Gesture input device and method

Abstract

In a first aspect of the present invention, a plurality of electromyogram signals are acquired from respective body parts corresponding to a plurality of specific muscles that make coordinated motion when a user makes a gesture, and myoelectric feature amounts are extracted from the respective electromyogram signals acquired. Then, in a learning phase, the plurality of myoelectric feature amounts extracted are used as learning data to construct a prediction model for predicting the gesture. In a prediction phase, the plurality of myoelectric feature amounts extracted are input into the prediction model to acquire a prediction result for the gesture from the prediction model, and input information corresponding to the acquired prediction result is output.

Description

Gesture input device and method

[0001] One aspect of the present invention relates to a gesture input device and method used to allow a user to input information into a computer using gestures, for example.

[0002] One method of inputting information is to input information such as commands using gestures. For example, Non-Patent Document 1 describes a method of capturing an image of a gesture using a camera and recognizing it when inputting information into a head-mounted device (HMD) or a wearable device using hand gestures, and a method of recognizing gestures by attaching acceleration sensors to the user's hands, arms, or legs. It also describes a method of recognizing hand gestures by attaching electromyographic sensors to muscle sites. Non-Patent Document 2 also describes a method of recognizing hand gestures by attaching electromyographic sensors.

[0003] Zhang, Xu, et al. "Hand recognition and virtual game control based on 3D accelerometer and EMG sensors." Proceedings of the 14th international conference on Intelligent user interfaces. 2009. Zhang, Zhen, Changxin He, and Kuo Yang. "A novel surface electromyographic signal-based hand gesture prediction using a recurrent neural network" Sensors 20.14 (2020): 3994.

[0004] However, inputting information using gestures requires larger physical movements than inputting information using a keyboard or mouse, which places a greater physical strain on the user. Also, since the entire gesture movement needs to be detected, it takes time to input information.

[0005] However, Non-Patent Documents 1 and 2 do not take into consideration such physical strain on the user or the time required for information input. Furthermore, when using electromyographic sensors to accurately recognize gestures, it is common to attach multiple electromyographic sensors to muscle areas that are directly involved in the gesture movement, but placing multiple electromyographic sensors close together places a heavy burden on the user and increasing the number of electromyographic sensors leads to increased costs.

[0006] The present invention has been made in light of the above circumstances, and its first object is to provide a technique that enables gesture recognition with high accuracy using a small number of electromyographic sensors.

[0007] A second object of the present invention is to provide a technique that can predict a gesture in a short time without waiting until the entire gesture is completed.

[0008] In order to solve the above problems, a first aspect of a gesture input device or method according to the present invention acquires electromyogram signals from body parts corresponding to specific muscles that move in coordination when a user makes a gesture, and extracts electromyogram features from the acquired electromyogram signals. Then, in a learning phase, the extracted electromyogram features are used as training data to construct a prediction model for predicting the gesture. Meanwhile, in a prediction phase, the extracted electromyogram features are input to the prediction model to acquire a prediction result of the gesture from the prediction model, and input information corresponding to the acquired prediction result is output.

[0009] According to a first aspect of the present invention, by focusing on the fact that a specific plurality of muscles move in coordination when a user makes a gesture, the gesture is predicted based on myoelectric feature quantities extracted from the electromyogram signals of the specific plurality of muscles. Therefore, it is sufficient to attach electromyographic sensors only to body parts corresponding to the specific plurality of muscles, and it is possible to predict gestures using only the minimum number of electromyographic sensors necessary without imposing a heavy burden on the user.

[0010] In addition, a second aspect of the present invention sets a prediction target window having a time width shorter than the entire movement period of the gesture for the electromyogram signal, and extracts the myoelectric feature based on the signal obtained in the prediction target window of the electromyogram signal.

[0011] According to a second aspect of the present invention, the myoelectric feature is extracted in a prediction window set to have a time width shorter than the entire movement period of the gesture, and the gesture is predicted based on the myoelectric feature. This makes it possible to predict the gesture without waiting until the entire movement is completed, thereby shortening the time required for information input. In other words, predictive input of information becomes possible.

[0012] That is, according to the first aspect of the present invention, it is possible to provide a technique that enables gesture recognition with high accuracy using a small number of electromyographic sensors.

[0013] Furthermore, according to the second aspect of the present invention, it is possible to provide a technique that can predict a gesture in a short time without waiting until the entire movement is completed.

[0014] Fig. 1 is a diagram showing an example of a system including a gesture input device according to an embodiment of the present invention. Fig. 2 is a block diagram showing an example of a hardware configuration of a gesture input device according to an embodiment of the present invention. Fig. 3 is a block diagram showing an example of a software configuration of a gesture input device according to an embodiment of the present invention. Fig. 4 is a flowchart showing an example of a processing procedure and processing content of a prediction model construction process executed by a control unit of the gesture input device shown in Fig. 3. Fig. 5 is a flowchart showing an example of a processing procedure and processing content of a gesture prediction process executed by a control unit of the gesture input device shown in Fig. 3.

[0015] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0016] [One Embodiment] (Configuration Example) (1) System FIG. 1 is a diagram showing an example of a system including a gesture input device according to one embodiment of the present invention.

[0017] In FIG. 1, a user US making a gesture has electromyographic sensors SS1 to SSn attached to body parts corresponding to a plurality of specific muscles that move in coordination with one another when making a gesture.

[0018] For example, when five types of gestures, namely, clicking, clenching, and reaching (forward, sideways, and upward), are performed, it has been confirmed that five muscles, namely, the index finger and little finger on the back of the hand, the forearm, the upper arm, and the shoulder, move in coordination, and electromyographic sensors are attached to the body parts corresponding to these five muscles. However, it is also possible to attach electromyographic sensors only to the body parts corresponding to the index finger and little finger on the back of the hand and the forearm out of the above five muscles.

[0019] The electromyogram sensors SS1 to SSn measure electromyograms (EMG), which are bioelectric potential signals generated by muscle activity, and output the measured EMG signals.

[0020] In one embodiment, the system receives electromyogram signals output from the electromyogram sensors SS1 to SSn using a gesture input device GT. The gesture input device GT recognizes gestures performed by a user US based on the electromyogram signals and generates input information corresponding to the recognized gestures. The gesture input device GT then transmits the input information to a server SV via a network NW, for example.

[0021] The server device SV has a function of managing a community in the metaverse, for example, and controls the movement of an avatar that acts as an avatar of the user US based on input information transmitted from the gesture input device GT.

[0022] The network NW includes a wide area network with the Internet at its core, and an access network for accessing the wide area network. Examples of the access network include, but are not limited to, a public communication network using wired or wireless connections, a local area network (LAN) using wired or wireless connections, and a cable television (CATV) network.

[0023] (2) Gesture Input Device GT FIGS. 2 and 3 are block diagrams showing an example of the hardware configuration and software configuration, respectively, of the gesture input device GT according to an embodiment of the present invention.

[0024] The gesture input device GT is configured by a portable terminal that can be carried by the user US, such as a smartphone or a wearable terminal, etc. Note that a personal computer or the like may also be used as the gesture input device GT.

[0025] The gesture input device GT includes a control unit 1 that uses a hardware processor such as a central processing unit (CPU), and a storage unit having a program storage unit 2 and a data storage unit 3, a sensor interface (hereinafter, interface will be referred to as I / F) unit 4, and a communication I / F unit 5 are connected to the control unit 1 via a bus 6.

[0026] The aforementioned plurality of myoelectric sensors SS1 to SSn are connected, for example, by signal cables, to the sensor I / F unit 4. The sensor I / F unit 4 receives the electromyogram signals output from the plurality of myoelectric sensors SS1 to SSn and converts them into digital signals at a predetermined sampling rate.

[0027] The communication I / F unit 5 performs data communication via the network NW in accordance with a communication protocol defined by the network NW under the control of the control unit 1, and is used to transmit input information from the gesture input device GT to the server device SV.

[0028] The program storage unit 2 is configured, for example, by combining a non-volatile memory such as a HDD (Hard Disk Drive) or SSD (Solid State Drive) as a storage medium that can be written to and read at any time, with a non-volatile memory such as a ROM (Read Only Memory), and stores various programs necessary to execute various control processes according to one embodiment of the present invention, in addition to middleware such as an OS (Operating System).

[0029] The data storage unit 3 is configured, for example, by combining a non-volatile memory such as an HDD or SSD, which can be written to and read from at any time, with a volatile memory such as a RAM (Random Access Memory) as a storage medium, and a prediction model storage unit 31 is provided in the storage area.

[0030] The prediction model storage unit 31 stores data constituting a prediction model constructed by the control unit 1 through the prediction model construction process.

[0031] The control unit 1 includes, as processing functions for executing one embodiment of the present invention, an electromyogram acquisition processing unit 11, an electromyogram feature extraction processing unit 12, a prediction model construction processing unit 13, a gesture prediction processing unit 14, and an input information output processing unit 15.

[0032] These processing units 11 to 15 are all realized by causing a hardware processor in the control unit 1 to execute application programs stored in the program storage unit 2. Note that some or all of the processing units 11 to 15 may be realized using hardware such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit).

[0033] During the learning phase, the electromyogram acquisition processing unit 11 acquires, as learning data via the communication I / F unit 5, electromyogram signals of specific muscles that have been measured in the past for each of multiple expected gestures, together with label information indicating the type of gesture, from, for example, a database on the Web.

[0034] In addition, the learning data may be electromyogram signals output from electromyogram sensors SS1 to SSn attached to the user US, and in this case, label information representing the type of gesture is input by the user US from the input device of the gesture input device.

[0035] In addition, in the prediction phase, the electromyogram acquisition processing unit 11 acquires electromyogram signals of specific muscles output from the electromyogram sensors SS1 to SSn of the user US as sampling data via the sensor I / F unit 4.

[0036] The myoelectric feature extraction processing unit analyzes the sampled data of the electromyogram signals of the specific muscles acquired by the electromyogram acquisition processing unit 11 using a band-pass filter and processing for calculating effective values, etc. Then, it extracts myoelectric feature values ​​of the multiple muscles based on the analyzed signals. An example of the myoelectric feature extraction processing will be described in the operation example.

[0037] In the learning phase, the prediction model construction processing unit 13 uses the myoelectric feature quantities of the specific muscles extracted by the myoelectric feature quantity extraction processing unit 12 as learning data and label information indicating the gesture type as training data to learn a prediction model. Then, data representing the learned prediction model is stored in the prediction model storage unit 31.

[0038] In the prediction phase, the gesture prediction processing unit 14 inputs the myoelectric feature extracted by the myoelectric feature extraction processing unit 12 as explanatory variables into the prediction model stored in the prediction model storage unit 31, and obtains information representing the gesture prediction result output from the prediction model as a target variable.

[0039] The input information output processing unit 15 generates input information represented by the gesture based on information representing the gesture prediction result output from the gesture prediction processing unit 14, and transmits the generated input information to the server device SV from the communication I / F unit 5 via the network NW. An example of the input information will be described in an operation example.

[0040] (Example of Operation) Next, an example of operation of the gesture input device GT configured as above will be described.

[0041] (1) Learning Phase FIG. 4 is a flowchart showing an example of a series of processing steps and processing contents related to the prediction model construction processing executed by the control unit 1 of the gesture input device GT.

[0042] (1-1) Acquisition of Learning Data When constructing a prediction model, the user US or the system administrator inputs a learning start request to the gesture input device GT.

[0043] When the control unit 1 of the gesture input device GT detects the learning start request in step S10, first, in step S11, it accesses, for example, a database on the Web under the control of the electromyogram acquisition processing unit 11. Then, from this database, electromyogram signals of specific muscles that have been measured and stored in the past for each of a plurality of expected gestures are acquired together with label information indicating the type of gesture via the communication I / F unit 5. In this case, the electromyogram signals may have already been sampled at a predetermined sampling rate (e.g., 1,000 Hz) when they are stored in the database.

[0044] The acquired sets of electromyogram signals of the specific muscles and label information are temporarily stored in a buffer area in the data storage unit 3 for each gesture type.

[0045] The learning data may be acquired as follows: That is, the user US actually performs a gesture, and the electromyogram signals output from the electromyogram sensors SS1 to SSn during the gesture are acquired via the sensor I / F unit 4. Then, in the input device of the gesture input device GT, label information indicating the type of gesture input by the user US via the input device is acquired and associated with the electromyogram signals.

[0046] (1-2) Extraction of Myoelectric Feature Amounts When electromyogram signals of specific muscles related to one gesture are acquired, the control unit 1 of the gesture input device GT, under the control of the myoelectric feature amount extraction processing unit 12, first analyzes the electromyogram signals in step S12.

[0047] That is, the myoelectric feature extraction processing unit 12 performs, for example, band-pass filtering on the acquired sampling data of the plurality of electromyogram signals to select electromyogram signals in the 20-450 Hz band, then sets a prediction target window of a predetermined time width from the start point of the electromyogram signals in the selected band, and calculates the root mean square (RMS) representing the effective value of the electromyogram signals in the set prediction target window.

[0048] Here, the time width of the prediction window is set to 200 msec in general myoelectric feature extraction, but in one embodiment, it is set taking into consideration both the time required to predict a gesture and the prediction accuracy. That is, the shorter the time width of the prediction window, the faster the gesture can be predicted, but the lower the prediction accuracy.

[0049] Therefore, for example, since it is assumed that a gesture takes about two seconds from start to finish, the time width of the prediction window is set to one second, which is half the total movement period of the gesture. By setting the time width of the prediction window in this way, it becomes possible to predict the gesture while maintaining accuracy in the middle of the movement.

[0050] If the movements immediately after the start of the gestures to be predicted are clearly different between the gestures, the time width of the prediction window may be set to a shorter value. Conversely, if the movements immediately after the start of the gestures to be predicted are similar between the gestures to be predicted, the time width of the prediction window may be set to a longer value.

[0051] Subsequently, in step S13, the myoelectric feature extraction processing unit 12 uses the RMS time series data of the electromyogram signal obtained in the prediction target window as it is as a myoelectric feature vector.

[0052] (1-3) Construction of Prediction Model Next, in step S14, the control unit 1 of the gesture input device GT constructs a prediction model under the control of the prediction model construction processing unit 13 as follows.

[0053] That is, the prediction model construction processing unit 13 uses the myoelectric feature quantities of the specific muscles extracted by the myoelectric feature quantity extraction processing unit 12 as training data and label information indicating the type of gesture corresponding to the myoelectric feature quantities as training data to cause the prediction model to learn. For example, a deep neural network (DNN) is used to construct a prediction model by having the deep neural network learn using the myoelectric feature quantity vectors of the specific muscles as training data.

[0054] Note that a prediction model may be constructed using another machine learning method such as an SVM (Support Vector Machine) or a kNN (k-Nearest Neighbor) algorithm, with the average or maximum value of the RMS as the feature. Also, a muscle synergy analysis may be performed to grasp the coordinated movement of multiple muscles, generating a feature vector, and the prediction model may be constructed using this feature vector.

[0055] (1-4) Learning Completion Determination In step S15, the predictive model construction processing unit 13 determines whether the learning process based on all past EMG signals for one gesture has been completed, and whether the surprise process for all gestures has been completed. If there are still unselected EMG signals or unselected gestures remaining, the process returns to step S11 and repeats the series of processes from step S11 to step S15. On the other hand, if the learning process for all gestures has been completed, the predictive model construction process ends.

[0056] (2) Prediction Phase With the prediction model constructed, the control unit 1 of the gesture input device GT executes gesture prediction processing as follows.

[0057] FIG. 5 is a flowchart showing an example of a series of processing steps and processing contents relating to gesture prediction processing executed by the control unit 1 of the gesture input device GT.

[0058] (2-1) Acquisition of Electromyogram Signals When the user US inputs a prediction start request and this prediction start request is detected in step S20, the control unit 1 of the gesture input device GT, while the user US is performing a gesture motion, acquires the electromyogram signals output from the electromyogram sensors SS1 to SSn via the sensor I / F unit 4 in step S21 under the control of the electromyogram acquisition processing unit 11. At this time, each electromyogram signal is sampled in the sensor I / F unit 4 at a sampling rate of, for example, 1,000 Hz.

[0059] (2-2) Extraction of Myoelectric Feature Amounts When each of the electromyogram signals is acquired, the control unit 1 of the gesture input device GT then analyzes each of the electromyogram signals in step S22 under the control of the electromyogram feature amount extraction processing unit 12. As in the learning phase, the analysis process is performed by extracting an electromyogram signal in the 20-450 Hz band by bandpass filtering, setting a prediction target window with a width of 1 second from the start point of the extracted electromyogram signal, and calculating the RMS of the electromyogram signal in the set prediction target window.

[0060] Subsequently, in step S23, the myoelectric feature extraction processing unit 12 extracts the RMS time series data of each electromyogram signal obtained in the prediction target window as a myoelectric feature vector, and provides these feature vectors to the gesture prediction processing unit 14.

[0061] (2-3) Gesture Prediction In step S24, the gesture prediction processing unit 14 inputs, as explanatory variables, feature vectors of electromyogram signals acquired from a plurality of specific muscles involved in the gesture made by the user US to the prediction model stored in the prediction model storage unit 31. Then, in step S25, the gesture prediction processing unit 14 acquires, from the prediction model, information representing a gesture prediction result to be output as a target variable in response to the input of the explanatory variables.

[0062] The prediction model predicts a gesture based on RMS time series data of each electromyogram signal calculated during the prediction window from the start of the gesture. That is, the prediction model outputs a gesture prediction result at the end of the prediction window without waiting for the end of the entire gesture.

[0063] (2-4) Output of Input Information Finally, in step S26, the control unit 1 of the gesture input device GT generates input information that the user US is attempting to input by gesture, based on information representing the gesture prediction result output from the prediction model, under the control of the input information output processing unit 15. For example, in one embodiment, an avatar control signal that controls the behavior of the user US's avatar in the metaverse in which the user US participates is generated as the input information.

[0064] The input information output processing unit 15 transmits the generated avatar control signal from the communication I / F unit 5 to the server device SV via the network NW.

[0065] As a result, the server SV controls the movement of the user US's avatar in the metaverse in response to the avatar control signal. As described above, the gesture prediction result is output at the end of the prediction window without waiting for the entire gesture to be completed. Therefore, the avatar control signal is also transmitted to the server SV before the gesture is completed. Therefore, the server SV can move the avatar prior to the gesture of the user US. Furthermore, even if a transmission delay occurs in the network NW, the avatar can be moved without any delay in response to the gesture of the user US.

[0066] (2-5) Prediction Completion Determination When the control unit 1 of the gesture input device GT completes the motion prediction for one gesture and the transmission of the avatar control signal based on the prediction result, it determines in step S27 whether all gestures have been completed. If the avatar motion control based on gestures is still ongoing, the process returns to step S21, and the series of prediction processes from step S21 to step S27 is repeated for the next gesture.

[0067] On the other hand, when a request to end the gesture is input by the user US or a period during which no gesture is made exceeds a preset time, the control unit 1 ends the prediction process and returns to a standby state.

[0068] (Effect) As described above, in one embodiment, electromyogram signals are acquired from electromyogram sensors SS1 to SSn attached to body parts corresponding to specific muscles that move in coordination when the user US makes a gesture, feature vectors are extracted from each acquired electromyogram signal, and these feature vectors are input as explanatory variables into a prediction model to predict the gesture, and input information corresponding to the prediction result is output.

[0069] Therefore, the gestures of the user US are predicted based on the myoelectric feature values ​​extracted from the electromyogram signals of a specific plurality of muscles. Therefore, it is sufficient to attach the myoelectric sensors SS1 to SSn only to the body parts corresponding to the specific plurality of muscles, and it is possible to predict gestures without imposing a large burden on the user by using only the minimum number of myoelectric sensors required.

[0070] Furthermore, a prediction window having a time width shorter than the entire movement period of the gesture is set, and the feature vector of the electromyogram signal extracted within this prediction window is input to the prediction model. This makes it possible to predict the gesture at the end of the prediction window without waiting until the entire movement is completed, thereby shortening the time required for gesture prediction. Then, by transmitting an avatar control signal to the server device SV based on the prediction result, it becomes possible to move the avatar prior to the gesture of the user US. Furthermore, even if a transmission delay occurs in the network NW, it becomes possible to move the avatar without delay from the gesture of the user US.

[0071] (Verification Example) The inventors conducted the following simulation to verify the effects of the invention. Five types of gestures, namely, clicking, clenching, and reaching (forward, sideways, and upward), were predicted by attaching electromyogram sensors to five locations that move in coordination during the gestures: the index finger and little finger on the back of the hand, the forearm, the upper arm, and the shoulder. As a result, it was confirmed that gestures can be predicted with an accuracy of 80% or more by using electromyogram signals obtained within one second from the start of the gesture.

[0072] Furthermore, when the EMG sensors were attached to only three locations, the index finger and little finger on the back of the hand and the forearm, the prediction accuracy dropped to about 60%. The prediction accuracy for reaching movements in particular dropped. This suggests that it is important to capture the movements of the upper arm and shoulder during reaching movements.

[0073] Therefore, it is important to select a specific number of muscles that move in coordination with each other, i.e., a sufficient number of muscles, depending on the type of gesture to be predicted, and to attach EMG sensors to the body parts corresponding to each of the selected muscles. It is desirable to attach at least one EMG sensor to each body part.

[0074] Other Embodiments (1) In one embodiment, the length of the prediction target window is fixed to half the length of the entire gesture duration (e.g., 1 second). However, the length of the prediction target window may be variably set depending on the type of gesture. For example, the length of the prediction target window may be adaptively set depending on the accuracy rate of the prediction result obtained by the prediction model. In this way, it is possible to automatically set an appropriate prediction target window for each of multiple types of gestures.

[0075] (2) In one embodiment, the description has been given taking as an example a case where the user's gestures control the behavior of an avatar in the metaverse. However, the present invention is not limited to this. For example, the present invention can also be applied to a case where various operations are performed by sending commands using gestures when browsing a web page. In this case, too, the command that the user US was trying to input by gesture can be recognized before the entire gesture operation is completed, allowing for smooth and responsive operation of the web page.

[0076] (3) The present invention can also be applied to remote control of the operation of a robot located in a remote location using gestures. In this case, the input information that was to be input by the gesture can be predicted during the gesture operation, and a robot control signal can be transmitted according to the prediction result. Therefore, even if there is a transmission delay due to the network, the robot can be controlled with good responsiveness without any operational delay.

[0077] (4) Furthermore, this invention can be applied when creating a document using gestures. In this case, it is possible to predict and input the word or phrase to be input during the gesture. In other words, it is possible to achieve the same effect as predictive conversion in text input.

[0078] (5) In the embodiment, the electromyographic sensors SS1 to SSn and the gesture input device GT are connected by a signal cable, but the connection may be made using a wireless interface such as Bluetooth (registered trademark). In addition, in the embodiment, the electromyographic signals output from the electromyographic sensors SS1 to SSn are analyzed in the gesture input device GT, but the signal analysis function may be provided in the electromyographic sensors.

[0079] (6) In addition, various modifications can be made to the type and functional configuration of the gesture input device, the processing procedure and processing content, the type of gesture to be predicted, and the like, without departing from the spirit of the present invention.

[0080] Although the embodiments of the present invention have been described in detail above, the above description is merely an example of the present invention in every respect. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. In other words, when implementing the present invention, specific configurations according to the embodiments may be appropriately adopted.

[0081] In short, this invention is not limited to the above-described embodiments, and in the implementation stage, the components can be modified and embodied without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining multiple components disclosed in the above-described embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined.

[0082] GT...gesture input device SS1 to SSn...electromyogram sensors SV...server device NW...network 1...control unit 2...program storage unit 3...data storage unit 4...sensor I / F unit 5...communication I / F unit 6...bus 11...electromyogram acquisition processing unit 12...electromyogram feature extraction processing unit 13...prediction model construction processing unit 14...gesture prediction processing unit 15...input information output processing unit 31...prediction model storage unit

Claims

1. A gesture input device comprising: a first processing unit that acquires electromyogram signals from body parts corresponding to specific muscles that move in coordination when a user makes a gesture; a second processing unit that extracts electromyogram features from each of the acquired electromyogram signals; a third processing unit that, in a learning phase, uses the extracted electromyogram features as learning data to construct a prediction model for predicting the gesture; a fourth processing unit that, in a prediction phase, inputs the extracted electromyogram features into the prediction model to acquire a prediction result of the gesture from the prediction model; and a fifth processing unit that outputs input information corresponding to the prediction result.

2. The gesture input device according to claim 1, wherein the second processing unit sets a prediction window for the EMG signal having a time width shorter than the entire movement period of the gesture, and extracts the EMG feature based on the signal obtained in the prediction window of the EMG signal, and the fourth processing unit inputs the extracted multiple EMG feature values ​​to the prediction model and obtains the prediction result from the prediction model.

3. The gesture input device according to claim 1, wherein the fifth processing unit generates, as the input information, control information for controlling the movement of an avatar that moves in accordance with the gesture of the user, and outputs the generated control information.

4. A gesture input method executed by an information processing device, comprising: a step of acquiring electromyogram signals from body parts corresponding to specific muscles that move in coordination when a user makes a gesture; a step of extracting electromyogram features from each of the acquired electromyogram signals; a step of constructing a prediction model for predicting the gesture using the extracted electromyogram features as learning data in a learning phase; a step of inputting the extracted electromyogram features into the prediction model and acquiring a prediction result of the gesture from the prediction model in a prediction phase; and a step of outputting input information corresponding to the prediction result.

Images