Golf swing analysis device and analysis method

Abstract

To provide an analysis system and an analysis method for a golf swing that can analyze the golf swing with a simple structure.SOLUTION: An analysis system 1 for a golf swing using a golf club 2 comprising a grip 21, a shaft 22, and a head 23 comprises a swing sensor 100 for measuring first data that is time-series data of at least one of angular velocity and / or acceleration of the golf club 2 during the golf swing for hitting a ball 4 with the golf club 2, an image generation unit for generating first image data on the basis of the first data, and an estimation unit for estimating an initial condition of the head during the golf swing by using the first image data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a golf swing analysis device and analysis method. [Background technology]

[0002] In order to analyze a golf swing, a system has been proposed in which a high-precision camera or the like is used to directly capture images of the behavior of the head of a golf club during a swing (see, for example, Patent Document 1 below). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2007-167549 DISCLOSURE OF THE INVENTION [Problem to be solved by the invention]

[0004] However, the analytical device of Patent Document 1 has a problem in that the equipment is large-scale.

[0005] The present disclosure has been devised in view of the above problems, and has as its main object to provide a golf swing analysis system and analysis method that are simple in configuration and capable of analyzing a golf swing. [Means for solving the problem]

[0006] The present disclosure relates to a system for analyzing a golf swing using a golf club having a grip, a shaft, and a head, the system including: a swing sensor for measuring first data, which is time series data of at least one of the angular velocity and acceleration of the golf club during a golf swing in which the golf club is used to strike a ball; an image generation unit for generating first image data based on the first data; and an estimation unit for estimating the initial condition of the head during the golf swing using the first image data. [Effects of the Invention]

[0007] By adopting the above-described configuration, the golf swing analysis system of the present disclosure is able to analyze a golf swing with a simple configuration. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a diagram illustrating the overall configuration of a golf swing analysis system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a block diagram of an analysis system according to an embodiment of the present invention. [Figure 3] 4 is a graph showing first data of angular velocity obtained by a swing sensor. [Figure 4] FIG. 2 is a block diagram of an image generating unit. [Figure 5] FIG. 1 is a diagram showing an area graph around the x-axis. [Figure 6] FIG. 1 is a diagram showing an area graph around the y-axis. [Figure 7] FIG. 1 is a diagram showing an area graph around the z-axis. [Figure 8] This is a line diagram obtained by converting an area graph around the x-axis into binary image data. [Figure 9] FIG. 4 is a diagram illustrating an example of first image data. [Figure 10] FIG. 10 is a block diagram showing a procedure for generating first image data (RGB color image data) in the first specific example. [Figure 11] FIG. 10 is a block diagram showing a procedure for generating primary x-axis composite image data in specific example 2. [Figure 12] FIG. 10 is a block diagram showing a procedure for generating first image data in a second specific example. [Figure 13] FIG. 11 is a block diagram showing a procedure for generating first image data in a third specific example. [Figure 14] FIG. 1 is a plan view illustrating a head speed, a face angle, and an approach angle of a golf club during a swing. [Figure 15] 1 is a side view illustrating a golf club during a swing, illustrating a dynamic loft angle and a blow angle. FIG. [Figure 16] FIG. 1 is a front view of a golf club illustrating left and right impact points and upper and lower impact points. [Figure 17] FIG. 2 is a block diagram illustrating an estimation unit. [Figure 18] 1 is a flowchart of a golf swing analysis method according to an embodiment of the present invention. [Figure 19] 10 is a graph showing correlations of the estimation unit of the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. The specific configurations shown in the embodiments and drawings are for the purpose of understanding the contents of the present disclosure, and the present disclosure is not limited to the specific configurations shown. Furthermore, in multiple embodiments, the same or common elements are designated by the same reference numerals throughout the specification, and redundant explanations will be omitted.

[0010] Fig. 1 shows an overall configuration diagram of a golf swing analysis system 1 of this embodiment. Fig. 2 shows a block configuration diagram of the analysis system 1 of this embodiment. The analysis system 1 of this embodiment analyzes the golf swing of a golfer 3, which is useful for, for example, effectively fitting a golf club 2 or giving lessons to the golfer 3.

[0011] As shown in FIGS. 1 and 2, the analysis system 1 of this embodiment is for analyzing a golf swing in which a golf club 2 is used to hit a ball 4, and includes, for example, a swing sensor 100 and a processing device 200.

[0012] [Golf Clubs] The golf club 2 has, for example, a grip 21, a shaft 22, and a head 23. In this embodiment, a wood-type golf club is shown as the golf club 2. In other embodiments, the golf club 2 may be an iron-type or a putter-type club.

[0013] [Swing sensor] The swing sensor 100 can measure first data, which is time-series data of at least one of the angular velocity and acceleration of the golf club 2 during a golf swing. As shown in FIG. 2 , the swing sensor 100 of this embodiment includes an angular velocity sensor 101 and an acceleration sensor 102. Therefore, the swing sensor 100 of this embodiment can measure, as the first data, time-series data of both the angular velocity and acceleration of the golf club 2 during a golf swing. In another aspect, the swing sensor 100 may measure, as the first data, only one of the angular velocity and acceleration of the golf club 2 during a golf swing.

[0014] 1, for example, it is desirable that the swing sensor 100 be attachable to the golf club 2. In a preferred embodiment, the swing sensor 100 of this embodiment is detachable from the grip 21 or shaft 22 of the golf club 2. It is desirable that the swing sensor 100 be small and lightweight so as not to interfere with the swing motion of the golfer 3.

[0015] The swing sensor 100 of this embodiment measures first data relating to a three-dimensional local coordinate system predefined for the golf club 2. This coordinate system has orthogonal x-, y-, and z-axes. The z-axis coincides with the axial direction of the shaft 22, and the direction from the head 23 to the grip 21 is considered positive. The y-axis is a direction along the target ball flight direction when the golf club 2 is addressed, and can also be considered the front-to-back direction of the head 23 (face-to-backface direction). On the y-axis, the direction toward the front (ball flight direction) is considered positive. The x-axis is a direction perpendicular to the y- and z-axes, and the direction from the heel side to the toe side of the head 23 is considered positive.

[0016] The swing sensor 100 of this embodiment can measure time series data of angular velocities around the three axes and accelerations in the three axial directions as the first data. That is, the swing sensor 100 can obtain a total of six types of time series data, three types of angular velocities and three types of accelerations.

[0017] FIG. 3 shows the output of the angular velocity sensor 101 as an example of the first data. FIG. 3 shows the results of a golf swing in which a golfer 3 hits a ball 4 using a golf club 2, with the horizontal axis representing time and the vertical axis representing the angular velocity of the golf club 2. The angular velocity is output around each of the x, y, and z axes. The zero mark on the horizontal axis indicates the moment when the head 23 hits the ball 4. The first data can be acquired in a time series at a predetermined sampling period (for example, 1 millisecond). Such first data is specific to an individual golf swing, This is data that represents the characteristics of the golf swing.

[0018] As shown in FIG. 2, swing sensor 100 of this embodiment may further include a communication unit 103 for providing first data to processing device 200. Communication unit 103 is preferably a wireless system so as not to interfere with the swing motion, but may also be a wired system using a cable. Swing sensor 100 may also include a removable storage medium (not shown) instead of or in addition to communication unit 103. In this case, the first data may be temporarily stored in the storage medium and then imported into processing device 200 via this storage medium.

[0019] [Processing equipment] The processing device 200 of this embodiment receives first data and executes predetermined processing. Such a processing device 200 may be realized, for example, as various general-purpose personal computers, tablet computers, smartphones, or even dedicated terminals. In this embodiment, FIG. 1 illustrates a desktop computer. In the example of FIG. 1, the processing device 200 includes a display serving as a display unit 204 and a keyboard, mouse, and the like serving as an input unit 205. The display unit 204 appropriately displays results of processing by the processing device 200, etc. Furthermore, the input unit 205 can be used to input information required to operate the processing device 200 and instructions required for the displayed results, etc.

[0020] 2, the processing device 200 includes at least an image generation unit 202 for generating first image data based on the above-mentioned first data, and an estimation unit 203 for estimating the initial condition of the head 23 during a golf swing using the first image data. As described above, the analysis system of this embodiment is characterized in that it converts the first data specific to each golf swing into first image data, and uses this first image data to estimate the initial condition of the head 23 during a golf swing (this will be described later).

[0021] The processing device 200 of this embodiment further includes a control unit 201 , a storage unit 206 , and a communication unit 207 .

[0022] The communication unit 207 can receive the first data from the communication unit 103 of the swing sensor 100. The communication unit 207 in this embodiment is a wireless system, but may be a wired system using a cable. Furthermore, the processing device 200 may be provided with a removable storage medium (not shown) instead of or in addition to the communication unit 207. In this case, the first data may be input to the processing device 200 via the storage medium.

[0023] The storage unit 206 stores, for example, a program for causing the processing device 200 to execute predetermined processing. The storage unit 206 includes, for example, a magnetic storage unit or a nonvolatile memory, and a volatile memory for temporarily storing working information.

[0024] The control unit 201 executes predetermined processing in accordance with a program stored in the storage unit 206. The control unit 201 can also store processing results and the like in the storage unit 206 or display them on the display unit 204.

[0025] [Image generation section] 4 shows a block diagram of the image generation unit 202 of the processing device 200. The image generation unit 202 generates first image data based on the received first data. As shown in FIG. 4, the image generation unit 202 of this embodiment includes, for example, an area graph creation unit 2021 and a first image data generation unit 2022. In this embodiment, the area graph creation unit 2021 and the first image data generation unit are realized by, for example, causing the control unit 201 to function using a program.

[0026] [Area Graph Creation Department] The area graph creating unit 2021 creates an area graph (surface graph) for a predetermined time range including the moment the ball is hit, out of the first data. The processing procedure of the area graph creating unit 2021 is, for example, as follows.

[0027] First, the area graph creation unit 2021 reads data for a time range for creating an area graph from the first data. This range is predetermined as a desired time range t1 including the moment when the ball 4 is hit (time 0 on the horizontal axis), as shown in FIG. 3 . The time when the ball 4 is hit can be determined by identifying a large change (peak) in the angular velocity of the golf club 2. Furthermore, such data reading is performed for each of the angular velocity and acceleration of the first data for a total of six axes, the x-axis, y-axis, and z-axis of the local coordinate system described above.

[0028] Next, the area graph creation unit 2021 normalizes each of the read data to align the scale of each data. Here, for example, the multiple types of read data corresponding to each of the x, y, and z axes are normalized by an arbitrarily determined angular velocity or acceleration value. In this embodiment, the angular velocity is normalized by ±4000 (deg / s), and the acceleration is normalized by ±320 (m / s2).

[0029] Next, the area graph creation unit 2021 creates six types of area graphs corresponding to the six types of normalized data. FIGS. 5 to 7 show examples of area graphs of angular velocity around the x-axis, y-axis, and z-axis as representative examples of the created area graphs. As can be seen from FIGS. 5 to 7, in the area graphs, the region surrounded by the horizontal axis and the angular velocity graph is colored a specific color (a single color in this embodiment). Textures may be used in place of colors in the area graphs. Similarly, the area graph creation unit 2021 can create area graphs (not shown) of acceleration in the x-axis, y-axis, and z-axis directions. Each of the created area graphs is stored in the storage unit 206.

[0030] As described above, the area graph creation unit 2021 of this embodiment creates multiple types of area graphs from the first data of one golf swing to be analyzed. These area graphs are based on the first data, and therefore contain information specific to each individual golf swing.

[0031] [First image data generation unit] The first image data generation unit 2022 generates one piece of first image data using the six types of area graphs created above. Since this first image data is also created using area data containing information specific to each golf swing, it contains characteristics specific to the golf swing being analyzed. Therefore, any image data generated using an area graph can be used as the first image data. In this embodiment, a case where RGB color image data is used will be described below as an example of the first image data.

[0032] As shown in FIG. 4, the first image data generation unit 2022 of this embodiment includes, for example, a first generation unit 2022A and a second generation unit 2022B.

[0033] [First generation part] The first generation unit 2022A generates a total of six types of image data, three types for angular velocity and three types for acceleration, corresponding to the six types of area graphs, three types for angular velocity and three types for acceleration, created by the area graph creation unit 2021. The first generation unit 2022A of this embodiment performs binarization processing on each area graph using a predetermined threshold value, and generates binary image data as image data.

[0034] FIG. 8 shows a typical example in which an area graph of the angular velocity of a golf club around the x-axis is converted into binary image data. This binary image data is a collection of pixels, m pixels wide by n pixels high. Each pixel is 1-bit data, assigned a gradation of black or white. Each generated binary image data is stored in the memory unit 206.

[0035] [Second generation part] The second generation unit 2022B generates one RGB color image data as the first image data from the multiple types of binary image data obtained by the first generation unit 2022A. Fig. 9 shows an example of such RGB color image data.

[0036] RGB color image data is a color model based on additive mixing, which reproduces a specific color by mixing the three primary colors of red (R), green (G), and blue (B). Similar to binary image data, the RGB color image data of this embodiment has m horizontal pixels and n vertical pixels. Each pixel has an R channel, a G channel, and a B channel as color channels that indicate the luminance information of each of the three primary colors. The RGB color image data of this embodiment uses 8 bits (256 gradations) x 3 = 24-bit data for the luminance information of each color channel.

[0037] The second generation unit 2022B assigns, for example, binary image data alone or composite image data obtained by combining binary image data to color channels of the RGB color model. Next, the second generation unit 2022B generates one piece of RGB color image data by combining the luminance information of each assigned color channel.

[0038] When generating RGB color image data, there are various methods for allocating the six types of binary image data to each color channel. In the present disclosure, such allocation methods are not particularly limited, and various methods can be adopted. Some specific examples are shown below.

[0039] [Example 1: Assigning one type (one axis) of image data to each color channel] Specific Example 1 is one of the simplest methods for generating RGB color image data. In Specific Example 1, to obtain one piece of RGB color image data, one type of image data (one axis in terms of the number of axial directions in the local coordinate system) is assigned (input) to each channel. For example, the second generation unit 2022B extracts three types (three axes) of binary image data from six types of binary image data, and assigns these to the R channel, G channel, and B channel to generate RGB color image data.

[0040] A more specific example is shown in Fig. 10. As the three types of binary image data, binary image data of angular velocities around the x-axis, y-axis, and z-axis are selected. These binary image data are then assigned to any one of the R channel, G channel, and B channel of the RGB color image data, i.e., a single primary color channel.

[0041] Next, the second generation unit 2022B synthesizes the single color channels of the three primary colors obtained above to generate one RGB color image data, as shown in the lower part of Fig. 10. The second generation unit 2022B stores the generated RGB color image data in the storage unit 206 as first image data.

[0042] In such a specific example 1, each color channel of the RGB color image data can include information of data of any one of the x-axis, y-axis, and z-axis of the local coordinate system (in this example, three types of image data of the angular velocity around the x-axis, y-axis, and z-axis). In the above example, three types of data regarding the angular velocity are used to generate the RGB color image data, but it is not limited thereto, and any three types of binary image data can be used from six types of binary image data. Therefore, in other embodiments, binary image data of the acceleration in the x-axis, y-axis, and z-axis directions may be used.

[0043] [Specific Example 2: Assigning Two Types (Two Axes) of Image Data to Any Color Channel] Specific Example 2 can include information of more types of image data in one RGB color image data compared to Specific Example 1. In Specific Example 2, in order to obtain one RGB color image data, two types (two axes in terms of the number of axis directions of the local coordinate system) of image data are assigned to at least one color channel. Here, an example of assigning two types (two axes in terms of the number of axis directions of the local coordinate system) of image data to each color channel is shown.

[0044] The second generation unit 2022B generates, for example, the following three composite image data based on the six types of binary image data with respect to the x-axis, y-axis, and z-axis of the local coordinate system. <X-axis Composite Image Data> Composite the binary image data of the angular velocity around the x-axis and the binary image data of the acceleration in the x-axis direction <Y-axis Composite Image Data> Composite the binary image data of the angular velocity around the y-axis and the binary image data of the acceleration in the y-axis direction <Z-axis Composite Image Data> Composite the binary image data of the angular velocity around the z-axis and the binary image data of the acceleration in the z-axis direction

[0045] Here, if two types of binary image data are simply combined, in areas where the black pixels overlap, the characteristics of each binary image data (i.e., the characteristics of the golf swing) are lost and cannot be distinguished. Therefore, in Specific Example 2, in order to generate combined image data for each axis, the second generation unit 2022B assigns the two types of binary image data to different color channels of the RGB color image data, as performed in Specific Example 1, and combines them to generate RGB image data, and then converts this RGB image data into grayscale image data.

[0046] Fig. 11 shows a more specific example of generating x-axis composite image data. As shown in Fig. 11, binary image data of angular velocity around the x-axis is assigned to, for example, the R channel and converted into R monochrome image data. Also, binary image data of acceleration in the x-axis direction is assigned to, for example, the G channel and converted into G monochrome image data. Note that a collection of blank pixels is assigned to the remaining B channel.

[0047] Next, the second generation unit 2022B generates primary x-axis composite image data by combining these single-color image data, as shown in the lower part of Fig. 11. Such primary x-axis composite image data includes information on the binary image data of the angular velocity around the x-axis and the binary image data of the acceleration in the x-axis direction in an identifiable manner.

[0048] Next, because the primary x-axis composite image data is RGB color image data, it cannot be assigned to the color channels of the RGB color image data when generating the final first image data. Therefore, the second generation unit 2022B converts the primary x-axis composite image data into, for example, 8-bit, 256-level grayscale image data. This results in x-axis composite image data as grayscale data, as shown in the upper left of FIG. 12. This x-axis composite image data can distinguishably contain binary image data of angular velocity around the x-axis and binary image data of acceleration in the x-axis direction. Furthermore, the second generation unit 2022B generates y-axis composite image data and z-axis composite image data as grayscale data, as shown in the upper left of FIG. 12, using the same procedure as described above, and stores them in the storage unit 206.

[0049] Next, the second generation unit 2022B assigns the x-axis composite image data, y-axis composite image data, and z-axis composite image data generated above as grayscale data to any of the color channels of the RGB color image data. In this embodiment, the x-axis composite image data, y-axis composite image data, and z-axis composite image data are assigned to the R channel, G channel, and B channel, respectively, and combined. As a result, one piece of RGB color image data is generated as the first image data, as shown in the lower part of FIG. 12. Then, the second generation unit 2022B stores the generated RGB color image data in the storage unit 206 as the first image data.

[0050] [Example 3: Assigning three types (three axes) of image data to one of the color channels] In Specific Example 3, more types of binary image data information can be included in one RGB color image data item than in Specific Example 1. In Specific Example 3, three types of image data (three axes in terms of the number of axial directions of the local coordinate system) are assigned to at least one color channel to obtain one RGB color image data item. In this embodiment, three types of image data (three axes in terms of the number of axial directions of the local coordinate system) are assigned to each of two color channels.

[0051] In the specific example 3, the second generating unit 2022B generates the following two pieces of composite image data using six types of binary image data based on angular velocity and acceleration. <Angular velocity composite image data> Combining binary image data of angular velocity around the x-axis, y-axis, and z-axis <Acceleration composite image data> Combining binary image data of acceleration in the x-axis, y-axis, and z-axis directions

[0052] As in Example 2, the method of combining data involves assigning each binary image data to a single-color channel of the RGB primary colors, and combining these. Next, the combined RGB image data is converted to grayscale data, and again assigned to one of the single-color channels of the RGB color image data primary colors. In this embodiment, as shown in FIG. 13, the angular velocity combined image data and the acceleration combined image data are assigned to the R channel and the G channel, respectively. Blank image data is assigned to the B channel. Next, these are combined to generate RGB color image data as the first image data.

[0053] [Estimation part] The estimation unit 203 of this embodiment can estimate the initial condition of the head 23 using the first image data generated above. Therefore, in the analysis system 1 of this embodiment, there is no need to directly observe the behavior of the head 23 of the golf club 2 during a swing using a camera or the like, and therefore the initial condition of the head can be estimated easily and with high accuracy.

[0054] In this specification, the "initial conditions of the head" refers to information that specifies the state of the head 23 during a swing or the state of hitting the ball, and includes, for example, at least one of the head speed, face angle, approach angle, dynamic loft angle, blow angle, left / right impact point, and up / down impact point. The details of these are as follows:

[0055] As shown in FIG. 14, "head speed" refers to the speed F of the head 23 immediately before striking the ball during a swing. This speed F is defined by velocities Fx, Fy, and Fz in the directions of the axes of an XYZ coordinate system defined in a three-dimensional swing space. The X axis is horizontal and aligned with the target ball flight direction D, the Z axis is vertical, and the Y axis is perpendicular to both the X and Z axes. "Face angle" refers to the angle α of the face 23a relative to a vertical plane VP perpendicular to the target ball flight direction D, as viewed from a planar perspective of the swing. "Entry angle" refers to the angle β of the path K of the head 23 during the swing relative to the target ball flight direction D, as viewed from a planar perspective of the swing. As shown in FIG. 15, "dynamic loft angle" refers to the angle γ of the face 23a relative to the vertical plane VP, as viewed from a side of the swing. "Blow angle" refers to the angle δ of the path K of the head 23 during the swing relative to a horizontal plane HP, as viewed from a side of the swing. Of these initial conditions of the head, the "face angle α" and the "dynamic loft angle γ" change during the golf swing of the head 23, and therefore may be defined as values at a specific position of the head 23 during the swing.

[0056] Also, as shown in Figure 16, the "left and right impact points" and "up and down impact points" are the positions where the ball 4 is hit with the face 23a of the head 23, and respectively mean the X coordinate value and the Y coordinate value in the XY coordinate system with the area center of gravity FC of the face 23a as the origin.

[0057] The estimation unit 203 is preferably configured using, for example, AI. The estimation unit 203 of this embodiment is a trained machine learning model. The trained machine learning model is trained in advance so that it can estimate the initial conditions of the head 23 from the first image data. As the machine learning model, a deep learning model equipped with a neural network is preferable. With such a model, the estimation unit 203 itself can extract feature amounts of the first image data and estimate the initial conditions of the head 23.

[0058] The training data for learning is prepared by collecting impact data for a plurality of actual golf swings. For example, the training data is a set of first image data for each golf swing and the initial condition of the head 23 at that time (e.g., one of the conditions such as the left or right impact point). The initial condition of the head 23 can be acquired in advance, for example, by capturing an image of the golf club 2 during a golf swing with a high-resolution camera.

[0059] The learning by the estimation unit 203 involves inputting training data into an initialized neural network to estimate the initial conditions of the head 23 as output values. The error between this estimated initial condition of the head 23 and the initial condition (correct answer) of the head 23 included in the training data is calculated, and various parameters of the neural network (weighting coefficients, bias, etc.) are updated to minimize this error. The parameters are then optimized by successively applying a large number of different patterns of training data. In addition, by changing the initial condition of the head 23 in the training data to any of the head speed, face angle, approach angle, dynamic loft angle, blow angle, or upper or lower impact point, a neural network for estimating each of these can be generated.

[0060] FIG. 17 is a conceptual diagram of a neural network model of the estimation unit 203. The configuration of the neural network model is not particularly limited, but in this embodiment, as shown in FIG. 17, a convolutional neural network (CNN) is adopted. More specifically, among CNNs, ResNet is preferable. ResNet can solve the problem of performance degradation when the number of network layers is made deeper by using a residual learning structure. The residual learning structure bypasses input to a certain layer and inputs it to a deeper layer across layers. This prevents gradient vanishing and divergence, making it possible to realize an ultra-multilayer network. In this embodiment, ResNet50 is adopted.

[0061] The neural network of this embodiment includes an input layer 2031 , an intermediate layer 2032 , and an output layer 2033 .

[0062] Information about each pixel of the first image data generated by the image generation unit 202 is input to the input layer 2031. Specifically, since the first image data is RGB color image data in this embodiment, three pieces of image data with the number of pixels (m×n) are input to the input layer 2031.

[0063] The intermediate layer 2032 includes, for example, a convolutional layer 2032A, a batch normalization layer 2032B, a pooling layer 2032C, a recursive convolutional layer 2032D, and a fully connected layer 2032E.

[0064] In the convolution layer 2032A, for example, 64 types of kernels (7 × 7 matrix filters in this example) are used to filter the first image data, thereby generating 64-channel image data (feature maps) of the first image data.

[0065] The batch normalization layer 2032B normalizes the 64-channel image data.

[0066] In the pooling layer 2032C, MAX pooling is sequentially performed on pixel groups of a matrix of a predetermined size a×b of the normalized image data, thereby further reducing the size of each image data.

[0067] In the repeated convolutional layer 2032D, a set of three convolutional layers is repeated 16 times to form a layer structure.

[0068] In addition, the data trained by ResNet50 is the residual, which is the difference between the input and output, so at the end of each set containing three convolutional layers, the skip connection x and the residual F(x) are added together.

[0069] The fully connected layer 2032E is connected from 2048 nodes to one output layer, which ultimately outputs the initial conditions of the head 23.

[0070] [How to analyze your golf swing] 18 is a flowchart illustrating a method for analyzing a golf swing according to this embodiment. The method for analyzing a golf swing according to this embodiment includes step S1 of measuring first data, which is time-series data of at least one of the angular velocity and acceleration of the golf club 2 during a golf swing in which the golf club 2 hits a ball, step S2 of generating first image data based on the first data, and step S3 of estimating an initial condition of the head 23 during the golf swing using the first image data. Steps S1 and S2 can be performed by the swing sensor 100 and the image generation unit 202 of the processing device 200, respectively, as already described. Furthermore, the estimating step S3 can be performed by an estimation unit of the processing device 200, as already described.

[0071] Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the above specific disclosure, and can be implemented with various modifications within the scope of the technical idea described in the claims. [Example]

[0072] A more specific example of the present disclosure will be described below, but it should be noted that the present disclosure is not limited to such an example. The main specifications of this example are as follows.

[0073] [Initial conditions of the head] The initial conditions for the head were head speed, left / right impact point, face angle, and approach angle, and an estimation section (machine learning model) for each was constructed.

[0074] [First data] The angular velocity and acceleration of the golf club during the golf swing were measured for each of the three axes of the local coordinate system.

[0075] [Estimation part] A neural network model using ResNet50 was constructed as the estimation section. Data from 661 actual swings was used as training data. This data was from 173 golfers, and when separated by golf club differences, there were 26 types. In addition, RGB color image data was generated as first image data from the first data according to the above-mentioned specific example 2. This RGB color image data was combined with information on the initial conditions of the head to create a training data set, and the parameters of each neural network model were optimized.

[0076] 19 is a graph showing the correlation between the true value and the estimated value of the head speed estimation unit. It is clear that the estimation accuracy of this neural network is high.

[0077] Table 1 shows the validation error (RMSE) for the left and right impact points, face angle, and approach angle estimated in this example. The validation error was calculated using the 80-division crossover method. As a comparative example, the results of an analysis system (comparative example) that captures images of the golf club head during a swing with a camera and analyzes the image data to calculate the initial conditions of the head are also shown.

[0078] [Table 1]

[0079] As is clear from Table 1, this embodiment was confirmed to be able to maintain the same level of analytical accuracy as conventional analytical systems using camera data, with a simple configuration, without using large-scale equipment such as high-resolution cameras.

[0080] [Note] The present disclosure includes the following aspects.

[0081] [Disclosure 1] A system for analyzing a golf swing using a golf club having a grip, a shaft, and a head, a swing sensor for measuring first data which is time-series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club is used to hit a ball; an image generating unit for generating first image data based on the first data; an estimation unit for estimating an initial condition of the head during the golf swing using the first image data, Golf swing analysis system. [Disclosure 2] the image generation unit includes an area graph creation unit for creating an area graph of a predetermined range of the first data including the moment when the ball is hit; A golf swing analysis system as described in Disclosure 1, including a first image data generation unit for generating the first image data using the area graph. [Disclosure 3] the swing sensor measures, as the first data, a plurality of types of time-series data corresponding to each axis of a three-dimensional local coordinate system associated with the golf club, The golf swing analysis system described in Disclosure 2, wherein the area graph creation unit creates multiple types of area graphs corresponding to each of the axes from the multiple types of time series data. [Disclosure 4] the first image data generation unit includes a first generation unit for generating a plurality of types of image data corresponding to each of the axes from the plurality of types of area graphs; A golf swing analysis system according to the third disclosure, including a second generation unit that generates one of the first image data from the plurality of types of image data. [Disclosure 5] The golf swing analysis system according to any one of Disclosures 1 to 4, wherein the first image data is RGB color image data. [Disclosure 6] The golf swing analysis system described in Disclosure 4 or 5, wherein the second generation unit assigns the multiple types of image data to the R channel, G channel, or B channel of an RGB color model to generate one RGB color image data. [Disclosure 7] The golf swing analysis system described in Disclosure 6, wherein the second generation unit assigns one type of image data from the plurality of types of image data to each of the R channel, G channel, and B channel of the RGB color model. [Disclosure 8] The golf swing analysis system described in Disclosure 6, wherein the second generation unit assigns composite image data that combines two or more types of image data from the multiple types of image data to at least one of the R channel, G channel, and B channel of the RGB color model. [Disclosure 9] The golf swing analysis system described in Disclosure 8, wherein the composite image data is grayscale data. [Disclosure 10] The golf swing analysis system according to any one of Disclosures 1 to 9, wherein the estimation unit is a trained machine learning model. [Disclosure 11] The golf swing analysis system described in Disclosure 10, wherein the machine learning model is trained to output the initial conditions of the head from the first image data. [Disclosure 12] 12. The golf swing analysis system of disclosure 10 or 11, wherein the machine learning model includes a deep learning model. [Disclosure 13] A golf swing analysis system according to any one of disclosures 1 to 12, wherein the initial conditions of the head include at least one of head speed, face angle, approach angle, dynamic loft angle, blow angle, left / right impact point, and up / down impact point. [Disclosure 14] A method for analyzing a golf swing using a golf club having a grip, a shaft, and a head, comprising: measuring first data which is time series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; generating first image data based on the first data; and estimating an initial condition of the head during the golf swing using the first image data. How to analyze your golf swing. [Disclosure 15] The golf swing analysis system described in Disclosure 14, wherein the estimating step is performed by a machine learning model trained to output the initial conditions of the head from the first image data. [Disclosure 16] A processing device for analyzing a golf swing using a golf club having a grip, a shaft, and a head, a communication unit for receiving first data, which is time-series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; a generating unit that generates first image data based on the first data; an estimation unit that estimates an initial condition of the head during the golf swing using the first image data, Processing device for analysis of golf swing. [Disclosure 17] A program for analyzing a golf swing using a golf club having a grip, a shaft, and a head, receiving first data that is time series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; generating first image data based on the first data; estimating an initial condition of the head during the golf swing using the first image data; A program for analyzing golf swings that causes a computer to execute the above. [Explanation of symbols]

[0082] 1. Analysis System 2. Golf clubs 4 balls 21 Grip 22 shaft 23 head 23a face 100 Swing Sensor 103, 207 Communications Department 200 Processing Equipment 202 Image Generation Unit 203 Estimation Department 2021 Area Graph Creation Department 2022 First Image Data Generation Unit 2022A 1st generation part 2022B 2nd generation part

Claims

1. A system for analyzing a golf swing using a golf club having a grip, a shaft, and a head, a swing sensor for measuring first data, which is time-series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club is used to hit a ball; an image generating unit for generating first image data based on the first data; an estimation unit for estimating an initial condition of the head during the golf swing using the first image data, the image generation unit includes an area graph creation unit for creating an area graph of a predetermined range of the first data including the moment when the ball is hit; a first image data generation unit for generating the first image data using the area graph; Golf swing analysis system.

2. The swing sensor measures, as the first data, multiple types of time series data corresponding to each axis of a three-dimensional local coordinate system associated with the golf club, 2. The golf swing analysis system according to claim 1, wherein the area graph creation unit creates a plurality of types of area graphs corresponding to the respective axes from the plurality of types of time series data.

3. The first image data generation unit includes: a first generation unit for generating multiple types of image data corresponding to each of the axes from the multiple types of area graphs; The golf swing analysis system according to claim 2 , further comprising: a second generation unit that generates one of the first image data from the plurality of types of image data.

4. A golf swing analysis system described in any one of claims 1 to 3, wherein the first image data is RGB color image data.

5. The golf swing analysis system described in Claim 3, wherein the second generation unit assigns the multiple types of image data to the R channel, G channel, or B channel of an RGB color model to generate one RGB color image data.

6. 6. The golf swing analysis system according to claim 5, wherein the second generation unit assigns one type of image data from the plurality of types of image data to each of the R channel, G channel, and B channel of the RGB color model.

7. The golf swing analysis system of claim 5, wherein the second generation unit assigns composite image data obtained by combining two or more types of image data from the plurality of types of image data to at least one of the R channel, G channel, and B channel of the RGB color model.

8. A golf swing analysis system as described in claim 7, wherein the synthetic image data is grayscale data.

9. A golf swing analysis system described in any one of claims 1 to 8, wherein the estimation unit is a trained machine learning model.

10. A golf swing analysis system as described in Claim 9, wherein the machine learning model is trained to output the initial conditions of the head from the first image data.

11. A golf swing analysis system as described in claim 9 or 10, wherein the machine learning model includes a deep learning model.

12. A golf swing analysis system as described in any one of claims 1 to 11, wherein the initial conditions of the head include at least one of head speed, face angle, approach angle, dynamic loft angle, blow angle, left / right impact point, and up / down impact point.

13. A method for analyzing a golf swing using a golf club having a grip, a shaft, and a head, comprising: measuring first data which is time series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; creating an area graph for a predetermined range of the first data that includes the moment when the ball is hit; generating first image data using the area graph; and estimating an initial condition of the head during the golf swing using the first image data. How to analyze your golf swing.

14. A golf swing analysis system as described in Claim 13, wherein the estimating step is performed by a machine learning model trained to output the initial condition of the head from the first image data.

15. A processing device for analyzing a golf swing using a golf club having a grip, a shaft, and a head, comprising: a communication unit for receiving first data, which is time-series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; an image generating unit that generates first image data based on the first data; an estimation unit that estimates an initial condition of the head during the golf swing using the first image data, the image generation unit includes an area graph creation unit for creating an area graph of a predetermined range of the first data including the moment when the ball is hit; a first image data generation unit for generating the first image data using the area graph; Processing device for analysis of golf swing.

16. A program for analyzing a golf swing using a golf club having a grip, a shaft, and a head, receiving first data, which is time series data of at least one of angular velocity and acceleration of the golf club during a golf swing in which the golf club hits a ball; creating an area graph for a predetermined range of the first data that includes the moment when the ball is hit; generating first image data using the area graph; estimating an initial condition of the head during the golf swing using the first image data; A program for analyzing golf swings that causes a computer to execute the above.

Images