JP5870011B2 - Point cloud analysis device, point cloud analysis method, and point cloud analysis program - Google Patents

Description

  The present invention relates to a point cloud analysis device, a point cloud analysis method, and a point cloud analysis program for analyzing point cloud data obtained by a sensor or the like.

  In recent years, with the widespread use of long-range laser range finders, environmental measurement has been realized in a wide range of spaces, both indoors and outdoors, and technology development has been carried out to extract useful objects (3D objects) from the measured 3D point cloud. It has been broken. In particular, it has become possible to measure the position of a linear object (hereinafter referred to as a wire) such as an electric wire or a communication overhead line by increasing the accuracy of a sensor such as a laser range finder.

  In Non-Patent Document 1, principal component analysis is performed on three-dimensional point cloud data to identify what three-dimensional object (straight line, plane, three-dimensional shape) each point belongs to. If the three-dimensional coordinate of each point is a vector and the distribution of a point group and its surrounding points is a linear shape, the first eigenvalue tends to be larger than other eigenvalues. Whether or not the attention point is on the overhead line can be determined by threshold processing. In Non-Patent Document 1, pattern recognition is performed using labels (attribute numbers for determining which objects belong) that are assigned in advance to each point, considering the first to third eigenvalues obtained by principal component analysis as feature quantities. It is determined which object (overhead line, building, road, etc.) each point belongs to. FIG. 8 is an explanatory diagram showing the linear approximation accuracy of the point group density and the overhead line (wire). As shown in FIG. 8A, when the wire can be measured with high density, the estimation accuracy of the straight line obtained by the principal component analysis is high.

  However, in general, in a wide-area measurement environment such as outdoors, subjects such as the ground near the measuring device can measure with high density, but wires such as overhead wires are far from the measuring device because of subjects in the sky. This causes the measurement points to be sparse. In addition, since there are a plurality of wires between the utility poles, occlusion occurs (an object becomes an obstacle on the way and an object far from the measuring instrument cannot be measured), and the measurement interval is further reduced. For this reason, it is difficult to determine whether or not the point of interest is a point that constitutes an overhead line (wire constituent point) unless the shape of a point of interest and a wide range of point groups around it are examined. However, if the range (processing target region) for examining the shape is expanded, the point group other than the object to which the attention point belongs is included, and there is a problem that the reliability of the result of the principal component analysis is lowered (FIG. 8B). )reference).

  On the other hand, in the prior art, unlike the above-described approach using principal component analysis, there is an approach that considers a wider distribution of point clouds. In Non-Patent Document 2, overhead line detection is performed using Hough transform. The Hough transform is performed by voting a point group estimated to belong to a model such as a straight line or a parabola, such as a position or inclination of the model.

  However, Non-Patent Document 2 only performs overhead line detection in a narrow area where the overhead line looks almost straight, and it is difficult to estimate a complex shape model in a wide area on the road. is there. In Non-Patent Document 2, model estimation is performed on an image (a two-dimensional space). However, when model estimation is performed on a three-dimensional space, the parameter voting space also increases, so that an accurate parameter There is a problem that estimation itself is difficult.

  As another approach different from the model estimation in the prior art, there is a method of correcting a point where the estimation accuracy of the object identification is considered to be low by the Markov Random Field. In Non-Patent Document 3, the shape of the point of interest and the surrounding point group is estimated by principal component analysis, and smoothing is performed using the result that seems to have high estimation accuracy of neighboring points. The estimation accuracy has been improved. This technique can be expected to improve the estimation accuracy of the part where the measurement density is partially sparse.

  However, there is a problem that a portion where points with low estimation accuracy of individual points are gathered cannot be corrected correctly because there is no point group with high estimation accuracy, and the estimation accuracy cannot be improved. For example, when a plurality of overhead lines are densely arranged and the overhead lines cross at the twisted position, the number of point groups with low identification accuracy increases.

  Also, as a problem of applying principal component analysis as in the existing technology, the point cloud distribution is examined after coordinate transformation so that the center of gravity is the origin. FIG. 9 is a diagram illustrating an example of a measurement result of a utility pole and an overhead wire (wire) between the utility poles. FIG. 9A represents a utility pole and an overhead line between the utility poles. FIG. 9B represents a three-dimensional point group obtained by measuring the utility pole and the overhead line between the utility poles with a laser range finder. As shown in FIG. 9B, if the density of the point group is biased, there is a possibility that the position indicated by “x” far away from the position of the point of interest p becomes the center of gravity. Therefore, a tangential direction at a position different from the position of the point p is obtained, and there is a problem that the estimation accuracy is lowered.

Natural Terrain Classification using 3-D Ladar Data, N. Vandapel, D. F. Huber, A. Kapuria and M. Hebert, Robotics and Automation, 2004. Proceedings.ICRA '04. 2004 IEEE International Conference on (2004) Knowledge-based Power Line Detection for UAV Surveillance and Inspection Systems, Z. Li, Y. Liu, R. Hayward, J. Zhang and J. Cai, Image and Vision Computing New Zealand, 2008. IVCNZ 2008. 23rd International Conference (2008 ) Directional Associative Markov Network for 3-D Point Cloud Classification, D. Munoz, N. Vandapel and M. Hebert, Fourth International Symposium on 3D Data Processing, Visualization and Transmission, June (2008)

  As described above, in the conventional technique for estimating the shape of individual points using principal component analysis, the occlusion and overhead line positions are far from the measuring device and the measurement density is low, or a plurality of overhead lines are dense. In such a case, there is a problem that the detection accuracy of the overhead line is lowered. Further, since the covariance matrix is obtained, the point cloud shape and the tangential direction at a position different from the point of interest are obtained, and there is a problem that the estimation accuracy is lowered.

  The present invention has been made in view of such circumstances, and a point group analysis apparatus, a point group analysis method, and a point that can accurately detect a linear object by analyzing the obtained three-dimensional point group The purpose is to provide a group analysis program.

  The present invention provides a three-dimensional point cloud acquisition means for acquiring three-dimensional point cloud data, one of the three-dimensional point cloud data as a point of interest, and a point within a certain distance from the point of interest as a surrounding point, An autocorrelation matrix calculating means for calculating the importance of the surrounding points with respect to the attention point, calculating a weighted autocorrelation matrix using the importance for the attention point, and an eigenvalue feature having eigenvalues of the autocorrelation matrix as elements It is characterized by comprising: a feature vector calculating means for calculating a vector; and an identifying means for determining whether or not the point of interest constitutes a linear object using the eigenvalue feature vector.

  The present invention is characterized in that the importance is increased when the point of interest exists in a direction in which the distribution of the point group in the vicinity region of each surrounding point is wide.

  The present invention is characterized in that the importance value increases when a point group exists in a wide range on a band centering on a straight line connecting the attention point and each surrounding point.

  The present invention is characterized in that the identifying means determines whether or not the point of interest constitutes a linear object based on whether or not the maximum eigenvalue exceeds a predetermined threshold value.

  In the present invention, the identification unit compares the distance between the feature vector and the dictionary vector, and determines that an object having a dictionary vector with a close distance is an object to which the attention point belongs, so that the attention point is It is characterized by determining whether to comprise a linear object.

  The present invention is a point cloud analysis method performed by a point cloud analysis apparatus including a 3D point cloud acquisition unit for acquiring 3D point cloud data, wherein one of the 3D point cloud data is used as a target point, and the target point And a point below a certain distance as a surrounding point, calculating the importance of the surrounding point relative to the point of interest, and calculating a weighted autocorrelation matrix using the degree of importance for the point of interest A feature vector calculation step of calculating an eigenvalue feature vector having the eigenvalue of the autocorrelation matrix as an element, and an identification step of determining whether or not the point of interest constitutes a linear object using the eigenvalue feature vector; It is characterized by having.

  The present invention is a point cloud analysis program for causing a computer to function as the point cloud analysis device.

  According to the present invention, there is an effect that a linear object can be detected with high accuracy from the obtained three-dimensional point group. In particular, even in the case where there is a point group in which a plurality of linear objects are densely arranged in parallel or a linear object intersecting at a twisted position, which is difficult with the prior art, Detection can be performed.

It is a block diagram which shows the structure of the 1st Embodiment of this invention. 2 is a flowchart showing the operation of the wire composing point detection apparatus 100 (wire composing point feature extracting means 105, wire composing point determining means 106) shown in FIG. FIG. 10 is an enlarged view of a point of interest _pi and a processing target point group in FIG. 9. From the point of interest p _i is a diagram showing a point group in the radius R. It is a flowchart which shows the operation | movement which the wire constituent point feature extraction means 105 calculates _| requires frequency distribution _Hq . It is explanatory drawing which shows the creation example of frequency distribution. It is a figure which shows the difference in weight distribution when the point p _i exists on the curved surface of the object (A) and when it exists on the wire (B). It is explanatory drawing which shows the point group density and the linear approximation precision of an overhead wire (wire). It is explanatory drawing which shows an example of the measurement result of the overhead wire (wire) between utility poles.

<First Embodiment>
Hereinafter, a point cloud analyzer according to a first embodiment of the present invention will be described with reference to the drawings. In this embodiment, a point group belonging to a linear object (wire) is detected from a large-scale three-dimensional point group, and at the same time, a tangent direction at the detected point (wire composing point) is calculated. In the following description, as a specific example, a point group analysis for analyzing a point group (wire constituent point group) constituting a linear object using a point group having position information (three-dimensional coordinates) acquired by a laser range finder The apparatus will be described. Here, the three-dimensional information may be latitude, longitude, sea level (height) information, or a three-dimensional Euclidean coordinate system or a polar coordinate system with a specific position set by the user as the origin. In the following description, a three-dimensional Euclidean coordinate system at the origin set by the user (each direction is assumed to be an X, Y, Z coordinate) is assumed, and the unit of each coordinate is meter. The term “three-dimensional point” means that each point is given measurement information such as the time when the above-mentioned three-dimensional coordinates and the group of points are photographed, the reflection intensity of the laser, and the attribute information of the point. There is no particular limitation on the information given to the three-dimensional point, but at least position information (X, Y, Z coordinates) is given, and the three-dimensional point group is a collection of a plurality of points.

  Next, the configuration of the point group analysis apparatus according to the embodiment will be described. FIG. 1 is a block diagram showing the configuration of the embodiment. Reference numeral 100 denotes a wire composing point detecting device configured by a computer device. Reference numeral 101 denotes a subject measurement unit that is configured by a laser range finder, an infrared sensor, an ultrasonic sensor, or the like and measures the distance to the subject. The subject measuring means 101 is equipped with, for example, a laser range finder on a vehicle equipped with a GPS (Global Positioning System) or on an airplane equipped with a GPS, and by measuring while moving, buildings, overhead lines, roads in the city This is a device that measures the three-dimensional position of an unspecified number of subjects. If the detection is not a columnar object around the road but a wire object at a specific position (such as an intersection), the measurement need only be performed from one location, and the GPS does not necessarily have to be mounted. In the present embodiment, it is assumed that the subject measuring unit 101 is an MMS (Mobile Mapping System) in which a GPS and a laser range finder are mounted on a vehicle.

  Reference numeral 102 denotes parameter input means for inputting parameters necessary for the wire composing point detection apparatus 100. The parameter input means 102 is a parameter data transfer device from a user interface such as a keyboard, mouse, or touch input device, or from an external storage device such as a DVD (Digital Versatile Disc) or USB (Universal Serial Bus) memory.

  Reference numeral 103 denotes a three-dimensional point group storage unit that stores measurement data such as laser reflection intensity, three-dimensional position information, and photographing time measured by the subject measurement unit 101. Reference numeral 104 denotes parameter storage means for storing parameters input by the parameter input means 102. The parameter is a parameter used for the identification method, for example, a threshold value or a value input to the classifier. The three-dimensional point cloud storage unit 103 and the parameter storage unit 104 are hardware storage devices such as HDD (Hard Disk Drive) and SSD (Solid State Drive), and may be the same or different hardware.

  Reference numeral 105 reads out a three-dimensional point group in the three-dimensional coordinates to be processed from the three-dimensional point group storage unit 103, and performs a feature extraction process for determining whether all three-dimensional points are wire constituent points. This is a wire constituent point feature extraction means to be performed. Reference numeral 106 denotes a wire composing point determination unit that obtains a parameter output from the result output from the wire composing point feature extracting unit 105 and the parameter storage unit 104 and identifies whether each point is a wire composing point. Reference numeral 107 denotes a wire configuration that acquires the result of the wire configuration point determination unit 106 and changes the attribute to the wire-specific attribute number acquired from the tangential direction assignment and parameter storage unit 104 for the point determined as the wire configuration point This is point information giving means. The attribute information is an object-specific number preset by the user for a building, ground, wire, tree, or the like.

  Reference numeral 108 denotes a three-dimensional point cloud output means for converting information to which a three-dimensional point cloud is assigned into text data such as ASCII code or binary data and outputting the data as a video signal for display display. The given information includes the three-dimensional position coordinates of each point, the reflection intensity of the laser, the attribute, the tangential direction, the measurement time, and the like. Reference numeral 109 denotes a display unit configured by a display device. The display means 109 is composed of, for example, a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display), a PDP (Plasma Display Panel), or the like.

  Next, the operation of the wire composing point detection apparatus 100 shown in FIG. 1 will be described with reference to FIG. FIG. 2 is a flowchart showing the operation of the wire composing point detection apparatus 100 (wire composing point feature extracting means 105, wire composing point determining means 106) shown in FIG. Here, it is assumed that the 3D point cloud storage unit 103 stores 3D point cloud data and the parameter storage unit 104 stores parameters.

First, the wire constituent point feature extraction unit 105 inputs point cloud data from the three-dimensional point cloud storage unit 103 (step S1). FIG. 9A is a diagram expressing the utility poles and the overhead lines between the utility poles. FIG. 9B is a diagram expressing a three-dimensional point group obtained by measuring the utility pole and the overhead line between the utility poles with a laser range finder. In the three-dimensional point group, a right-handed X, Y, Z coordinate system as shown in FIG. 9B is applied. The following processing is performed independently for all three-dimensional points. When the number for distinguishing the point of interest p is represented by i and the number of all point groups is I, the processing of steps S2 to S4 is performed independently for each point p _i .

Wire Configuration feature point extracting unit 105 calculates the weighted autocorrelation matrix at the point of interest p _i (step S2). First, a point q _j (jε1, 2, 3,..., N) within a radius R from the point p _i is set as a processing target point group. Here, N is the total number of points within the radius R, and j is a number for distinguishing each of the points to be processed. When the importance of the point q _j with respect to the point p _i is w _j , the weighted autocorrelation matrix P is obtained by the equations (1), (2), and (3).

Here, ‖q _j −p _iユ ー is the Euclidean distance (two norms of the vector q _j −p _i ) between the point q _j and the point P _i , and “·” in the equation (3) is an inner product of the vectors. u _l point from the point _{q j} within a radius of r is _{q m (m∈1,2,3, ..., M} ) for to determine the autocorrelation matrix Q relative to the point _{q j} by eigenvalue expansion eigenvectors Δ _k ² is the eigenvalue of Q (δ ₁ ² > δ ₂ ² > δ ₃ ² ). When M is the number of point groups having a distance within a radius r from the point q _j , Q based on q _j is obtained by equation (4).

Here, when the point q _j belongs to the same wire as the point p _i , it is assumed that the direction of the vector (q _j −p _i ) has a relatively larger number of point groups than the other directions. . That is, by considering the dispersion direction of the point group near the point q _j, it can be considered the possibility that the point p _i and the point q _j is on the same object. The importance ratio (weight w _j ) for the direction (q _j −p _i ) is the ratio of the Mahalanobis distance ML using the eigen value and the eigen vector determined by the equation (4) to the distance between the vector (q _j −p _i ).

FIG. 3 is an enlarged view of the attention point p _i and the processing target point group in FIG. 9. In the example shown in FIG. 3B, the dispersion of the point group and the first eigenvector (the direction in which the dispersion is the largest) for the points q ₁ and q ₂ are shown. When two points are compared, since the angle formed by the first eigenvector of the point q ₂ and the vector (q _j −p _i ) is small, the weight w ₂ is larger than the weight w ₁ , and the point q ₂ is on the same object Can be considered as. Thereby, even when there is a wire to which the target point does not belong in the processing target area within the radius R, the importance of the point in the direction of the different wire can be suppressed, so that the estimation accuracy can be improved.

  The parameter r and the radius R are determined experimentally. Here, r = 0.2 and R = 0.5. Increasing r increases the estimation accuracy even for a point group having a low density. However, increasing r increases the possibility of including a wire to which the point of interest does not belong, resulting in the influence of noise during estimation. Specifically, r is preferably the shortest distance or the average distance between different wires.

Next, the wire composing point feature extraction unit 105 expands the eigenvalues of the autocorrelation matrix P to calculate eigenvalue feature quantities λ ₁ , λ ₂ , λ ₃ (λ ₁ > λ ₂ > λ ₃ ) (step S3). . Hereinafter, a vector [λ ₁ , λ ₂ , λ ₃ ] ^{T in} which eigenvalue elements are vertically arranged is referred to as a feature vector and is referred to as an eigenvalue feature vector. Further, the first eigenvector u ₁ calculated at this time is set as the tangential direction at the point p _i .

閾値Ｖｔｈは実験的に決めるパラメータであり、閾値Ｖｔｈ以上の点をワイヤ構成点と判定する。ただし、閾値Ｖｔｈは０〜１の値をとり、ここではＶｔｈ＝０．９とした。そして、ワイヤ構成点判定手段１０６は、物体識別結果を出力する（ステップＳ５）。 Then, the wire constituting point determination unit 106 performs the object identification processing of each point p _i using eigenvalue feature quantity (step S4). The simplest processing method is threshold processing according to equation (5).

The threshold value Vth is a parameter determined experimentally, and a point equal to or higher than the threshold value Vth is determined as a wire constituent point. However, the threshold value Vth takes a value of 0 to 1, and here, Vth = 0.9. Then, the wire composing point determination means 106 outputs the object identification result (step S5).

  Further, when there are a plurality of wires within the radius R as indicated by a point p ′ in FIG. 9B, there is a tendency that the first eigenvalue of the equation (1) becomes small due to noise. Therefore, a determination process including eigenvalue information other than the first eigenvalue is performed. Specifically, the eigenvalue combination of the wire and the object other than the wire is registered in advance as a dictionary vector, and pattern matching such as the nearest neighbor (NN) method, the k-NN method, and the SVM method is used. By performing the identification, a determination that is more robust than the threshold processing can be performed.

In the present embodiment, a method for identifying wire composing points using the NN method will be described. Eigenvalues are obtained from the autocorrelation matrix P of the equation (1) for the point group belonging to the wire (class 1) and the other object (class 2). Next, feature vectors f ^{c} = [λ ₁ , λ ₂ , λ ₃ ] ^T (classε1,2) are obtained for the point cloud of class 1 and class 2. If the feature vector of the point p _i is f _pi , the likelihood for each class is sim (f, c), and the numbers that distinguish the points belonging to each class are t _c (cε1,2), respectively, the discriminant function F (F) can be obtained by equations (6) and (7).

Here, sim in the equation (7) outputs the shortest distance between the feature vector f _{pi at the} point p _i and the dictionary vector of each class as a likelihood. The discriminant function G compares the shortest distance of the dictionary vector of each class and outputs the class number of the smaller distance for the class 1 and class 2 for the feature vector f _pi of the point of interest p _i and the dictionary vector of each class. Function. In this embodiment, the wire and the other two-class identification have been described, but a plurality of objects (classes) such as a wire, a building, a tree, a car, and a utility pole can be identified by applying the same processing. .

Second Embodiment
Next, a point cloud analysis device according to a second embodiment of the present invention will be described. Since the apparatus configuration in the second embodiment is the same as the apparatus configuration shown in FIG. 1, detailed description thereof is omitted here. In the second embodiment, among the processing operations shown in FIG. 2, the processing in step S2 is different, and the processing operations other than step S2 shown in FIG. 2 are the same. In step S2 in the second embodiment, the importance w _j of the point p _i against point q _j, determined by the area of the distribution of points existing within a distance Δd from a straight line connecting the points p _i and the point q _j. FIG. 4 is a diagram showing a point group within the radius R from the point of interest p _i . As shown in FIG. 4B, the importance w _i is obtained from the distribution of points existing in two parallel lines separated by Δd from the straight line connecting the points P _i and q ₁ . Also, as shown in FIG. 4C, the importance W _i is obtained from the distribution of points existing in two parallel lines separated by Δd from the straight line connecting the points P _i and q ₂ .

Wire Configuration feature point extraction unit 105, in step S2 of the second embodiment, in order to examine the distribution of points, first create a frequency distribution H _q points on a straight line to make the point p _i and the point q _j. Processed point a point in the area so can be distinguished and _{q j s K (k∈1,2, ..} , K) is expressed as the point _{s K} and the straight line _(p i, _{q j)} the distance between the D (s _K ), and Δh is a value (parameter) for dividing the straight line (q _j , p _i ) discretely. The bin of the frequency distribution is represented by bin. Δd and Δh are parameters determined experimentally. Here, Δh = 0.03 (M) and Δd = 0.06 (M).

Here, with reference to FIG. 5, an operation in which the wire composing point feature extracting unit 105 obtains the frequency distribution _Hq will be described. FIG. 5 is a flowchart showing an operation in which the wire composing point feature extracting unit 105 obtains the frequency distribution _Hq . First, the wire composing point feature extracting unit 105 initializes the sum of the frequency distributions H _q to be 0 (step S11), and then sets all points s _k (kε1, 2,. , with N), determined from the frequency distribution _{H q} (step S12 to S16).

The wire composing point feature extraction unit 105 calculates D (s _k ) and L (s _k ) (step S13), and calculates bin (s _k ) from L (s _k ) (step S14). Subsequently, the wire composing point feature extracting unit 105 sets H _q (bin) ← H _q (bin) +1 when Gd (D (s _k ))> 0 (step S15). Then, the wire configuration feature point extraction unit 105 outputs the frequency distribution _{H q} determined (step S17). By this process, the frequency distribution _Hq is obtained (see FIG. 6). FIG. 6 is a diagram illustrating an example of creating the frequency distribution _Hq .

The processing operation shown in FIG. 5 is expressed by equations (8), (9), and (10).

ただし、‖・‖はベクトルの２ノルムを表し、θは点ｓ_Ｋと直線（ｑ_ｊ，ｐ_ｉ）との成す角を表している。 Here, the function G _d is a function that outputs 1 if the distance D ((s _K )) is less than or equal to the threshold value Δd, and the function ceil is a function that outputs a value obtained by rounding down the input value to the decimal point. The range of the bin value of the histogram is the range from the minimum value to the maximum value when the processing target point group q _j is input to the equation (9). The distance D (s _K ) and the variable L (s _K ) are obtained by the equations (11), (12), and (13).

Here, ‖ / ‖ represents the 2-norm of the vector, and θ represents the angle formed by the point s _K and the straight line (q _j , p _i ).

Next, the wire composing point feature extraction means 105 uses the frequency distribution H _q to calculate the importance w _j in the direction of (q _j −p _i ) using the equations (14) and (15).

FIG. 7 is a diagram illustrating a difference in weight distribution between the case where the point p _i exists on the curved surface of the object (FIG. 7A) and the case where the point p _i exists on the wire (FIG. 7B). . Compared to the case of FIG. 7A, in the case of belonging to the point _pi wire (FIG. 7B), the direction of the point where the specific direction (q ₃ , p _i ) exists due to the distribution of the point distribution The weight of the projecting increases. On the other hand, for the points p _i existing on the curved surface, weights in various directions tend to increase. Therefore, by determining whether there is a weight peak in a specific direction in the three-dimensional space, it can be determined whether or not the point of interest p _i is on the wire.

The existence of this peak is determined by using an autocorrelation matrix using the weights w _j. When the point of interest p _i is in the shape of a wire, the weight in a specific direction becomes large as shown in FIG. 7B, so that the autocorrelation matrix also has a component in a specific direction, and the eigenvalue expansion of the autocorrelation matrix Only the first eigenvalue increases. That is, whether or not there is a peak is determined by checking whether or not the first eigenvalue is larger than the threshold value of the expression (5) without checking the weights in the direction of many (q _j , p _i ). Can do. Further, by utilizing the property that the weight in a specific direction becomes large, the eigenvalue expansion is performed after the weighting coefficient of the point q _j not on the same wire as the point p _i is set to 0 by the threshold processing, so that the tangential direction The estimation accuracy can also be improved. For example, as shown in FIG. 7B, a threshold value Tw may be set such that a value less than half of the peak value is zero. Thus, consider a point not on the same wire and the point p _i is a point on the less important than the threshold value Tw, without adding to the autocorrelation matrix, as a result it so as not to affect the determination of the peak existence. The weighted autocorrelation matrix P is obtained by the equation (1) using the weight of the equation (14).

  As described above, in the conventional attention point identification technology, principal component analysis is performed using a point group within a certain distance (processing target region) from the attention point, and the object to which the point belongs is identified using its eigenvalue. Had gone. On the other hand, in the present embodiment, an object to which the attention point belongs is identified using an autocorrelation matrix that places importance on the direction of the point highly related to the attention point. Specifically, a weighted autocorrelation matrix is calculated by increasing the weight in the direction of a point estimated to belong to the same object as the target point. Next, eigenvalues and eigenvectors are obtained from the autocorrelation matrix. By appropriately obtaining the weight, the influence of the point of the three-dimensional object that is not the same as the target point can be reduced, and even if the processing target area is enlarged, it is possible to stably identify whether the object is a wire or other object. .

  As a result, when calculating the points that make up an elongated object (wire) such as a parabola or straight line from the 3D point cloud and the tangential direction of the constituent points, there are measurement noises in the measurement results or the measurement density is sparse. In some cases, and even when a plurality of wires are close to each other, the detection of wire composing points and the tangential direction of each composing point can be calculated with high accuracy.

  A three-dimensional point cloud is obtained by recording a program for realizing the functions of the processing unit in FIG. 1 on a computer-readable recording medium, causing the computer system to read and execute the program recorded on the recording medium. Analysis processing may be performed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Accordingly, additions, omissions, substitutions, and other changes of the components may be made without departing from the technical idea and scope of the present invention.

  By analyzing the obtained three-dimensional point group, it can be applied to applications in which it is indispensable to accurately detect an object.

  DESCRIPTION OF SYMBOLS 101 ... Subject measuring means, 102 ... Parameter input means, 103 ... Three-dimensional point group storage means, 104 ... Parameter storage means, 105 ... Wire constituent point feature extraction means, 106 ... Wire composing point determination means, 107... Wire composing point information giving means, 108... 3D point group output means, 109.

Claims (7)

3D point cloud acquisition means for acquiring 3D point cloud data;
One of the three-dimensional point cloud data is set as a point of interest, a point within a certain distance from the point of interest is set as a surrounding point, and the importance of the surrounding point with respect to the point of interest is calculated. Autocorrelation matrix calculating means for calculating a weighted autocorrelation matrix using
A feature vector calculation means for calculating an eigenvalue feature vector whose elements are eigenvalues of the autocorrelation matrix;
A point group analysis apparatus comprising: identification means for determining whether or not the point of interest constitutes a linear object using the eigenvalue feature vector.   The point cloud analysis device according to claim 1, wherein the importance level increases when the point of interest exists in a direction in which a distribution of a point cloud in a region near each surrounding point is wide. .   2. The point according to claim 1, wherein the importance value increases when a point group exists in a wide range on a band centering on a straight line connecting the attention point and each surrounding point. 3. Group analysis device.   The said identification means determines whether the said attention point comprises a linear object based on whether the largest eigenvalue exceeds a predetermined threshold value, The any one of Claim 1 to 3 characterized by the above-mentioned. The point cloud analyzer described in the paragraph.   The identification means compares the distance between the feature vector and the dictionary vector, and determines that an object having a dictionary vector with a close distance is an object to which the attention point belongs, so that the attention point is a linear object. 4. The point cloud analysis apparatus according to claim 1, wherein it is determined whether or not to configure. A point cloud analysis method performed by a point cloud analysis apparatus including a 3D point cloud acquisition unit for acquiring 3D point cloud data,
One of the three-dimensional point cloud data is set as a point of interest, a point within a certain distance from the point of interest is set as a surrounding point, and the importance of the surrounding point with respect to the point of interest is calculated. An autocorrelation matrix calculating step for calculating a weighted autocorrelation matrix using
A feature vector calculation step of calculating an eigenvalue feature vector whose elements are eigenvalues of the autocorrelation matrix;
An identification step of determining whether or not the point of interest constitutes a linear object using the eigenvalue feature vector.   A point cloud analysis program for causing a computer to function as the point cloud analysis device according to any one of claims 1 to 5.

Images