Movement improvement device, movement improvement method, movement improvement program, and movement improvement system

Abstract

To provide a motion improvement device that can present suggestions for improving a motion in accordance with a playing motion of a user.SOLUTION: A motion improvement device comprises: a learned model 19b which has been trained to receive, as input, time-series information on a specific motion and output result information on a result which is caused by a motion; a receiving unit 12 which receives target time-series information on a target motion; a generation unit 13 which generates motion data so that a degree of achievement for a predetermined goal may be higher than an estimation result obtained by inputting the target time-series information to the learned model 19b, based on the received target time-series information; and an output unit 14 which outputs information for improving the target motion based on the generated motion data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a motion improvement device, a motion improvement method, a motion improvement program, and a motion improvement system. [Background technology]

[0002] In training sites for various sports and production sites such as factories, the movements of athletes and workers are analyzed to evaluate their skills and proficiency, and feedback is provided to the athletes and workers to help improve their abilities and proficiency. Patent Document 1 listed below discloses a learning system for improving proficiency in sports, etc.

[0003] In this learning system, standard movements obtained from the movements of multiple experts are acquired from a server, and the acquired standard movements and the user's movements are displayed side by side or overlaid, allowing the user to check how to move the body using the standard movements as a model. The standard movements to be compared with the user's movements are created by averaging the data of multiple experts, and are prepared for each attribute, such as gender, age, weight, and height. Therefore, the user can select and learn standard movements that suit their own attributes. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2017-64095 A Summary of the Invention [Problem to be solved by the invention]

[0005] By the way, when playing sports, etc., one often feels that one's play was bad but that one does not know how to improve. In addition, the content of the improvement varies depending on the situation and movement at the time of playing, and it is difficult to determine it uniformly. For example, when playing table tennis, if one fails to hit a ball that is coming towards the player, the content of the improvement will be different from if one fails to hit a ball that is away from the player.

[0006] The learning system in Patent Document 1 creates standard movements for target actions by averaging the data of multiple experts, making it difficult to prepare standard movements that match the various situations and movements that users may experience when playing, and there is room for improvement.

[0007] The present invention provides a motion improvement device, a motion improvement method, a motion improvement program, and a motion improvement system that can present motion improvement points that correspond to the user's movements when playing. [Means for solving the problem]

[0008] A motion improvement device according to one aspect of the present disclosure includes a trained model that is trained to receive time series information related to a specific motion as input and to output result information related to a result resulting from the motion; a receiving unit that receives target time series information related to a target motion; a generating unit that generates motion data based on the received target time series information, the motion data having a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; and an output unit that outputs information for improving the target motion based on the generated motion data.

[0009] According to this aspect, based on result information estimated by inputting time-series information regarding the target movement into a learned model, movement data that has a higher degree of achievement of a specified objective than the result of the result information is generated starting from the target motion data, and information for improving the target movement can be output based on the generated movement data.

[0010] In the above aspect, the generation unit may calculate the operational data from the target time series information based on an optimization problem algorithm that minimizes an error between a result estimated by inputting the target time series information into a trained model and a specified objective.

[0011] According to this aspect, when calculating movement data from the target time-series information, it is possible to generate and present movement data indicating movements that have a higher degree of achievement of a specified objective than the target movement according to the algorithm of the optimization problem.

[0012] In the above aspect, the algorithm for the optimization problem may be a gradient descent method.

[0013] According to this aspect, it is possible to improve the accuracy of minimizing the error with respect to a predetermined objective by using a gradient descent algorithm.

[0014] In the above aspect, the time-series information and the motion data may be skeletal data.

[0015] According to this aspect, when performing a target movement, it becomes possible to objectively recognize movements for improving the target movement while viewing the skeletal data.

[0016] A motion improvement method according to another aspect of the present disclosure is a method executed by a processor, and includes training a model to receive time series information regarding a specific motion as input and output result information regarding a result resulting from the motion, thereby creating a trained model; receiving target time series information regarding a target motion; generating motion data based on the received target time series information, the motion data having a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; and outputting information for improving the target motion based on the generated motion data.

[0017] According to this aspect, based on result information estimated by inputting time-series information regarding the target movement into a learned model, movement data that has a higher degree of achievement of a specified objective than the result of the result information is generated starting from the target motion data, and information for improving the target movement can be output based on the generated movement data.

[0018] A motion improvement program according to another aspect of the present disclosure causes a computer to function as a trained model that is trained to receive time series information regarding a specific motion as input and to output result information regarding a result resulting from the motion; a receiving unit that receives target time series information regarding a target motion; a generating unit that generates motion data based on the received target time series information, the motion data achieving a higher degree of achievement of a specified objective than a result estimated by inputting the target time series information into the trained model; and an output unit that outputs information for improving the target motion based on the generated motion data.

[0019] According to this aspect, based on result information estimated by inputting time-series information regarding the target movement into a learned model, movement data that has a higher degree of achievement of a specified objective than the result of the result information is generated starting from the target motion data, and information for improving the target movement can be output based on the generated movement data.

[0020] A motion improvement system according to another aspect of the present disclosure is a motion improvement system including a sensor terminal and a motion improvement device, wherein the sensor terminal includes a detection unit that detects a specific motion and outputs time series information related to the motion, and a transmission unit that transmits the output time series information as target time series information related to a target motion, and the motion improvement device includes a trained model that is trained to receive the time series information as input and to output result information related to a result resulting from the motion of the time series information, a receiving unit that receives the target time series information, a generation unit that generates motion data based on the received target time series information that has a higher degree of achievement of a specified objective than a result estimated by inputting the target time series information into the trained model, and an output unit that outputs information for improving the target motion based on the generated motion data.

[0021] According to this aspect, based on result information estimated by inputting time-series information regarding the target movement into a learned model, movement data that has a higher degree of achievement of a specified objective than the result of the result information is generated starting from the target motion data, and information for improving the target movement can be output based on the generated movement data. Effect of the Invention

[0022] According to the present invention, it is possible to provide a movement improvement device, a movement improvement method, a movement improvement program, and a movement improvement system that can present movement improvements that correspond to the user's movements when playing. [Brief description of the drawings]

[0023] [Figure 1] 1 is a diagram illustrating a configuration of an action improvement system according to an embodiment of the present invention. [Diagram 2] FIG. 13 is a diagram showing an example of a screen for inputting result information. [Diagram 3] 11A and 11B are diagrams for explaining a concept when calculating action data from target motion data. [Figure 4] FIG. 13 is a diagram showing an example of an image displayed on a display of a sensor terminal based on operation data. [Diagram 5] FIG. 13 is a diagram showing an example of an image displayed on a display of a sensor terminal based on operation data. [Figure 6] FIG. 2 is a diagram illustrating an example of a hardware configuration of the operation improvement device according to the present embodiment. [Figure 7] 10 is a flowchart of a learning data storage process executed by the action improvement device according to the present embodiment. [Figure 8] 10 is a flowchart of a model learning process executed by the action improving device according to the present embodiment. [Figure 9] 10 is a flowchart of an action improvement process executed by the action improvement device according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] Hereinafter, an embodiment according to one aspect of the present invention (hereinafter, referred to as "the present embodiment") will be described with reference to the drawings. In each drawing, the same reference numerals denote the same or similar configurations.

[0025] §1 Examples of application First, an example of a scene to which the present invention is applied will be described with reference to Fig. 1. In the action improving system 100 according to this embodiment, for example, the action of the user A is photographed by a camera of a sensor terminal 20 such as a smartphone, and motion data (an example of time-series information) extracted from the photographed video is transmitted to the action improving device 10 as an action to be analyzed. Then, the action improving device 10 generates action data that achieves a higher degree of achievement of a predetermined objective than the result estimated by inputting the action to be analyzed into the learned model 19b, and outputs information for improving the action to be analyzed to the sensor terminal 20 based on the action data.

[0026] The above-mentioned action of user A corresponds to, for example, the action of throwing a dart. The above-mentioned estimated result corresponds to, for example, the location where the dart has landed and whether the dart has landed in a specific location. The above-mentioned predetermined purpose corresponds to, for example, sticking the dart in the center of the dartboard, sticking the dart in a specific location on the dartboard, etc.

[0027] The trained model 19b is an AI model trained to receive motion data representing a user's action as input and output result information relating to the result of the action. For example, the trained model 19b can receive motion data representing the action of throwing a dart as input and output result information representing whether the dart has been stuck in the center of a dartboard.

[0028] Here, the closer the accuracy of the trained model 19b is to 100%, the better. If the accuracy is low, it becomes doubtful whether the result optimized by the estimation of the AI ​​model is truly optimal in the real world. Details of the trained model 19b will be described later.

[0029] By providing the above-described action improving device 10, it is possible to present points for improving actions that correspond to the user's movements when playing.

[0030] §2 Configuration Example [Function configuration] Next, an example of the functional configuration of the action improvement system 100 according to this embodiment will be described with reference to Fig. 1. The action improvement system 100 includes a sensor terminal 20 and an action improvement device 10. The sensor terminal 20 and the action improvement device 10 are connected via a communication network N.

[0031] <Sensor terminal 20> The sensor terminal 20 is, for example, a smartphone with a camera, which captures a scene in which user A is performing a motion with the camera, extracts motion data from the captured video, and transmits the motion improving device 10.

[0032] The motion data includes, for example, skeletal data indicating the movement of the user's skeleton, coordinate value data indicating the user's movement, and sensor data indicating the user's movement, but what is transmitted to the motion improvement device 10 is not limited to motion data. Any time-series information indicating the user's movement may be transmitted, and such time-series information includes not only motion data but also videos capturing the user's movement.

[0033] The skeletal data can be detected, for example, by a skeletal detection application installed on a smartphone. The coordinate value data can be measured, for example, by a motion capture device provided in place of the sensor terminal 20. The sensor data can be measured, for example, by an acceleration sensor or a gyro sensor provided in place of the sensor terminal 20. It is preferable that the acceleration sensor or the gyro sensor is worn by the user. The video can be captured, for example, by a smartphone camera. When using a video, the motion analysis device 10 that receives the video may detect motion data such as the skeletal data.

[0034] In this embodiment, a case will be described in which skeletal data is used as motion data, as an example.

[0035] The sensor terminal 20 has, as its functional configuration, for example, a detection unit, an input reception unit, and a transmission unit. The detection unit detects the movement of the user A, and outputs motion data indicating the movement of the user A's skeleton.

[0036] The input receiving unit receives input of result information regarding the results caused by the action of user A. Note that the input of result information is received at the stage of training the AI ​​model, and it is not necessary to receive the input of result information at the stage of analyzing the target action.

[0037] The transmission unit transmits the motion data output by the detection unit and the result information accepted by the input acceptance unit to the action improvement device 10. The motion data transmitted to the action improvement device 10 is transmitted as motion data related to the analysis target action (hereinafter referred to as "target motion data").

[0038] 2 shows an example of a screen for inputting result information during the learning stage. The screen in the figure is a screen for inputting the result of throwing a dart at a dartboard. The display of the sensor terminal 20 is provided with a field Fa for inputting the coordinates of the location where the dart landed, and a field Fb for inputting a label ID for identifying the result of throwing the dart.

[0039] The coordinates can be expressed, for example, by XY coordinates with the center of the dartboard as the origin. The label ID can be set appropriately for each result of throwing a dart. For example, the label can be a label indicating the location where the dart has landed, or a label indicating whether the dart has landed in a specific location on the dartboard.

[0040] <Operation improvement device 10> 1 has, as functional components, for example, a learning unit 11, a receiving unit 12, a generating unit 13, an output unit 14, and a storage unit 19. The storage unit 19 stores, for example, learning data 19a and a trained model 19b. Each functional component will be described below.

[0041] <Learning Data 19a> The learning data 19a includes motion data, which is time-series information on a specific motion, and result information on the result caused by the motion. The motion data and the result information are stored in association with each other.

[0042] Examples of the specific motion include the motion of throwing a dart, the motion of hitting a table tennis ball, the motion of putting a golf ball, and the like.

[0043] The results of the above actions can be set for each action, for example, as follows: For the action of throwing a dart, it is possible to set the location where the dart hits and whether or not the dart hits a specific location. For the action of hitting a table tennis ball, it is possible to set the location on the table where the ball lands and whether or not the ball lands in a specific location on the table. For the action of puttingt a golf ball, it is possible to set the location where the ball stops and whether or not the ball goes into the cup.

[0044] In the learning stage, after the user performs a specific action, the user inputs the result of the action into the sensor terminal 20. Then, the input result information is transmitted to the action improving device 10.

[0045] <Trained model 19b> The trained model 19b is an AI model trained using the training data 19a, and is created by the training unit 11 described below.

[0046] <Study Section 11> The learning unit 11 creates a learned model 19b by training an AI model so that the motion data of the learning data 19a is input and the result information of the learning data 19a is output. For example, GCN (Graph Convolutional Networks) and ST-GCN (Spatial Temporal GCN) can be used as a method for training the AI ​​model.

[0047] For example, when motion data showing the action of throwing a dart is input to trained model 19b, result information showing whether or not the dart hits the center of the dartboard is output from trained model 19b.

[0048] <Receiving section 12> The receiving unit 12 receives the target motion data and result information from the sensor terminal 20. Note that the result information is information received during the learning stage.

[0049] <Generation part 13> Based on the target motion data received by the receiving unit 12, the generating unit 13 generates action data that has a higher degree of achievement of a specified purpose than the result estimated by inputting the target motion data into the trained model 19b.

[0050] The estimated result is, for example, the result that is estimated to be the most likely to occur among multiple results set as the results that may occur due to the action. Specifically, a case where "the dart hit the center of the board" and "the dart did not hit the center of the board" are set as the results of throwing a dart will be described below.

[0051] For example, if it is estimated that the probability of the dart hitting the center of the board is 40% and the probability of the dart not hitting the center of the board is 60% as the result of throwing the dart, the result "not hitting the center of the board", which has the highest probability, is output from the trained model 19b as the result of throwing the dart.

[0052] The above-mentioned predetermined objective can be defined for each action, for example, as follows. For the action of throwing a dart, the objective can be defined as "sticking the dart in the center of the dartboard" or "sticking the dart in a specific spot on the dartboard". For the action of hitting a table tennis ball, the objective can be defined as "making the ball land in a specific spot on the table". For the action of putting a golf ball, the objective can be defined as "making the ball go in the cup".

[0053] For example, if the predetermined objective is to "stick the dart in the center of the dartboard" and the estimated result is "not to stick the dart in the center of the board," the generation unit 13 generates motion data that "sticks the dart in the center of the dartboard" to a higher degree than the motion of the target motion data when the user plays. The motion data may be skeletal data or data for generating skeletal data.

[0054] Here, the generation unit 13 generates the movement data starting from the target motion data when the user played. In other words, the generation unit 13 creates improved movement data from the motion data when the user failed to approach success (achieve a predetermined objective). This makes it possible to present improvements to the movement that match the user's movements when playing. Specifically, the generation unit 13 generates the movement data as follows.

[0055] The generation unit 13 calculates action data from the target motion data based on an optimization problem algorithm that inputs the target motion data into the learned model 19b and minimizes an error function that indicates the difference between the estimated result and a specified objective.

[0056] As an algorithm for the optimization problem, for example, an algorithm for minimizing a function value such as gradient descent, steepest descent, stochastic gradient descent, mini-batch gradient descent, Adagrad, RMSProp, Adadelta, Adam, etc. As an algorithm for the optimization problem, it is preferable to use gradient descent, and it is more preferable to use steepest descent among gradient descent methods.

[0057] The concept of calculating action data from target motion data based on the steepest descent method will be described with reference to Fig. 3. This figure is a conceptual diagram showing the user's motion on the horizontal axis and the error between the motion and a predetermined objective on the vertical axis.

[0058] This conceptual diagram illustrates a case where the range RA of motion that can be realized by person A is different from the range RB of motion that can be realized by person B, and the range RC of motion that can be realized by person C. In reality, various user motions corresponding to all the motion data used when learning the learned model 19b are learned.

[0059] Point Aa indicates the result of estimating the target motion data of person A by inputting it into the learned model 19b, point Ba indicates the result of estimating the target motion data of person B by inputting it into the learned model 19b, and point Ca indicates the result of estimating the target motion data of person C by inputting it into the learned model 19b. When these points Aa, Ba, and Ca are optimized based on the steepest descent method so as to minimize the error, points Ab, Bb, and Cb are calculated as the respective motion data. The calculated points Ab, Bb, and Cb are the minimum values ​​that can be taken as the error between the motion that each user can realize and the predetermined objective.

[0060] In this way, by optimizing the results estimated from the target motion data of each user using the local solutions formed by the learned motions of each user, it is possible to effectively calculate the optimal solution (improvement point) that matches the movements of each user.

[0061] <Output section 14> The output unit 14 outputs information for improving the analysis target motion, based on the motion data generated by the generation unit 13.

[0062] The sensor terminal 20 displays the skeletal data by superimposing it on the movement of the user captured by the camera, based on the information for improving the analysis target motion output from the output unit 14. This will be specifically described below.

[0063] As shown in FIGS. 4 and 5, on the display of the sensor terminal 20, skeletal data Mb based on information for improving the analysis target motion is displayed superimposed on an image Ma of a user repeatedly throwing a dart.

[0064] Figure 4 shows skeletal data Mb of the user holding a dart in his or her right hand with his or her right elbow bent, superimposed on the user's image Ma, and Figure 5 shows skeletal data Mb of the user throwing a dart with his or her right elbow extended.

[0065] By providing such footage, it becomes possible to objectively recognize areas for improvement in movements that correspond to the user's movements when actually throwing a dart.

[0066] In addition, when displaying the skeletal data based on the information for improving the analysis target motion, the skeletal data may be displayed in a manner not limited to being superimposed on the user's motion, but may be displayed in a manner arranged around the user. In addition, the skeletal data may be displayed in accordance with the user's motion captured by a camera. For example, a video used for extracting the motion data may be saved, and when the video is played, the skeletal data may be displayed in accordance with the user's motion in the video.

[0067] [Hardware configuration] Next, an example of the hardware configuration of the behavior improvement device 10 according to this embodiment will be described with reference to FIG. 6. The behavior improvement device 10 has a CPU (Central Processing Unit) 10a corresponding to the arithmetic device, a RAM (Random Access Memory) 10b corresponding to the storage unit 19, a ROM (Read only Memory) 10c corresponding to the storage unit 19, a communication device 10d, an input device 10e corresponding to the receiving unit 12, and an output device 10f corresponding to the output unit 14. These components are connected to each other via a bus so that they can transmit and receive data to each other. Note that, although a case where the behavior improvement device 10 is configured by one computer will be described in this embodiment, the behavior improvement device 10 may be realized using multiple computers.

[0068] The CPU 10a executes programs stored in the RAM 10b or the ROM 10c, and functions as a control unit that performs calculations and processing of data. The CPU 10a receives various input data from the input device 10e or the communication device 10d, and outputs the results of calculations of the input data from the output device 10f or stores them in the RAM 10b or the ROM 10c. The CPU 10a in this embodiment executes a program (a behavior improvement program) that generates behavior data that achieves a higher degree of achievement of a predetermined purpose than the results estimated from the behavior of the motion data, and outputs information for improving the behavior of the motion data. The calculation device is not limited to a CPU, and may be another processor such as a GPU (Graphics Processing Unit).

[0069] The RAM 10b is made up of, for example, a semiconductor memory device, and stores rewritable data.The ROM 10c is made up of, for example, a semiconductor memory device, and stores readable but non-rewritable data.

[0070] The communication device 10d is an interface that connects the action improving device 10 to an external device. The communication device 10d is connected to the sensor terminal 20 via a communication network such as a LAN (Local Area Network) or the Internet, and receives motion data from the sensor terminal 20.

[0071] The input device 10e is an interface that accepts data input from a user, and can include, for example, a keyboard, a mouse, and a touch panel.

[0072] The output device 10f is an interface that outputs the results of calculations performed by the CPU 10a, etc. The output device 10f may be an interface that visually displays information. In this case, the output device 10f may be configured, for example, with an LCD (Liquid Crystal Display).

[0073] The behavior improvement program may be provided by being stored in a computer-readable storage medium such as RAM 10b or ROM 10c, or may be provided via a communication network connected by a communication device 10d. In the behavior improvement device 10, the CPU 10a executes the behavior improvement program, thereby realizing the operations of the learning unit 11, receiving unit 12, generating unit 13, and output unit 14 shown in FIG. 1. Note that these physical configurations are merely examples, and do not necessarily have to be independent configurations. For example, the behavior improvement device 10 may be provided with an LSI (Large-Scale Integration) in which the CPU 10a is integrated with the RAM 10b and ROM 10c.

[0074] §3 Example of operation An example of a learning data storage process executed by the action improving device 10 according to this embodiment will be described with reference to FIG.

[0075] First, the receiving unit 12 of the motion improving device 10 receives motion data relating to a specific motion from the sensor terminal 20 (step S101).

[0076] Next, the receiving unit 12 of the motion improving device 10 receives, from the sensor terminal 20, result information corresponding to the motion data received in the above step S101 (step S102).

[0077] Next, the storage unit 19 of the action improving device 10 associates the motion data received in the above step S101 with the result information received in the above step S102, and stores them as learning data 19a (step S103). Then, the learning data storage process ends.

[0078] An example of a model learning process executed by the action improving device 10 according to this embodiment will be described with reference to FIG.

[0079] First, the learning unit 11 of the action improving device 10 trains an AI model so that the motion data of the learning data 19a is input and the result information of the learning data 19a is output (step S201).

[0080] Next, the storage unit 19 of the action improving device 10 stores the trained model 19b created by the training in the above step S201 (step S202), and the model training process is then terminated.

[0081] An example of the action improving process executed by the action improving device 10 according to the present embodiment will be described with reference to FIG.

[0082] First, the receiving unit 12 of the action improving device 10 receives target motion data related to the action to be analyzed from the sensor terminal 20 (step S301).

[0083] Next, the generation unit 13 of the action improvement device 10 generates action data based on the target motion data received in the above step S301, which has a higher degree of achievement of the objective than the result estimated by inputting the target motion data into the learned model 19b (step S302).

[0084] Next, the output unit 14 of the action improving device 10 outputs information for improving the action to be analyzed based on the action data generated in the above step S302 (step S303), and the action improving process ends.

[0085] This makes it possible to display on the display of the sensor terminal 20 skeletal data based on information for improving the action to be analyzed.

[0086] As described above, according to the motion improvement device 10 of this embodiment, based on result information estimated by inputting the target motion data into the learned model 19b, motion data that will achieve a higher degree of achievement of a specified objective than the result of that result information is generated starting from the target motion data, and based on the generated motion data, it is possible to output information for improving the motion to be analyzed.

[0087] In other words, starting from the target motion data obtained when the user plays, movement data that minimizes the difference from a specified objective is generated, and based on that movement data, information for improving the movements made by the user when playing can be presented to the user.

[0088] Therefore, the action improving device 10 according to the present embodiment can present to the user action improvements that correspond to the user's actions when playing.

[0089] The embodiment of the present invention may be described as in the following appendices. However, the embodiment of the present invention is not limited to the form described in the appendices. In addition, the embodiment of the present invention may be a form in which the descriptions between the appendices are replaced or combined.

[0090] [Appendix 1] A trained model (19b) that is trained to input time series information regarding a specific action and output result information regarding the result caused by the action; A receiving unit (12) that receives target time-series information related to a target action; a generation unit (13) that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model (19b); an output unit (14) that outputs information for improving the target motion based on the generated motion data; A behavior improvement device (10) comprising:

[0091] [Appendix 2] In the above aspect, the generation unit (13) may calculate the operation data from the target time series information based on an optimization problem algorithm that minimizes an error between a result estimated by inputting the target time series information into the trained model (19b) and the specified objective. 2. The behavior improvement device (10) of claim 1.

[0092] [Appendix 3] The algorithm for the optimization problem is gradient descent. 3. The behavior improvement device (10) of claim 2.

[0093] [Appendix 4] The time-series information and the motion data are skeletal data. 4. A behavior improving device (10) according to any one of appendix 1 to 3.

[0094] [Appendix 5] A method executed by a processor (10a), comprising: A model is trained so that time-series information regarding a specific action is input and result information regarding the result caused by the action is output, thereby creating a trained model (19b); receiving target time series information relating to a target motion; generating, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model (19b); outputting information for improving the target motion based on the generated motion data; Methods for improving performance, including:

[0095] [Appendix 6] Computer, A trained model (19b) that is trained to input time series information regarding a specific action and output result information regarding the result caused by the action; A receiving unit (12) for receiving target time-series information relating to a target action; A generation unit (13) that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model (19b); an output unit (14) that outputs information for improving the target motion based on the generated motion data; A behavior improvement program that acts as a.

[0096] [Appendix 7] A movement improvement system (100) including a sensor terminal (20) and a movement improvement device (10), The sensor terminal (20) A detection unit that detects a specific motion and outputs time-series information related to the motion; A transmission unit that transmits the output time-series information as target time-series information related to a target action; Equipped with The movement improving device (10) comprises: A trained model (19b) trained to input the time series information and output result information related to the result caused by the operation of the time series information; A receiving unit (12) for receiving the target time series information; a generation unit (13) that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model (19b); an output unit (14) that outputs information for improving the target motion based on the generated motion data; Equipped with A behavior improvement system (100). [Explanation of symbols]

[0097] 10...motion improvement device, 10a...CPU, 10b...RAM, 10c...ROM, 10d...communication device, 10e...input device, 10f...output device, 11...learning unit, 12...receiving unit, 13...generation unit, 14...output unit, 19...storage unit, 19a...learning data, 19b...learned model, 20...sensor terminal, 100...motion improvement system, N...communication network, A...user, Ma...user's image, Mb...skeletal data

Claims

1. A trained model that is trained to input time-series information about a specific action and output result information about the result caused by the action; A receiving unit that receives target time-series information regarding a target action; A generation unit that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; and an output unit that outputs information for improving the target motion based on the generated motion data; Equipped with The generation unit calculates the motion data from the target time series information based on an algorithm for an optimization problem that minimizes an error between a result estimated by inputting the target time series information into the trained model and the predetermined objective. Movement improvement device.

2. The algorithm for the optimization problem is gradient descent. The motion improving device according to claim 1 .

3. The time-series information and the motion data are skeletal data.

3. The operation improvement device according to claim 1 or 2.

4. 1. A processor-implemented method, comprising: A model is trained so that time-series information regarding a specific action is input and result information regarding the result caused by the action is output, thereby creating a trained model; receiving target time series information relating to a target motion; generating, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; outputting information for improving the target motion based on the generated motion data; Including, The generating of the motion data includes calculating the motion data from the target time series information based on an algorithm for an optimization problem that minimizes an error between a result estimated by inputting the target time series information into the trained model and the predetermined objective. How to improve operation.

5. Computer, A trained model that is trained to input time series information regarding a specific action and output result information regarding the result caused by the action; a receiving unit for receiving target time-series information regarding a target action; A generation unit that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; an output unit that outputs information for improving the target motion based on the generated motion data; Function as a The generation unit calculates the motion data from the target time series information based on an algorithm for an optimization problem that minimizes an error between a result estimated by inputting the target time series information into the trained model and the predetermined objective. Performance improvement program.

6. A behavior improvement system including a sensor terminal and a behavior improvement device, The sensor terminal includes: A detection unit that detects a specific motion and outputs time-series information related to the motion; A transmission unit that transmits the output time-series information as target time-series information related to a target action; Equipped with The operation improvement device includes: A trained model trained to input the time series information and output result information regarding the result caused by the operation of the time series information; A receiving unit that receives the target time series information; A generation unit that generates, based on the received target time series information, motion data that achieves a higher degree of achievement of a predetermined objective than a result estimated by inputting the target time series information into the trained model; and an output unit that outputs information for improving the target motion based on the generated motion data; Equipped with The generation unit calculates the motion data from the target time series information based on an algorithm for an optimization problem that minimizes an error between a result estimated by inputting the target time series information into the trained model and the predetermined objective. Performance improvement system.

Images