Human body action recognition method and system

Abstract

The invention provides a human body action recognition method and system. The human body action recognition method comprises the following steps: collecting a multi-dimensional signal influenced by human body actions, wherein the multi-dimensional signal is sent out by a radio frequency electronic tag arranged in front of a human body; the human body action is determined based on the fused multi-dimensional signals, and the multi-dimensional signals comprise at least two of a received signal strength indication signal, a phase signal and a Doppler frequency shift signal; and the number of the radio frequency electronic tags is the same as that of the human bodies. According to the human body action identification method and system provided by the invention, non-wearable action identification based on the radio frequency identification technology is realized, on the basis of protecting privacy and not being influenced by ambient light, the reliability of single-person single-signature in the radio frequency identification technology is ensured by fusing multi-dimensional signals, and multi-person action identification can be realized, so that the human body action identification efficiency is improved. The system has high practical value in single-person application scenes and multi-person application scenes including classrooms, meeting rooms, study rooms, libraries and the like.

Description

Technical Field

[0001] The present application relates to the technical field of human motion recognition, and more specifically, to a human motion recognition method and a human motion recognition system. Background Art

[0002] With the development of the Internet of Things, human motion recognition has been widely used in all aspects of social life and plays a key role in various fields. In the field of health care, human motion recognition can provide real-time feedback and evaluation tools for rehabilitation training of patients with movement disorders, helping doctors better monitor the recovery process of patients. In the field of human-computer interaction, human motion recognition enables machines to understand human intentions and needs more naturally, thereby providing a better interactive experience. In security monitoring systems, human motion recognition can effectively identify and warn potential security threats, such as intrusion detection and abnormal behavior analysis.

[0003] There are many solutions for human motion recognition based on sensors, computer vision, and wireless signals. The sensor-based method requires the use of multiple high-precision sensors to collect human-related data for motion recognition. Since it is only applicable to specific scenarios, such as scenarios that require users to wear or deploy in advance in the environment, it has limitations. Although the computer vision-based method has good recognition effects and a quick deployment method, it is sensitive to light. In addition, there are problems such as privacy leakage and failure due to obstruction of line of sight. Human body recognition based on wireless signals not only has no requirements for ambient light, but also can realize human motion recognition under the premise of user privacy protection.

[0004] However, taking RFID (Radio Frequency Identification) technology as an example, due to its relatively weak signal, in order to improve the accuracy and reliability of motion recognition, users are required to wear radio frequency electronic tags (Tags), which deteriorates the user experience and requires multiple tags or even multiple antennas, making deployment more troublesome. In addition, in multi-person scenarios, multiple tags interfere with each other, which results in poor accuracy and reliability of wireless signal human body recognition.

[0005] Therefore, there is an urgent need for a wearable, easy-to-deploy, accurate and reliable human motion recognition method and system that can realize both single-person motion recognition and multi-person motion recognition. Summary of the invention

[0006] The present application provides a human motion recognition method and system that can at least partially solve the above-mentioned problems.

[0007] On the one hand, the present application provides a method for human motion recognition, the method comprising: collecting a multidimensional signal affected by human motion, wherein the multidimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body; and determining the human motion based on the fused multidimensional signal, wherein the multidimensional signal includes at least two of a received signal strength indication signal, a phase signal, and a Doppler frequency shift signal; and the number of radio frequency electronic tags is the same as the number of human bodies.

[0008] In one embodiment of the present application, determining human body movements based on the fused multi-dimensional signal includes: inputting the fused multi-dimensional signal into a trained guided aggregation model for classification to determine the human body movements, wherein the guided aggregation model includes a random forest model.

[0009] In one embodiment of the present application, collecting multidimensional signals affected by human body movements includes: using a dynamic antenna system to collect multidimensional signals, including: controlling the dynamic antenna system to move along a first direction, a second direction, and a third direction, and rotating in a plane intersecting with any one of the first direction, the second direction, and the third direction to collect multidimensional signals, wherein the first direction, the second direction, and the third direction intersect with each other.

[0010] In one embodiment of the present application, when the number of human bodies is greater than or equal to 2, the number of dynamic antenna systems is 1.

[0011] In one embodiment of the present application, the method also includes: when the number of human bodies is greater than or equal to 2, the dynamic antenna system at the lowest transmission power determines the optimal receiving angle of the antenna by moving and rotating, wherein the optimal receiving angle is the angle with the least number of radio frequency electronic tags.

[0012] In one embodiment of the present application, determining human body movement based on the fused multi-dimensional signal includes: determining human body movement based on the difference between a first signal and a second signal, wherein the first signal is a signal received by a radio frequency electronic tag; and the second signal is a signal sent by the radio frequency electronic tag and affected by human body movement.

[0013] In one embodiment of the present application, before determining human body movements based on the fused multidimensional signal, the method also includes: performing a preprocessing operation on at least one of the multidimensional signals, including: performing phase restoration processing, wavelet threshold noise reduction processing, and normalization processing to enhance the features of the multidimensional signal affected by human body movements.

[0014] In one embodiment of the present application, the method further includes: dividing the multiple human body movements into different movement groups based on the waveform of the multidimensional signal, wherein the waveforms of the multidimensional signals of the multiple human body movements in the same movement group are similar.

[0015] In one embodiment of the present application, the start and end signal intensities of the multi-dimensional signals of multiple human body movements in the same movement group are different in one acquisition cycle.

[0016] On the other hand, the present application provides a human motion recognition system, which includes: an acquisition module, configured to acquire multi-dimensional signals affected by human motion, wherein the multi-dimensional signals are emitted by radio frequency electronic tags arranged in front of the human body; and an identification module, configured to determine human motion based on the fused multi-dimensional signals, wherein the multi-dimensional signals include at least two of a received signal strength indication signal, a phase signal, and a Doppler frequency shift signal; and the number of radio frequency electronic tags is the same as the number of human bodies.

[0017] In one embodiment of the present application, the recognition module is further configured to input the fused multi-dimensional signal into a trained guided aggregation model for classification to determine human body movements, wherein the guided aggregation model includes a random forest model.

[0018] In one embodiment of the present application, the acquisition module is further configured to utilize a dynamic antenna system to collect multi-dimensional signals, including: controlling the dynamic antenna system to move along a first direction, a second direction, and a third direction, and rotating in a plane intersecting with any one of the first direction, the second direction, and the third direction to collect multi-dimensional signals, wherein the first direction, the second direction, and the third direction intersect with each other.

[0019] In one embodiment of the present application, when the number of human bodies is greater than or equal to 2, the number of dynamic antenna systems is 1.

[0020] In one embodiment of the present application, the acquisition module is further configured to determine the optimal receiving angle of the antenna by moving and rotating the dynamic antenna system at the lowest transmission power when the number of human bodies is greater than or equal to 2, wherein the optimal receiving angle is the angle with the least number of radio frequency electronic tags.

[0021] In one embodiment of the present application, the identification module is further configured to determine human motion based on the difference between a first signal and a second signal, wherein the first signal is a signal received by the radio frequency electronic tag; and the second signal is a signal sent by the radio frequency electronic tag and affected by human motion.

[0022] In one embodiment of the present application, the system also includes a data processing module, which is configured to perform preprocessing operations on at least one of the multidimensional signals, including: phase restoration processing, wavelet threshold noise reduction processing and normalization processing to enhance the features of the multidimensional signals affected by human body movements.

[0023] In one embodiment of the present application, the recognition module is further configured to classify multiple human body movements into different movement groups based on the waveform of the multidimensional signal, wherein the waveforms of the multidimensional signals of multiple human body movements in the same movement group are similar.

[0024] In one embodiment of the present application, the recognition module is further configured so that the intensities of the start and end signals of the multi-dimensional signals of multiple human body movements in the same movement group in one acquisition cycle are different.

[0025] According to the human motion recognition method and system provided by at least one embodiment of the present application, a multi-dimensional signal affected by human motion is collected, wherein the multi-dimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body, and the human motion is determined based on the fused multi-dimensional signal. Thus, non-wearable motion recognition based on radio frequency identification technology is realized, and the reliability of single-person single signature in radio frequency identification technology is guaranteed by fusing multi-dimensional signals while protecting privacy and not being affected by ambient light, and multi-person motion recognition can be realized, which has strong practical value in single-person application scenarios as well as multi-person application scenarios including classrooms, conference rooms, study rooms and libraries. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings, in which:

[0027] Figure 1 is a flow chart of a human action recognition method according to one embodiment of the present application;

[0028] Figure 2 is an application scenario diagram of a human action recognition method according to an embodiment of the present application;

[0029] Figure 3 is an application scenario diagram of a human motion recognition method according to another embodiment of the present application;

[0030] Figure 4A is an application scenario diagram of a human motion recognition method according to another embodiment of the present application;

[0031] Figure 4B is a three-dimensional schematic diagram of a dynamic antenna system according to one embodiment of the present application;

[0032] Figure 5 is a schematic diagram of a random forest model of a human action recognition method according to an embodiment of the present application;

[0033] Figure 6 is a table showing the influence of different multi-dimensional signals on the accuracy of the action recognition method according to one embodiment of the present application;

[0034] Figure 7-Figure 13 They are the confusion matrix analysis diagrams of different signals as input signals of the classification model;

[0035] Fig.14 is a table showing the influence of different multi-dimensional signals and different classification models on the accuracy of the action recognition method according to one embodiment of the present application;

[0036] Fig.15 is a schematic diagram of dividing a plurality of human body movements into different movement groups according to one embodiment of the present application; and

[0037] Fig.16 It is a schematic diagram of a human motion recognition system according to one embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to better understand the present application, a more detailed description will be made of various aspects of the present application with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of exemplary embodiments of the present application, and are not intended to limit the scope of the present application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0039] It should be noted that in this specification, the expressions first, second, third, etc. are only used to distinguish one feature from another feature region, but do not represent any limitation on the features, and especially do not represent any order of precedence.

[0040] In the drawings, the thickness, size and shape of the components have been slightly adjusted for ease of illustration. The drawings are for illustration only and are not drawn strictly to scale. As used herein, the terms "substantially," "approximately," and similar terms are used as terms of approximation, not as terms of degree, and are intended to account for the inherent deviations in measurements or calculations that would be recognized by one of ordinary skill in the art.

[0041] It should also be understood that expressions such as "include", "including", "have", "contain" and / or "comprising" are open rather than closed expressions in this specification, which indicate the presence of the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features rather than just the individual elements in the list. In addition, when describing embodiments of the present application, "may" is used to mean "one or more embodiments of the present application". And, the term "exemplary" is intended to refer to an example or illustration.

[0042] Unless otherwise specified, all words (including engineering terms and scientific and technological terms) used in this article have the same meaning as those commonly understood by ordinary technicians in the field to which this application belongs. It should also be understood that, unless clearly stated in this application, words defined in common dictionaries should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.

[0043] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. In addition, unless explicitly limited or contradictory to the context, the specific steps included in the method described in the present application are not necessarily limited to the order described, but can be performed in any order or in parallel. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0044] Figure 1 FIG. 1 is a flow chart of a method for human motion recognition according to an embodiment of the present application. Figure 1 As shown, the present application provides a human action recognition method 1000 including:

[0045] S1, collecting multi-dimensional signals affected by human body movements, wherein the multi-dimensional signals are emitted by radio frequency electronic tags arranged in front of the human body, the number of radio frequency electronic tags is the same as the number of human bodies, and the multi-dimensional signals include at least two of a received signal strength indication signal, a phase signal, and a Doppler frequency shift signal.

[0046] S2, determines the human body motion based on the fused multi-dimensional signals.

[0047] The above-mentioned human action recognition method 1000 will be described in detail below.

[0048] With the development of the Internet of Things, human motion recognition has been widely used in all aspects of social life and plays a key role in various fields. In the field of health care, human motion recognition can provide real-time feedback and evaluation tools for rehabilitation training of patients with movement disorders, and help doctors better monitor the recovery process of patients. In the field of human-computer interaction, human motion recognition enables machines to understand human intentions and needs more naturally, thereby providing a better interactive experience. In security monitoring systems, human motion recognition can effectively identify and warn potential security threats, such as intrusion detection and abnormal behavior analysis.

[0049] There are many solutions for human motion recognition based on sensors, computer vision, and wireless signals.

[0050] The sensor-based human motion recognition method can obtain data information through different sensors worn by users or deployed in the environment in advance, and recognize human motion based on this data information. For example, by wearing multiple IMU (Inertial Measurement Unit) sensors on the user and using deep neural networks, the whole body posture of the human body can be reconstructed in real time; or, by collecting data through the built-in accelerometer, gyroscope, etc. of the mobile phone, and by establishing a recognition model, human motion classification can be performed. However, the sensor-based human motion recognition method requires the use of multiple high-precision sensors to collect human-related data for motion recognition. Since it is only applicable to specific scenarios, such as scenarios that require users to wear or deploy in advance in the environment, it has limitations.

[0051] Human action recognition methods based on computer vision can use cameras deployed in the environment to obtain relevant data on human actions, and classify and recognize human actions based on these relevant data. For example, the key frames of each action are extracted from the video, and transfer learning is used to obtain deep image features and well-trained machine learning classifiers to classify human actions in the data set using machine learning models; or, a recognition framework based on Fisher vector (FV, a coding method similar to the bag-of-words model) encoding and a multi-class support vector machine are used to classify human actions. However, although the human action recognition method based on computer vision has good recognition effects and a quick deployment method, it is sensitive to light, and there are also problems such as privacy leakage and failure due to obstruction of line of sight.

[0052] Wireless signals include WiFi (mobile hotspots), sound waves, millimeter waves, and radio frequency signals, etc. The human motion recognition method based on wireless signals not only has no requirements for ambient light, but also can realize the recognition of human motion under the premise of user privacy and security.

[0053] Specifically, the WiFi-aware human motion recognition method can accurately recognize human posture and motion based on the human body's ability to absorb and reflect wireless signals. On the one hand, although WiFi recognition is not affected by ambient light, it relies on CSI (Channel State Information) data, which is only supported by a few computer network cards, making it difficult to popularize the method. On the other hand, WiFi signals are easily affected by the environment, causing problems with large amounts of data processing. In addition, due to its own limitations, the WiFi-aware human motion recognition method still has many deficiencies in the human body recognition process of multiple users.

[0054] The human motion recognition method based on acoustic wave perception can accurately recognize human posture and motion by analyzing and processing acoustic wave signals. For example, the speaker and microphone in a smartphone can be used to recognize the motion of a single person, or a single speaker and microphone on a smart device can be used to accurately classify multiple human motions. However, in the above process, multiple mobile devices often need to be deployed, which is not conducive to practical application. In addition, the microphone-speaker combination needs to face the human body and is relatively close to the human body, which limits the application scenarios and deployment methods of the human motion recognition method based on acoustic wave perception.

[0055] The human motion recognition method based on millimeter-wave radar perception can use the penetration and reflection capabilities of millimeter waves to detect the human body in a non-contact manner. For example, by collecting human motion data and extracting features using clustering algorithms and neural networks, human motion can be recognized; or human motion data can be extracted from a noisy background environment, and then a custom-designed lightweight neural network can be used to classify the motion. However, the cost of millimeter-wave radar is too high, and data processing in multi-person scenarios is difficult, which is not conducive to development.

[0056] Table 1 shows the comparative data of various action recognition methods in terms of cost, user privacy protection, portability, and deployment cost.

[0057] As shown in Table 1, due to the "diminishing marginal effect", the above-mentioned multiple methods of human motion recognition based on wireless signals need to deploy devices greater than or equal to the number of subjects in order to effectively recognize the motions of multiple people, which limits the practical application of the above methods.

[0058] Action Recognition Methods cost Protecting User Privacy Removable Additional deployment costs Camera-based Low No No medium Sensor-based Low Can Can medium RFID higher Can Can Low Millimeter wave higher Can No higher ultrasound Low Can No medium

[0059] Table 1

[0060] RFID is a communication technology that can identify specific targets and read and write related data through radio signals. RFID can be used for motion recognition. For example, when the human body to be tested moves within the communication range of the RF (Radio Frequency) signal, the movements it makes will affect the RF signal, and different movements will have different effects on the RF signal. By establishing a relationship between the affected signal and the corresponding movement, the purpose of human movement recognition can be achieved.

[0061] Optionally, the RFID system may include three basic components, a radio frequency electronic tag (Tag), a radio frequency identification reader and an antenna. The radio frequency identification reader can communicate with the radio frequency electronic tag through radio waves via the antenna, thereby realizing the reading or writing operation of the electronic code of the radio frequency electronic tag and its memory data.

[0062] In some embodiments, a tag array consisting of multiple tags can be placed behind the user, and a variety of sitting postures can be identified by using the correlation between phase changes and sitting postures. In addition, the tag array attached to the human body can also be used to estimate the three-dimensional human posture, for example, two-person motion recognition can be achieved through multiple antennas and multiple tags. However, although these human motion recognition methods using RFID are effective, they are only suitable for specific scenarios, such as scenarios that require users to wear or deploy in advance in the environment, and therefore have limitations. In addition, there is serious mutual interference between multiple tag arrays in multi-person scenarios, which leads to the deterioration of the accuracy and reliability of the multi-person human motion recognition method using RFID.

[0063] According to the human motion recognition method provided by at least one embodiment of the present application, a multi-dimensional signal affected by human motion is collected, wherein the multi-dimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body, and the human motion is determined based on the fused multi-dimensional signal. Thus, non-wearable motion recognition based on radio frequency identification technology is realized, and the reliability of single-person single signature in radio frequency identification technology is guaranteed by fusing multi-dimensional signals while protecting privacy and not being affected by ambient light, and multi-person motion recognition can be realized, which has strong practical value in single-person application scenarios as well as multi-person application scenarios including classrooms, conference rooms, study rooms and libraries.

[0064] Step S1

[0065] Figure 2 This is an application scenario diagram of a human motion recognition method according to an embodiment of the present application. Figure 3 This is an application scenario diagram of a human motion recognition method according to another embodiment of the present application. Figure 4A This is an application scenario diagram of a human motion recognition method according to another embodiment of the present application.

[0066] Specifically, in Figure 2 or Figure 3 In the single-person application scenario 1 or 2 shown, the radio frequency electronic tag (Tag) 200 can be set in front of the human body 100. In addition, the antenna (Antenna) 301 can be set on the side of the radio frequency electronic tag 200 away from the human body 100 and aligned with the radio frequency electronic tag 200 at a certain distance (Distance), wherein the distance can be between 0.5-3m.

[0067] Exemplarily, a radio frequency identification reader 400 is further provided in the single-person application scenario 1 or 2. Optionally, the communication frequency of the radio frequency identification reader 400 may be 920.625 MHz.

[0068] As an option, in some embodiments of the present application, the human body 100 can move at a natural speed, wherein the duration of the action collection can be, for example, 4 seconds, and the collected radio frequency electronic tag information can include the radio frequency electronic tag number and multi-dimensional signal, and the multi-dimensional signal can include at least two of the received signal strength indication signal (RSSI), phase signal (Phase angle) and Doppler shift signal (Doppler Shift). Optionally, in single-person application scenario 1 or 2, the antenna 301 can be at the lowest transmission power, and at this power, the radio frequency electronic tag 200 is identified.

[0069] In some embodiments of the present application, the RFID reader / writer 400 may first transmit a first signal to the RFID electronic tag 200, and after receiving the first signal, the RFID electronic tag 200 may transmit a second signal identical to the first signal to the RFID reader / writer 400. In other words, the first signal is a signal received by the RFID electronic tag 200, and the second signal is a signal sent by the RFID electronic tag 200, wherein both the first signal and the second signal may be multi-dimensional signals, and the second signal should be identical to the first signal, but affected by the movement of the human body 100, the second signal changes, resulting in a difference between the first signal and the second signal. Therefore, based on the change of the second signal due to the movement of the human body 100, the movement of the human body can be determined by determining the difference between the first signal and the second signal.

[0070] Taking the example that both the first signal and the second signal include RSSI and Phase angle, the first signal including RSSI and Phase angle is the signal received by the radio frequency electronic tag 200, and the second signal including RSSI and Phase angle is the signal sent by the radio frequency electronic tag 200, and the two should be the same. However, affected by the movement of the human body 100, the RSSI in the second signal has changed compared with the RSSI in the first signal; in addition, the Phase angle in the second signal has also changed compared with the Phase angle in the first signal. Based on the changes in the two signals caused by the movement of the human body 100, the movement of the human body 100 can be distinguished. In addition, due to the connectivity between the multi-dimensional signals, the differences between the above-mentioned different movements can also be amplified, which can realize the recognition of human body movements.

[0071] In addition, reference Figure 4AIn some embodiments of the present application, step S1 collects a multidimensional signal affected by human body movements, wherein the multidimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body, the number of radio frequency electronic tags is the same as the number of human bodies, and the multidimensional signal includes at least two of a received signal strength indication signal, a phase signal, and a Doppler frequency shift signal, which may include, for example: using a dynamic antenna system 302 to collect the above-mentioned multidimensional signal, wherein the dynamic antenna system 302 is controlled to move along a first direction (x direction), a second direction (y direction), and a third direction (z direction), and rotate in a plane intersecting any one of the x direction, the y direction, and the z direction, so that the above-mentioned multidimensional signal can be collected, wherein the x direction, the y direction, and the z direction intersect with each other.

[0072] Figure 4B It is a three-dimensional schematic diagram of a dynamic antenna system 302 according to an embodiment of the present application.

[0073] Specifically, Figure 4A and Figure 4B As shown, the dynamic antenna system 302 can rotate on a three-dimensional hemispherical surface. Optionally, the dynamic antenna system 302 may include an antenna 3201, a two-degree-of-freedom servo platform and a fixed frame 3204, wherein the servo platform is matched with a retractable fixed frame 3204 to adjust the height of the entire dynamic antenna system 302 (which can be understood as the size in the z direction). In addition, the servo platform can be bound to the antenna 3201 through the fixed frame. The servo platform may include a servo (e.g., a vertical servo 3202 and a horizontal servo 3203), a platform bracket, a servo control board and a host computer. The servo can realize rotation in multiple directions, for example, the vertical servo 3202 can realize rotation around the z direction, and the horizontal servo 3203 can rotate in the xy plane. The servo control board may include a power module, a communication module and a control module. The power module is used to power the servo, and the control module can be powered by, for example, a step-down circuit. The communication module can use serial communication and be connected to the host computer through a Type-C interface. The control module can control the rotation angle and speed of the servo by outputting PWM signals.

[0074] Therefore, the dynamic antenna system 302 can move along the x-direction, the y-direction, and the z-direction, and can rotate in a plane intersecting any one of the x-direction, the y-direction, and the z-direction, so as to collect multi-dimensional signals emitted by the radio frequency electronic tag 200 and affected by the human body motion emitted by the human body 100 through the above movement and rotation. Optionally, the x-direction, the y-direction, and the z-direction can be perpendicular to each other.

[0075] In addition, refer again Figure 4AIn multi-person application scenario 3, the dynamic antenna system 302 can also be at the lowest transmission power, and identify the radio frequency electronic tag 200 at this power. In other words, when the number of human bodies is greater than or equal to 2, the dynamic antenna system 302 at the lowest transmission power (Low Power) can move along the x direction, the y direction, and the z direction, and rotate in a plane intersecting any one of the x direction, the y direction, and the z direction to determine the best receiving angle of the antenna in the dynamic antenna system 302, wherein the best receiving angle is the angle with the least number of radio frequency electronic tags 200. The best receiving angle is the angle with the least number of multi-dimensional signals sent by the radio frequency electronic tags 200 received among all angles.

[0076] Since the multi-dimensional signal emitted by the RFID tag 200 is affected by the angle and distance between the RFID tag 200 and the antenna and the overall test environment, it takes a lot of time to determine the best receiving angle. The dynamic antenna system 302 can solve the above problem by moving along the x-direction, y-direction and z-direction at the lowest transmission power and rotating in a plane intersecting any one of the x-direction, y-direction and z-direction.

[0077] For example, Figure 4A In the illustrated multi-person application scenario 3, the multiple human bodies 100 include human body No. 1 to be tested 101 and human body No. 5 to be tested 105, etc. The radio frequency electronic tag 200 can be set in front of the human body 100, for example, a corresponding radio frequency electronic tag 200 is set in front of each human body to be tested, wherein the radio frequency electronic tag No. 1 201 is set in front of human body No. 101 to be tested, the radio frequency electronic tag No. 5 205 is set in front of human body No. 5 105 to be tested, and so on.

[0078] Optionally, when the number of human bodies is greater than or equal to 2, the number of dynamic antenna systems 302 may be 1.

[0079] In addition, in this case, the number of the RFID reader / writer 400 may be 1, thereby reducing the cost of motion recognition and improving the efficiency of motion recognition.

[0080] In other words, in a multi-person application scenario, it is usually necessary to set up multiple Tag arrays and multiple antennas to achieve multi-person motion recognition. However, this deployment method is cumbersome, and the number of tags in multiple Tag arrays will increase dramatically, which seriously interferes with the recognition effect of a single Tag array. In addition, due to mutual interference between different Tag arrays, a different frequency band needs to be set for each Tag array, which in turn requires the deployment of new RFID readers and antennas, which not only increases the cost of motion recognition, but also reduces the efficiency of motion recognition. During the recognition process, it is necessary to switch the antenna path and obtain the response of different channels at different frequency bands, which is not conducive to the application of motion recognition methods.

[0081] In at least one embodiment of the present application, by collecting multi-dimensional signals affected by human body movements, wherein the multi-dimensional signals are emitted by a radio frequency electronic tag disposed in front of the human body, and based on the fused multi-dimensional signals, the distinction between different movements can be enhanced to facilitate determination of human body movements, thereby realizing non-wearable movement recognition based on radio frequency identification technology. In addition, by fusing multi-dimensional signals, the reliability of single-person single-signature in radio frequency identification technology can be guaranteed, and in multi-person application scenarios, accurate and reliable human movement recognition can also be realized by fusing multi-dimensional signals and using a dynamic antenna system of one.

[0082] In addition, refer again Figure 4A In multi-person application scenario 3, the dynamic antenna system 302 and the multiple radio frequency electronic tags 200 cannot be arranged directly opposite each other. As an option, the angle between the dynamic antenna system 302 and the multiple radio frequency electronic tags 200 can be changed by a predetermined value. For example, the relative angles between the dynamic antenna system 302 and the multiple radio frequency electronic tags 200 can be -60°, -30°, 0°, 30°, and 60°, respectively. In this case, the recognition effects of the actions performed by the human body corresponding to different radio frequency electronic tags are not much different.

[0083] In addition, reference Figure 2 and Figure 3 Optionally, in single-person application scenario 1 or 2, the antenna 301 may not be directly opposite to the radio frequency electronic tag 200. For example, the relative angle between the two may be selected between 0° and 80°, for example, the relative angle may be 60°.

[0084] In addition, in some embodiments of the present application, the radio frequency electronic tag 200 may be a bar-shaped radio frequency electronic tag. Taking a square radio frequency electronic tag as an example, the square radio frequency electronic tag has a larger area, a stronger backscatter signal, and is overly sensitive to human body movements. In addition, in a multi-person application scenario, the distance between multiple square radio frequency electronic tags is too close, which is more likely to cause signal interference.

[0085] Step S2

[0086] Figure 5 It is a schematic diagram of a random forest model of a human motion recognition method according to an embodiment of the present application.

[0087] Alternatively, if Figure 5 As shown, step S2 determines the human body movement based on the fused multi-dimensional signal, which may for example include: inputting the fused multi-dimensional signal into a trained guided aggregation model for classification to determine the human body movement, wherein the guided aggregation model may include a random forest model.

[0088] Specifically, the fused multidimensional signal can be understood as a signal formed after the fusion of multidimensional signals. For example, a signal formed after the fusion of at least two of the multidimensional signals including RSSI signals, Phase angle signals, and Doppler Shift signals. In view of the fact that there is less data in the data set formed by the fused multidimensional signal, the data set selection can be based on the boostrap sampling method. For example, sampling the initial data set TD (Trainingset) containing m samples can generate a data set TD' (Training set_m), in which a sample is randomly selected from TD each time, put into TD', and then put back into TD, so that the sample may still be sampled at the next sampling. This process can be repeated m times to obtain TD' containing m samples.

[0089] Machine learning methods are widely used in classification and regression tasks. Optionally, the fused multi-dimensional signal can be input into a trained classification model, and the classification model can be used to classify and determine the human body movements.

[0090] In addition, the selection of classification models should take into account the dependencies between the motion features represented by the multidimensional signals. For example, the Naive Bayes (NB) algorithm model is a classifier method based on the Bayesian definition and the conditional independence assumption of features. It assumes that the motion features are independent of each other. Therefore, when the same fused multidimensional signal is used as input, the effect of using the Naive Bayes algorithm model to determine human motion will be poor.

[0091] As an option, in order to solve the problems of overfitting caused by fewer training samples and data noise in complex environments, after extracting effective features from the multidimensional signal sequence, the present application can use ensemble learning to complete the learning and classification tasks by constructing and combining multiple decision trees. In addition, during the model training stage, due to the randomness of the samples, each decision tree can be different, increasing the diversity of the model.

[0092] The pseudo code for training a decision tree is as follows:

[0093]

[0094] The random forest model can effectively deal with the overfitting problem and has good prediction accuracy and generalization ability. Its advantages lie in randomness and integration. Randomness refers to sample randomness and feature randomness, which makes each decision tree different and increases the diversity of the model. Integration is to combine the prediction results of multiple decision trees (for example, decision tree A, decision tree B, decision tree C, and decision tree D) by voting to reduce variance and improve the stability and accuracy of the model. In the model testing phase, the prediction results of multiple decision trees are combined by voting to reduce variance and improve the stability and accuracy of the model. As shown in formula (1), Figure 5 One way to implement the combiner in is to output the prediction result with the most votes.

[0095]

[0096] In addition, in order to achieve the best effect of the Bagging model, the grid search (GridSearchCV) in the Scikit-Learn library can be used to optimize the model, and cross-validation can be used to maximize the use of the data set to train and test the Bagging model.

[0097] In conventional RFID human motion recognition methods, although multidimensional signals are also used, different signals in the multidimensional signals are modeled or feature extracted separately, and then the results are merged, which loses the correlation between different signals. At least one embodiment of the present application uses the fused multidimensional signal to determine human motion, which can complement the motion features and improve the recognition rate of the motion recognition method.

[0098] In some embodiments of the present application, different types of signals can be combined to form one-dimensional (pure signal), two-dimensional (combination of two types of signals), and three-dimensional (combination of three types of signals) data, and the data of different dimensions can be used as inputs of the classifier to evaluate the impact of different inputs on the recognition effect.

[0099] Specifically, the RSSI signal can be used to determine the link quality and is expressed using formula (2):

[0100]

[0101] Where η is the path loss exponent; X σ It can be expressed as a mean of 0 and a variance of σ 2 Gaussian random variable; and d is the distance between the radio frequency electronic tag and the antenna.

[0102] The Phase angle signal can represent the angle of antenna transmission and reception, as well as the inherent phase offset caused by the backscattering of the RFID tag on the antenna. The Phase angle can be expressed using formula (3):

[0103]

[0104] Among them, θ, θ T ,θ R and θ Tag They represent Phase angle, antenna transmission angle, receiving angle and inherent phase shift caused by backscattering of radio frequency electronic tags on the antenna.

[0105] The Doppler Shift signal can reflect the relative movement between the human body and the RFID tag, and can be calculated using formula (4):

[0106] express:

[0107] Δf=f′-f (4)

[0108] Where f′ is the received frequency, f is the antenna transmission frequency, and Δf is the difference between the received frequency and the transmission frequency.

[0109] When detecting human motion, the position of the RFID tag and the antenna remains unchanged, so the RSSI, Phrase, Doppler Shift and other signals should also remain unchanged. Therefore, the changes in RSSI, Phrase, Doppler Shift and other signals reflect the impact of human motion on the signal emitted by the RFID tag.

[0110] Optionally, after different one-dimensional or fused multi-dimensional signals are input into the model, the corresponding confusion matrix output by the model can be used to evaluate and optimize the model.

[0111] Figure 6 This is a table showing the impact of different signals on the accuracy of the action recognition method according to one embodiment of the present application.

[0112] Figure 7-Figure 13 These are the confusion matrix analysis diagrams of different signals as input signals of the classification model.

[0113] like Figure 6-Figure 11 As shown, for the convenience of recording, the RSSI signal, Phase signal and one-dimensional signal of Doppler Shift are abbreviated as R, P and D respectively, and the fused two-dimensional signal of RSSI signal and Phase signal is recorded as R+P, the fused two-dimensional signal of Phase signal and Doppler Shift signal is recorded as P+D, and the fused three-dimensional signal of RSSI signal, Phase signal and Doppler Shift signal is recorded as R+P+D.

[0114] By comparison, it is not difficult to find that the fused multi-dimensional signal has better action recognition effect than the one-dimensional signal (for example, R, P or D). For example, the action recognition effect of R+P and P+D is nearly 20% higher than that of R and P alone, and doubled that of D alone. This is because, although the model recognition rates of R and P are not much different, the information they carry is not exactly the same. Therefore, the action recognition rate can be improved by complementing the action features of different signals. This is different from the conventional method, which only models the signals separately and then merges the results. Therefore, the conventional method cuts off the connection between different signals.

[0115] In addition, the more fused multi-dimensional signals, the better. For example, R+P+D is not as good as R+P, P+D in motion recognition, and R+D is not as good as R, P in motion recognition. This is because the D signal is not stable and will interfere with the judgment of the classification model. If unreliable signal data is added to the input of the classification model, the performance of the classification model may be reduced.

[0116] In addition, there are conflicts between different signals. For example, the action recognition effect of R+P and R+D is good, but the action recognition effect of R+P+D is poor, and the action recognition effect of R+D is even worse. This shows that there is a conflict between the two signals R and D. If both are used as data input for the classification model, the recognition effect of the classification model will be weakened.

[0117] Therefore, in at least one embodiment of the present application, the multidimensional signal can be a signal obtained by merging different signals that represent complementary motion features, merging different signals that represent complementary motion features to obtain a fused multidimensional signal, and using the fused multidimensional signal as data input for a classification model can improve the recognition effect of motion recognition. Alternatively, different signals that represent dependent motion features are fused, and the fused multidimensional signal is used as data input for a classification model to improve the recognition effect of motion recognition.

[0118] Fig.14 This is a table showing the influence of different multi-dimensional signals and different classification models on the accuracy of the action recognition method according to one embodiment of the present application.

[0119] like Fig.14As shown, the classification models may include Random Forest model (RF, K-Nearest Neighbor algorithm model (KNN), Support Vector Machine (SVM), Decision Tree model (DT), Naive Bayes algorithm model (NB), Quadratic Discriminant Analysis algorithm model (QDA), Extreme Gradient Boosting Decision Tree model (XGBoost), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM).

[0120] Table 2 shows the impact of different multidimensional signals and different classification models on the accuracy of action recognition methods.

[0121] R P D R+P R+D P+D R+P+D DT 39.0 40.2 23.9 57.1 32.8 53.5 36.1 KNN 38.8 40.5 26.7 61.0 37.2 54.2 44.1 NB 22.6 21.7 21.4 29.4 25.6 25.8 26.8 QDA 34.0 42.0 25.4 39.2 28.6 26.7 28.7 RF 59.9 63.1 41.3 81.0 54.2 79.1 69.9 SVM 36.2 33.3 20.3 50.0 26.3 33.3 30.7 CatBoost 42.0 43.2 34.6 64.2 41.6 60.1 48.1 XGBoost 52.8 55.3 34.7 81.0 43.6 70.4 54.9 LightGBM 55.0 56.4 34.9 82.9 47.7 76.5 57.3

[0122] Table 2

[0123] Combining Table 2 and Fig.14 , the fused two-dimensional signal R+P performs better than the fused two-dimensional signal P+D under all classification models. In addition, the performance of different classification models varies greatly, such as the difference between the NB classification model and the RF classification model is more than 2.5 times. The NB model has the lowest recognition rate, less than 30%. The NB classification model assumes that the motion features are independent of each other. Therefore, when the same fused multi-dimensional signal is used as input, the effect of using the NB classification model to determine human motion will be poor.

[0124] In addition, the fine-grained signals R and P are better than the coarse-grained signal D, and the fine-grained signal P is slightly better than the signal R. Compared with a single signal, the classification model converges faster and has a higher recognition rate for action recognition using the fused two-dimensional signals R+P and P+D as inputs. However, the higher the data dimension, the better. Taking the decision tree model as an example, the model effect of the fused three-dimensional signal R+P+D is only 36.1%, even lower than the 39.0% using only the R signal. In addition, the fused two-dimensional signals R+P and P+D are much more accurate than the classification model that only uses one-dimensional data as input data.

[0125] Fig.15 It is a schematic diagram of dividing multiple human body movements into different movement groups according to one embodiment of the present application.

[0126] like Fig.15 As shown, in some embodiments of the present application, the human motion recognition method 1000 further includes: dividing the multiple human motions into different motion groups based on the waveform of the multidimensional signal, wherein the waveforms of the multidimensional signals of the multiple human motions in the same motion group are similar. In addition, in different motion groups, the waveforms of the multidimensional signals of the multiple human motions are different.

[0127] For example, in some embodiments of the present application, multiple human body movements are divided into four different movement groups, among which standing up (Stand) and sitting down (Sit) are the first group (Group 1); raising hand (Raisehand) and dropping hand (Drophand) are the second group (Group 2); sleeping (Sleep) and waking up (Wake) are the third group (Group 3); and static (Static) and walking (Walk) are the fourth group (Group 4).

[0128] Fig.15 The RSSI signals and Phase angle signals of the above eight actions are shown, wherein the RSSI signal is located above the Phase angle signal. The waveforms of the multidimensional signals of multiple human body actions in the same action group are similar. For example, in the first group, the waveforms of the RSSI signals of standing up and sitting down are similar. In addition, the Phase angle signals of the two are also similar. Similarly, in the second group, the waveforms of the RSSI signals of raising hands and lowering hands are similar, and the Phase angle signals of the two are also similar.

[0129] In addition, the waveforms of the multi-dimensional signals of multiple human body movements are different in different action groups. For example, in the first group and the second group, the waveforms of the RSSI signals of standing up and raising hands are completely different, and the Phase angle signals of the two are also completely different.

[0130] Optionally, refer again to Fig.15 , the multi-dimensional signals of multiple human body movements in the same action group have different start and end signal strengths in one acquisition cycle. For example, in the first group, the start and end signal strengths of the RSSI signals of standing up and sitting down in one acquisition cycle are different. In addition, the start and end signal strengths of the Phase angle signals of the two in one acquisition cycle are also different.

[0131] By dividing multiple human body movements into different action groups, the waveform changes of the multi-dimensional signals of different actions in the group are relatively similar, which can improve the efficiency and accuracy of the action recognition method. For example, compared with the standing up of the first group and the raising of the hands of the second group, the signal changes of the former are more complex. The second group is the arm movements, the first and third groups are the torso movements, and the fourth group reflects the changes in the position of the human body. This can make the human body movements in the same group have more similar action features, and the multi-dimensional signals of the same group of actions have a strong correlation, which is conducive to improving the efficiency and accuracy of the action recognition method.

[0132] In addition, in some embodiments of the present application, the human motion recognition method 1000 also includes: before determining the human motion based on the fused multidimensional signal, performing a preprocessing operation on at least one of the multidimensional signals, and the preprocessing operation may include: performing phase restoration processing, wavelet threshold denoising processing and normalization processing to enhance the features in the multidimensional signal affected by the human motion.

[0133] The multi-dimensional signal data collected by the RFID reader may contain noise caused by various factors, such as environmental noise, data blanks caused by equipment defects, and phase wrapping, etc. Therefore, the collected multi-dimensional signal data can be subjected to phase restoration processing, wavelet threshold noise reduction processing, normalization processing, etc., so as to enhance the characteristics of the multi-dimensional signal affected by human motion.

[0134] Optionally, phase wrapping means that when the actual phase is less than 0 or greater than 2π, the phase value jumps. From Formula 3, it can be seen that the phase range is [0, 2π], so a remainder operation is performed when the phase is calculated. Some embodiments of the present application restore the phase curve to make it a continuous curve, thereby improving the accuracy and reliability of the human motion recognition method.

[0135] In addition, wavelet threshold denoising can be performed on multidimensional signal data. For example, noise components in multidimensional signal data can be removed by decomposition and selective reconstruction. Since wavelet threshold denoising has a strong data-removal correlation, the amplitude of the wavelet coefficient of the original signal after wavelet decomposition is greater than the amplitude of the noise coefficient. Therefore, the coefficients of the part less than the threshold function are removed as interference noise, and then wavelet reconstruction is performed to retain the characteristics of the multidimensional signal affected by human body movements.

[0136] Considering that there are obvious differences in data types and numerical ranges in multidimensional signals, this not only increases the training time of the classification model, but also may cause the classification model to fail to converge. In addition, the base value of the antenna's reading of the multidimensional signal emitted by the RFID tag is related to multiple factors. For example, the distance, angle, and test environment between the RFID reader and the RFID tag. Considering that the signal emitted by the RFID tag changes due to the influence of human body movements, the human body movement can be determined by determining the difference between the signals received and emitted by the RFID tag. It is necessary to eliminate the above base value influence and remove the base value.

[0137] Normalization processing can map multidimensional signal data to the range of [-1,1], which not only amplifies the signal changes caused by different actions, but also weakens the impact of inconsistent base values. i (1≤i≤n), the normalized value can be expressed as formula (5):

[0138]

[0139] Among them, S i , S max , S min They respectively represent the average value, maximum value and minimum value of a certain signal in the multidimensional signal.

[0140] Therefore, according to the human motion recognition method provided by at least one embodiment of the present application, a multi-dimensional signal affected by human motion is collected, wherein the multi-dimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body, and the human motion is determined based on the fused multi-dimensional signal. Therefore, non-wearable motion recognition based on radio frequency identification technology is realized, and the reliability of single-person single signature in radio frequency identification technology is guaranteed by fusing multi-dimensional signals on the basis of protecting privacy and not being affected by ambient light, and multi-person motion recognition can be realized, which has strong practical value.

[0141] It should be noted that the fused multi-dimensional signal can be understood as a signal formed after the above pre-processing, including at least two of the received signal strength indication signal, the phase signal and the Doppler frequency shift signal are combined.

[0142] However, those skilled in the art should understand that, without departing from the technical solution claimed in the present application, the specific steps of the human motion recognition method can be changed to obtain the various results and advantages described in this specification.

[0143] Fig.16 is a schematic diagram of a human motion recognition system 2000 according to another embodiment of the present application. Fig.16As shown, another aspect of the present application also provides a human motion recognition system 2000. The human motion recognition system 2000 may include: an acquisition module 2100 and an identification module 2300. The acquisition module 2100 is configured to acquire a multi-dimensional signal affected by human motion, wherein the multi-dimensional signal is emitted by a radio frequency electronic tag arranged in front of the human body. The identification module 2300 is configured to determine the human motion based on the fused multi-dimensional signal, wherein the multi-dimensional signal includes at least two of a received signal strength indication signal, a phase signal, and a Doppler frequency shift signal; and the number of radio frequency electronic tags is the same as the number of human bodies.

[0144] In addition, machine learning methods are widely used in classification and regression tasks. Optionally, the fused multi-dimensional signal can be input into a trained classification model, and the classification model can be used to classify and determine the human body movements.

[0145] In order to solve the problems of overfitting caused by fewer training samples and data noise in complex environments, the present application can use a guided aggregation model to classify different human body movements after extracting effective features from the multidimensional signal sequence. Therefore, the recognition module 2300 can be further configured to input the fused multidimensional signal into the trained guided aggregation model for classification to determine the human body movement, wherein the guided aggregation model includes a random forest model.

[0146] In addition, the selection of the classification model should take into account the dependencies between the motion features represented by the multidimensional signals. For example, different signals that are stable and represent complementary motion features can be fused, and the fused multidimensional signal can be used as the data input of the classification model to improve the recognition effect of motion recognition. Alternatively, different signals that represent dependencies between motion features can be fused, and the fused multidimensional signal can be used as the data input of the classification model to improve the recognition effect of motion recognition.

[0147] Optionally, combined Figure 4A and Fig.16 , the acquisition module 2100 may be further configured to acquire multi-dimensional signals using the dynamic antenna system 302. Specifically, by controlling the dynamic antenna system 302 to move along the x-direction, the y-direction, and the z-direction, and to rotate in a plane intersecting any one of the x-direction, the y-direction, and the z-direction, the multi-dimensional signals may be acquired, wherein the x-direction, the y-direction, and the z-direction intersect with each other. In addition, in a multi-person application scenario, the number of human bodies is greater than or equal to 2, and the number of dynamic antenna systems 302 may be 1.

[0148] Usually, in multi-person application scenarios, it is necessary to set up multiple tag arrays and multiple antennas to achieve multi-person motion recognition. However, this deployment method is cumbersome, and the number of tags in multiple tag arrays will increase dramatically, which seriously interferes with the recognition effect of a single tag array. In addition, due to mutual interference between different tag arrays, different frequency bands need to be set for each tag array, which in turn requires the deployment of new RFID readers and antennas, which not only increases the cost of motion recognition, but also reduces the efficiency of motion recognition. During the recognition process, it is necessary to switch the antenna path and obtain the response of different channels at different frequency bands, which is not conducive to the application of motion recognition methods.

[0149] In at least one embodiment of the present application, by collecting multi-dimensional signals affected by human body movements, wherein the multi-dimensional signals are emitted by a radio frequency electronic tag arranged in front of the human body, and based on the fused multi-dimensional signals, human body movements can be determined. Non-wearable motion recognition based on radio frequency identification technology is realized, and the reliability of single-person single-signature in radio frequency identification technology is ensured by fusing multi-dimensional signals. In multi-person application scenarios, accurate and reliable human body motion recognition can also be realized by fusing multi-dimensional signals and using a dynamic antenna system with a quantity of 1.

[0150] In addition, the acquisition module 2100 can be further configured so that when the number of human bodies is greater than or equal to 2, the dynamic antenna system 302 at the lowest transmission power can determine the best receiving angle of the antenna by moving and rotating, wherein the best receiving angle is the angle with the least number of radio frequency electronic tags 200. The best receiving angle is the angle with the least number of multi-dimensional signals emitted by the radio frequency electronic tags 200 received among all angles.

[0151] Since the multi-dimensional signal emitted by the RFID tag 200 is affected by the angle and distance between the RFID tag 200 and the antenna and the overall test environment, it takes a lot of time to determine the best receiving angle. The dynamic antenna system 302 can solve the above problem by moving along the x-direction, y-direction and z-direction at the lowest transmission power and rotating in a plane intersecting any one of the x-direction, y-direction and z-direction.

[0152] Combination Figure 2-4A as well as Fig.16 In one embodiment of the present application, the identification module 2300 can be further configured to determine human motion based on the difference between a first signal and a second signal, wherein the first signal is a signal received by the radio frequency electronic tag 200; and the second signal is a signal sent by the radio frequency electronic tag 200 and affected by human motion.

[0153] In other words, the first signal is a signal received by the radio frequency electronic tag 200, and the second signal is a signal sent by the radio frequency electronic tag 200, wherein both the first signal and the second signal can be multi-dimensional signals, and the second signal should be the same as the first signal, but the second signal changes due to the movement of the human body 100, resulting in a difference between the first signal and the second signal. Therefore, based on the change of the second signal due to the movement of the human body 100, the movement of the human body can be determined by determining the difference between the first signal and the second signal.

[0154] In addition, in one embodiment of the present application, the human motion recognition system 2000 further includes a data processing module 2200, which can be configured to perform a preprocessing operation on at least one of the multidimensional signals. The preprocessing operation can include: performing phase restoration processing, wavelet threshold noise reduction processing, and normalization processing to enhance the features in the multidimensional signal affected by human motion.

[0155] The multi-dimensional signal data collected by the RFID reader may contain noise caused by various factors, such as environmental noise, data blanks caused by equipment defects, and phase wrapping, etc. Therefore, the collected multi-dimensional signal data can be subjected to phase restoration processing, wavelet threshold noise reduction processing, normalization processing, etc., so as to enhance the characteristics of the multi-dimensional signal affected by human motion.

[0156] like Fig.15 and Fig.16 As shown, in one embodiment of the present application, the recognition module 2300 can be further configured to divide multiple human body movements into different action groups based on the waveform of the multidimensional signal, wherein the waveforms of the multidimensional signals of multiple human body movements in the same action group are similar. In addition, in different action groups, the waveforms of the multidimensional signals of multiple human body movements are different.

[0157] For example, in some embodiments of the present application, multiple human body movements are divided into four different movement groups, among which standing up (Stand) and sitting down (Sit) are the first group (Group 1); raising hand (Raisehand) and dropping hand (Drophand) are the second group (Group 2); sleeping (Sleep) and waking up (Wake) are the third group (Group 3); and static (Static) and walking (Walk) are the fourth group (Group 4).

[0158] Fig.15The RSSI signals and Phase angle signals of the above eight actions are shown, wherein the RSSI signal is located above the Phase angle signal. The waveforms of the multidimensional signals of multiple human body actions in the same action group are similar. For example, in the first group, the waveforms of the RSSI signals of standing up and sitting down are similar. In addition, the Phase angle signals of the two are also similar. Similarly, the RSSI signals and Phase angle signals of raising and lowering hands in the second group and sleeping and waking up in the third group also have similar situations.

[0159] In addition, the waveforms of the multi-dimensional signals of multiple human body movements are different in different action groups. For example, in the first group and the second group, the waveforms of the RSSI signals of standing up and raising hands are completely different, and the Phase angle signals of the two are also completely different.

[0160] Optionally, the identification module 2300 can be further configured so that the start and end signal strengths of the multi-dimensional signals of multiple human body movements in the same action group are different in one acquisition cycle. For example, in the first group, the start and end signal strengths of the RSSI signals of standing up and sitting down in one acquisition cycle are different. In addition, the start and end signal strengths of the Phase angle signals of the two in one acquisition cycle are also different.

[0161] By dividing multiple human body movements into different action groups, the waveform changes of the multi-dimensional signals of different actions in the group are relatively similar, which can improve the efficiency and accuracy of the action recognition method. For example, compared with the standing up of the first group and the raising of the hands of the second group, the signal changes of the former are more complex. The second group is the arm movements, the first and third groups are the torso movements, and the fourth group reflects the changes in the position of the human body. This can make the human body movements in the same group have more similar action features, and the multi-dimensional signals of the same group of actions have a strong correlation, which is conducive to improving the efficiency and accuracy of the action recognition method.

[0162] The human motion recognition system provided in the present application has the same beneficial effects as the above-mentioned human motion recognition method because it adopts the human motion recognition method provided in at least one embodiment of the present application, and will not be elaborated here.

[0163] The above description is only a preferred implementation method of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of protection involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the technical concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A human motion recognition method, characterized in that: The method comprises: collecting a multi-dimensional signal affected by human body movements, wherein the multi-dimensional signal is emitted by a radio frequency electronic tag disposed in front of the human body; and Based on the fused multi-dimensional signal, determining the human body movement, Wherein, the multi-dimensional signal includes at least two of a received signal strength indication signal, a phase signal and a Doppler frequency shift signal; and The number of the radio frequency electronic tags is the same as the number of the human bodies.

2. The method according to claim 1, wherein: Determining the human body motion based on the fused multi-dimensional signal includes: The fused multi-dimensional signal is input into the trained guided aggregation model for classification to determine the human body action, Wherein, the guided aggregation model includes a random forest model.

3. The method according to claim 1, wherein: Collecting the multi-dimensional signal affected by human body motion includes: using a dynamic antenna system to collect the multi-dimensional signal, including: The dynamic antenna system is controlled to move along a first direction, a second direction, and a third direction, and to rotate in a plane intersecting with any one of the first direction, the second direction, and the third direction to collect the multi-dimensional signal, wherein the first direction, the second direction, and the third direction intersect with each other.

4. The method according to claim 3, wherein: When the number of the human bodies is greater than or equal to 2, the number of the dynamic antenna systems is 1.

5. The method according to claim 3, wherein: The method further comprises: when the number of the human bodies is greater than or equal to 2, the dynamic antenna system at the lowest transmission power determines the best receiving angle of the antenna through the movement and the rotation, The optimal receiving angle is the angle at which the number of radio frequency electronic tags is the least.

6. The method according to claim 1, wherein: Determining the human body motion based on the fused multi-dimensional signal includes: determining the human body motion based on a difference between the first signal and the second signal, Wherein, the first signal is a signal received by the radio frequency electronic tag; and The second signal is a signal sent by the radio frequency electronic tag and affected by the human body movement.

7. The method according to claim 1, wherein: Before determining the human body motion based on the fused multi-dimensional signal, the method further includes: Performing a preprocessing operation on at least one of the multidimensional signals comprises: Phase restoration processing, wavelet threshold noise reduction processing and normalization processing are performed to enhance the features of the multidimensional signal affected by the human body motion.

8. The method according to claim 1, wherein: The method further comprises: Based on the waveform of the multi-dimensional signal, the plurality of human body movements are divided into different movement groups, Among them, the waveforms of the multi-dimensional signals of the multiple human body movements in the same movement group are similar.

9. The method according to claim 8, wherein: The multi-dimensional signals of the plurality of human body movements in the same movement group have different intensities of the start and end signals in one acquisition cycle.

10. A human motion recognition system, characterized in that: The system comprises: a collection module configured to collect multi-dimensional signals affected by human body movements, wherein the multi-dimensional signals are emitted by a radio frequency electronic tag disposed in front of the human body; and A recognition module is configured to determine the human body action based on the fused multi-dimensional signal, The multi-dimensional signal includes at least two of a received signal strength indication signal, a phase signal and a Doppler frequency shift signal; and The number of the radio frequency electronic tags is the same as the number of the human bodies.

Images