A wireless signal segmentation method and system for real-time state perception

Abstract

The present invention discloses a wireless signal segmentation method and system for real-time state perception, which is implemented based on millimeter wave transceiver equipment. The method comprises the following steps: collecting human target perception signal data; using the MTI algorithm to remove multipath effects; processing the signal data at each moment to obtain distance information and incident azimuth information, constructing a distance-angle spectrum matrix representing the target position, and realizing real-time judgment of position attributes. For walking status, Doppler data is obtained based on the constructed distance-Doppler spectrum, and after segmentation, it is input into a gait recognition network to realize real-time gait recognition. For in-situ activities, Doppler data is obtained based on the constructed distance-Doppler spectrum, and is input into an action recognition network through dynamic window segmentation to realize real-time action recognition. For in-situ stillness, phase information at the range bin is obtained to obtain the frequency of phase change, and the different frequency characteristics of micro-motion interference and respiratory movement are used for segmentation to realize respiratory monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless perception and state segmentation, and in particular relates to a wireless signal segmentation method and system for real-time state perception. Background Art

[0002] Wireless sensing technology is the foundation of numerous intelligent applications and is crucial for improving people's living standards and promoting the development of intelligent society. For example, in smart homes, wireless sensing technology can detect users' location and movements in real time, enabling intelligent control of home appliances like air conditioners and lighting. In smart healthcare, it can monitor patients' location, movements, and vital signs in real time, facilitating personalized treatment and health management. Furthermore, in the rescue and security fields, this technology can achieve contactless, non-cooperative motion and vital sign sensing of obscured objects, supporting emergency rescue and safety monitoring. Wireless sensing technology can provide these applications with a variety of human status information, including gesture recognition, gait analysis, and vital sign monitoring, enabling more refined intelligent management and services. Its broad application prospects can effectively enhance the level of intelligence in various fields, bringing a more convenient, safe, and efficient life experience.

[0003] In practical application scenarios, real-time perception of human target states is a key technology. Achieving this requires the ability to determine state changes in real time and perform wireless signal segmentation when a state change occurs and the conditions for single-state perception are met, enabling effective real-time perception of human target states. For example, gait recognition can be performed when a target is moving; gesture recognition can be performed when a target is stationary; and vital sign monitoring can be performed when a target is stationary. However, existing wireless sensing research mostly focuses on single-state perception, collecting state-related data within constrained perception scenarios. For example, gait recognition research requires subjects to remain walking; gesture recognition research requires subjects to only perform gestures while stationary; and vital sign monitoring research requires subjects to remain stationary facing the device. However, in real-life situations, target states are dynamic and difficult to maintain. Therefore, achieving real-time and accurate perception of multiple human target states in dynamic environments remains a major challenge in the wireless sensing field. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the prior art, and proposes a wireless signal segmentation method for real-time state perception, and also discloses a wireless signal segmentation system for real-time state perception.

[0005] To achieve the above objectives, the present invention proposes a wireless signal segmentation method for real-time status perception, which is implemented based on a millimeter wave transceiver device. The method includes:

[0006] Step 1) collecting human target perception signal data;

[0007] Step 2) Using the MTI algorithm to remove the multipath effect caused by static objects in the environment;

[0008] Step 3) Process the signal data at each moment to obtain distance information and incident azimuth information, and construct a distance-angle spectrum matrix representing the target position;

[0009] Step 4) Tracking the trajectory of the human target based on the distance-angle spectrum matrix to achieve real-time judgment of the human target's position attributes, including walking state, in-situ movement, and in-situ stillness;

[0010] For walking conditions, a range-Doppler spectrum is constructed based on signal data to obtain Doppler data. The Doppler data is segmented with the same duration as gait recognition samples of human targets and input into the trained gait recognition network to achieve real-time gait recognition.

[0011] For stationary activities, a range-Doppler spectrum is constructed based on signal data to obtain Doppler data. A sliding window is used to detect large-scale motion information. The Doppler data segmented by the dynamic window is used as the action recognition sample for human targets and input into the trained action recognition network to achieve real-time action recognition.

[0012] For stationary state, the phase information at the Range bin is obtained according to the signal data, and the frequency of phase change is obtained. The different frequency characteristics of micro-motion interference and respiratory movement are used for segmentation to achieve respiratory monitoring.

[0013] Preferably, the millimeter wave transceiver device is a 2-transmit and 4-receive antenna millimeter wave transceiver device.

[0014] Preferably, the step 3) comprises:

[0015] The signal frame data S after removing static interference is processed by FFT along the fast time axis to obtain the distance information R between the human target and the i-th receiving antenna i :

[0016] R i =RangeFFT(S i ),

[0017] Among them, S i is the data received by the i-th receiving antenna;

[0018] Perform FFT processing along the antenna direction to obtain the target's incident azimuth information;

[0019] Use the distance information of multiple antennas to perform angle FFT processing and construct the distance-angle spectrum matrix RAM:

[0020] RAM=AngleFFT(R1,...,R i ,...,R I ),

[0021] Wherein, the subscript I is the total number of receiving antennas.

[0022] Preferably, the step 4) tracks the trajectory of the human target based on the distance-angle spectrum matrix, which previously includes: removing environmental noise in the RAM by a constant false alarm rate algorithm.

[0023] Preferably, the step 4) tracks the trajectory of the human target based on the distance-angle spectrum matrix to achieve real-time judgment of the position attributes of the human target, including walking state, in-situ activity and in-situ stillness; including:

[0024] The distance-angle spectrum matrices of multiple consecutive moments are accumulated through the matrix accumulation method. If the position of the human target changes within the set time period, it is determined to be in a walking state; if the position of the human target does not change within the set time period and the human target position information exists in the distance-angle spectrum, it is determined to be active in situ; if the human position information disappears from the distance-angle spectrum, it is determined to be stationary in situ.

[0025] Preferably, step 4) constructs a distance-Doppler spectrum according to the signal data for the walking state to obtain Doppler data; divides the Doppler data into gait recognition samples of the human target with the same duration, and inputs the data into a trained gait recognition network to realize real-time gait recognition; comprising:

[0026] For the walking state, 2D-FFT operations are performed along the fast time axis and slow time axis of the frame data S to construct the range-Doppler spectrum RDM data of the human target;

[0027] For each frame moment, the RDM data is summed according to the data within the range bin where the human target trajectory is located, and the sum is used as the Doppler data of the current frame;

[0028] Segment the Doppler data with the same duration as the gait recognition samples of human targets;

[0029] Input the trained gait recognition network to achieve real-time gait recognition; the gait recognition network is CNN.

[0030] Preferably, step 4) constructs a range-Doppler spectrum based on signal data for stationary activities to obtain Doppler data; uses a sliding window to detect large-scale motion information, and uses the Doppler data segmented by the dynamic window as a motion recognition sample for the human target, which is input into a trained motion recognition network to achieve real-time motion recognition; comprising:

[0031] For a human target with a distance of r and an incident azimuth of θ, narrowband filtering is performed in the distance dimension and beamforming is performed in the antenna dimension to obtain focused data.

[0032] Edge Data The fast time axis and slow time axis are subjected to 2D-FFT operation to construct the range-Doppler spectrum RDM data of the human target;

[0033] For each frame moment, the RDM data is summed according to the data within the range bin where the human target trajectory is located, and the sum is used as the Doppler data of the current frame;

[0034] A sliding window is used to detect large-scale motion information. In a sliding window, if the data of a set number of frames contains motion information, the current window is judged to be in motion state. The duration of the dynamic window is automatically increased until the sliding window detects a non-motion state, and the dynamic window representing motion information ends.

[0035] The Doppler data segmented by dynamic windows is used as the action recognition samples of human targets and input into the trained action recognition network to realize real-time action recognition. The action recognition network is CNN.

[0036] Preferably, the step 4) acquires phase information at the range bin according to the signal data for stationary state, obtains the frequency of phase change, and uses the different frequency characteristics of micro-motion interference and respiratory movement to perform segmentation to achieve respiratory monitoring; comprising:

[0037] For stationary motion, RangeFFT processing is performed on the average chirp continuous frames along the fast time axis of the frame data S to obtain the Range bin of the micro-scale motion of the human target.

[0038] Extract the phase change of the Range bin and use a bandpass filter to suppress clutter interference;

[0039] The displacement change representing microscale motion is obtained according to the following formula:

[0040]

[0041] Among them, Δφ b is the phase change caused by the rise and fall of the chest cavity during human breathing, ΔR is the displacement change of the chest cavity during human breathing, and λ is the signal wavelength.

[0042] On the other hand, the present invention provides a wireless signal segmentation system for real-time status perception, which is implemented based on a millimeter wave transceiver device, and the system includes:

[0043] Acquisition module, used to collect human target perception signal data;

[0044] Multipath effect removal module, used to remove the multipath effect caused by static objects in the environment using the MTI algorithm;

[0045] The RAM matrix construction module is used to process the signal data at each moment, obtain the distance information and the incident azimuth information, and construct the distance-angle spectrum matrix representing the target position;

[0046] The position attribute judgment module is used to track the trajectory of human targets based on the distance-angle spectrum matrix, and realize real-time judgment of the position attributes of human targets, including walking status, in-place activity and in-place stillness;

[0047] The real-time gait recognition module is used to construct a range-Doppler spectrum based on the signal data for the walking state, and obtain Doppler data. The Doppler data is segmented into equal durations as gait recognition samples for the human target, and input into the trained gait recognition network to achieve real-time gait recognition.

[0048] The real-time action recognition module is used to construct a range-Doppler spectrum based on signal data for stationary activities and obtain Doppler data; a sliding window is used to detect large-scale motion information, and the Doppler data segmented by the dynamic window is used as the action recognition sample of the human target, which is input into the trained action recognition network to achieve real-time action recognition; the breathing monitoring module is used to obtain the phase information at the Range bin based on the signal data for stationary activities, and obtain the frequency of phase change. The different frequency characteristics of micro-motion interference and respiratory movement are used for segmentation to achieve breathing monitoring.

[0049] Compared with the prior art, the advantages of the present invention are:

[0050] The present invention overcomes the application limitations of existing wireless state perception research scenarios, and realizes real-time segmentation of human target state signals according to the position attribute requirements and state characteristic attributes when different human target states occur, so that the segmented signals can be used for human target state perception; when the target is performing any activity, the real-time judgment of the human target position attributes can be realized through trajectory tracking, including walking state, in-place activity, and in-place stillness; according to the position attribute requirements for different human target state perception, the characteristic attributes of different human target states are utilized to realize real-time segmentation of human target state signals for human target state perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of a wireless signal segmentation method for real-time state perception according to the present invention;

[0052] Figure 2 This is a schematic diagram of the real-time segmentation principle of gait motion signals;

[0053] Figure 3 This is a schematic diagram of the real-time segmentation principle of large-scale motion signals;

[0054] Figure 4 This is a schematic diagram of the real-time segmentation principle of micro-scale respiratory motion signals. DETAILED DESCRIPTION

[0055] The core idea of ​​the present invention is to realize the real-time segmentation of the human target state signal according to the position attribute requirements and state characteristic attributes when the human target occurs in different states, so that the segmented signal can be used for the state perception of the human target.

[0056] The method flow of the present invention is as follows Figure 1 As shown, the overall process is divided into a real-time determination phase for human target location attributes and a real-time segmentation phase for status signals. During this phase, the present invention uses a 2-transmitter, 4-receiver millimeter-wave transceiver to collect human target perception signal data. First, the present invention employs the Moving Target Indication (MTI) algorithm to remove multipath effects caused by static objects in the environment. Then, at each moment, an FFT (Fast Fourier Transformation) is performed along the fast time axis to obtain the target's Range information, and an FFT is performed along the antenna direction to obtain the target's Angle of Arrival (AoA) information, constructing RAM information representing the target's position. Furthermore, a Constant False-Alarm Rate (CFAR) algorithm is used to remove environmental noise from the RAM (Range-AoA Map) to obtain the human target's location information. Finally, a matrix accumulation method is used to accumulate the RAM matrices at each moment to track the human target's trajectory. Based on the human target's trajectory, real-time determination of the human target's location attributes can be achieved. The present invention divides the state of the human body when it moves to a target position into a walking state; the state of the human body when it moves in place into a large-scale motion state, such as gestures and other actions; and the state of the human body when it is stationary into a micro-scale motion state, such as chest movement during breathing.

[0057] During the real-time segmentation of human target state signals, the present invention implements real-time segmentation of human target state signals based on the characteristic attributes of different motion states, allowing the segmented signals to be used for human target state perception. For the walking state, the present invention performs a 2DFFT on each frame along the fast and slow time axes to extract information about the target's Doppler changes over time. Doppler data of the same duration are then segmented as samples for gait recognition. During the real-time segmentation of large-scale and micro-scale motion, to address interference from other dynamic targets in the environment, the present invention utilizes beamforming and narrowband filtering to focus on in-situ moving targets. For large-scale motion, a 2DFFT is performed on each frame along the fast and slow time axes to extract information about the target's Doppler changes over time. The present invention then utilizes a dynamic window mechanism to detect large-scale motion. When large-scale motion is detected in the Doppler data, the detection window begins to increase until the motion disappears, completing the real-time segmentation of the large-scale motion state. For micro-scale motion, the present invention exploits the sensitivity of the intermediate frequency signal phase of FMCW millimeter-wave transceivers to tiny displacements to extract phase information from the range bin where the human target resides. This phase information is then subjected to an FFT to obtain its frequency information. This frequency information is then used to segment the effective periodic motion within the micro-scale motion in real time to remove micro-motion interference, thereby enabling vital sign monitoring. This invention enables real-time segmentation of sensory signals for different motion states of a human target. The following describes the technical solution in detail with reference to the accompanying figures and examples.

[0058] Example 1

[0059] Embodiment 1 of the present invention provides a wireless signal segmentation method for real-time status perception, which is implemented based on a millimeter wave transceiver device. The method includes:

[0060] Step 1) collecting human target perception signal data;

[0061] Step 2) Using the MTI algorithm to remove the multipath effect caused by static objects in the environment;

[0062] Step 3) Process the signal data at each moment to obtain distance information and incident azimuth information, and construct a distance-angle spectrum matrix representing the target position;

[0063] Step 4) Track the trajectory of the human target based on the distance-angle spectrum matrix to achieve real-time judgment of the human target's position attributes. After judging it as a walking state, perform real-time segmentation of the gait motion perception signal.

[0064] The system configuration is as follows:

[0065] 1. The system operates on a Texas Instrument AWR1443 FMCW millimeter wave transceiver;

[0066] 2. The transceiver operates at 77 GHz and provides a range resolution of 6.7 cm;

[0067] 3. The transceiver uses two transmit and four receive antennas, and with the help of TDM-MIMO mode, it can form an 8-element virtual array and provide 15° AoA resolution.

[0068] Task: Based on real-time judgment of human target position attributes, complete real-time segmentation of gait motion perception signals.

[0069] In order to determine the position attributes of the human target in real time, the present invention tracks the trajectory of the human target in real time. The present invention uses a 2-transmitter and 4-receiver millimeter wave transceiver to collect human target perception signal data, and uses the MTI algorithm to remove the multipath effect caused by static objects in the environment. Then, the present invention obtains the position information of the human target in space at the current moment by constructing a RAM graph. Since the distance between the human target and the transceiver is linearly related to the frequency of the intermediate frequency signal, the present invention performs FFT processing on the signal frame data S after removing static interference along the fast time axis, and can obtain the distance information between the human target and each receiving antenna, which is expressed as:

[0070] R i =RangeFFT(S i ),

[0071] Among them S i is the data received by the i-th receiving antenna, R i represents the distance between the human target and the i-th receiving antenna. Next, the multi-antenna distance information is used to perform an angle FFT, that is, an FFT is performed on each range bin along the receiving antenna axis to construct a RAM, which is expressed as:

[0072] RAM=AngleFFT(R1,...,R i ,...,R I ),

[0073] Where I is the total number of receiving antennas. Furthermore, the present invention removes the environmental noise in the RAM by the CFAR algorithm to obtain the position information of the human target. In order to obtain the trajectory of the human target, the present invention uses the matrix accumulation method to accumulate the RAM matrices of multiple consecutive moments to achieve trajectory tracking of the human target. According to the trajectory of the human target, the real-time judgment of the position attribute of the human target can be achieved. If the trajectory of the human target over a period of time is as follows Figure 2As shown in (a), the position is constantly changing, which indicates that the target is walking. In order to obtain the signal of the human target in the walking state for gait recognition, the present invention first performs 2D-FFT operations along the fast time axis and slow time axis of the frame data S to construct the range-Doppler spectrum RDM (Range-Doppler Map) of the human target, which is expressed as:

[0074] D=2DFFT(S).

[0075] At a frame time, the RDM is as follows Figure 2 As shown in (b) in the figure. Then, the data within the range bin of the human target trajectory at each frame moment is summed up and used as the Doppler data of the current frame. The Doppler data is divided into the same duration as the gait recognition sample of the human target, as shown in Figure 2 As shown in (c) in the figure, the sample is fed into the trained gait recognition network to achieve real-time gait recognition.

[0076] Example 2

[0077] Embodiment 1 of the present invention provides a wireless signal segmentation method for real-time status perception, which is implemented based on a millimeter wave transceiver device. The method includes:

[0078] Step 1) collecting human target perception signal data;

[0079] Step 2) Using the MTI algorithm to remove the multipath effect caused by static objects in the environment;

[0080] Step 3) Process the signal data at each moment to obtain distance information and incident azimuth information, and construct a distance-angle spectrum matrix representing the target position;

[0081] Step 4) Track the trajectory of the human target based on the distance-angle spectrum matrix to achieve real-time judgment of the human target's position attributes. If it is judged to be in-situ activity, the motion perception signal is segmented in real time.

[0082] The system configuration is as follows:

[0083] 1. The system operates on a Texas Instrument AWR1443 FMCW millimeter wave transceiver;

[0084] 2. The transceiver operates at 77 GHz and provides a range resolution of 6.7 cm;

[0085] 3. The transceiver uses two transmit and four receive antennas, and with the help of TDM-MIMO mode, it can form an 8-element virtual array and provide 15° AoA resolution.

[0086] Task: Based on real-time judgment of human target position attributes, complete real-time segmentation of motion perception signals.

[0087] After the human target is determined to be in an in-situ active state through trajectory tracking, the present invention performs large-scale motion signal segmentation for action recognition. In the real-time judgment stage of position attributes, the present invention realizes the judgment of the target position attribute based on RAM trajectory tracking. After determining that the human target is in a large-scale motion state, the action signal segmentation is performed for action recognition, such as Figure 3 Specifically, after removing static interference, the signal frame data S is subjected to FFT along the fast time axis and the antenna axis to construct RAM, and the position information of the human target can be extracted. Figure 3 As shown in (a), when the trajectory of a human target remains unchanged over a period of time, it can be determined that the target is in a large-scale motion state. The present invention can suppress the interference of targets in other directions by focusing on the human target. For a human target with a distance of r and an AoA of θ, narrowband filtering is performed in the distance dimension and beamforming is performed in the antenna dimension to obtain focused data. In order to obtain the motion signal of the human target, the present invention firstly The fast time axis and slow time axis are subjected to 2D-FFT operation to construct RDM. The RDM of a frame moment is as follows Figure 3 As shown in (b) in the figure. Then, the data within the Range bin where the human target is located is summed for the RDM at each frame moment, which can be used as the Doppler data of the current frame. The present invention designs a dynamic window mechanism and uses a sliding window to detect large-scale motion information. In a sliding window, if the data of 4 / 5 frames contain action information, the present invention determines that the current window is in an action state. When the sliding window detects an action state, the duration of the dynamic window automatically increases until the sliding window detects a non-action state, and the dynamic window representing the action information ends, as shown in FIG. Figure 3 As shown in (c) in the figure, the Doppler data segmented by the dynamic window is used as the action recognition sample of the human target and fed into the trained action recognition network to achieve real-time action recognition.

[0088] Example 3

[0089] Embodiment 1 of the present invention provides a wireless signal segmentation method for real-time status perception, which is implemented based on a millimeter wave transceiver device. The method includes:

[0090] Step 1) collecting human target perception signal data;

[0091] Step 2) Using the MTI algorithm to remove the multipath effect caused by static objects in the environment;

[0092] Step 3) Process the signal data at each moment to obtain distance information and incident azimuth information, and construct a distance-angle spectrum matrix representing the target position;

[0093] Step 4) Track the trajectory of the human target based on the distance-angle spectrum matrix to achieve real-time judgment of the human target's position attributes. If it is judged to be stationary, the breathing perception signal is segmented in real time.

[0094] The system configuration is as follows:

[0095] 1. The system operates on a Texas Instrument AWR1443 FMCW millimeter wave transceiver;

[0096] 2. The transceiver operates at 77 GHz and provides a range resolution of 6.7 cm;

[0097] 3. The transceiver uses two transmit and four receive antennas, and with the help of TDM-MIMO mode, it can form an 8-element virtual array and provide 15° AoA resolution.

[0098] Task: Based on real-time judgment of human target position attributes, complete real-time segmentation of respiratory perception signals.

[0099] After the human target is determined to be in a stationary state through trajectory tracking, the present invention performs micro-scale motion signal segmentation for real-time respiratory monitoring, such as Figure 4 In the real-time judgment stage of the human target position attribute, if the target position disappears in the RAM, it can be judged that the current target is in a micro-scale motion state, such as Figure 4 As shown in (a) in the figure, during the real-time state signal segmentation stage, the present invention first extracts the phase information of the intermediate frequency signal of the FMCW millimeter-wave transceiver, which is sensitive to small displacements. The present invention then determines the presence of micro-motion interference based on the frequency information of the phase change, thereby enabling real-time segmentation of the effective periodic motion signal from micro-scale motion to achieve respiration monitoring.

[0100] In order to obtain the range bin where the target is located, the present invention improves the frequency resolution and signal-to-noise ratio by combining multi-frame analysis. Specifically, the chirp contained in a frame is defined as:

[0101] chirp i =r(t)+n i (t),i=1,2,...,N,

[0102] Among them, chirp i Is a frame i th chirp, N is the total number of chirps in a frame, r(t) is the intermediate frequency signal, ni (t) is the Gaussian noise in the chirp. r(t) is considered to be fixed within a frame, and the average value of all chirps in a frame is calculated along each range bin to represent the average chirp of each frame, which is defined as:

[0103] AC(t)=r(t)+n(t).

[0104] Then, the average chirps of different frames are combined along time to form an observation time window. An observation time window can be defined as:

[0105] window={AC k (t)|k=1,2,...,F},

[0106] Among them, AC k k th The average chirp of the frame in the time window, F is the number of frames in the time window. By performing RangeFFT on the average chirp continuous frames along the fast time axis of each frame, the rangebin of the target micro-scale motion can be obtained, such as Figure 4 By extracting the phase change of the range bin, the displacement change of the micro-scale motion can be expressed as

[0107]

[0108] Among them, Δφ b is the phase change caused by the chest cavity when the human body breathes, ΔR is the displacement of the chest cavity when the human body breathes, and λ is the signal wavelength. Since the original phase change extracted contains obvious clutter interference, a bandpass filter is used to suppress the clutter interference and obtain the phase change caused by micro-scale motion, such as Figure 4 In order to distinguish the micro-scale chest motion caused by breathing from the micro-motion interference, the frequency of phase change is obtained by performing FFT on each observation time window. When there are multiple frequency components, it is considered that there is micro-motion interference, such as Figure 4 When the frequency component comes only from the micro-scale motion of the chest caused by breathing, it is considered to be micro-scale respiratory motion, such as Figure 4 As shown in (f) in the figure, the phase signal caused by the splitting of breathing can be used to realize breathing monitoring.

[0109] Example 4

[0110] Embodiment 4 of the present invention provides a wireless signal segmentation system for real-time status perception, which is implemented based on a millimeter wave transceiver device. The system is implemented using the methods of Embodiments 1-3, including:

[0111] Acquisition module, used to collect human target perception signal data;

[0112] Multipath effect removal module, used to remove the multipath effect caused by static objects in the environment using the MTI algorithm;

[0113] The RAM matrix construction module is used to process the signal data at each moment, obtain the distance information and the incident azimuth information, and construct the distance-angle spectrum matrix representing the target position;

[0114] The position attribute judgment module is used to track the trajectory of human targets based on the distance-angle spectrum matrix, and realize real-time judgment of the position attributes of human targets, including walking status, in-place activity and in-place stillness;

[0115] The real-time gait recognition module is used to construct a range-Doppler spectrum based on the signal data for the walking state, and obtain Doppler data. The Doppler data is segmented into equal durations as gait recognition samples for the human target, and input into the trained gait recognition network to achieve real-time gait recognition.

[0116] The real-time action recognition module is used to construct a range-Doppler spectrum based on signal data for stationary activities and obtain Doppler data. It uses a sliding window to detect large-scale motion information and uses the Doppler data segmented by the dynamic window as the action recognition sample for human targets. This data is input into the trained action recognition network to achieve real-time action recognition.

[0117] The respiratory monitoring module is used to obtain the phase information at the Range bin based on the signal data when the patient is stationary, obtain the frequency of the phase change, and use the different frequency characteristics of micro-motion interference and respiratory movement to segment and realize respiratory monitoring.

[0118] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. A wireless signal segmentation method for real-time status perception, implemented based on a millimeter wave transceiver device, comprising: Step 1) collecting human target perception signal data; Step 2) Using the MTI algorithm to remove the multipath effect caused by static objects in the environment; Step 3) Process the signal data at each moment to obtain distance information and incident azimuth information, and construct a distance-angle spectrum matrix representing the target position; Step 4) Accumulate the distance-angle spectrum matrix at each moment to track the trajectory of the human target and realize real-time judgment of the position attributes of the human target, including walking state, in-situ movement and in-situ stillness; For walking conditions, a range-Doppler spectrum is constructed based on signal data to obtain Doppler data. The Doppler data is segmented with the same duration as gait recognition samples of human targets and input into the trained gait recognition network to achieve real-time gait recognition. For stationary activities, a range-Doppler spectrum is constructed based on signal data to obtain Doppler data. A sliding window is used to detect large-scale motion information. The Doppler data segmented by the dynamic window is used as the action recognition sample for human targets and input into the trained action recognition network to achieve real-time action recognition. For stationary state, the phase information at the range bin is obtained based on the signal data to obtain the frequency of phase change. The different frequency characteristics of micro-motion interference and respiratory movement are used for segmentation to achieve respiratory monitoring. The step 4) accumulates the distance-angle spectrum matrix at each moment to track the trajectory of the human target, and realizes real-time judgment of the position attributes of the human target, including walking state, in-situ activity and in-situ stillness, including: The distance-angle spectrum matrices of multiple consecutive moments are accumulated through the matrix accumulation method. If the position of the human target changes within the set time period, it is determined to be in a walking state; if the position of the human target does not change within the set time period and the human target position information exists in the distance-angle spectrum, it is determined to be in-situ activity; if the human position information disappears from the distance-angle spectrum, it is determined to be stationary.

2. The wireless signal segmentation method for real-time state perception according to claim 1, characterized in that: The millimeter wave transceiver device is a 2-transmit and 4-receive antenna millimeter wave transceiver device.

3. The wireless signal segmentation method for real-time state perception according to claim 2, characterized in that: The step 3) comprises: The signal frame data S after removing static interference is processed by FFT along the fast time axis to obtain the distance information R between the human target and the i-th receiving antenna i : R i =RangeFFT(S i ), Among them, S i is the data received by the i-th receiving antenna; RangeFFT() represents the distance between the human target and the i-th receiving antenna; Perform FFT processing along the antenna direction to obtain the target's incident azimuth information; Use the distance information of multiple antennas to perform angle FFT processing and construct the distance-angle spectrum matrix RAM: RAM=AngleFFT(R1,...,R i ,...,R I ), Wherein, the subscript I is the total number of receiving antennas, and AngleFFT() represents the angle information between the human target and the i-th receiving antenna.

4. The wireless signal segmentation method for real-time state perception according to claim 3, characterized in that: The step 4) tracks the trajectory of the human target based on the distance-angle spectrum matrix, which previously includes: removing environmental noise in the RAM by a constant false alarm rate algorithm.

5. The wireless signal segmentation method for real-time state perception according to claim 3, characterized in that: In step 4), for the walking state, a range-Doppler spectrum is constructed according to the signal data to obtain Doppler data; Doppler data with the same duration is segmented as gait recognition samples of human targets and input into the trained gait recognition network to achieve real-time gait recognition; including: For the walking state, 2D-FFT operations are performed along the fast time axis and slow time axis of the frame data S to construct the range-Doppler spectrum RDM data of the human target; For each frame moment, the RDM data is summed according to the data within the range bin where the human target trajectory is located, and the sum is used as the Doppler data of the current frame; Segment the Doppler data with the same duration as the gait recognition samples of human targets; Input the trained gait recognition network to achieve real-time gait recognition; the gait recognition network is CNN.

6. The wireless signal segmentation method for real-time state perception according to claim 3, characterized in that: Step 4) constructs a range-Doppler spectrum based on the signal data for stationary activities to obtain Doppler data; uses a sliding window to detect large-scale motion information, and uses the Doppler data segmented by the dynamic window as a motion recognition sample for the human target, which is input into a trained motion recognition network to achieve real-time motion recognition; including: For a human target with a distance of r and an incident azimuth of θ, narrowband filtering is performed in the distance dimension and beamforming is performed in the antenna dimension to obtain focused data. Edge Data The fast time axis and slow time axis are subjected to 2D-FFT operation to construct the range-Doppler spectrum RDM data of the human target; For each frame moment, the RDM data is summed according to the data within the range bin where the human target trajectory is located, and the sum is used as the Doppler data of the current frame; A sliding window is used to detect large-scale motion information. In a sliding window, if the data of a set number of frames contains motion information, the current window is judged to be in motion state. The duration of the dynamic window is automatically increased until the sliding window detects a non-motion state, and the dynamic window representing motion information ends. The Doppler data segmented by dynamic windows is used as the action recognition samples of human targets and input into the trained action recognition network to realize real-time action recognition. The action recognition network is CNN.

7. The wireless signal segmentation method for real-time state perception according to claim 3, characterized in that: Step 4) for stationary state, obtaining phase information at the Range bin based on signal data, obtaining the frequency of phase change, and segmenting using different frequency characteristics of micro-motion interference and respiratory movement to achieve respiratory monitoring; including: For stationary motion, RangeFFT processing is performed on the average chirp continuous frames along the fast time axis of the frame data S to obtain the Range bin of the micro-scale motion of the human target. Extract the phase change of the Range bin and use a bandpass filter to suppress clutter interference; The displacement change representing microscale motion is obtained according to the following formula: Among them, Δφ b is the phase change caused by the rise and fall of the chest cavity during human breathing, ΔR is the displacement change of the chest cavity during human breathing, and λ is the signal wavelength.

8. A system based on the wireless signal segmentation method for real-time state perception according to claim 1, implemented based on a millimeter wave transceiver device, characterized in that: The system comprises: Acquisition module, used to collect human target perception signal data; Multipath effect removal module, used to remove the multipath effect caused by static objects in the environment using the MTI algorithm; The RAM matrix construction module is used to process the signal data at each moment, obtain the distance information and the incident azimuth information, and construct the distance-angle spectrum matrix representing the target position; The position attribute judgment module is used to accumulate the distance-angle spectrum matrix at each moment to track the trajectory of the human target and realize real-time judgment of the position attributes of the human target, including walking state, in-place activity and in-place stillness; A real-time gait recognition module is used to construct a range-Doppler spectrogram based on the signal data for the walking state to obtain Doppler data; the Doppler data is segmented into gait recognition samples of the human target with the same duration, and input into the trained gait recognition network to achieve real-time gait recognition; and The real-time action recognition module is used to construct a range-Doppler spectrum based on signal data for stationary activities to obtain Doppler data; a sliding window is used to detect large-scale motion information, and the Doppler data segmented by the dynamic window is used as the action recognition sample of the human target, which is input into the trained action recognition network to achieve real-time action recognition; and the respiratory monitoring module is used to obtain the phase information at the Range bin based on the signal data for stationary activities, obtain the frequency of phase change, and use the different frequency characteristics of micro-motion interference and respiratory movement for segmentation to achieve respiratory monitoring.

Images