POINT CLOUD IDENTIFICATION DEVICE, LEARNING DEVICE, POINT CLOUD IDENTIFICATION PROCEDURE AND LEARNING PROCEDURE

Abstract

Point cloud identification device (300), comprising: a point cloud acquisition unit (320) to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit (310) to acquire a model that has a learning parameter; a rotation invariance conversion unit (330) to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotation invariance feature using data after orthogonalization; an inference unit (340) to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and a result output unit (350) to output a classification result by identification by the inference unit (340).

Description

TECHNICAL AREA

[0001] The present disclosure technology relates to a point cloud identification technology for identifying a point cloud specified by point cloud information. BACKGROUND ON THE STATE OF THE TECHNOLOGY

[0002] For example, the acquisition result of a sensor such as LiDAR or radar can be expressed as point cloud information containing a large number of points. A technique exists for extracting a feature from a point cloud specified by such point cloud information. Patent literature 1 discloses a feature display device that represents a feature of a point cloud. In particular, a feature display device of patent literature 1 is a feature display device for the feature display of three-dimensional point cloud data (point cloud information) and comprises a distance field conversion unit that converts a set of points into a distance field containing the coordinates s x , s y , and s za spatial sampling point group s around the points, and specifies a nearest-neighbor distance ϕ(s) from the spatial sampling point s to a nearest-neighbor point, a regular projection unit that receives a conversion into a standard coordinate system by performing a singular value decomposition of a matrix M which gives the coordinates s x , s y , and s z of the spatial sampling point s and the nearest-neighbor distance ϕ(s), and a parameterization unit that outputs a weight β as a feature vector of the three-dimensional point cloud data by training an expression learning machine that takes as input a coordinate L in of the spatial sampling point s, which has been converted into the standard coordinate system, and outputs the nearest neighbor distance ϕ(s).

[0003] The feature display device from patent literature 1 is designed to align, for example, a point cloud and a point cloud obtained by rotating or turning the point cloud, based on a nearest-neighbor distance to a spatial sampling point group around the point cloud. REFERENCE LIST PATENT LITERATURE

[0004] Patent Literature 1: JP 2019-133545 A SUMMARY OF THE INVENTIONAL PROBLEM

[0005] However, the feature representation device from patent literature 1 has the problem that, depending on the point cloud, an error occurs at the time of positioning using the nearest neighbor distance between the point cloud and the spatial sampling point, and that a point cloud with the same shape can be identified as a point cloud with a different shape, and that the accuracy of the point cloud identification tends to be low.

[0006] The present disclosure solves the above problem, and one task is to improve the accuracy of identifying a point cloud. SOLUTION TO THE PROBLEM

[0007] A point cloud identification device within the meaning of the present disclosure comprises: a point cloud acquisition unit to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit to acquire a model that has a learning parameter; a rotational invariance conversion unit to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotational invariance feature using data after orthogonalization; an inference unit to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and A result output unit to output a classification result by identification through the inference unit. ADVANTAGEOUS EFFECTS OF THE INVENTION

[0008] According to the present disclosure, it is possible to improve the accuracy of identifying a point cloud. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a representation which shows an example of a configuration of a point cloud identification system including a point cloud identification device according to a first embodiment. Fig. Figure 2 is a representation that shows an example of a configuration of a rotation invariance conversion unit in the point cloud identification device. Fig. Figure 3 is a flowchart that illustrates an example of processing by the point cloud identification device. Fig. Figure 4 is a flowchart that presents a concrete example of rotation invariance conversion processing in the processing of the point cloud identification device. Fig. Figure 5 is a representation that explains the rotation of a point cloud. Fig. 6A is a representation that describes a relationship between the rotation of point clouds that show the same shapes and subspaces. Fig. 6B is a representation that describes a relationship between point clouds that show different shapes and subspaces. Fig. Figure 7 is a representation which shows an example of a configuration of a point cloud learning system 2 including a point cloud learning device according to a second embodiment. Fig. Figure 8 is a flowchart that illustrates an example of processing by the point cloud learning device. Fig. Figure 9 is a representation showing a first example of a hardware configuration for implementing functions of the point cloud identification device or the point cloud learning device in the present disclosure. Fig. Figure 10 is a representation showing a second example of a hardware configuration for implementing the functions of the point cloud identification device or the point cloud learning device in the present disclosure. DESCRIPTION OF THE EXECUTION FORMS

[0009] In order to explain the present disclosure in more detail below, embodiments of the present disclosure will be described with reference to the accompanying drawings. First embodiment.

[0010] In a first embodiment, a form of point cloud identification device is described.

[0011] Fig. Figure 1 is a representation illustrating an example of a configuration of a point cloud identification system 1 including a point cloud identification device 300 according to the first embodiment.

[0012] Fig. Figure 2 is a representation that shows an example of a configuration of a rotation invariance conversion unit 330 in the point cloud identification device 300.

[0013] Fig. Figure 3 is a flowchart that illustrates an example of processing by the point cloud identification device 300.

[0014] Fig. Figure 4 is a flowchart that presents a concrete example of rotation invariance conversion processing in the processing of the point cloud identification device 300.

[0015] Fig. Figure 5 is a representation that depicts a rotation of point clouds 1100 and 1200.

[0016] Fig. Figure 6A is a representation that explains a relationship between the rotation of point clouds that have the same shape and the same subspaces 2100 and 2200. Fig. Figure 6B is a representation that explains the relationship between point clouds, which indicate different shapes and subspaces 3100 and 3200.

[0017] An example configuration of the point cloud identification system 1, including the point cloud identification device 300, is explained.

[0018] The point cloud identification system 1 is a system that includes a point cloud identification device 300 which identifies a point cloud.

[0019] The point cloud identification system 1, shown in Fig. 1, comprises a point cloud input device 100, a storage device 200, a point cloud identification device 300 and a result output device 900.

[0020] The point cloud input device 100 retrieves point cloud information (point cloud data), for example from a sensor not shown, and outputs the point cloud information to the point cloud identification device 300.

[0021] The sensor, which is not shown, is, for example, a LiDAR (Light Detection and Ranging) or a radar.

[0022] The point cloud information displays a multitude of (N ≥ 2) points representing position coordinates or features captured by the unspecified sensor.

[0023] The point cloud information represents each point contained in the point cloud, for example, in the form of a coordinate value in a k (k ≥ 2) dimension. The storage facility 200 comprises a storage unit 210.

[0024] Storage unit 210 contains information used for identification processing to identify a point cloud. Specifically, storage unit 210 contains, for example, a training parameter as a model for identification.

[0025] The point cloud identification device 300 identifies the point cloud using a learned model for k-dimensional point cloud identification. The learned model includes, for example, a model that was learned as described in a second embodiment.

[0026] k-dimensional point cloud identification is used to perform class identification of point cloud data that specify a point cloud represented in k dimensions, based on a feature such as the shape of the point cloud.

[0027] A configuration example for the Point Cloud Identification Device 300 is explained.

[0028] The point cloud identification device 300 comprises a model acquisition unit 310, a point cloud acquisition unit 320, a rotation invariance conversion unit 330, an inference unit 340, and a result output unit 350.

[0029] Model acquisition unit 310 acquires a model that has a learning parameter.

[0030] The learning parameter is a parameter that is learned by the learning device and used in the identification of the point cloud.

[0031] The point cloud acquisition unit 320 acquires k-dimensional point cloud information from N points.

[0032] Specifically, the point cloud acquisition unit acquires 320 point cloud information items that specify N (N ≥ 2) points in k (k ≥ 2) dimensions.

[0033] The rotation invariance conversion unit 330 performs an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information and computes a rotation invariance feature using data after the orthogonalization.

[0034] The rotational invariance feature is a feature that is unique for the shape of the point cloud, which is not affected even when the point cloud is rotated.

[0035] The inference unit 340 uses the rotational invariance feature and the model to identify the point cloud specified in the point cloud information.

[0036] Specifically, inference unit 340 calculates a point cloud feature, which specifies a characteristic of the point cloud, using the rotational invariance feature and the model, and identifies the point cloud based on this feature. The point cloud feature is a unique characteristic for each category of the point cloud and denotes, or specifies, a feature that is effective for classifying the point cloud into classes. The point cloud feature is a feature that can be identified as a category in a point cloud of a horse and as a category in a point cloud of a bird, for example, when the point cloud feature is identified by an animal species.

[0037] Furthermore, the inference unit 340 can be configured to extract the shape of the point cloud and to identify the point cloud using a method that employs a filter to extract object-specific information from the point cloud's shape. In this case, for example, when distinguishing between a horse and a bird, a filter is designed to extract features such as head shape, number of feet, presence or absence of wings, presence or absence of a beak, and the like. The inference unit 340 then extracts a point cloud feature using the filter and distinguishes between the two objects using that point cloud feature.

[0038] The result output unit 350 outputs a classification result through identification by the inference unit 340.

[0039] Specifically, the result output unit 350 outputs the classification result to the result output device 900.

[0040] An example of an internal configuration of the rotational invariance conversion unit 330 is explained. The in Fig. The rotational invariance conversion unit 330 shown comprises an orthogonalization plane or layer unit 331 and a projection plane or layer unit 332.

[0041] The orthogonalization plane unit 331 computes a basis vector for each of k coordinates that specify a point (each of N points) contained in the point cloud information, and performs a conversion in such a way that the k basis vectors are orthogonal to each other in order to produce an (N × k) orthonormal basis matrix.

[0042] The orthogonalization plane unit 331 performs orthogonalization processing, for example, according to the following procedure.

[0043] A point cloud P, specified in point cloud data (k-dimensional point cloud data) and containing N (N ≥ 2) points N in k (k ≥ 2) dimensions, is expressed by an N × k matrix, for example as given in expression (1). In this case, each point in the point cloud P is represented by k coordinate values ​​such as P N1 ,..., and P Nk expressed as specified in expression (1). P=(p11⋯p1k⋮⋱⋮pN1⋯pNk)∈ℝN×k

[0044] It is assumed that the center of mass of the point cloud P is transferred to the origin of the coordinate system.

[0045] The orthogonalization plane unit 331 calculates a basis vector for each of the k coordinates.

[0046] The orthogonalization plane unit 331 performs an orthogonalization orth on the point cloud P, as shown in expression (2), to carry out a conversion such that the k basis vectors are orthogonal to each other.

[0047] The orthogonalization plane unit 331 receives an orthonormal basis matrix X through orthogonalization orth.

[0048] Specifically, the Gram-Schmidt orthogonalization method X = GramSchmit(P) can be used for orthogonalization, a QR decomposition (X, r = QR(P)) can be used, or a singular value decomposition (X, σ, V) can be used. T = SVD(P)) can be used, or an eigenvalue decomposition (X, Λ = EIG(P)) can be used. orth(P)=X,stXTX=I_k In expression (2) “Ik” represents a k × k identity matrix.

[0049] The projection plane unit 332 calculates a projection matrix that specifies the rotational invariance feature, using the orthonormal basis matrix and the model.

[0050] The projection plane unit 332 extracts a projection matrix M, which is a k-dimensional rotational invariance feature, from the orthonormal basis matrix obtained by the orthogonalization plane unit 331.

[0051] The projection matrix M is calculated using the following expression (3). M=XXT

[0052] If the projection matrix is ​​computed using a matrix XR obtained by a k-dimensional rotation (rotation R) of the orthonormal basis matrix X, the following expression (4) is obtained. XR(XR)T=XXT=M

[0053] For this reason, the projection matrix M is invariant with respect to the rotation R (rotationally invariant).

[0054] The projection plane unit 332 calculates and outputs a matrix “Y” obtained by multiplying the projection matrix M by a learning parameter W contained in the model, as in the following expression (5). In the present disclosure, the learning parameter W represented in expression (5) is a parameter that specifies a geometrically meaningful initial weight and eventually becomes a constant through learning. Y=MW

[0055] The projection matrix M is geometrically a projection matrix onto a space spanned by the projection matrix M.

[0056] “Y”, expressed by expression (5), represents a point cloud (shape) in a case where “W” is projected onto a space spanned by “M” by multiplying the projection matrix M by “W” from the right. “Y” is thus shape information that is geometrically independent of rotation and denotes a rotational invariance feature.

[0057] Using this information, the projection plane unit 332 can then obtain the same "Y" from the point cloud P before rotation (orthonormal basis matrix X) and a point cloud PR after rotation (orthonormal basis matrix XR) as later with reference to the Fig. 6A and Fig. 6B described.

[0058] In contrast, the projection plane unit can obtain 332 different “Y” and “Y_hat” (this means that “Y” obtained by expression (5) is different) from the point cloud P1 (orthonormal basis matrix X1) and the point cloud P2 (orthonormal basis matrix X2) with different shapes.

[0059] The in Fig. For example, the rotation invariance conversion unit 330 shown performs an orthogonalization of each basis vector for each of the points specified in the point cloud information as explained above, and calculates the rotation invariance feature using the data after orthogonalization and the model (the learning parameter W specifies the first weighting).

[0060] In a case where the rotation invariance conversion unit 330 is configured as described above, the inference unit 340 identifies the point cloud specified in the point cloud information using the projection matrix and the model that specify the rotation invariance feature.

[0061] An example of an internal configuration of the Inference Unit 340 is described.

[0062] Inference unit 340 extracts a point cloud feature based on the result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, using the projection matrix and the model that specify the rotational invariance feature, and identifies the point cloud using this feature. Specifically, the point cloud is classified.

[0063] The point cloud feature constructed by the inference unit 340 is evaluated with reference to the Fig. 5 and Fig. 6 described.

[0064] First, the rotation of the point cloud is explained.

[0065] In Fig. 5 represents a ◯ point cloud 1100 (white circle) a two-dimensional point cloud X before rotation, and the • point cloud 1200 (black circle) represents a two-dimensional point cloud XR(π) after rotation.

[0066] The two-dimensional point cloud X in Fig. 5 is represented by a matrix X in which x 11 equals -0.5, x 12 equals -1.625, x 21 equal to 0.5, x 22 equals -0.625, x 31 equals -1.5, x 32 equals 3.375, x 41 equals 1.5 and x 42 is equal to -1.125.

[0067] The two-dimensional point cloud XR(π) in Fig. 5 is represented by a matrix XR(π) in which xr 11 equal to 0.5, xr 12 equal to 1.625, xr 21 equals -0.5, xr 22 equal to 0.625, xr 31 equals 1.5, xr 32 equals -3,375, xr 41 equal to -1.5 and xr 42 which is equal to 1.125.

[0068] As in Fig. As shown in Figure 5, the position coordinates change when the point cloud data is rotated, and thus the information before and after the rotation is different as numerical data.

[0069] Although the o-group (white circle point) and the •-group (black circle point) are expressed by different coordinate values ​​due to the rotation, since the object indicated by the point cloud is the same as the object, even if the coordinates change before and after the rotation, the rotated point cloud can also be identified with the same accuracy (object of the same category) without the essential information changing.

[0070] In this context, the feature construction based on the projection matrix M can be carried out by the inference unit 340 with reference to Fig. 6 can be described geometrically.

[0071] The orthonormal basis matrix X is a matrix in which the k basis vectors of each of the N points are arranged as in the following expression (6). X=(x11⋯x1k⋮⋱⋮xN1⋯xNk)

[0072] The orthonormal basis matrix X spans a k-dimensional subspace spanX in an N-dimensional space. That is, the subspace spanX is a space generated by the orthonormal basis matrix X.

[0073] In Fig. 6A represents a subspace 2100 (spanX) a column space of an orthonormal basis matrix X obtained from the point cloud P. The orthonormal basis matrix X in Fig. 6A indicates that x 11 -0.5, x 12 -1.625, x 21 0.5, x 22 -0.625, x 31 -1.5, x 32 3,375, x 41 1.5 and x 42 -1.125 is.

[0074] In Fig. 6A represents a subspace 2200 (spanXR) a column space of an orthonormal basis matrix XR obtained from the point cloud PR. The orthonormal basis matrix XR(π) in Fig. 6A shows that xr 11 equal to 0.5, xr 12 equal to 1.625, xr 21 equals -0.5, xr 22 equal to 0.625, xr 31 equals 1.5, xr 32 equals -3,375, xr 41 equal to -1.5 and xr 42 which is equal to 1.125.

[0075] The orthonormal basis matrices X and XR are orthonormal basis matrices generated from point clouds of the same shape. In this case, the subspace 2100 (spanX) and the subspace 2200 (spanXR) coincide.

[0076] In Fig. In 6B, subspace 3100 (spanX1) represents a column space of the orthonormal basis matrix X1, which was obtained from the point cloud P1. The orthonormal basis matrix X1 in Fig. 6B indicates that x1 11 -0.5, x112 -1.625, x1 21 0.5, x1 22 -0.625, x1 31 -1.5, x1 32 3,375, x1 41 1.5 and x1 42 -1.125 is.

[0077] In Fig. 6B represents the subspace 3200 (spanX2) as a column space of the orthonormal basis matrix X2, which was obtained from the point cloud P2. The orthonormal basis matrix X2 in Fig. 6B indicates that x2 11 -4,676, x2 12 3,1419615, x2 21 3,892, x2 22 5,1419615, x2 31 1,892, x2 32 -7.4258845, x2 41 -1,108 and x2 42 - 0.8580385 is.

[0078] The orthonormal basis matrices X1 and X2 are orthonormal basis matrices generated from point clouds with different shapes. In this case, the subspace 3100 (spanX1) and the subspace 3200 (spanX2) do not coincide.

[0079] Based on this fact, the inference unit 340 extracts a point cloud feature, which is a feature that is effective for identifying a point cloud, using an orthonormal basis matrix for each point cloud.

[0080] The inference unit 340 extracts a point cloud feature Z by using, for example, the following expression (7). Z=Φ(X)=σ(WY)+B

[0081] In expression (7) “Y” is obtained by multiplying the projection matrix M with the parameter W, which specifies the first weighting as expressed in expression (5).

[0082] In expression (7) the learning parameter W specifies a weight (second weight), and a learning parameter B (second learning parameter) specifies a bias.

[0083] Expression (7) indicates that a feature extraction function Φ as feature extractor has a learning parameter (a parameter W specifying a first weight, a parameter W specifying a second weight, and a parameter B specifying a bias) and a nonlinear conversion function σ for nonlinear conversion, and the point cloud feature Z is extracted by inputting the orthonormal basis matrix X into the feature extraction function Φ.

[0084] The extraction of the point cloud feature Z by Φ(X) in expression (7) corresponds to performing an orthogonal projection onto the subspace spanX (subspace 2100) spanned by the orthonormal basis matrix X.

[0085] In the orthogonal projection onto the subspace spanX (subspace 2100) and the subspace spanXR (subspace 2200), the parameter W is projected onto the same coordinates, since these subspaces coincide. This corresponds to the extraction of a rotational invariance feature that is unique for point cloud form.

[0086] In contrast, orthogonal projections onto the subspace spanX1 (subspace 3100) and the subspace spanX2 (subspace 3200) project the parameter W onto different coordinates, since these subspaces do not coincide.

[0087] In this way, the point cloud information can be identified using the point cloud shape-specific features that are not dependent on the coordinates of the point cloud.

[0088] It should be noted that in the description, the inference unit 340 is configured to receive "Y", which is calculated by multiplying the projection matrix (projection matrix M) by the learning parameter (parameter W, which specifies the first weight) contained in the model and output by the projection plane unit 332. However, the same applies if the inference unit 340 is configured to multiply the projection matrix (projection matrix M) by the learning parameter (parameter W, which specifies the first weight) contained in the model. In such a configuration, the projection plane unit 332 outputs the projection matrix M unchanged.

[0089] The result output device 900 receives and outputs a classification result from the point cloud output by the point cloud identification device 300. The result output device 900 only needs to be a device that uses the point cloud classification result and can, for example, be a display device that simply shows the result, or it can be a control device for the automated driving of a vehicle.

[0090] The processing by the point cloud identification device 300 is carried out with reference to the Fig. 3 and Fig. 4 explained.

[0091] After processing starts, the point cloud identification device 300 performs the model acquisition processing (step ST110).

[0092] Specifically, the model acquisition unit 310 in the point cloud identification device 300 acquires a model with a learning parameter from the storage unit 210 of the storage device 200. The model acquisition unit 310 outputs the model that has the learning parameter.

[0093] The point cloud identification device 300 performs the point cloud acquisition processing (step ST120).

[0094] Specifically, the point cloud acquisition unit 320 acquires point cloud information from the point cloud identification unit 300, specifying N (N ≥ 2) points in k (k ≥ 2) dimensions, from the point cloud input unit 100. The point cloud acquisition unit 320 outputs the acquired point cloud information.

[0095] The point cloud identification device 300 performs a rotation invariance conversion processing (step ST130).

[0096] In particular, the rotation invariance conversion unit 330 in the point cloud identification device 300 performs an orthogonalization of each basis vector for each of the points specified in the point cloud information and calculates the rotation invariance feature using data after orthogonalization.

[0097] A specific example of rotational invariance conversion processing is explained.

[0098] After starting the rotation invariance conversion processing, the rotation invariance conversion unit 330 performs the orthogonalization processing (step ST131), as shown in Fig. 4 shown.

[0099] In particular, the orthogonalization plane unit 331 in the rotation invariance conversion unit 330 computes a basis vector for each of k coordinates specifying a point (each of N points) contained in the point cloud information and performs a conversion in such a way that the k basis vectors are orthogonal to each other in order to produce a (N × k) orthonormal basis matrix.

[0100] The Orthogonalization Plane Unit 331, for example, uses the Gram-Schmidt orthogonalization procedure, QR decomposition, singular value decomposition, or eigenvalue decomposition to perform the conversion in such a way that the basis vectors are orthogonal to each other, in order to generate an (N × k) orthonormal basis matrix. The Orthogonalization Plane Unit 331 outputs an orthonormal basis matrix.

[0101] The rotation invariance conversion unit 330 performs projection processing (step ST132).

[0102] Specifically, the projection plane unit 332 acquires the orthonormal basis matrix from the orthogonalization plane unit 331 within the rotational invariance conversion unit 330. Using this orthonormal basis matrix, the projection plane unit 332 computes a projection matrix that specifies the rotational invariance feature. The projection plane unit 332 then acquires the model from the model acquisition unit 310, multiplies the projection matrix (projection matrix M) by the learning parameter (learning parameter W, which specifies the initial weight) contained in the model, and outputs a result.

[0103] The point cloud identification device 300 performs inference processing (step ST140).

[0104] Specifically, in the point cloud identification unit 300, the inference unit 340 first acquires the projection matrix from the projection plane unit 332 and acquires the model (the learning parameter W, which specifies a second weight, and the learning parameter B, which specifies a bias) from the model acquisition unit 310. Next, the inference unit 340 uses the rotation invariance feature and the model to identify the point cloud specified in the point cloud information.More specifically, as described above, the inference unit 340 extracts a point cloud feature based on a result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, using the projection matrix and the model that specifies the rotational invariance feature, and identifies the point cloud using the point cloud feature.

[0105] It should be noted that in a case where the projection plane unit 332 in the rotation invariance conversion unit 330 is designed to output the projection matrix without multiplying the learning parameter (parameter W), the projection matrix is ​​multiplied by the learning parameter (parameter W) in the inference unit 340.

[0106] The point cloud identification device 300 performs result output processing (step ST150).

[0107] Specifically, the result output unit 350 in the point cloud identification unit 300 outputs a classification result to the result output unit 900 by identifying the inference unit 340.

[0108] The point cloud identification device 300 determines whether the processing sequence is terminated (step ST160).

[0109] If it is determined not to terminate processing (step ST160: "No"), the point cloud identification device 300 proceeds to process step ST110 and repeats the processing of step ST110.

[0110] When it is determined to stop processing (step ST160 “YES”), the point cloud identification device 300 stops processing.

[0111] An example of the effect of the point cloud identification device of the present disclosure is described.

[0112] To perceive a vehicle's immediate surroundings, automated driving technology, for example, uses point cloud data to identify other vehicles and obstacles in the periphery. In this case, real-time processing of the point cloud data and point cloud identification make it possible to check for the presence of a vehicle ahead to avoid a collision, or to detect an obstacle, rocks, or similar object that the automated driving system is approaching. In some cases, the point cloud is rotated to reveal a state where it is viewed from different perspectives, and this typically requires separate processing such as data augmentation through rotation and orientation adjustment.

[0113] In contrast, the present disclosure, as described above, explains the configuration in which the point cloud data, expressed by the orthonormal basis matrix, are treated as a single subspace, and the projection matrix is ​​treated as data to acquire rotational invariance. Since the point cloud is expressed as rotationally invariant data in the present disclosure, learning only a single layer or position is equivalent to learning every single layer, and it is possible to acquire an identification model that is robust against any rotation. This not only makes learning more efficient but also allows for the same identification accuracy before and after rotation.

[0114] The point cloud identification device within the meaning of this disclosure is configured as follows.

[0115] Point cloud identification device, including: a point cloud acquisition unit to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit to acquire a model that has a learning parameter; a rotational invariance conversion unit to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotational invariance feature using data after orthogonalization; an inference unit to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and A result output unit to output a classification result by identification through the inference unit.

[0116] Thus, the present disclosure has the effect of providing a point cloud identification device which improves the accuracy of identifying a point cloud.

[0117] The point cloud identification method as defined in this disclosure is configured as follows.

[0118] Point cloud identification methods, including: a point cloud acquisition step of initiating a point cloud acquisition unit to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition step of initiating a model acquisition unit, a model that has a learning parameter, to acquire; a rotational invariance conversion step of causing a rotational invariance conversion unit to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotational invariance feature using the information after the orthogonalization; an inference step of initiating an inference unit to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and a result output step of causing a result output unit to output a classification result by identification by the inference unit.

[0119] Thus, the present disclosure has the effect of providing a point cloud identification method which improves the accuracy of identifying a point cloud.

[0120] The point cloud identification device within the meaning of this disclosure is further configured as follows.

[0121] Point cloud identification device, in which which has a rotational invariance conversion unit: an orthogonalization plane unit to compute a basis vector for each of k coordinates specifying a point contained in the point cloud information, and to perform a conversion in such a way that the k basis vectors are orthogonal to each other in order to generate an orthonormal basis matrix; and a projection plane unit to compute a projection matrix that specifies the rotational invariance feature using the orthonormal basis matrix, and the inference unit a point cloud specified in the point cloud information is identified using the projection matrix, which specifies the rotational invariance feature, and the model.

[0122] Thus, the present disclosure has the effect of providing a point cloud identification device that is capable of efficiently calculating a rotational invariance feature of a point cloud.

[0123] Furthermore, the present disclosure shows a similar effect to the effect described above when the configuration described above is applied to the point cloud identification method.

[0124] The point cloud identification device within the meaning of this disclosure is further configured as follows.

[0125] In the point cloud identification facility, extracts the inference unit, using the projection matrix, which specifies the rotational invariance feature, and the model, a point cloud feature is extracted based on a result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, and a point cloud is identified using the point cloud feature.

[0126] Thus, the present disclosure has the effect that it is possible to provide a point cloud identification device that identifies point cloud information using the point cloud shape-specific features that are not dependent on the coordinates of the point cloud.

[0127] Furthermore, the present disclosure shows a similar effect to the effect described above when the configuration described above is applied to the point cloud identification method. Second embodiment.

[0128] A second embodiment describes a form of learning facility.

[0129] In the second embodiment, a detailed description of the same content as that already explained may be omitted.

[0130] Fig. Figure 7 is a representation illustrating an example of a configuration of a point cloud learning system 2 including a point cloud learning device 400 according to the second embodiment.

[0131] The point cloud learning system 2 includes a point cloud learning device 100, a storage device 200A and a point cloud learning device 400.

[0132] The point cloud input device 100 is similar to the previously described point cloud input device 100, so a detailed description of it is omitted here.

[0133] The 200A storage system includes a 210A storage unit.

[0134] The 210A storage unit contains information used for the identification processing of a point cloud. Specifically, the 210A storage unit has, for example, a learning parameter as a model for identification. This model is appropriately trained and updated by the 400 point cloud learning unit.

[0135] The Point Cloud Identification Unit 400 trains a rotational invariance model in k-dimensional point cloud identification and identifies a point cloud using the learned model. The k-dimensional point cloud identification is used to perform class identification of point cloud data, specifying a point cloud represented in k dimensions, based on a feature such as the shape of the point cloud.

[0136] A configuration example of the point cloud learning device 400 is explained.

[0137] The point cloud learning facility 400 comprises a model learning unit 410, a point cloud acquisition unit 420, a rotation invariance conversion unit 430, an inference unit 440, a result output unit 450, an evaluation unit 460 and a model update unit 470.

[0138] The model acquisition unit 410 acquires a model that has a learning parameter.

[0139] The learning parameter is a parameter that is trained by the point cloud learning device 400 and is a parameter that is used when the point cloud is identified.

[0140] The point cloud acquisition unit 420 acquires k-dimensional point cloud information from N points.

[0141] The point cloud acquisition unit 420 acquires point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions, similar to the point cloud acquisition unit 320 described above.

[0142] Similar to the rotation invariance conversion unit 330 described above, the rotation invariance conversion unit 430 performs an orthogonalization of each basis vector for each of the points specified in the point cloud information and calculates a rotation invariance feature using data after the orthogonalization.

[0143] Similar to the inference unit 340 described above, the inference unit 440 uses the rotational invariance feature and the model to identify the point cloud specified in the point cloud information. Specifically, it performs the class identification of the point cloud.

[0144] The result output unit 450 outputs a classification result through identification by the inference unit.

[0145] Specifically, the result output unit 450 outputs the classification result to the evaluation unit 460.

[0146] Furthermore, the result output unit 450 can be configured to output the classification result to an external device. In this case, the external device can, for example, be the result output device 900 described in the first embodiment.

[0147] The evaluation unit 460 evaluates the model using the classification result.

[0148] Various assessment methods are suitable for assessment unit 460, provided the assessment method is an index for measuring an error between an output label and a correct answer label. For example, a basic assessment method could use a cross-entropy error function or a quadratic error function.

[0149] The evaluation unit 460 outputs an evaluation result to the model update unit 470.

[0150] The model update unit 470 updates the model using the evaluation result from the evaluation unit 460.

[0151] In particular, the model update unit 470 updates the learning parameter as the model stored in the memory unit 210A of the memory device 200A by rewriting the learning parameter.

[0152] For example, when using the evaluation result by the cross-entropy error function, the Model Update Unit 470 updates the parameter W in a direction in which the value of the error function is minimized. The direction in which the function value is minimized is the direction of a gradient in which the function decreases when the error function is differentiated by the model's parameter W. It should be noted that the Model Update Unit 470 can continue the update until the value of the error function converges, or it can stop the update halfway through if a predetermined condition is met.

[0153] Furthermore, the model update unit 470 performs the update using a general optimization method, such as a stochastic gradient descent method or a Newton method. It should be noted that at the time of gradient calculation, the gradients of all model parameters are calculated sequentially, for example, using a method called backpropagation.

[0154] An example of an internal configuration of the rotational invariance conversion unit 430 is explained.

[0155] Here, the internal configuration of the rotational invariance conversion unit 430 is obtained by including the orthogonalization plane unit 331 and the projection plane unit 332 in the rotational invariance conversion unit 330, which is already in Fig. 2, which was described, is replaced by an orthogonalization plane unit 431 and a projection plane unit 432, and is not shown.

[0156] The rotational invariance conversion unit 430 comprises an orthogonalization plane unit 431 and a projection plane unit 432.

[0157] Similar to the orthogonalization plane unit 331 described above, the orthogonalization plane unit 431 calculates a basis vector for each of k coordinates that specify a point (each of N points) in the point cloud information and performs a conversion in such a way that the k basis vectors are orthogonal to each other in order to produce an (N × k) orthonormal basis matrix.

[0158] Similar to the projection plane unit 332 described above, the projection plane unit 432 calculates a projection matrix that specifies a rotational invariance feature, using an orthonormal basis matrix.

[0159] The orthogonalization processing by the orthogonalization plane unit 431 is similar to the orthogonalization processing by the orthogonalization plane unit 331 described above, and its more detailed description is omitted here.

[0160] In a case where the rotation invariance conversion unit 430 is configured in this way, the inference unit 440 identifies the point cloud specified in the point cloud information using the projection matrix and the model that specify the rotation invariance feature.

[0161] The following describes an internal configuration of the Inference Unit 440.

[0162] The inference unit 440 extracts a point cloud feature based on the result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, using the projection matrix and the model that specify the rotational invariance feature, and identifies the point cloud using this feature. Specifically, the point cloud is classified.

[0163] The point cloud feature constructed by inference unit 440 is similar to the point cloud feature constructed by inference unit 340 described above, so a more detailed description is omitted here.

[0164] In this way, the point cloud information in the present disclosure can be identified based on the point cloud shape-specific features that do not depend on the coordinates of the point cloud.

[0165] The processing of the point cloud identification device is carried out with reference to Fig. 8 explained.

[0166] Fig. Figure 8 is a flowchart that illustrates an example of processing by the point cloud learning device 400.

[0167] After processing begins, the Point Cloud Learning Unit 400 first determines whether learning will be performed (step ST100). Specifically, the Point Cloud Learning Unit 400 selects whether the learning parameter should be learned from the beginning. For example, the Point Cloud Learning Unit 400 checks predefined setting information and determines whether to learn from the beginning or to use the learning parameters of the previously saved model.

[0168] If it is determined that the learning should be carried out (step ST100 “YES”), the point cloud learning facility 400 continues with the processing of step ST120.

[0169] In the case of learning from the beginning, the point cloud learning device 400 can initialize the learning parameters randomly or can initialize the learning parameters using any commonly used initialization method.

[0170] If the point cloud learning facility 400 determines not to perform learning (step ST100 “NO”), the point cloud learning facility 400 performs model acquisition processing (step ST110).

[0171] After determining whether to perform learning (step ST100 “YES”) or to execute model acquisition processing (step ST110), the point cloud learning facility 400 performs the point cloud acquisition processing (step ST120).

[0172] Specifically, the point cloud acquisition unit 420 in the point cloud learning unit 400 acquires point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions from the point cloud input unit 100. The point cloud acquisition unit 420 outputs the acquired point cloud information.

[0173] The point cloud learning device 400 performs rotation invariance conversion processing (step ST130).

[0174] In particular, the rotation invariance conversion unit 430 in the point cloud learning unit 400 performs an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and calculates the rotation invariance feature using data after the orthogonalization.

[0175] A specific example of rotational invariance conversion processing is explained. It should be noted that the rotational invariance conversion processing by the Point Cloud Learning Facility 400 is similar to the rotational invariance conversion processing of Fig. 4 is, and a further illustration is omitted here.

[0176] After starting the rotation invariance conversion processing, the rotation invariance conversion unit 430 performs the orthogonalization processing (step ST131), as shown in Fig. 4 shown.

[0177] Specifically, the orthogonalization plane unit 431 in the rotational invariance conversion unit 430 computes a basis vector for each of k coordinates that specify a point (each of N points) in the point cloud information and performs the conversion in such a way that the k basis vectors are orthogonal to each other, in order to produce an (N × k) orthonormal basis matrix. The orthogonalization plane unit 431 outputs an orthonormal basis matrix.

[0178] The rotational invariance conversion unit 430 performs a projection processing (step ST132).

[0179] Specifically, the projection plane unit 432 acquires the orthonormal basis matrix from the orthogonalization plane unit 431 in the rotation invariance conversion unit 430. The projection plane unit 432 computes a projection matrix that specifies the rotation invariance feature using the orthonormal basis matrix.

[0180] The point cloud learning device 400 performs inference processing (step ST140).

[0181] Specifically, in the point cloud learning unit 400, the inference unit 440 first acquires the projection matrix from the projection plane unit 432 and acquires the model from the model acquisition unit 410. Next, the inference unit 440 uses the rotation invariance feature and the model to identify the point cloud specified in the point cloud information.

[0182] More specifically, the inference unit 440 extracts a point cloud feature based on a result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, using the projection matrix and the model that specify the rotational invariance feature, and identifies the point cloud using the point cloud feature.

[0183] The point cloud learning device 400 performs a result output processing (step ST150).

[0184] Specifically, the result output unit 450 in the point cloud learning unit 400 outputs the classification result to the result output unit 900 by identifying the inference unit 440.

[0185] The point cloud learning device 400 determines whether the processing sequence should be terminated (step ST160).

[0186] When it is determined to stop processing (step ST160 “YES”), the point cloud learning facility 400 stops processing.

[0187] If it is determined not to terminate processing (step ST160 “NO”), the point cloud learning facility 400 performs the model evaluation processing (step ST170).

[0188] Specifically, the evaluation unit 460 in the point cloud learning unit 400 evaluates the model using the classification result. The evaluation unit 460 then outputs an evaluation result to the model update unit 470.

[0189] The point cloud learning facility 400 performs a model update processing (step ST180).

[0190] The model update unit 470 updates the model using the evaluation result from the evaluation unit 460.

[0191] In particular, the model update unit 470 updates the learning parameter as the model stored in the memory unit 210A of the memory device 200A by rewriting the learning parameter.

[0192] After the model update processing (step ST180) has been completed, the point cloud learning device 400 continues with the processing of step ST110 and repeats the processing from step ST110 onwards.

[0193] The learning facility (point cloud learning facility) within the meaning of the present disclosure is configured as follows.

[0194] Learning facility, comprehensive: a point cloud acquisition unit to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit to acquire a model that has a learning parameter; a rotational invariance conversion unit to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotational invariance feature using data after orthogonalization; an inference unit to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; a result output unit to output a classification result through identification by the inference unit; a rating unit to evaluate the model based on the classification result; and a model update unit to update the model using an evaluation result from the evaluation unit.

[0195] Thus, the present disclosure has the effect of providing a point cloud learning device which improves the accuracy of point cloud identification.

[0196] The learning method (point cloud learning method) as defined in this disclosure is configured as follows. Learning method, comprising: a point cloud acquisition step of initiating a point cloud acquisition unit to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition step of initiating a model acquisition unit, a model that has a learning parameter, to acquire; a rotational invariance conversion step of causing a rotational invariance conversion unit to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotational invariance feature using the information after the orthogonalization; an inference step of causing an inference unit to identify a point cloud specified in the point cloud information using the rotation invariance feature and the model; a result output step of causing a result output unit to output a classification result by identification by the inference unit; an evaluation step of instructing an evaluation unit to evaluate the model using the classification result; and a model update step of causing a model update unit to update the model using an evaluation result from the evaluation unit.

[0197] Thus, the present disclosure has the effect of providing a learning method that improves the accuracy of identifying a point cloud.

[0198] The learning facility (point cloud learning facility) within the meaning of the present disclosure is further configured as follows.

[0199] Learning institution in which which has a rotational invariance conversion unit: an orthogonalization plane unit to compute a basis vector for each of k coordinates specifying a point contained in the point cloud information, and to perform a conversion in such a way that the k basis vectors are orthogonal to each other in order to generate an orthonormal basis matrix; and a projection plane unit to compute a projection matrix that specifies the rotational invariance feature using the orthonormal basis matrix, and the inference unit a point cloud specified in the point cloud information is identified using the projection matrix, which specifies the rotational invariance feature, and the model.

[0200] Thus, the present disclosure has the effect of providing a learning facility capable of efficiently calculating a rotational invariance feature of a point cloud.

[0201] Furthermore, the present disclosure achieves a similar effect to the effect described above when the configuration described above is applied to the learning procedure described above.

[0202] The learning facility (point cloud learning facility) within the meaning of the present disclosure is further configured as follows.

[0203] The inference unit in the learning facility features: extracts the inference unit, using the projection matrix, which specifies the rotational invariance feature, and the model, a point cloud feature is extracted based on a result of the orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, and a point cloud is identified using the point cloud feature.

[0204] Thus, the present disclosure has the effect that it is possible to provide a learning device that identifies point cloud information using the point cloud shape-specific features that are not dependent on the coordinates of the point cloud.

[0205] Furthermore, the present disclosure achieves a similar effect to the effect described above when the configuration described above is applied to the learning procedure described above.

[0206] This document describes a hardware configuration that implements the functions of the Point Cloud Identification Device 300 and the Point Cloud Learning Device 400 as disclosed herein.

[0207] Fig. Figure 9 is a representation showing a first example of a hardware configuration for implementing the functions of the point cloud identification device 300 and the point cloud learning device 400 according to the present disclosure.

[0208] Fig. Figure 10 is a representation showing a second example of a hardware configuration for implementing the functions of the point cloud identification device 300 and the point cloud learning device 400 according to the present disclosure.

[0209] The point cloud identification device 300 and the point cloud learning device 400 of the present disclosure are implemented by hardware, as shown in Fig. 9 or Fig. 10 shown.

[0210] As in Fig. As shown in Figure 9, the point cloud identification device 300 and the point cloud learning device 400, for example, comprise a processor 10001, a memory 10002, an input / output interface 10003 and a communication circuit 10004.

[0211] The processor 10001 and the RAM 10002 are, for example, installed in a computer.

[0212] The main memory 10002 stores a program that causes the computer to operate as a model acquisition unit 310, as a point cloud acquisition unit 320, as a rotation invariance conversion unit 330, as an inference unit 340, as a result output unit 350, as a model acquisition unit 410, as a point cloud acquisition unit 420, as a rotation invariance conversion unit 430, as an inference unit 440, as a result output unit 450, as an evaluation unit 460, as a model update unit 470, and as a control unit (not shown).When the processor 10001 reads and executes the program stored in the main memory 10002, the functions of the model acquisition unit 310, the point cloud acquisition unit 320, the rotation invariance conversion unit 330, the inference unit 340, the result output unit 350, the model acquisition unit 410, the point cloud acquisition unit 420, the rotation invariance conversion unit 430, the inference unit 440, the result output unit 450, the evaluation unit 460, the model update unit 470 and the control unit (not shown) are implemented.

[0213] Furthermore, a storage unit, which is not shown, is implemented by the main memory 10002 or another main memory, which is not shown.

[0214] For example, the 10001 processor uses a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, a microcontroller, a digital signal processor (DSP), or the like.

[0215] The working memory 10002 can, for example, be a non-volatile or volatile semiconductor memory, such as random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or the like, a magnetic disk such as a hard disk or a flexible disk, an optical storage disk such as a compact disc (CD), or a digital versatile disc (DVD), or a magneto-optical disk.

[0216] The processor 10001 and the main memory 10002 are connected in a state in which data can be transferred to each other. Furthermore, the processor 10001 and the main memory 10002 are connected in a state in which data can be transferred bidirectionally with other hardware via an input / output interface 10003.

[0217] Furthermore, a communication unit (not shown) is implemented by the communication circuit 10004.

[0218] Alternatively, the functions of the model acquisition unit 310, the point cloud acquisition unit 320, the rotation invariance conversion unit 330, the inference unit 340, the result output unit 350, the model acquisition unit 410, the point cloud acquisition unit 420, the rotation invariance conversion unit 430, the inference unit 440, the result output unit 450, the evaluation unit 460, the model update unit 470, and the control unit, which is not shown, can be replaced by a Fig. The dedicated processing circuit 20001 shown in the diagram must be implemented.

[0219] For example, the processing circuit 20001 uses a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a programmable logic device (PLD), a field-programmable gate array (FPGA), a system-on-a-chip (SoC), a large-scale system integration (LSI), or the like.

[0220] Furthermore, a storage unit, which is not shown, is implemented by the main memory 20002 or another main memory, which is not shown.

[0221] The main memory 20002 can be, for example, a non-volatile or volatile semiconductor memory, such as random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or the like, a magnetic disk such as a hard disk or a flexible disk, an optical storage disk such as a compact disc (CD), or a digital versatile disc (DVD), or a magneto-optical disk.

[0222] The processing circuit 20001 and the main memory 20002 are connected in a state in which data can be transferred to each other. Furthermore, the processor 20001 and the main memory 20002 are connected in a state in which data can be transferred bidirectionally with other hardware via an input / output interface 20003.

[0223] Furthermore, a communication unit (not shown) is implemented by the communication circuit 20004.

[0224] It should be noted that the functions of the model acquisition unit 310, the point cloud acquisition unit 320, the rotation invariance conversion unit 330, the inference unit 340, the result output unit 350, the model acquisition unit 410, the point cloud acquisition unit 420, the rotation invariance conversion unit 430, the inference unit 440, the result output unit 450, the evaluation unit 460, the model update unit 470 and the control unit, which is not shown, may be implemented by different processing circuits, or collectively by one processing circuit.

[0225] Alternatively, some functions of the model acquisition unit 310, the point cloud acquisition unit 320, the rotation invariance conversion unit 330, the inference unit 340, the result output unit 350, the model acquisition unit 410, the point cloud acquisition unit 420, the rotation invariance conversion unit 430, the inference unit 440, the result output unit 450, the evaluation unit 460, the model update unit 470 and the control unit, which is not shown, can be implemented by the processor 10001 and the main memory 10002, and the remaining functions can be implemented by the processing circuit 20001.

[0226] It should be noted that the present disclosure allows for free combination of the respective embodiments, modification of any component of any embodiment, or omission of any component in any embodiment within the scope of the invention. For example, the first embodiment and the second embodiment can be combined such that the result output unit outputs the result of the point cloud identification via the evaluation unit to the result output device and the model update unit. In this way, the effects of both the first embodiment and the second embodiment can be achieved. INDUSTRIAL APPLICABILITY

[0227] Both the point cloud identification device and the learning device as described in this disclosure can improve the accuracy of point cloud identification and are therefore suitable for use in point cloud identification in technologies such as vehicle control and driver assistance. REFERENCE MARK LIST

[0228] 1: Point Cloud Identification System, 2: Point Cloud Learning System, 100: Point Cloud Input Unit, 200, 200A: Storage Unit, 210, 210A: Storage Unit, 300: Point Cloud Identification Unit, 310: Model Acquisition Unit, 320: Point Cloud Acquisition Unit, 330: Rotational Invariance Conversion Unit, 331: Orthogonalization Plane Unit, 332: Projection Plane Unit, 340: Inference Unit, 350: Result Output Unit, 400: Point Cloud Learning Unit, 410: Model Acquisition Unit, 420: Point Cloud Acquisition Unit, 430: Rotational Invariance Conversion Unit, 431: Orthogonalization Plane Unit, 432: Projection Plane Unit, 440: Inference Unit 450: Result output unit, 460: Evaluation unit, 470: Model update unit, 900: Result output device

Claims

[1] Point cloud identification device (300), comprising: a point cloud acquisition unit (320) to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit (310) to acquire a model that has a learning parameter; a rotation invariance conversion unit (330) to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotation invariance feature using data after orthogonalization; an inference unit (340) to identify a point cloud specified in the point cloud information using the rotational invariance feature and the model; and a result output unit (350) to output a classification result by identification by the inference unit (340). [2] Point cloud identification device (300) according to claim 1, wherein the rotation invariance conversion unit (330) comprises: an orthogonalization plane unit (331) to compute a basis vector for each of k coordinates specifying a point contained in the point cloud information, and to perform a conversion in such a way that the k basis vectors are orthogonal to each other in order to produce an orthonormal basis matrix; and a projection plane unit (332) to compute a projection matrix that specifies the rotational invariance feature using the orthonormal basis matrix, and an inference unit (340) that identifies a point cloud specified in the point cloud information using the projection matrix which specifies the rotational invariance feature and the model. [3] Point cloud identification device (300) according to claim 2, wherein the inference unit (340), using the projection matrix, which specifies the rotational invariance feature, and the model, a point cloud feature is extracted based on a result of orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, and the point cloud is identified using the point cloud feature. [4] Learning facility (400), comprising: a point cloud acquisition unit (420) to acquire point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; a model acquisition unit (410) to acquire a model that has a learning parameter; a rotation invariance conversion unit (430) to perform an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and to compute a rotation invariance feature using data after orthogonalization; an inference unit (440) to identify a point cloud specified in the point cloud information using the rotation invariance feature and the model; an output unit (450) to output a classification result by identification by the inference unit (440); a rating unit (460) to evaluate the model using the classification result; and a model update unit (470) to update the model using an evaluation result by the evaluation unit (460). [5] Learning device (400) according to claim 4, wherein the rotation invariance conversion unit (430) comprises: an orthogonalization plane unit (431) to compute a basis vector for each of k coordinates specifying a point contained in the point cloud information, and to perform a conversion in such a way that the k basis vectors are orthogonal to each other in order to produce an orthonormal basis matrix; and a projection plane unit (432) to compute a projection matrix that specifies the rotational invariance feature using the orthonormal basis matrix, and an inference unit (440) that identifies a point cloud specified in the point cloud information using the projection matrix which specifies the rotational invariance feature and the model. [6] Learning facility (400) according to claim 5, wherein the inference unit (440), using the projection matrix, which specifies the rotational invariance feature, and the model, a point cloud feature is extracted based on a result of orthogonal projection of a component specified by the learning parameter onto a subspace spanned by the orthonormal basis matrix in an N-dimensional space, and the point cloud is identified using the point cloud feature. [7] Point cloud identification procedure performed by a point cloud identification device, comprising: Acquire, through a point cloud acquisition unit of the point cloud identification device, point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; Acquire, through a model acquisition unit of the point cloud identification device, a model that has a learning parameter; Perform, by a rotation invariance conversion unit of the point cloud identification device, an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and calculate, by the rotation invariance conversion unit of the point cloud identification device, a rotation invariance feature using information after the orthogonalization; Identify, by an inference unit of the point cloud identification device, a point cloud specified in the point cloud information using the rotational invariance feature and the model; and Output, through a result output unit of the point cloud identification device, of a classification result by identification by the inference unit. [8] Learning procedures for a learning device, comprising: Acquire, through a point cloud acquisition unit of the learning device, point cloud information specifying N (N ≥ 2) points in k (k ≥ 2) dimensions; Acquisition, through a model acquisition unit of the learning device, a model that has a learning parameter; Perform, by a rotational invariance conversion unit of the learning device, an orthogonalization of each of the basis vectors for each of the points specified in the point cloud information, and compute, by the rotational invariance conversion unit of the learning device, a rotational invariance feature using information after the orthogonalization; Identify, by an inference unit of the learning device, a point cloud specified in the point cloud information using the rotation invariance feature and the model; Output, through a result output unit of the learning device, of a classification result by identification by the inference unit; Evaluate, by means of an evaluation unit of the learning device, the model, using the classification result; and Update, by means of a model update unit of the learning device, the model using an evaluation result by the evaluation unit.

Images