CN113952707A

CN113952707A - Motion sensing game action recognition method, scoring method and system based on RFID

Info

Publication number: CN113952707A
Application number: CN202111536803.5A
Authority: CN
Inventors: 陈政霖; 郑飞州; 陈胜俭
Original assignee: Guangzhou Youkegu Technology Co ltd
Current assignee: Guangzhou Youkegu Technology Co ltd
Priority date: 2021-12-16
Filing date: 2021-12-16
Publication date: 2022-01-21
Anticipated expiration: 2041-12-16
Also published as: CN113952707B

Abstract

The invention relates to a motion sensing game action recognition method based on RFID, which comprises the following steps: arranging and fixing a plurality of RFID tags around a user's limb; the user makes corresponding actions according to the game instructions and obtains signal change data of the RFID tags; performing feature extraction on the signal change data of the plurality of RFID tags; and inputting the extracted features into a pre-trained motion recognition model, and outputting a motion recognition result by the motion recognition model. The invention improves the original fixed radio frequency signal transmitting equipment, arranges and fixes a plurality of RFID labels around the limbs of the user, amplifies the interference of the micro-actions of the user on the radio frequency signals, makes the characteristics of the user actions more obvious, improves the identification accuracy of the radio frequency signals on the micro-actions, and makes the motion sensing game based on the radio frequency signals possible.

Description

Motion sensing game action recognition method, scoring method and system based on RFID

Technical Field

The invention relates to the technical field of Internet of things, in particular to a motion sensing game action recognition method, a scoring method and a system based on RFID.

Background

With the development of society, the living standard of people is continuously improved, and more people come into contact with the motion sensing game. As an emerging game form, the motion sensing game can bring more real immersive experience to a user compared with the traditional game depending on a display screen, the interestingness of the game is increased, and meanwhile, the motion and the game are combined together, so that the healthy game is really realized. In recent years, a motion sensing game spanning multiple platforms, such as "full dance" and "big adventure in fitness circle" has rapidly become popular and has become a new fashion.

The motion recognition modes of the motion sensing game widely used at present mainly comprise two modes: based on the camera and based on wearable rechargeable peripheral hardware. The camera-based method collects external images through a monocular camera or a multi-view camera, and realizes action recognition through a machine vision method; this approach relies on a powerful computer game and a sufficiently clear camera, which is expensive; and the camera-based method is difficult to use in a dark or dark environment; the problem of privacy and safety is also brought by using a camera to collect images on the household equipment. The mode based on the wearable rechargeable peripheral often utilizes sensors such as an infrared emitter or a gyroscope on external equipment to respond to user actions to identify the actions, but the method requires a user to additionally purchase rechargeable wearable hardware equipment, higher cost is required, the sensors have certain weight, discomfort can be caused after wearing, cervical vertebra ache can be caused by the head-mounted peripheral, the handheld peripheral must be continuously held, and the like, so that the immersive experience of a game player is influenced.

There is currently research on motion recognition based on radio frequency signals. Existing research often employs one or more fixed rf signal transmission sources and a fixed rf signal reception source to match corresponding actions through rf signal variations between the one or more transmission sources and the reception source. However, such methods have the problem that the use of one or more fixed transmission sources results in a coarse granularity of motion perception. In the existing method, in order to prevent signal interference, the fixed distance between a plurality of radio frequency signal emission sources and a plurality of radio frequency signal receiving sources is often larger than 6cm, and the diffraction effect of electromagnetic waves causes the electromagnetic waves to tend to bypass objects with the dimension smaller than the wavelength of the electromagnetic waves, so that the system is difficult to identify the action with the action amplitude smaller than 6 cm. Although some studies have implemented detection below the scale of 6cm by measuring small changes caused by regular motion, this exploits the periodicity of motion and the identification of non-periodic small motions is still limited by the wavelength limitations of common radio frequency devices. For example, in a non-interference detection area with a plurality of fixed rf emission sources and a fixed rf receiving source, if the action implementer performs the action between the emission source and the receiving source by turning the arm, most fresnel regions of the emission sources do not change greatly, the propagation path of the emitted rf signals is also almost unchanged, and the changes of the generated channel state information are often filtered out as gaussian noise in the rf signals. In the field of motion sensing games, most of motions to be recognized are aperiodic, and due to the judgment requirement on motion accuracy, the recognition requirement on micro motions is higher, and the existing method is difficult to meet the motion recognition requirement of the motion sensing games.

Disclosure of Invention

The invention provides a motion sensing game motion recognition method based on RFID, which can remarkably improve the recognition accuracy of micro motion.

A second object of the present invention is to provide a motion recognition system for a motion sensing game using the motion recognition method.

A third object of the present invention is to provide a motion sensing game motion scoring method using the motion recognition method.

A fourth object of the present invention is to provide a motion sensing game motion scoring system to which the motion recognition method is applied.

In order to realize the first invention, the technical scheme is as follows:

the motion sensing game action identification method based on the RFID comprises the following steps:

arranging and fixing a plurality of RFID tags around a user's limb;

the user makes corresponding actions according to the game instructions and obtains signal change data of the RFID tags;

performing feature extraction on the signal change data of the plurality of RFID tags;

and inputting the extracted features into a pre-trained motion recognition model, and outputting a motion recognition result by the motion recognition model.

In order to realize the second purpose, the adopted technical scheme is as follows:

the motion sensing game action recognition system based on the RFID comprises a display, a plurality of RFID labels and a motion sensing game machine; the somatosensory game machine comprises an RFID reading module, a feature extraction module and an action recognition module;

the display is used for giving game instructions to the user; receiving and displaying the action recognition result sent by the action recognition module;

the RFID reading module is used for acquiring signal change data of a plurality of RFID labels;

the characteristic extraction module is used for extracting the characteristics of the signal change data of the plurality of RFID labels;

the action recognition module is used for inputting the extracted features into a pre-trained action recognition model, and the action recognition model outputs an action recognition result.

In order to achieve the third object, the technical scheme is as follows:

a motion sensing game action scoring method based on RFID comprises the following steps:

arranging and fixing a plurality of RFID tags around a user's limb;

inputting the extracted features into a pre-trained action recognition model, and outputting an action recognition result by the action recognition model; the action recognition result output by the action recognition model is the prediction probability of the action belonging to a specific action category;

and scoring the action by taking the prediction probability as the matching degree between the action and a standard action.

In order to realize the fourth invention purpose, the adopted technical scheme is as follows:

the motion sensing game action scoring system based on the RFID comprises a display, a plurality of RFID labels and a motion sensing game machine; the motion sensing game machine comprises an RFID reading module, a feature extraction module, an action recognition module and a score acquisition module;

the display is used for giving game instructions to the user; receiving and displaying the score sent by the score acquisition module;

the action recognition module is used for inputting the extracted features into a pre-trained action recognition model, and the action recognition model outputs an action recognition result; the action recognition result output by the action recognition model is the prediction probability of the action belonging to a specific action category;

and the score acquisition module is used for scoring the action by taking the prediction probability as the matching degree between the action and a standard action.

Compared with the prior art, the invention has the beneficial effects that:

1) the accuracy rate of the radio frequency signals for identifying the micro actions is improved: the invention improves the original fixed radio frequency signal transmitting equipment, arranges and fixes a plurality of flexible RFID labels around the limbs of the user, amplifies the interference of the micro-actions of the user on the radio frequency signals, makes the characteristics of the user actions more obvious, improves the identification accuracy of the micro-actions of the radio frequency signals, and makes the motion sensing game based on the radio frequency signals possible.

2) The user experience is increased: by the invention, the action recognition time in the motion sensing game is greatly reduced. The delay experienced by the player is greatly reduced. Meanwhile, the flexible RFID tag is used as a wearable peripheral, so that the weight is light, the wearing is simple and comfortable, and the experience of the user in the motion sensing game can be greatly improved.

3) And the motion recognition cost is reduced: according to the invention, expensive external equipment such as a camera and an infrared sensor is not required to be purchased additionally, and only cheap flexible RFID tags and readers are required, so that the hardware cost of the motion sensing game can be greatly reduced, and the large-scale popularization of the motion sensing game is facilitated.

4) Can be operated in dark environment: the invention adopts RFID signals, namely electromagnetic wave signals, to carry out action recognition, has no requirement on the brightness of the environment, and has low requirement on the use environment compared with a method based on a camera.

5) Avoid privacy problems: the action recognition is carried out by adopting the RFID signal, compared with the high risk brought to the privacy of the user by the camera equipment, the action recognition method only responds to the specific action which is collected in advance, other privacy information is not captured at all, and the privacy safety of the user is protected.

Drawings

Fig. 1 is a system schematic diagram of a motion recognition method of a motion sensing game based on RFID in embodiment 1.

Fig. 2 is a schematic flow chart of the motion sensing game motion recognition method based on RFID in embodiment 1.

Fig. 3 is a system schematic diagram of the motion sensing game motion scoring method based on RFID in embodiment 2.

Fig. 4 is a schematic diagram of an application example of the motion sensing game motion scoring method based on RFID in embodiment 3.

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the patent;

the invention is further illustrated below with reference to the figures and examples.

Example 1

The embodiment provides a motion sensing game motion recognition method based on an RFID, and the method needs to use a flexible RFID ring, a motion sensing game machine and a display when being implemented, and is specifically shown in FIG. 1. Wherein a plurality of groups of flexible RFID labels are arranged on the flexible RFID ring; the motion sensing game machine comprises an RFID reading module, a signal analyzing module, a noise reduction module, a feature extraction module, an action recognition module, a display module and an operating system.

The method has the basic idea that a CNN-XGboost model designed by relevant characteristic training is extracted by detecting the change of RFID signals transmitted by a plurality of groups of RFID tags fixed on a human body when the human body does standard movement, reflecting the distance change between the RFID tags and an RFID reading module by using the signal intensity change and reflecting the angle change between the RFID tags and the RFID reading module by using the arrival angle change. And identifying the action of a user wearing the flexible RFID ring by using the trained CNN-XGboost model. The motion sensing game has high requirement on the recognition precision of the user fine actions, and the user experience and cost are more emphasized, so that the requirements of the motion sensing game can be effectively met.

The parts are explained as follows:

1) flexible RFID ring

1.1) flexible RFID ring: the wearable peripheral is used for fixing a plurality of RFID tags on limbs or other parts which need to be sensed of a user and is necessary for realizing motion sensing game action recognition;

1.2) RFID tag: the RFID reader module is responsible for receiving signals sent by the body sensing game machine and feeding back RFID signals, and active and passive tags can be used. Each RFID tag has a number and can be recognized by the motion sensing game machine. The RFID tag sets are wrapped around the user's limbs, such as the wrist and ankle, in a loop.

2) RFID reading module

2.1) reader: the RFID label receiving module is responsible for receiving RFID signals from a plurality of RFID labels, packaging information into a data packet after preliminary analysis, and forwarding the data packet to the signal analysis module.

3) Signal analysis module

3.1) data packet analysis: the RFID reading module is responsible for analyzing the data packet with the specified format obtained by the RFID reading module into a channel state information packet and sending the channel state information packet to the noise reduction module;

4) noise reduction module

4.1) noise reduction filtering treatment: and filtering the obtained channel state information packet by using a filtering algorithm, wherein the purpose of the filtering algorithm is to reserve the fluctuation caused by the human body motion in a certain frequency range and eliminate the fluctuation caused by other factors except the human body motion. In the invention, the filtering method is as follows: firstly, preliminarily filtering the waveform of a channel state information packet by using an average filtering method, retaining low-frequency fluctuation generated by human motion by using a Butterworth filter, and further obtaining the human motion fluctuation with most of noise removed by using wavelet transform.

Daubechines wavelets have excellent processing power for non-stationary time series. The present invention uses the db1 wavelet transform. The number of decomposition layers is mainly related to the signal-to-noise ratio, when the signal-to-noise ratio is low, the input signal mainly takes noise as the main, and a larger number of decomposition layers should be selected to be beneficial to removing the noise; when the signal-to-noise ratio is high, the input signal is mainly signal-based, and a smaller number of decomposition layers should be selected, otherwise the signal distortion is severe. The number of decomposition layers can be dynamically selected according to the signal quality, signals with high signal-to-noise ratio can be decomposed by one layer, and signals with low signal-to-noise ratio can be decomposed by more than three layers. The db1 wavelet transform is implemented using the dwt function in matlab:

[cA,cD]=dwt(X,‘db1’)

[ cA, cD ] = dwt (X, 'db 1') the signal X is decomposed using a specified wavelet basis function db1, with cA, cD being the approximation and detail components, respectively. And measuring the stationary degree of the detail component by using the variance S, wherein if S is greater than a preset threshold, the detail component is not stationary enough, and more layers of signal decomposition are required. When the detail component is stable enough, the corresponding approximate component is extracted as the waveform after noise reduction, and the detail component is considered as the environmental noise to be discarded.

5) Feature extraction module

5.1) feature extraction: according to the method, firstly, the motion sample is divided into a series of time segments with the same length which can be used for identification according to the motion sensing game content. And obtaining a two-dimensional matrix of the initial action information according to the sampling times in each time segment and the corresponding signal fluctuation amplitude. And 7 characteristics of the average value, the variance, the fluctuation peak position, the short-time energy coefficient, the number of peaks exceeding the average value, the maximum value and the minimum value of the two-dimensional matrix are extracted as identification characteristics. As an example given here, if 10 RFID chips are used, the above-mentioned 7 features of signal strength and arrival angle are extracted, the time slice is 2s, and 20 time stamps (100 ms per time stamp, sampling frequency is 10 times/s) exist in the time slice, then the extracted feature matrix is a four-dimensional matrix of 10 × 2 × 7 × 20. Where 10 represents 10 chips, 2 represents 2 physical quantities of signal strength and angle of arrival, 7 represents 7 features, and 20 represents 20 time stamps.

6) Action recognition module

6.1) training of a CNN-XGboost model: the part is carried out in the development process of the motion sensing game, and the trained model is directly loaded into the system when the game runs without repeated training. The model training process is as follows: and obtaining a plurality of groups of RFID channel characteristic information under standard actions and nonstandard actions to form a training data set, and training the designed CNN-XGboost model by using the training data set.

The training process is described as follows:

firstly, feature extraction is carried out on the acquired arrival angle data and the acquired signal intensity data. Because the duration of different actions is different, taking a dance game as an example, the action information in the same time segment of each beat is collected, so as to ensure that the sizes of two-dimensional matrixes constructed by the collected data are the same. Each data set is labeled with a digitized label as data according to different actions (e.g., 1,2,3 …). After recording multiple sets of standard actions and multiple sets of nonstandard actions from multiple testers, the standard action label is set to 1, the label of the nonstandard action is set to 0-1 according to the standard degree, and the label is marked as L. And acquiring the signal intensity data and arrival angle data of all the actions of 10 RFID chips corresponding to the 7 characteristics to obtain a 4-dimensional characteristic matrix [10 multiplied by 2 multiplied by 7 multiplied by 20] and a label L. For multi-dimensional data, a 4-dimensional feature matrix [10 × 2 × 7 × 20] is input into the CNN-XGboost model by using the basic idea of image classification. In the invention, considering that the collected characteristic matrix has response to micro actions, is easy to overfit, and has a positive value and a negative value simultaneously, the designed CNN-XGboost network structure is a sampling layer with 3 layers of convolution layers and 2 layers of activation functions as ELUs, a sampling layer with +1 layer of maximum pooling layer and +1 layer of Sigmoid full-link layer and +1 layer of XGboost layer, the Loss function is cross entropy, and the weight of each neuron is updated by adopting a stochastic gradient descent algorithm and back propagation.

The specific training process of the model comprises the steps of firstly defining initial parameters of the CNN-XGboost network, adopting a random gradient descent algorithm, inputting a part of randomly selected processed four-dimensional feature matrix into the CNN-XGboost model as a training set, outputting a classification result and probability of the action, obtaining the global Loss of the CNN-XGboost model according to the corresponding relation between the four-dimensional feature matrix and the label L, taking the global Loss as the Loss of the CNN-XGboost model, after reversely propagating the gradient to obtain new CNN-XGboost model parameters, selecting the part of processed four-dimensional feature matrix as the training set again for training, and iterating until the global Loss value tested by the test set is smaller than a target threshold value. The input of the model is a four-dimensional feature matrix [10 × 2 × 7 × 20], and the prediction probability P of the action type to which the action belongs can be output.

6.2) model use: and the four-dimensional feature matrix [10 multiplied by 2 multiplied by 7 multiplied by 20] of the change features of the RFID signals is input into the CNN-XGboost model, and the prediction probability P of the action belonging to the action A is output as the recognition result.

7) Display module

7.1) score display: the device is responsible for feeding back the action recognition result to the user through a display and other equipment;

7.2) game running: the game system is responsible for outputting game contents, such as displaying the game contents, prompting, next-step game and the like;

8) operating system

8.1) global management: the system is responsible for realizing all functions of the motion sensing game machine and making decisions of the next game content;

9) display device

9.1) display: and the game machine is responsible for the feedback of the action recognition result and the feedback of picture or sound information which the motion sensing game machine hopes to feed back to the user. Including but not limited to televisions, projectors, etc.

As shown in fig. 2, the specific method steps of the motion sensing game motion recognition method based on RFID provided in this embodiment are as follows:

s201: a user wears a flexible RFID ring as required, opens the motion sensing game machine and the display and starts a game;

s202: the motion sensing game machine loads a corresponding motion sensing game and loads a pre-trained CNN-XGboost model;

s203: when a game starts, a user firstly does not move, and waits until an RFID reading module of the motion sensing game machine obtains initial signal information of a flexible RFID ring;

s204: the user makes corresponding actions according to the guidance of the display, and the relative position of the flexible RFID ring and the motion sensing game machine is changed;

s205: the RFID reading module of the motion sensing game machine continuously acquires the signal changes of different RFID labels on the flexible RFID ring, packages the signals into a data packet and sends the data packet to the signal analysis module;

s206: the signal analysis module extracts the channel information of each RFID label from the data packet and sends the channel information to the noise reduction module;

s207: the noise reduction module removes noise contained in the information and sends the noise removed noise to the feature extraction module;

s208: after extracting the corresponding characteristics of the signals, the characteristic extraction module sends the characteristics to the action recognition module;

s209: the action recognition module inputs a change characteristic matrix of the RFID signal by using a pre-loaded CNN-XGboost model, outputs a corresponding action recognition result and sends the corresponding action recognition result to the display module and the operating system;

s210: the operating system of the game machine judges the change of the game content caused by the action according to the identification result, determines whether to carry out the next game process and feeds back the game content to the user through the display;

s211: and the user performs the next action according to the display content of the display to continue to finish the motion sensing game.

Example 2

The embodiment provides a motion sensing game action scoring method based on an RFID, and the method needs to use a flexible RFID ring, a motion sensing game machine and a display when being implemented, and is specifically shown in FIG. 3. Wherein a plurality of groups of flexible RFID labels are arranged on the flexible RFID ring; the motion sensing game machine comprises an RFID reading module, a signal analyzing module, a noise reduction module, a feature extraction module, an action recognition module, a score acquisition module, a display module and an operating system.

The functional roles and principles of the flexible RFID ring, the display, the RFID reading module, the signal analyzing module, the noise reduction module, the feature extraction module, the action recognition module, the display module and the operating system are consistent with those of corresponding modules in the motion sensing game action recognition method of the embodiment 1, and the motion sensing game action scoring method based on the RFID provided by the embodiment is mainly based on the embodiment 1, and is additionally provided with a step of scoring the actions of the user by using the action recognition result and a score acquisition module.

When the game score is used specifically, the recognition result output by the action recognition module is the probability that the corresponding action belongs to the specific action category, the score acquisition module takes the probability as the matching degree of the action, and the game action of the user is scored based on the matching degree. For example, the predicted probability P that the corresponding action output by the action recognition module belongs to the action a is 0.921, and after forensics, the predicted probability P is 0.95, and the result is multiplied by the maximum score 100 in the game, so that the score of the action of the user is 95.

In specific implementation, the scoring result is fed back to the user through a display module, a display and other equipment.

Example 3

The embodiment provides an application example of the motion sensing game action scoring method based on the RFID, and as shown in fig. 4, the flexible RFID ring is made into a bracelet and a foot ring and is worn on the left wrist and the right ankle of a user. The RFID tags in the flexible RFID ring on the user's arm are numbered sequentially in a clockwise order. The RFID signal receiving end is an intelligent motion sensing game machine, signal receiving, signal processing, noise reduction, feature extraction and action identification are completed by the intelligent motion sensing game machine, a CNN-XGboost model training phase is completed during game development, a CNN-XGboost model to be used during game loading is loaded into the intelligent motion sensing game machine, and a display is connected with the intelligent motion sensing game machine and is responsible for displaying pictures output by the intelligent motion sensing game machine. When a user stands at a designated position and sends a game start instruction to the smart motion sensing game machine by using a gamepad or other methods, a motion sensing game can be started. At the moment, the user plays a body feeling dance game. The specific process of application is as follows:

1) a user stands at a designated position, a game starting instruction is sent to the intelligent motion sensing game machine through a game handle, a display displays corresponding dance guide actions along with the progress of music and shows the dance guide actions to the user, and meanwhile, initial signal information transmitted by a flexible RFID ring of the user is collected;

2) the player takes the corresponding action: the arm rotates left and then right. Meanwhile, the flexible RFID ring sends an identification data packet to the intelligent motion sensing game machine;

3) an RFID reading module of the intelligent motion sensing game machine continuously receives a data packet containing information and transmits the data packet to a signal analysis module;

4) the signal analysis module converts the data packet into channel state information of the RFID signal and transmits the channel state information to the noise reduction module;

5) the noise reduction module filters within the selected frequency range and transmits clean fluctuation to the feature extraction module;

6) the feature extraction module performs time slice dividing operation on the information, and finds that the signal intensity of the No. 2 label is reduced and the signal intensity of the No. 1 label is increased, and then the signal intensity of the No. 1 label is reduced and the signal intensity of the No. 3 label is increased. Transmitting the characteristic to an action recognition module;

7) the action recognition module loads a pre-trained CNN-XGboost model, inputs RFID information to predict action prediction probability, sends the action prediction probability to the score acquisition module to score a user, the score acquisition module finds that the matching degree of the action of the user and the rotation of an arm from left to right is 90%, sets the score corresponding to the action to be 90%, transmits the score to a display and feeds the score back to the user;

8) at the moment, the music just goes to the next action, and after the user receives feedback, the user continues to follow the music to make the next dance action, and the steps are repeated.

It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A motion sensing game action recognition method based on RFID is characterized in that: the method comprises the following steps:

arranging and fixing a plurality of RFID tags around a user's limb;

2. The motion sensing game motion recognition method based on the RFID as claimed in claim 1, wherein: after the signal change data of the plurality of RFID tags are obtained, before feature extraction is performed on the signal change data of the plurality of RFID tags, the following operations are performed:

primarily analyzing the signal change data of the plurality of RFID labels, and packaging the information into a data packet;

parsing the data packet into a channel state information packet;

and carrying out filtering and noise reduction processing on the channel state information packet.

3. The motion sensing game motion recognition method based on the RFID as claimed in claim 2, wherein: the filtering and denoising processing on the channel state information packet comprises:

carrying out preliminary filtering by using a mean filtering method;

performing secondary filtering by using a Butterworth filter;

the cubic filtering is performed using wavelet transform.

4. The motion sensing game motion recognition method based on the RFID as claimed in claim 3, wherein: the cubic filtering using wavelet transform includes:

1) the channel state information packet is decomposed using the dwt function in matlab:

[cA,cD]=dwt(X,‘db1’)

db1 represents a wavelet basis function, cA and cD are an approximate component and a detail component respectively, and X represents a channel state information packet;

2) calculating the variance S of the detail components;

3) if the variance S of the detail component is larger than a preset threshold value, increasing the number of layers for decomposing the channel state information packet, and then executing the step 1); and if the variance S of the detail component is less than or equal to a preset threshold value, extracting the corresponding approximate component as a waveform after filtering and denoising, and discarding the detail component.

5. The motion sensing game motion recognition method based on the RFID according to any one of claims 2-4, wherein: the feature extraction of the signal change data of the plurality of RFID tags comprises extracting signal strength features and arrival angle features, and the extraction processes of extracting the signal strength features and the arrival angle features are as follows:

dividing the signal intensity sequence in the channel state information packet into a series of segments with the same length in sequence, and obtaining a two-dimensional matrix of the signal intensity according to the sampling times in each segment and the corresponding signal fluctuation amplitude; extracting the average value, the variance, the fluctuation peak position, the short-time energy coefficient, the number of peaks exceeding the average value, the maximum value and the minimum value of the two-dimensional matrix as signal intensity characteristics;

dividing the arrival angle sequence in the channel state information packet into a series of segments with the same length in sequence, and obtaining a two-dimensional matrix of the arrival angle according to the sampling times in each segment and the corresponding signal fluctuation amplitude; and extracting the average value, the variance, the fluctuation peak position, the short-time energy coefficient, the number of peaks exceeding the average value, the maximum value and the minimum value of the two-dimensional matrix as the arrival angle characteristics.

6. The motion sensing game motion recognition method based on the RFID as claimed in claim 5, wherein: the pre-trained action recognition model is a CNN-XGboost model, and the training process is as follows:

1) acquiring training data;

2) defining initial parameters of a CNN-XGboost model;

3) inputting randomly selected training data into a CNN-XGboost model by adopting a random gradient descent algorithm;

4) the CNN-XGboost model outputs an action recognition result;

5) calculating a global loss function of the CNN-XGboost model, judging whether the global loss function of the CNN-XGboost model is smaller than a set threshold value or not, and if so, ending the training process; otherwise, updating the parameters of the CNN-XGboost model by using a back propagation gradient algorithm based on the global loss function;

6) step 3) is performed.

7. The motion sensing game motion recognition method based on the RFID as claimed in claim 6, wherein: the specific process of acquiring the training data is as follows:

disposing and securing a plurality of RFID tags around a limb of a subject;

enabling the subject to sequentially make standard motions and nonstandard motions of the game motions, and obtaining signal change data of a plurality of RFID tags corresponding to the standard motions and the nonstandard motions of the game motions;

and extracting the characteristics of the signal change data of the plurality of RFID labels corresponding to the standard action and the non-standard action of the game action, and marking the labels by using the obtained characteristics as training data.

8. An RFID-based motion sensing game motion recognition system, which applies the RFID-based motion sensing game motion recognition method according to any one of claims 1 to 7, and is characterized in that: the system comprises a display, a plurality of RFID labels and a motion sensing game machine; the somatosensory game machine comprises an RFID reading module, a feature extraction module and an action recognition module;

9. A motion sensing game action scoring method based on RFID is characterized in that: the method comprises the following steps:

arranging and fixing a plurality of RFID tags around a user's limb;

10. The motion sensing game motion scoring system based on the RFID and the motion sensing game motion scoring method based on the RFID as claimed in claim 9 are applied, and the system is characterized in that: the system comprises a display, a plurality of RFID labels and a motion sensing game machine; the motion sensing game machine comprises an RFID reading module, a feature extraction module, an action recognition module and a score acquisition module;