Object shape feature amount calculation / storage device

Abstract

To more quickly calculate a feature quantity by selecting the feature quantity of the shape of an arbitrary object which can be calculated as quickly as possible from point group data expressing the shape of the object.SOLUTION: An apparatus for calculating and storing a feature amount of a shape of an object acquires point cloud data including coordinate values of points given to each part of the object, wherein the points are given such that a distance between the points decreases as a curvature in the shape of the object increases, and selects other points in the point cloud data included in a predetermined range from the point for each point in the point cloud data; For each point in the point cloud, any one of a sum, a product, and a dispersion of any one of distances between the point and other points included in a predetermined range, reciprocals of the distances, differences between coordinate values, reciprocals of the differences, an exponentiation value of the differences, and reciprocals of the average values is calculated as a feature amount of the point, and the feature amount is stored for each point in the point cloud.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a technique for calculating feature quantities of an object's shape, and more specifically to an apparatus for calculating and storing feature quantities of an object's shape using coordinate values ​​of a point cloud (point cloud data) that are assigned to the surface of the object's shape so as to represent the shape. [Background technology]

[0002] Attempts have been made to estimate any physical quantity of an object, such as stress, strain, flow velocity, and their distribution, based on the shape characteristics of the object using a machine learning model constructed according to an algorithm such as a neural network (see FIG. 7). To this end, a technique has been proposed for calculating index quantities (feature quantities) that represent the shape characteristics of any object. For example, Patent Document 1 proposes that when the shape of an object is represented by point cloud data consisting of coordinate values ​​of points assigned to each part of the object's surface, a principal component analysis is performed on the point cloud data that exists within a predetermined distance from each of multiple points included in the point cloud data to calculate eigenvectors, and the curvature of a multidimensional function that takes the point located nearest to the calculated eigenvector as an extreme point is calculated, and the shape feature quantities are calculated based on the calculated curvature. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2022-186156 Summary of the Invention [Problem to be solved by the invention]

[0004] In the above-described machine learning model that calculates any physical quantity of an object from feature quantities of the object's shape, the feature quantities of the shape may be quantities representing the shape features obtained in any manner. In this regard, when it is desired to obtain the physical quantity of an object from its shape as quickly as possible, for example, in real time, it is preferable that the feature quantities of the object's shape can also be calculated as quickly as possible. For example, when the shape of an object is represented by point cloud data consisting of coordinate values ​​of points assigned to each portion of the surface of the object's shape, if the feature quantities are set to values ​​that can be calculated by a relatively simple calculation of the coordinate values ​​of each point, the calculation of the feature quantities can be more quickly performed, and the physical quantity of the object can also be estimated more quickly.

[0005] Thus, the main objective of the present invention is to enable faster calculation of features by selecting features of the shape of an arbitrary object that can be calculated as quickly as possible from point cloud data representing the shape of the object. [Means for solving the problem]

[0006] According to one aspect of the present invention, the above problem is solved by an apparatus for calculating and storing feature quantities of an object shape, comprising: a point cloud data acquisition means for acquiring point cloud data consisting of coordinate values ​​of points assigned to each portion of the shape of the object so as to represent the shape, the points being assigned so that the intervals between the points become smaller as the curvature of the shape of the object increases; another point selecting means for selecting, for each point in the point cloud data, another point in the point cloud data whose distance from the point is included within a predetermined range; a feature amount calculation means for calculating, for each point in the point cloud data, as a feature amount of the point, any of the following: a distance between the point and other points included in the predetermined range, its inverse, a difference between coordinate values, its inverse, a sum, a sum product, and a degree of dispersion of any of the powers of the difference and their inverses; a feature amount storage means for storing the feature amount for each point in the point cloud data; This is achieved by an apparatus comprising:

[0007] In the above configuration, the "object" may be any object, such as a machine tool or its parts. The "shape feature" is a quantity representing the shape feature and may be, but is not limited to, a feature used as input for a machine learning model that calculates any physical quantity of an object. As described above, the "point cloud data" is data consisting of coordinate values ​​of points assigned to each part of the object's shape (in the case of a three-dimensional object, the coordinate values ​​may be coordinate values ​​(x, y, z) on three-dimensional axes appropriately set in the shape of the object). In particular, in the configuration of the present invention, points in the point cloud data are assigned to represent the shape of the object; the greater the curvature of the shape, the smaller the point spacing; i.e., the more drastic the changes in the shape of the object, the more densely the points are assigned. Such point cloud data may be data assigned in STL (Standard Triangulated Language) format. The "predetermined range" in the other point selection means may be set appropriately. The feature calculated by the feature calculation means is calculated for each point in the point cloud data, as described above. Specifically, the feature amount is any one of the distance between each point in the point cloud data and other points whose distance to each point is within a predetermined range, its reciprocal, the difference in coordinate values, the reciprocal of the difference, the sum of any of the powers of the difference and their reciprocals, the summation, and the degree of dispersion (variance, standard deviation, etc.). The feature amount calculated for each point is stored in a storage device, storage medium, database, etc. Each means of the device of the present invention described above may be realized by a computer device operating according to a program.

[0008] In the above configuration, for each point in point cloud data representing the shape of an arbitrary object, the sum, summation, and dispersion of any of the following are calculated as feature quantities: the distance from that point to other points within a predetermined distance range, the reciprocal of that distance, the difference in coordinate values, the reciprocal of the difference, the power of the difference, and the reciprocal of the difference. As already mentioned, in point cloud data, the greater the curvature of the object's shape, the smaller the spacing between points; that is, the more drastic the shape of the object, the denser the points are assigned. Therefore, the more drastic the shape of the object changes, the higher the density of assigned points, resulting in a larger value of the feature quantity. Therefore, the feature quantity reflects the characteristics of the object's shape. Furthermore, in each case, the feature quantity is a relatively simple combination of addition, multiplication, or division of the differences in coordinate values ​​between each point and other points within the predetermined distance, so the time required to calculate the feature quantity is short, and therefore the feature quantity can be calculated more quickly.

[0009] In the above configuration, the spacing between points in the point cloud data may be set between a predetermined minimum value that may be set as appropriate and a predetermined maximum value that may be set as appropriate. Here, the predetermined minimum value is set so as to prevent excessive computational load due to an excessively small spacing between points, and the predetermined maximum value is set so as to prevent excessively low accuracy in capturing shape features due to an excessively large spacing between points. Typically, the predetermined maximum value may be set so that an appropriate number of points exist in a flat shape, and the predetermined minimum value may be set so that the difference between the curvature obtained from the point cloud data and the curvature of the actual shape is within an acceptable range, taking into account the computational load. The predetermined range in the other point selection means may be set to the predetermined maximum value. As a result, the feature value will be at its minimum value in an area where the point spacing is at the predetermined maximum value, preventing the feature value from becoming unnecessarily high.

[0010] In the above configuration, for each point in the point cloud data, the distance between that point and other points referenced when selecting other points whose distances from that point fall within a predetermined range may be either the Euclidean distance or the Manhattan distance. In the case of the Euclidean distance, the number of other points falling within the predetermined range is reduced compared to the case of the Manhattan distance, so the computational load of the feature is reduced, while the computational load of the Manhattan distance is lighter than that of the Euclidean distance, so the computational load required to select other points is reduced. [Effects of the Invention]

[0011] Thus, the device of the present invention uses point cloud data representing the shape of an object to calculate and store feature quantities representing the shape characteristics of each part of the object, which are obtained by a relatively simple combination of addition, multiplication, or division of the coordinate differences between each point in the point cloud data and other points whose distances from each point fall within a predetermined range. This configuration eliminates the need for complex and high-load calculations, such as principal component analysis of point cloud data, singular value decomposition, inverse matrix calculation, and eigenvector calculation, thereby enabling rapid calculation of feature quantities for each part of the object. The configuration of the present invention is expected to be advantageously used in applications such as shape generation technology, which requires rapid calculation of physical quantities of an object and real-time performance, and moving object detection during autonomous driving.

[0012] Other objects and advantages of the present invention will become apparent from the following description of preferred embodiments of the invention. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a diagram schematically illustrating a computer in which the present embodiment is implemented. [Figure 2] FIG. 2(A) is an example of point cloud data assigned to an object (a pipe chair), and FIG. 2(B) is a partially enlarged view thereof. [Figure 3]3A and 3B are diagrams showing the manner in which points are assigned in regions with small curvature and regions with large curvature, respectively. 3C and 3D are diagrams explaining the difference in the number of points s that are within a predetermined distance from a certain point a in regions with small curvature and regions with large curvature, respectively. [Figure 4] FIG. 4 is a schematic diagram illustrating how feature amounts are calculated in order for each point assigned to the shape of an object. [Figure 5] FIG. 5 is a flowchart showing the processing steps of the device of this embodiment. [Figure 6] FIG. 6 is an example of a distribution map of the feature amounts of an object calculated by the device of this embodiment. [Figure 7] FIG. 7 is a schematic diagram of a machine learning model that estimates the physical quantity of an object using the feature quantity of the object calculated by the device of this embodiment. [Explanation of symbols]

[0014] 1...Computer main body, 2...Computer terminal, 3...Monitor, 4...Keyboard, mouse (input device), database 10 BEST MODE FOR CARRYING OUT THE INVENTION

[0015] Computer equipment configuration The apparatus for calculating and storing feature quantities of an object's shape according to this embodiment may be realized by a computer 1, typically used in this field, operating according to a program, as illustrated in FIG. 1 . The computer 1 is equipped with a CPU, a storage device, and an input / output (I / O) device interconnected in a typical manner by a bidirectional common bus. The storage device includes a memory PM that stores programs for executing the calculations used in this embodiment, as well as a work memory WM and a data memory DM used during the calculations. The operator's instructions to the computer 1 and the display and output of calculation results and other information are provided through a computer terminal 2 connected to the computer 1. The computer terminal 2 is typically equipped with a monitor 3 and input devices 4, such as a keyboard and a mouse. When the program is started, the operator can use the input device 4 to give various instructions and inputs to the computer 1 according to the program's procedures and the display on the monitor 3, and can visually confirm the calculation status and results from the computer 1 on the monitor 3. Although not shown, a camera may also be provided to capture an image of the object and transmit the image to the computer 1. The point cloud data representing the shape of the object obtained by the computer device 1 and the calculated feature quantities of the shape of the object may be stored in the database 10.

[0016] Calculating feature values ​​of object shape In short, the device for calculating and storing feature quantities of an object's shape according to this embodiment executes (1) a process for constructing or acquiring point cloud data representing the shape of the object, and (2) a process for calculating and storing feature quantities representing the shape characteristics of the object around each point in the point cloud data. Each process will be explained below.

[0017] (1) Construction or acquisition of point cloud data representing the shape of an object In the apparatus of this embodiment, point cloud data representing the shape of an arbitrary object is first constructed or acquired. The shape of the object is typically a shape defining the surface of a three-dimensional object, but may also be a two-dimensional shape or a shape of other dimensions. Point cloud data is data consisting of coordinate values ​​of points assigned to each portion of the object's surface so as to represent the shape of the object. In the case of a three-dimensional object, the coordinate values ​​(x, y, z) of points assigned to each portion of the surface of the shape of the object in a space spanned by three-dimensional axes, as illustrated in Figures 2(A) and 2(B). The point cloud representing the shape of such an object may be assigned to an image defining the object's contour using a program employing any algorithm in a computer device. The image of the object may be captured by a camera and imported into a computer, or may be created using a CAD tool in a computer device. Note that the point cloud data may be data assigned in STL format.

[0018] In this embodiment, the points in the point cloud data described above are arranged so that the distance between each point, i.e., the spacing between points, is shorter in areas where the object's shape changes more significantly than in areas where the object's shape changes less significantly. That is, as shown in FIGS. 3A and 3B, points are arranged more densely in areas where the object's surface has a greater curvature (FIG. 3B: curvature κ2) than in areas where the curvature is smaller (FIG. 3A: curvature κ1). Therefore, as can be seen in FIG. 2B, the density of points in curved areas (α) is higher than in flat areas (β). This allows for more accurate representation of significant shape changes by more densely arranging the points, while reducing the amount of data in areas where the shape changes less significantly by reducing the number of points. The spacing between points may be set to vary between a minimum value that may be set as appropriate and a maximum value that may be set as appropriate. The minimum value may be set appropriately so that the change in the shape of the object image can be accurately represented without increasing the number of points unnecessarily, resulting in an excessive amount of data. The maximum value may be set appropriately so that the change in the shape of the object image is not ignored due to an excessive number of points, resulting in an unnecessary increase in the amount of data.

[0019] (2) Calculating the feature values ​​for each point As described above, once point cloud data (coordinate values ​​of points assigned to each portion of the shape) representing the shape of an object is acquired, feature quantities representing the characteristics of the shape near each point in the point cloud are calculated. That is, feature quantities are calculated in order for each of the points (1 to 13) arranged along the shape, as schematically illustrated in FIG. 4, and a feature quantity is assigned to each point. Furthermore, the feature quantity of each point increases the greater the change in the shape around that point, more specifically, the greater the curvature around that point. In particular, in this embodiment, the feature quantity is set to be calculated using relatively simple calculations. In this regard, as already mentioned, in this embodiment, the number of points assigned to the shape increases as the curvature increases. 3(C) and 3(D), the feature value that increases with increasing curvature at each point may be any of the following: the distance from each point a to another point s within a predetermined range (δ) that may be set as appropriate; the reciprocal of the distance; the difference in coordinate values; the reciprocal of the difference; the sum, summation, or dispersion of any of the following: the power of the difference; and the reciprocal of the difference. For these feature values, the greater the curvature near point a, the greater the number of points s near point a, i.e., points s within the predetermined range (δ). (The number of points within the range δ is greater in FIG. 3(D), where the curvature is greater, than in FIG. 3(C), where the curvature is smaller.) Therefore, the value of the feature value increases, and the feature value reflects the shape characteristics.

[0020] The distance defining the predetermined range δ may be either the Euclidean distance or the Manhattan distance from point a. In the former case, the number of points included in the predetermined range δ is smaller than in the latter case, reducing the amount of data, but increasing the load of calculating the distance. Furthermore, the distance within the predetermined range δ may be, for example, the maximum inter-point distance, but is not limited to this. Setting the distance within the predetermined range δ to the maximum inter-point distance ensures that other points exist within the predetermined range δ, ensuring that significant feature values ​​are calculated. This is advantageous in that the feature values ​​in the region where the inter-point distance is at its maximum (flat region) are at their minimum values, allowing changes in the shape of the object to be more effectively reflected in the feature values.

[0021] Specifically, the feature amount C may be calculated using, for example, any of the following formulas. Note that xa, ya, and za are the coordinate values ​​(x, y, z) of each point, and xi, yi, and zi are the coordinate values ​​(x, y, z) of other points that are within a predetermined distance δ from each point. n is the number of other points that exist within the predetermined distance.

number

number

number

number

number

number

number

number

number

number

number

number

number

[0022] Operation of the device In the operation of the apparatus of this embodiment, as shown in the processing step SC of FIG. 5, first, using an arbitrary algorithm, points are assigned to each portion of an image of the shape of an arbitrary object so as to represent the shape, as described above, to construct point cloud data (S1: step 1). Note that point cloud data may also be used that has already been constructed. Once point cloud data is acquired, one point (principal point) from the point cloud is selected (S2: step 2), and all other points (neighboring points) that are within a predetermined distance from the principal point are selected (S3: step 3). Features are calculated using the distances or coordinate differences between the principal point and the neighboring points (S4: step 4). If there is a point for which features have not been calculated (Y in S5: step 5), that point is selected as a new principal point, and steps 2-4 are repeated. Then, when there are no more points in the point cloud for which feature values ​​have not been calculated (N in S5: step 5), the calculated feature values ​​of each point are associated with the coordinate values ​​of each point and stored in the database 10 (S6: step 6).

[0023] Thus, when the above series of processes is completed, feature amounts are calculated for each point in the image of the object's shape, and the distribution of feature amounts in the image of the object can be adjusted, as shown in Fig. 6. As can be seen from the figure, it is observed that the feature amounts (calculated using the above formula (1) with the predetermined range δ set to the maximum point interval) are higher in areas where the shape changes significantly (curved areas) than in areas where the shape changes less significantly (flat areas), and it can be seen from this that the feature amounts represent the characteristics of the shape.

[0024] Use of features The feature quantities calculated as described above for each part of an object may be used to estimate any physical quantity of the object based on the shape features of the object, such as stress, strain, flow velocity, and their distribution, using a machine learning model 100 constructed according to an algorithm such as a neural network, as shown schematically in FIG. 7. Such machine learning model 100 is configured to calculate any physical quantity σ using the coordinate values ​​(x, y, z) of each part representing the shape of the object and the feature quantities (C1, C2, ...) associated therewith as inputs. In training the machine learning model 100, the coordinate values ​​(x, y, z) and feature quantities (C1, C2, ...) of each part representing the shape of the object are used as input values, and training data is used in which the physical quantity σ obtained from the shape of the object is used as correct values, so that correct values ​​corresponding to each input value are output. When the machine learning model 100 is used, the coordinate values ​​(x, y, z) of each part representing the shape of an arbitrary object and the feature values ​​(C1, C2, ...) are input, and the physical quantity σ exhibited by the object is estimated. The feature values ​​of this embodiment are feature values ​​used as input values ​​when learning or using the machine learning model 100.

[0025] Thus, in the device of this embodiment, feature amounts are calculated using only a relatively simple combination of addition, subtraction, multiplication, and division of the distances or coordinate value differences between each point in the point cloud data and its surrounding points, making it possible to quickly calculate feature amounts of the shape of each part of an object without the need to perform complex and heavy calculations. Therefore, the configuration of this embodiment is expected to be advantageously used in applications such as shape generation technology that requires rapid calculation of physical quantities of objects and real-time performance, or moving object detection during autonomous driving.

[0026] The above description has been made in relation to the embodiments of the present invention, but it will be apparent that many modifications and changes will be readily apparent to those skilled in the art, and the present invention is not limited to the above-exemplified embodiments, but can be applied to various devices without departing from the concept of the present invention.

Claims

1. An apparatus for calculating and storing feature quantities of an object shape, comprising: a point cloud data acquisition means for acquiring point cloud data consisting of coordinate values ​​of points assigned to each portion of the shape of the object so as to represent the shape, the points being assigned so that the intervals between the points become smaller as the curvature of the shape of the object increases; another point selecting means for selecting, for each point in the point cloud data, another point in the point cloud data whose distance from the point is included within a predetermined range; a feature amount calculation means for calculating, for each point in the point cloud data, as a feature amount of the point, any of the following: a distance between the point and other points included in the predetermined range, its inverse, a difference between coordinate values, its inverse, a sum, a sum product, and a degree of dispersion of any of the powers of the difference and their inverses; a feature amount storage means for storing the feature amount for each point in the point cloud data; An apparatus comprising:

2. 2. The apparatus of claim 1, wherein the spacing between the points in the point cloud data is set between a predetermined minimum value and a predetermined maximum value, and the predetermined range is set to the predetermined maximum value.

3. 2. The apparatus according to claim 1, wherein, for each point in the point cloud data, points whose Euclidean distance or Manhattan distance to the point falls within the predetermined range are selected as the other points.

Images