the high-precision map data is generated by matching the point clouds of (i.e., the source point cloud 110 and the target point cloud 120), a road shadow section not included in one point cloud can be restored by another point cloud, and the Changes in road facilities (eg, lanes, crosswalks, signs, etc.) between a plurality of points of view may be detected, and the changed road facilities may be updated to the latest road facilities.

10 is a block diagram illustrating a point cloud matching device for matching a source point cloud and a target point cloud according to an embodiment of the present invention.

Referring to FIG. 10 , a point cloud matching apparatus 1000 for matching a source point cloud and a target point cloud according to an embodiment of the present invention includes a point cloud preprocessing block 1010, a section division block 1020, and a section-by-section point cloud matching block 1030. can include

The point cloud preprocessing block 1010 receives the source point cloud obtained at a first point in time through the lidar device 1060 of the mobile mapping system 1050, and through the lidar device 1060 of the mobile mapping system 1050. The target point cloud obtained at a second time point after the first time point may be further received. For example, the lidar device 1060 may be a Velodyne LiDAR, an HDL-32 channel LiDAR, or the like, but is not limited thereto.

The point cloud preprocessing block 1010 may remove overlapping points and noise for each of the source point cloud and the target point cloud. In an embodiment, the point cloud preprocessing block 1010 may perform downsampling on each of the source point cloud and the target point cloud to remove the overlapping points. For example, in the point cloud preprocessing block 1010, for each of the source point cloud and the target point cloud, a plurality of voxels having a size corresponding to the error range of the LIDAR device 1060 are arranged in a space where each point cloud is located. ), and if two or more points of each point group exist in one voxel of the plurality of voxels, at least one of the two or more points is removed so that one point among the two or more points remains. can Also, in an embodiment, the point cloud preprocessing block 1010 may perform filtering on each of the source point cloud and the target point cloud to remove the noise. In one example, the point cloud preprocessing block 1010 searches for a predetermined number of neighboring points adjacent to each of a plurality of points of the source point group and the target point group, and extends from each point to the neighboring points. Calculate an average distance of the distances of the plurality of points, calculate an average value and standard deviation values for the average distances for the plurality of points, determine a reference distance based on the average value and the standard deviation value, and determine the plurality of points. Among the points in , points having the average distance greater than the reference distance may be removed. Such filtering may be referred to as Statistical Outlier Removal (SOR) filtering. In another example, the point cloud preprocessing block 1010 searches for neighboring points having a distance equal to or less than a predetermined distance from each of the plurality of points of the source point cloud and the target point cloud, and searches for a minimum distance from the neighboring points. A neighboring point plane having a distance may be determined, a distance of each point from the neighboring point plane may be calculated, and each point having a calculated distance greater than or equal to a reference distance may be removed. Such filtering may be referred to as noise filtering. According to embodiments, the SOR filtering may be performed, the noise filtering may be performed, or both the SOR filtering and the noise filtering may be performed on each of the source point group and the target point group.

The section division block 1020 may receive trajectory data acquired at the second point in time through the trajectory collecting device 1070 of the mobile mapping system 1050 . For example, the trajectory collection device 1070 may be a Global Positioning System (GPS) device, and/or a Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) device, but is not limited thereto.

The section division block 1020 divides the driving trajectory from which the source point cloud and the target point cloud are acquired into a plurality of sections based on the locus data, and divides the source point cloud and the target point cloud into sections belonging to each of the plurality of sections. It can be divided into a source point cloud for each segment and a target point cloud for each section. For example, the section division block 1020 sets a plurality of division points having predetermined intervals in the driving trajectory from which the source point cloud and the target point cloud are obtained, and sets the plurality of division points of the driving trajectory at the plurality of division points. A plurality of vertical planes perpendicular to the traveling direction are determined, spaces between the plurality of vertical planes are determined as the plurality of sections, and the source point cloud is a source point cloud for each section belonging to each of the plurality of sections. segmentation, and the target point cloud may be divided into target point clouds for each section belonging to each of the plurality of sections.

The section-by-section point cloud matching block 1030 may determine corresponding point pairs of source points included in the section-by-section source point cloud and target points included in the section-by-section target point cloud. For example, the section-by-section point cloud matching block 1030 determines, for each source point among the source points, the source point and a target point having a minimum distance from the source point among the target points as a pair of neighboring points; If one of the target points belongs to two or more of the neighboring point pairs, at least one of the two or more neighboring point pairs is removed, and the left neighbor point pairs are set as the corresponding point pairs, and the corresponding point pairs Correspondence point pairs having a distance equal to or greater than a predetermined correspondence pair reference distance may be removed as outlier correspondence point pairs.

Also, the section-by-section point cloud matching block 1030 may match the section-by-section source point cloud and the section-by-section target point cloud so that the sum of the distances of the corresponding point pairs is minimized. For example, the section-by-section point cloud matching block 1030 rotationally moves at least one of the source point cloud for each section and the target point cloud for each section, and parallelly moves at least one of the source point cloud for each section and the target point cloud for each section. , the source point cloud for each section and the target point cloud for each section can be matched. In this way, the matched point clouds for each section are merged, the source point cloud and the target point cloud are matched with respect to the entire driving route, and high-precision map data can be generated.

As described above, in the point cloud matching apparatus 1000 for matching the source point cloud and the target point cloud according to embodiments of the present invention, since duplicate points and noise for each of the source point cloud and the target point cloud are removed, the Accuracy of each of the source point cloud and the target point cloud may be improved. In addition, in the point cloud matching apparatus 1000 according to embodiments of the present invention, since point cloud matching is performed section by section, position errors of matched point clouds are reduced, and more accurate high-precision map data can be generated. In addition, in the point cloud matching apparatus 1000 according to embodiments of the present invention, a plurality of points (of a vehicle equipped with the mobile mapping system 1050) at a plurality of viewpoints, rather than a point cloud obtained by a single driving Since the high-precision map data is generated by matching a plurality of point clouds (i.e., the source point cloud and the target point cloud) acquired as Changes in road facilities (eg, lanes, crosswalks, signs, etc.) between the plurality of points of view may be detected, and the changed road facilities may be updated to the latest road facilities.

Although the point cloud matching method and the point cloud matching device according to the embodiments of the present invention have been described above with reference to the drawings, the above description is exemplary and requires common knowledge in the related art to the extent that it does not deviate from the technical spirit of the present invention. It can be modified and changed by those who have it.

The present invention can be applied to a method and apparatus for constructing high-precision map data. Although the above has been described with reference to the embodiments of the present invention, those skilled in the art can variously modify and change the present invention without departing from the spirit and scope of the present invention described in the claims below. You will understand that you can.

1000: point cloud matching device
1010: point cloud preprocessing block
1020: Section division block
1030: Sectional point cloud matching block
1050: mobile mapping system
1060: lidar device
1070: trajectory collection device

Claims (20)

In the point cloud matching method for matching a source point cloud and a target point cloud,
receiving the source point cloud obtained at a first point in time and the target point cloud and trajectory data obtained at a second point after the first point in time;
removing overlapping points and noise from each of the source point cloud and the target point cloud;
Based on the locus data, the driving trajectory from which the source point cloud and the target point cloud are obtained is divided into a plurality of sections, and the source point cloud and the target point cloud are obtained by section-by-section source point cloud and section-by-section target Dividing each into point groups;
determining corresponding point pairs between source points included in the source point cloud for each section and target points included in the target point cloud for each section; and
matching the source point cloud for each section with the target point cloud for each section so that the sum of the distances of the pairs of corresponding points is minimized;
The step of dividing the source point cloud and the target point cloud into the source point cloud for each section and the target point cloud for each section belonging to each of the plurality of sections, respectively.
setting a plurality of dividing points having predetermined intervals in the driving trajectory from which the source point cloud and the target point cloud are obtained;
determining a plurality of vertical planes perpendicular to a traveling direction of the driving trajectory at the plurality of dividing points;
determining spaces between the plurality of vertical planes as the plurality of sections;
dividing the source point cloud into source point clouds for each section belonging to each of the plurality of sections; and
and dividing the target point cloud into target point clouds for each section belonging to each of the plurality of sections. The method of claim 1, wherein each of the source point cloud and the target point cloud is obtained through a LiDaR device of a Mobile Mapping System (MMS),
The point cloud matching method, characterized in that the trajectory data is obtained through a trajectory collecting device of the mobile mapping system. The method of claim 1 , wherein the removing of the overlapping point and the noise for each of the source point group and the target point group comprises:
and performing downsampling on each of the source point cloud and the target point cloud to remove the overlapping point. The method of claim 3, wherein the downsampling step comprises:
For each of the source point cloud and the target point cloud, dividing a space where each point cloud is located into a plurality of voxels having a size corresponding to an error range of a LIDAR device; and
and removing at least one of the two or more points so that one of the two or more points remains when two or more points of each point group exist in one voxel of the plurality of voxels. The point group matching method with . The method of claim 1 , wherein the removing of the overlapping point and the noise for each of the source point group and the target point group comprises:
and performing filtering on each of the source point cloud and the target point cloud to remove the noise. The method of claim 5, wherein the filtering step comprises:
searching for a predetermined number of nearby points adjacent to each point with respect to a plurality of points of each of the source point group and the target point group;
calculating an average distance of distances from each point to the neighboring points;
calculating an average value and a standard deviation value for the average distances for the plurality of points;
determining a reference distance based on the average value and the standard deviation value; and
and removing a point having the average distance greater than the reference distance from among the plurality of points. The method of claim 5, wherein the filtering step comprises:
searching for neighboring points having a distance less than or equal to a predetermined distance from each of the plurality of points of the source point group and the target point group;
determining a nearby point plane having a minimum distance from the neighboring points;
calculating the distance of each point from the nearby point plane; and
and removing each point having the calculated distance equal to or greater than the reference distance. delete The method of claim 1 , wherein the step of determining corresponding point pairs of the source points included in the source point cloud for each section and the target points included in the target point group for each section comprises:
determining, for each source point among the source points, the source point and a target point having a minimum distance from the source point among the target points as a pair of neighboring points;
if one of the target points belongs to two or more of the neighboring point pairs, setting the neighboring point pairs left by removing at least one of the two or more neighboring point pairs as the corresponding point pairs; and
and removing corresponding-point pairs having a distance greater than or equal to a predetermined correspondence-pair reference distance among the corresponding-point pairs as outlier corresponding-point pairs. The method of claim 1 , wherein the matching of the source point cloud for each section and the target point cloud for each section comprises:
rotating and moving at least one of the source point cloud for each section and the target point cloud for each section; and
and parallel-moving at least one of the source point cloud for each section and the target point cloud for each section. A point cloud matching device for matching a source point cloud and a target point cloud,
Receiving the source point cloud obtained at a first time point, further receiving the target point cloud obtained at a second time point after the first time point, and removing overlapping points and noise for each of the source point cloud and the target point cloud point cloud preprocessing block;
Trajectory data obtained at the second point in time is received, and based on the locus data, a driving trajectory from which the source point cloud and the target point cloud are obtained is divided into a plurality of sections to obtain the source point cloud and the target point cloud. a section dividing block that divides each section into a source point cloud and a target point group for each section belonging to each section; and
Corresponding point pairs of the source points included in the source point cloud for each section and the target points included in the target point cloud for each section are determined, and the source point group for each section and the target point cloud for each section are determined so that the sum of distances between the corresponding point pairs is minimized. Including a point cloud matching block for each section to be matched,
The section division block,
Setting a plurality of dividing points having predetermined intervals in the driving trajectory from which the source point cloud and the target point cloud are obtained;
Determining a plurality of vertical planes perpendicular to the traveling direction of the driving trajectory at the plurality of dividing points;
determining spaces between the plurality of vertical planes as the plurality of sections;
Dividing the source point cloud into source point clouds for each section belonging to each of the plurality of sections;
and dividing the target point cloud into target point clouds for each section belonging to each of the plurality of sections. The method of claim 11, wherein each of the source point cloud and the target point cloud is obtained through a LiDaR device of a Mobile Mapping System (MMS),
The point cloud matching device, characterized in that the trajectory data is obtained through a trajectory collecting device of the mobile mapping system. The method of claim 11, wherein the point cloud preprocessing block,
wherein downsampling is performed on each of the source point cloud and the target point cloud to remove the overlapping point. The method of claim 13, wherein the point cloud preprocessing block,
For each of the source point cloud and the target point cloud, the space where each point cloud is located is divided into a plurality of voxels having sizes corresponding to the error range of the LIDAR device,
When two or more points of each point group exist in one voxel among the plurality of voxels, at least one of the two or more points is removed so that one point among the two or more points remains. Device. The method of claim 11, wherein the point cloud preprocessing block,
and performing filtering on each of the source point cloud and the target point cloud to remove the noise. The method of claim 15, wherein the point cloud preprocessing block,
For each of the plurality of points of the source point group and the target point group, a predetermined number of neighboring points adjacent to each point are searched;
Calculate the average distance of the distances from each point to the neighboring points,
Calculate an average value and a standard deviation value for the average distances for the plurality of points;
determine a reference distance based on the mean value and the standard deviation value;
The point cloud matching device characterized in that removing a point having the average distance greater than the reference distance from among the plurality of points. The method of claim 15, wherein the point cloud preprocessing block,
For each of the plurality of points of the source point group and the target point group, searching for neighboring points having a distance less than or equal to a predetermined distance from each point;
determining a nearby point plane having a minimum distance from the neighboring points;
Calculate the distance of each point from the nearby point plane;
The point cloud matching device characterized in that for removing each point having the calculated distance equal to or greater than the reference distance. delete The method of claim 11, wherein the point cloud matching block for each section comprises:
For each source point among the source points, determine the source point and a target point having a minimum distance from the source point among the target points as a pair of neighboring points;
If one of the target points belongs to two or more pairs of neighboring points among the pairs of neighboring points, at least one of the two or more pairs of neighboring points is removed, and the remaining pairs of neighboring points are set as the pairs of corresponding points;
The point cloud matching device characterized in that, among the corresponding point pairs, a corresponding point pair having a distance greater than or equal to a predetermined correspondence pair reference distance is removed as an outlier corresponding point pair. The method of claim 11, wherein the point cloud matching block for each section comprises:
rotating and moving at least one of the source point cloud for each section and the target point cloud for each section;
The point cloud matching device characterized in that at least one of the source point cloud for each section and the target point cloud for each section is moved in parallel.

Images