KR101822185B1 - Method and apparatus for poi detection in 3d point clouds - Google Patents

Abstract

A 3D point cloud POI detection method and apparatus are provided. The 3D point cloud POI detection apparatus extracts the POI (point of interest) which is the key point in the 3D point cloud by using the shape point description vector describing the shape of the surface including the pixel point of the 3D point cloud and the neighboring point adjacent to the pixel point can do.

Description

METHOD AND APPARATUS FOR POI DETECTION IN 3D POINT CLOUDS [0002]

The following description relates to computer graphics or computer vision technology.

Recently, with the use of depth cameras or 3D scanners, interest in collection and processing techniques for 3D point cloud data is increasing. In the field of computer vision and smart robotics, point of interest (POI) detection algorithms in 3D point clouds are used in techniques such as surface registration or object recognition.

The POI detection algorithm has three important features. (1) Sparseness: The number of POIs must be small to efficiently perform the description matching algorithm; (2) Distinctiveness: POIs should be able to distinguish the information of each surface structure; (3) Repeatability: POIs should be able to be repeatedly detected in different data conversion conditions.

A 3D point cloud POI detecting apparatus according to an exemplary embodiment includes a 3D point cloud data obtaining unit for obtaining 3D point cloud data; A shape description unit for creating a shape description vector describing a shape of a surface including a pixel point of the 3D point cloud and a neighboring point adjacent to the pixel point; And a POI extracting unit for extracting a POI based on the shape description vector.

In the 3D point cloud POI detection apparatus according to an embodiment, the shape description unit may include: a local reference frame generation unit that generates a local reference frame for a pixel point of the 3D point cloud; A fiducial distance dispersion calculator for calculating a fiducial distance dispersion from a neighboring point adjacent to the pixel point to a tangent plane to which the pixel point belongs; And a shape description vector generation unit for generating the shape description vector by expressing the dihedral distance variance as a vector of a finite dimension.

In the 3D point cloud POI detection apparatus according to an embodiment, the POI extraction unit may include: a profit value calculation unit for calculating a profit value based on the generated shape description vector; An edge point identification unit for determining the number of zero components in the shape description vector and determining whether the pixel point belongs to an edge point or an edge; And a POI identification unit for identifying the POI based on the calculated profit value.

A 3D point cloud POI detection method according to an embodiment includes: obtaining 3D point cloud data; Generating a shape description vector describing a shape of a surface including a pixel point of a 3D point cloud and a neighboring point adjacent to the pixel point; And extracting the POI based on the generated shape description vector.

In a 3D point cloud POI detection method according to an embodiment, generating the shape description vector may include generating a local reference frame for a pixel point of the 3D point cloud; Calculating a drift distance dispersion from a neighboring point adjacent to the pixel point to a tangent plane to which the pixel point belongs; And generating the shape description vector by expressing the forward distance variance as a vector of finite dimensions.

In the 3D point cloud POI detection method according to an embodiment, extracting the POI may include calculating a profit value based on the generated shape description vector; Determining a number of zero components in the shape description vector, and determining whether a pixel point belongs to an edge point or an edge; And identifying the POI based on the calculated profit value.

1 is a diagram showing a configuration of a 3D point cloud POI (point of interest) detecting apparatus according to an embodiment.
2 is a view showing a detailed configuration of a shape description unit according to an embodiment.
3 is a diagram illustrating an example of a local reference frame and neighboring points of an internal point according to one embodiment.
4 is a diagram illustrating an example of a local reference frame and neighbor points of an edge point according to one embodiment.
5 is a diagram illustrating an example of a directed distance scatter diagram of an internal point according to an embodiment.
6 is a diagram showing an example of a directional dispersion of edge points of an edge point according to an embodiment.
7 is a view showing an example of a shape description diagram of an internal point according to an embodiment.
8 is a view showing an example of a shape description diagram of an edge point according to an embodiment.
9 is a diagram illustrating a detailed configuration of a POI extracting unit according to an embodiment.
10 is a diagram showing an example of the detection result of the edge point performed by the edge point identifying unit according to the embodiment.
11 is a flowchart illustrating a 3D point cloud POI detection method according to an embodiment.
12 is a flowchart illustrating a process of generating a shape description vector according to an embodiment.
13 is a flowchart illustrating a process of extracting a POI according to an embodiment.
FIG. 14 is a diagram illustrating an example of a result of detecting a POI in a 3D point cloud according to an embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. The specific structural or functional descriptions below are merely illustrative for purposes of illustrating embodiments of the invention and are not to be construed as limiting the scope of the invention to the embodiments described in the text. Like reference symbols in the drawings denote like elements.

1 is a diagram illustrating a configuration of a 3D point cloud POI detecting apparatus 100 according to an embodiment.

The 3D point cloud POI detecting apparatus 100 can detect a point of interest (POI) as a key point in the 3D point cloud. The 3D point cloud may include a plurality of pixel points related to the shape of the object. A pixel point can be distinguished by an edge point located in the edge region of the shape and an internal point located in an area other than the edge region.

Referring to FIG. 1, the 3D point cloud POI detecting apparatus 100 may include a 3D point cloud data obtaining unit 110, a shape description unit 120, and a POI extracting unit 130.

The 3D point cloud data acquisition unit 110 may acquire 3D point cloud data. For example, the 3D point cloud data acquisition unit 110 may acquire various types of 3D points such as a point cloud obtained by a 3D scanner, a depth image acquired by a depth camera, a 3D grid model generated by software, Cloud data can be acquired. The 3D point cloud data may include position information of the object surface.

According to another embodiment, the 3D point cloud data acquiring unit 110 may include a 3D point cloud preprocessing unit (not shown). The 3D point cloud preprocessor can perform a preprocessing process on the obtained 3D point Clyde data. The 3D point cloud preprocessor can perform processes such as noise elimination, outlier point elimination, and down sampling processing on the 3D point cloud data. For example, the 3D point cloud preprocessor can identify noise points and sparseness points by analyzing the dispersion state of neighboring points. The 3D point cloud preprocessor can filter the noise using a Gaussian filter or the like. The 3D point cloud preprocessor can remove the outlier point using a threshold or an existing point cloud splitting method. The 3D point cloud preprocessing unit can remove the outlier points based on the statistical analysis of the adjacent areas of each point. The 3D point cloud preprocessor can fill small holes based on the Moving Least Square technique. The 3D point cloud preprocessor may perform downsampling on the 3D point cloud data using a 3D filter algorithm to improve the computation speed. If downsampling has been performed, subsequent processing may be performed based on downsampled 3D point cloud data. Here, any method can be used to perform the desired preprocessing process. In addition, the processor executed in the preprocessing process is not limited to the examples described herein, and various processes can be performed as the preprocessing of the 3D point cloud data.

The shape description unit 120 may generate a shape description vector describing a shape of a surface including a pixel point of the 3D point cloud and a neighborhood point adjacent to the pixel point. The shape description unit 120 may generate the shape description vector using the variance information about the directed distance from the neighboring point of the pixel point to the tangent plane including the pixel point.

The shape description vector may reflect a local change in the surface including the pixel point and the neighboring point. The shape description vector may be constructed by statistical analysis of the drift distance to the tangent plane of the neighboring points. The shape description vector reflects the protrusion change of the local surface, and the larger the change of the surface, the larger the profit value and the more likely it is the POI. It can be easily determined whether or not the pixel point belongs to an edge point through the shape description vector. In addition, the shape description vector can be set as a description operator of the POI and applied to the feature point matching algorithm.

The POI extracting unit 130 can extract the POI based on the shape description vector. The POI extracting unit 130 may calculate the profit value based on the shape description vector, and may identify the POI in the 3D point cloud data based on the calculated profit value.

Through the above process, the 3D point cloud POI detecting apparatus 100 quickly detects the POI from the 3D point cloud data, describes the characteristic of the POI, and accurately identifies the edge point. Also, the 3D point cloud POI detecting apparatus 100 can detect the POI without calculating the differential of higher order (HOD) information such as the curvature by statistically analyzing the position information of the 3D point. Thus, the 3D point cloud POI detection apparatus 100 provides high stability when processing 3D point cloud data in which noise, data missing holes, or point density variation exists, Thereby improving the efficiency of calculation.

2 is a view showing a detailed configuration of the shape description unit 120 according to an embodiment.

Referring to FIG. 2, the shape description unit 120 may include a local reference frame generation unit 210, a forwarding distance variance calculation unit 220, and a shape description vector generation unit 230.

The local reference frame generation unit 210 may generate a local reference frame for the pixel point of the 3D point cloud. The local reference frame represents the coordinate system X-axis, Y-axis, and Z-axis at the center of the pixel point.

For example, the local reference frame generation unit 210 may determine the z-axis direction (normal direction) of the tangent plane using pixel points in a small area adjacent to the center point located at the center of the target area. The local reference frame generation unit 210 can determine the X-axis direction using the neighbor point in the adjacent region of the pixel point. The local reference frame generation unit 210 calculates the yaw distance from the plurality of neighbor points to the tangent plane, projects the neighbor point corresponding to the largest yaw distance to the tangent plane, It can be determined in the X-axis direction. The local reference frame generation unit 210 can determine the Y-axis direction based on the determined Z-axis and X-axis.

3 is a diagram illustrating an example of a local reference frame and neighboring points of an internal point according to one embodiment. 3, an X-axis 310, a Y-axis 320, and a Z-axis 330 are represented by a center point 350 in a local reference frame. Points indicated in deep color represent neighboring points 340. [ The criteria (e.g., radius) for determining the range of neighboring points 340 may be specified by the user. For example, the range of radii for determining the range of neighboring point 340 may be set to 4mr-10mr. Points included in an area between radius 4 and 10 mr with respect to the center point may be determined as neighbor point 340. Where mr is the Mesh resolution of the 3D point cloud and represents the average distance between points adjacent to each other in the 3D point cloud. 4 is a diagram illustrating an example of a local reference frame and neighbor points of an edge point according to one embodiment.

Referring back to FIG. 2, the drift dispersion calculator 220 may calculate the drift distance dispersion from the neighboring point adjacent to the pixel point to the tangent plane to which the pixel point belongs. The drift dispersion may include information about at least one of an angle between the projection direction from the neighboring point to the tangent plane and the direction of the X axis and a drift distance to the tangent plane. For example, each neighbor point may correspond to a mapping relationship as: < EMI ID = 1.0 >

는 픽셀 포인트에서 접평면으로의 투영 방향과 X축의 방향 간의 각도를 나타내고(범위: -

~ +

), d는 픽셀 포인트에서 접평면까지의 유향 거리를 나타낸다. 유향 거리는 양(plus)의 값 또는 음(minus)의 값을 가질 수 있다.here,

Represents the angle between the projection direction from the pixel point to the tangent plane and the direction of the X axis (range:

~ +

), and d represents the drift distance from the pixel point to the tangent plane. The fiducial distance may have a value of plus or a value of minus.

5 is a diagram illustrating an example of a distance-angle map of an internal point according to an embodiment. The abscissa represents the angle (unit: radian) between the projection direction to the tangent plane and the direction of the X axis, and the ordinate represents the yang distance (unit: mr) from the neighboring point to the tangent plane. 6 is a diagram showing an example of a directional dispersion of edge points of an edge point according to an embodiment.

2, the shape description vector generation unit 230 may generate the shape description vector based on the direction distance variance. The shape description vector generation unit 230 may generate the shape description vector by expressing the direction distance variance as a vector of a finite dimension. The shape description vector generation unit 230 may display the entire change in the vector of the finite dimension.

The shape description vector generation unit 230 may divide the direction distance variance into N arbitrary natural number intervals or bins and calculate the number of points in each interval and the sum of the directions distance in each interval. For example, as in the shape description diagram 700 shown in FIG. 7, the shape description vector generation unit 230 divides the directional dispersion degree into 18 sections (N = 18), and calculates the number of points (710) and the sum of the yaw distances (720) in each section. The shape description vector generation unit 230 can determine the shape description vector based on the number of points 710 of each section and the sum 720 of the directions distance 720 in each section. Fig. 5 shows an example of the shape description diagram of the internal points, and Fig. 8 shows an example of the shape description diagram of the edge points. In the shape description diagram 800 shown in FIG. 8, the number of points 810 in each section and the sum 820 of the directions distances in each section are shown.

로 정의되고, 각 구간 내의 유향 거리의 합(720)들은 벡터

으로 정의될 수 있다. 여기서, n_i는 제i 번째 구간의 포인트 개수를 나타내고, d_i는 제i 번째 구간의 유향 거리의 합을 나타낸다. 형상 기술 벡터는 N 차원의 벡터

로 정의될 수 있다. 형상 기술 벡터의 각 컴포넌트는 각 구간에 대응하는 평균 유향 거리를 나타낸다. 여기서,

이면, 대응하는 벡터 컴포넌트는 0이고, 형상 기술 벡터를 정규화한 최종의 형상 기술 벡터는

로 나타낼 수 있다.
Since the local reference frame is clear, the shape description vector generating unit 230 can expand the statistical information (the number of points of each section and the sum of the directions distances in each section) shown in the directional dispersion diagram to the shape description vector. For example, the number of points 710 in each section may be a vector

, And the sum of the drift distances 720 in each section is defined as a vector

. ≪ / RTI > Here, n _i represents the number of points of the i-th section, and d _i represents the sum of the diversion distances of the i-th section. The shape description vector is an N-dimensional vector

. ≪ / RTI > Each component of the shape description vector represents an average directional distance corresponding to each section. here,

, The corresponding vector component is 0, and the final shape description vector obtained by normalizing the shape description vector is

.

FIG. 9 is a diagram showing a detailed configuration of the POI extracting unit 130 according to an embodiment.

9, the POI extracting unit 130 may include an interest value calculating unit 910, an edge point identifying unit 920, and a POI identifying unit 930. The larger the profit value, the more likely it is to be a POI. The POI can be located in an area where the variation of the local surface is large and the shape can be clearly identified.

)에 기초하여 이익 값을 계산할 수 있다. 예를 들어, 이익 값 계산부(910)는 아래의 수학식 2 내지 수학식 4에 기초한 3가지 방법을 이용하여 이익 값을 계산할 수 있다.The profit value calculation unit 910 calculates the profit value of the shape description vector (for example,

). ≪ / RTI > For example, the profit value calculation unit 910 may calculate the profit value using the three methods based on the following Equations (2) to (4).

는 형상 기술 벡터의 컴포넌트

의 평균값이고,

는 형상 기술 벡터의 컴포넌트

의 분산값이다.In the method (1) using Equation (2), the profit value can be defined as the product of the average value of the shape description vector D _i and the variance. here,

Is a component of the shape description vector

≪ / RTI >

Is a component of the shape description vector

.

쌍이 가지는 가장 큰 변화율로 정의될 수 있다. 여기서,

,

는 형상 기술 벡터의 컴포넌트를 나타내고, N은 형상 기술 벡터의 차원을 나타낸다.In the method (2) using Equation (3), the profit value is calculated using the components included in the shape description vector

The pair can be defined as the greatest rate of change. here,

,

Denotes the component of the shape description vector, and N denotes the dimension of the shape description vector.

,

는 형상 기술 벡터의 컴포넌트를 나타내고, N은 형상 기술 벡터의 차원을 나타낸다.In the method (3) using Equation (4), the profit value can be defined as a value obtained by accumulating differences between adjacent components among the components of the shape description vector. here,

,

Denotes the component of the shape description vector, and N denotes the dimension of the shape description vector.

The edge point identification unit 920 can determine the number of zero components in the shape description vector and determine whether the pixel point is near the edge point or edge. For example, the edge point identification unit 920 can determine the number of zero components in the N-dimensional shape description vector, and compare the number of determined zero components with a threshold value. The edge point identification unit 920 can determine that the pixel point belongs to an edge point or belongs to an edge if it is determined that the number of zero components is larger than the threshold value. The threshold value may be set by the user or arbitrarily changed. 10 is a diagram showing an example of the detection result of the edge point performed by the edge point identification unit 920 according to the embodiment. In Fig. 10, an area 1010 represents an edge point or an area to which an edge vicinity belongs.

The POI identification unit 930 can identify the POI based on the calculated profit value. The POI identification unit 930 can determine a POI having a large profit value within the local neighborhood range. For example, the POI identifier 930 may select a point that satisfies the sparseness criterion of the POI within the local neighborhood range and determine the selected point as a POI.

The POI identification unit 930 may perform smoothing on the profit value calculated by the profit value calculation unit 910 using a smoothing technique such as Gauss smooth. It is possible to prevent occurrence of a local calculation error due to noise by the smoothing process.

The POI identification unit 930 can select a point having the largest profit value within the local range by using the non-maximum suppression technique. The POI identification unit 930 can control the distance between the POIs according to the neighbor radius. The POI identifying unit 930 can determine the final POI by mapping the point having the greatest profit value to the closest pixel point in the initial 3D point cloud. The POI identification unit 930 can determine the final POI in the initial 3D point cloud that is closest to the selected point using the non-maximum suppression technique.

Here, whether to set an edge point or a point near the edge to the candidate POI can be determined by the necessity of an actual application. When setting an edge point or a point near an edge to a candidate POI in data whose internal characteristics are unclear, it is necessary to calculate and calculate a profit value for all points to identify and position the POI. Alternatively, the profit value may be calculated and any POI may be identified and positioned in any conventional manner.

11 is a flowchart illustrating a 3D point cloud POI detection method according to an embodiment.

In step 1110, the 3D point cloud POI detection device may obtain 3D point cloud data. 3D point cloud data can include 3D point cloud data in various formats.

According to another embodiment, the 3D point cloud POI detection apparatus may perform a preprocessing process (not shown) on the 3D point cloud data. For example, a 3D point cloud POI detection apparatus can perform processes such as noise elimination, outlier detection, and downsampling processing on 3D point cloud data.

In step 1120, the 3D point cloud POI detection device may generate a shape description vector that describes the shape of the surface including the pixel points of the 3D point cloud and neighboring points adjacent to the pixel points. The 3D point cloud POI detection apparatus can generate the shape description vector using the variance information about the yaw distance from the neighboring point of the pixel point to the tangent plane containing the pixel point.

In step 1130, the 3D point cloud POI detection device may extract the POI based on the shape description vector. The 3D point cloud POI detection apparatus can calculate the profit value based on the shape description vector, and can identify the POI in the 3D point cloud data based on the calculated profit value.

12 is a flowchart illustrating a process of generating a shape description vector according to an embodiment.

In step 1210, the 3D point cloud POI detection device may generate a local reference frame for the pixel point of the 3D point cloud. The local reference frame represents the coordinate system X-axis, Y-axis, and Z-axis at the center of the pixel point.

In step 1220, the 3D point cloud POI detection device may calculate the drift distance variance from the neighboring point of the pixel point to the tangent plane to which the pixel point belongs. The drift dispersion may include information about at least one of an angle between the projection direction from the neighboring point to the tangent plane and the direction of the X axis and a drift distance to the tangent plane.

In step 1230, the 3D point cloud POI detection device may generate the shape description vector based on the directional distance variance. The 3D point cloud POI detection apparatus can generate the shape description vector by expressing the direction distance variance as a vector of the finite dimension. The 3D point cloud POI detection apparatus can divide the drift dispersion into N arbitrary natural number intervals, and calculate the number of points in each section and the sum of the drift distances in each section. The 3D point cloud POI detection apparatus can determine the shape description vector based on the number of points in each section and the sum of the directions distances in each section.

13 is a flowchart illustrating a process of extracting a POI according to an embodiment.

In step 1310, the 3D point cloud POI detection device may calculate the profit value based on the shape description vector. The 3D point cloud POI detection apparatus calculates the difference between the mean value of the shape description vector and the product of variance, the largest change rate of the pair of components included in the shape description vector, and the accumulated value of the differences between adjacent components among the components of the shape description vector. One can be calculated as a profit value.

In step 1320, the 3D point cloud POI detection device may determine the number of zero components in the shape description vector and determine whether the pixel point belongs to an edge point or edge.

In step 1330, the 3D point cloud POI detection device may identify the POI based on the calculated profit value. The 3D point cloud POI detection apparatus can determine a POI having a large profit value within the local neighborhood range. For example, the 3D point cloud POI detection device may select a point that satisfies the scarcity criterion of the POI within the local neighborhood range and determine the selected point as a POI. FIG. 14 is a diagram illustrating an example of a result of detecting a POI in a 3D point cloud according to an embodiment.

Operations according to embodiments may be performed by one or more processors, or may be implemented in the form of program instructions that may be executed on various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks, and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (18)

A 3D point cloud data acquisition unit for acquiring 3D point cloud data;
A shape description unit for generating a shape description vector describing a shape of a surface including a pixel point of a 3D point cloud and a neighborhood point adjacent to the pixel point; And
A point of interest (POI) extraction unit,
The POI extracting unit,
Determining whether a pixel point is near an edge point or edge based on the shape description vector,
Determining whether the pixel point belongs to an edge point or an edge, identifying a POI based on a calculated profit value based on the generated shape description vector,
Wherein the gain value represents a change in the surface. The method according to claim 1,
The shape-
And generates the shape description vector using variance information about a directed distance from the neighbor point to a tangent plane including the pixel point. The method according to claim 1,
The shape-
A local reference frame generator for generating a local reference frame for a pixel point of the 3D point cloud;
Calculating a directed distance scatter diagram from a neighboring point adjacent to the pixel point to a tangent plane to which the pixel point belongs; And
And a shape description vector generation unit for generating shape description vectors by expressing the directivity dispersion degree as a vector of a finite dimension (limited dimension)
And a 3D point cloud POI detection device. The method of claim 3,
Wherein the local reference frame represents a coordinate system X-axis, Y-axis, and Z-axis at the center of the pixel point,
Wherein the drift dispersion diagram includes information about at least one of an angle between a projection direction from the neighboring point to a tangent plane and a direction of the X axis and a drift distance to the tangent plane. The method of claim 3,
The shape description vector generation unit may include:
Wherein the shape description vector is determined based on a sum of the number of points in each section and the number of directions in each section, and divides the FRDT into N arbitrary natural number intervals. The method according to claim 1,
The POI extracting unit,
A profit value calculator for calculating the profit value based on the generated shape description vector;
An edge point identification unit for determining the number of zero components in the shape description vector and determining whether the pixel point belongs to the edge point or the edge; And
A POI identification unit for identifying a POI based on the calculated profit value,
And a 3D point cloud POI detection device. The method according to claim 6,
The profit value calculation unit may calculate,
One of the product of the average value and the variance of the shape description vector, the largest change rate of the pair of components included in the shape description vector, and the value of the difference between adjacent components among the components of the shape description vector is calculated as a profit value 3D point cloud POI detection device. The method according to claim 1,
Wherein the 3D point cloud data obtaining unit comprises:
A 3D point cloud data preprocessing unit for preprocessing the obtained 3D point cloud data,
And a 3D point cloud POI detection device. 9. The method of claim 8,
The 3D point cloud data pre-
And performs at least one of noise removal, outlier point removal, and downsampling on the obtained 3D point cloud data. Obtaining 3D point cloud data;
Generating a shape description vector describing a shape of a surface including a pixel point of a 3D point cloud and a neighborhood point adjacent to the pixel point; And
Identifying a point of interest (POI)
Wherein identifying the POI comprises:
Determining whether a pixel point is near an edge point or edge based on the shape description vector,
Determining whether the pixel point belongs to an edge point or an edge, identifying a POI based on a calculated profit value based on the generated shape description vector,
Wherein the profit value is indicative of a change in the surface of the 3D point cloud. 11. The method of claim 10,
Wherein the generating the shape description vector comprises:
Generating the shape description vector using variance information about a directed distance from the neighbor point to a tangent plane containing the pixel point;
Gt; POI < / RTI > 11. The method of claim 10,
Wherein the generating the shape description vector comprises:
Generating a local reference frame for a pixel point of the 3D point cloud;
Calculating a directed distance scatter diagram from a neighboring point adjacent to the pixel point to a tangent plane to which the pixel point belongs; And
Generating the shape description vector by expressing the directional dispersion index as a vector of a finite dimension
Gt; POI < / RTI > 13. The method of claim 12,
Wherein the local reference frame represents a coordinate system X-axis, Y-axis, and Z-axis at the center of the pixel point,
Wherein the directional scatter diagram includes information about at least one of an angle between a projection direction from the neighbor point to a tangent plane and a direction of the X axis and a directional distance to the tangent plane. 13. The method of claim 12,
Wherein the generating the shape description vector comprises:
Dividing the drift dispersion into N arbitrary natural number intervals;
Calculating the sum of the number of points in each section and the number of directions in each section; And
Determining a shape description vector based on the number of points of each section and the sum of the directions distances within each section
Gt; POI < / RTI > 11. The method of claim 10,
Wherein identifying the POI comprises:
Calculating the profit value based on the generated shape description vector;
Determining a number of zero components in the shape description vector and determining whether the pixel point belongs to the edge point or near the edge; And
Identifying the POI based on the calculated profit value
Gt; POI < / RTI > 16. The method of claim 15,
Wherein the calculating the profit value comprises:
One of the product of the average value and the variance of the shape description vector, the largest change rate of the pair of components included in the shape description vector, and the value of the difference between adjacent components among the components of the shape description vector is calculated as a profit value step
Gt; POI < / RTI > 11. The method of claim 10,
Wherein acquiring the 3D point cloud data comprises:
And performing a preprocessing process on the obtained 3D point cloud data,
Wherein the pre-processing step performs at least one of noise removal, outlier point removal, and down-sampling on the obtained 3D point cloud data. A computer-readable recording medium on which a program for executing the method of claim 10 is recorded.

Images