Sleep scene-oriented bimodal fusion radar dynamic attitude identification method and device

Abstract

The invention discloses a sleep scene-oriented bimodal fusion radar dynamic attitude identification method and device. The method comprises the following steps of: S1, acquiring and demodulating radar original data; s2, double-window feature construction of a velocity mode and a structural mode; s3, carrying out speed-guided dual-mode time alignment and feature reweighting; s4, the dynamic sleeping postures are output in a classified mode. According to the method, the limitation of an existing single-mode radar processing method in the aspects of weak action visibility, bedding shielding robustness and directivity discrimination capability is broken through, and a new implementation mode is provided for whole-process dynamic posture monitoring of complex sleep behaviors.

Description

Technical Field

[0001] This invention belongs to the field of human posture recognition and sleep monitoring technology, and more specifically, relates to a dual-modal fusion radar dynamic posture recognition method and device for sleep scenarios. Background Technology

[0002] Existing sleep monitoring technologies mostly rely on wearable devices, cameras, or pressure sensor arrays, but all of these solutions have significant technical limitations.

[0003] While millimeter-wave radar offers advantages such as being non-contact and penetrating bedding, current data processing methods based on single-mode radar signals still cannot address key challenges related to complex sleep behaviors. Typical problems include:

[0004] (1) The bedding obstructs the reflection structure, causing further attenuation of the micro-Doppler energy;

[0005] (2) Sleep posture changes usually evolve slowly and continuously in multiple stages. Traditional models based on single-frame features or single sequences cannot fully characterize directional changes.

[0006] (3) Weak movements (such as slow rolling over or fine-tuning of posture) are easily overlooked due to their low energy, causing recognition gaps;

[0007] (4) Existing single-modal methods cannot simultaneously handle velocity changes, structural changes and occlusion compensation, resulting in a significant decrease in accuracy in the determination of turning direction and occlusion scenarios.

[0008] Therefore, how to make high-speed dynamic changes (micro-movements) and low-speed structural changes (rolling-over structures) visible simultaneously in a sleep scenario, and maintain directional discrimination stability under occlusion conditions, is a key challenge of existing technologies.

[0009] Existing methods based on single-modal microDoppler cannot determine the direction of the rollover; existing methods based on single-sequence structure vectors fail in weak motion or occlusion scenarios; existing multimodal fusion does not address time scale mismatch, leading to attention weight misalignment. Summary of the Invention

[0010] To address the aforementioned problems, this invention proposes a dual-modal fusion radar dynamic attitude recognition method and device for sleep scenarios. This invention constructs a collaborative fusion mechanism composed of velocity mode (micro-Doppler features) and structural mode (range structural features), enabling the system to simultaneously acquire velocity and structural change information in real sleep behaviors such as bedding occlusion, weak motion energy, gradual posture changes, and significant turning over, achieving continuous, stable, and accurate recognition of posture direction changes. This invention overcomes the limitations of existing single-modal radar processing methods in terms of "weak motion visibility," "robustness to bedding occlusion," and "directional discrimination capability," providing a new technical approach and implementation method for full-process dynamic attitude monitoring of complex sleep behaviors.

[0011] To address at least one of the aforementioned technical problems, according to one aspect of the present invention, a dual-modal fusion radar dynamic attitude recognition method for sleep scenarios is provided, comprising the following steps:

[0012] S1. Radar raw data acquisition and demodulation:

[0013] The I / Q echo signals during sleep are continuously acquired using millimeter-wave radar at a fixed frame period. The signals are demodulated and deinterleaved to obtain a multi-channel raw data sequence arranged in frames. The human body principal reflection range unit and its neighborhood are determined based on the range dimension energy distribution of the first frame or multiple frames.

[0014] S2. Construction of dual-window features for velocity and structural modes:

[0015] S21. Short window construction of velocity modes:

[0016] A weighting factor is constructed for the velocity signal in the neighborhood of the main reflection range cell according to the combination of the mean amplitude, energy trend factor and distance Gaussian weight, and the neighborhood velocity signal is weighted and fused.

[0017] The fusion results are sequentially subjected to DC removal, trend suppression and robust filtering, and a short-time Fourier transform with a short time window and high overlap rate is used to generate velocity mode features, wherein the short time window is used to capture micro-motion velocity changes with a duration of tens of milliseconds.

[0018] S22. Long window construction of structural modes:

[0019] The radar amplitude vectors of consecutive frames are flattened to form a structure vector sequence. Based on the statistical characteristics of the duration of sleep turning over (1.8–2.5 seconds), the structure vectors are sliced ​​according to a long time window covering 50–80 frames, with adjacent slices sliding with an overlap rate of 80%–90%.

[0020] Each structural slice is input into a multilayer perceptron for encoding, resulting in structural segment labels of a unified dimension. The long-time window is used to represent the slow posture change phase, while the short-time window used in the velocity mode is used to characterize micro-motion velocity changes; the two have different time scales. The short-time window used in the velocity mode and the long-time window used in the structural mode are not interchangeable in time scale. The short-time window is used to maintain the visibility of micro-motion velocity peaks, while the long-time window is used to maintain the integrity of the structural semantics during the rollover phase.

[0021] S3. Velocity-guided bimodal temporal alignment and feature reweighting:

[0022] Based on the location of the velocity peak and energy abrupt change in the short-time-window velocity mode, multiple time anchor points are determined;

[0023] Locate the sequence of structural segments corresponding to each time anchor point on a unified time axis;

[0024] Based on the velocity change amplitude, the energy of the structural segment, and the time distance between them, the reweighting coefficient of each structural segment is calculated. The structural segments are selectively enhanced or suppressed to obtain time-aligned structural enhancement features. The structural enhancement features are then residually fused with the velocity modal features to obtain dual-modal enhancement features for subsequent sleeping posture classification.

[0025] During the fusion process, the velocity peak detected in the velocity mode is used as the time anchor point to align and weight the time position of the structural segment sequence.

[0026] S4. Dynamic Sleeping Posture Classification Output:

[0027] By inputting bimodal enhanced features into a classification network and outputting the probability of dynamic sleeping posture in consecutive frames through a sliding window approach, continuous posture change recognition throughout the night can be achieved.

[0028] The millimeter-wave radar is positioned 0.2 to 3 meters from the side of the bed to cover supine, lateral, prone positions and their dynamic transition areas.

[0029] Furthermore, the weighting factors in step S21 include at least two combinations of the mean amplitude, the energy trend factor, and the distance Gaussian weight.

[0030] Furthermore, the length of the structural slice in step S22 is set to cover 50 to 80 frames based on the duration of the sleep turning-over action, so as to fully cover a turning-over phase lasting about 1.8 to 2.5 seconds, so that each structural slice sequence can represent a stable posture phase.

[0031] Furthermore, the overlap rate of adjacent structural slices is set to 80%–90% to maintain the continuity of structural modes in the time dimension and reduce attitude phase jumps.

[0032] Furthermore, step S3 employs a velocity-information-guided bimodal time alignment and feature reweighting mechanism. By using the velocity change peak as the time anchor point, structural features on different time slices are aligned and weights are redistributed to achieve correspondence and semantic consistency between velocity modes and structural modes in multiple subspaces.

[0033] Furthermore, the sliding window in step S4 is used to output pose predictions for consecutive frames, and probability smoothing is performed between different windows to improve the stability of overnight monitoring.

[0034] The velocity-guided feature reweighting mechanism automatically increases the weight of velocity modalities in the fusion process to achieve compensation for structural modalities and stable recognition of dynamic sleeping postures in occluded scenarios.

[0035] According to another aspect of the present invention, a dual-modal fusion radar dynamic attitude recognition device for sleep scenarios is provided for performing the above-described method.

[0036] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the dual-modal fusion radar dynamic attitude recognition method and apparatus for sleep scenarios of the present invention.

[0037] According to another aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the dual-modal fusion radar dynamic attitude recognition method and apparatus for sleep scenarios of the present invention.

[0038] Compared with existing technologies, the beneficial effects of the above-described method of the present invention are as follows:

[0039] The dual-window system and the cross-timescale dual-modal fusion mechanism based on velocity anchor points of the present invention have the following advantages:

[0040] 1. The dual timescale design is based on both physical and physiological principles; it is strictly matched with the radar sampling period and the human body turning over time constant, and is not an empirical parameter setting.

[0041] 2. Effectively solves the problem of modal mismatch; This invention provides instantaneous action cues through velocity mode and provides phased stable structure through structural mode. In the fusion stage, the structural segments are time-aligned and weighted according to velocity changes, which effectively avoids the mismatch of the two types of modes in time scale, so that weak actions and the turning stage can be accurately represented, and the two compensate for each other.

[0042] 3. Significantly improved stability under occlusion conditions; structural modes are greatly affected by bedding, but velocity modes are not affected by occlusion, and this complementary mechanism improves overall robustness. When occlusion causes significant energy decay in structural modes, the fusion mechanism of this invention can automatically increase the weight of velocity modes in the fusion process based on their stability, and use the high signal-to-noise ratio instantaneous information provided by the velocity peak to compensate for the decay segments in structural modes affected by occlusion, thereby avoiding deviations in attitude orientation recognition and significantly improving stability under occlusion conditions.

[0043] 4. Weak and minute movements are no longer ignored; the short-window STFT can capture minute speed changes and has high recognition continuity.

[0044] 5. The inter-class boundaries of sleeping posture changes are clear; the fused features have stronger separability in the feature space, avoiding misjudgment.

[0045] 6. Output is seamless and continuous throughout the night; probability smoothing combined with cross-window recognition ensures continuity. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of the present invention and are not intended to limit the present invention.

[0047] Figure 1 This is a general flowchart of a preferred embodiment of the present invention;

[0048] Figure 2 The following are structural diagrams of modal modules according to a preferred embodiment of the present invention: (a) a short-window velocity modal module structure diagram, and (b) a long-window structure modal module structure diagram.

[0049] Figure 3 This is a visual illustration of a preferred embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention.

[0051] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0052] Example 1:

[0053] A dual-modal fusion radar dynamic attitude recognition method for sleep scenarios, such as Figure 1 As shown, it includes the following steps:

[0054] Step 1: The AWR1642BOOST millimeter-wave radar was used as the front-end data acquisition device. The radar adopted a MIMO structure with 1 transmit and 4 receive, acquiring data at 128 chirps / frame, 512 ADC points, a 4 MHz sampling rate, and a frame period of approximately 30 ms. By continuously acquiring echo signals, the entire rolling motion process could be covered.

[0055] The radar equipment was placed at the same height as the bed, 1m away from the side of the bed (this is the test environment for 2100 sample data; in reality, the system can operate normally at a distance of 0.2-3m from the side of the bed), and collected bin files.

[0056] Step 2: Decode bin data, generate micro-Doppler data, and load distance structure features;

[0057] like Figure 2 As shown in (a), this is the short window speed mode module.

[0058] Specifically, the continuously acquired echo signals Inputting the Short Time Fourier Transform (STFT), the time-frequency distribution of sleep actions is obtained:

[0059]

[0060] This feature can sensitively capture fine-grained information such as turning over, limb adjustments, and changes from stillness to movement, which are important signals unique to the sleep scenario. This invention generates a stable dynamic micro-Doppler image by normalizing the amplitude of the STFT results, suppressing background, and stitching the time frame, to describe the user's continuous sleeping posture changes.

[0061] like Figure 2 (b) is the long-window structural modality generation module. To reflect the continuous changes in the overall body structure during sleep (e.g., from side-lying to prone), this device decodes the original radar echo frame by frame into a range structure feature vector. The continuity of sleep movements requires that the range structure feature vector cannot be used as a static global description. Therefore, this invention adopts a continuous slicing mechanism to reconstruct the range structure feature vector into a sequence of segments of equal length:

[0062]

[0063] Each slice after continuous slicing It characterizes the overall structural changes in human posture over a certain period of time, complementing the velocity change characteristics in micro-Doppler.

[0064] This "continuous structure fragmentation" is specifically designed for sleeping posture recognition and is completely different from the single-frame distance structure features used in ordinary action recognition. It is a necessary structural module of this technical system.

[0065] In a preferred embodiment of the invention, the slice length of the structural mode is... With sliding step size Further optimization was performed based on the temporal distribution of sleep turning movements under the radar acquisition conditions of this invention. Based on statistics from 2100 sleep samples, the duration of a single turning movement was determined under the condition of continuous acquisition by millimeter-wave radar at a frame period of approximately 30 ms. The data is mainly concentrated between approximately 1.8 s and 2.5 s, corresponding to approximately 60 to 80 frames of data. Therefore, this invention adjusts the structural slice length. Selecting to cover 50-80 frames enables:

[0066]

[0067] This ensures that each structural fragment can fully represent a semantically stable stage in the process of turning over, rather than fragmented segments.

[0068] To improve the stability of continuous output, this invention increases the sliding step size. Set as One-tenth to one-fifth of the time corresponds to 80%–90% time overlap between adjacent slices, consistent with the above.

[0069]

[0070] By comparing with the duration of turning over Matching , The design ensures that the structural modalities can reflect the slow evolution of posture in the time dimension, while avoiding semantic fragmentation due to excessively short windows or aliasing of different posture stages due to excessively long windows.

[0071] It should be emphasized that the above and The configuration is specifically optimized for the time constant of turning over in sleep scenarios, rather than the conventional window settings used in general action recognition technologies. If traditional short windows or fixed empirical parameters are used, it is difficult to form a stable correspondence with micro-Doppler modes on a time scale, and even conventional fusion methods cannot obtain reliable semantic alignment results.

[0072] Step 3: Perform dual-modal feature fusion based on the obtained micro-Doppler reflection of dynamic change speed (micro-motion) and distance structure characteristics reflecting posture structure changes (macro-position).

[0073] To achieve semantic alignment between the two, this device uses velocity change features as time anchors to guide the selective updating of distance structure features during the fusion process, enabling micro-motion changes and slow structure evolution to be modeled synchronously on a unified time reference.

[0074] Step 4: Finally, input the fused features into the classification head, and output the dynamic changes in seven sleeping positions, as shown in Table 1:

[0075] Table 1. Comparison Table of Seven Posture Types

[0076] Attitude coding Dynamic sleeping position categories SUSI Supine → Side-lying SUPR supine → prone SISI Side-lying → Side-lying SISU Side-lying → Supine SIPR Side-lying → Prone PRSI prone → side lying PRSU prone → supine

[0077] The entire system, from hardware to algorithms, revolves around "dynamic sleeping posture recognition," forming a complete closed loop:

[0078] The radar collects continuous reflected signals of the human body during sleep; the STFT extracts velocity changes and generates dynamic micro-Doppler; continuous slices of distance structural features form a dynamic sequence reflecting the body structure; the motion anchors provided by velocity changes guide the alignment of the overall posture structure, and finally outputs continuous sleeping posture status to achieve dynamic monitoring of sleeping posture throughout the night.

[0079] This process demonstrates strong scenario relevance:

[0080] If we leave the sleep environment, the micro-Doppler amplitude is extremely small and the continuity of distance structure features is weak, so the algorithm module of this system cannot work properly.

[0081] Furthermore, this embodiment uses a feature visualization method to reduce the dimensionality of the fused high-dimensional features, clearly showing that the seven poses form relatively separate cluster structures in the feature space, indicating that the bimodal fusion model has strong inter-class discrimination ability. Figure 3 As shown, this is the identification result of 2100 samples, including 1400 old data and 700 new data.

[0082] In summary, this invention achieves real-time, stable, and continuous monitoring of dynamic sleeping postures by combining the micro-Doppler image processing unit, the original sequence feature unit, and their fusion module in a structural manner. The system has a clear structure, strong adaptability, and can work stably in scenarios such as home bedside, hospital wards, and long-term bedridden care, and has good practical value.

[0083] Example 2:

[0084] The computer-readable storage medium of this embodiment stores a computer program that, when executed by a processor, implements the steps in the dual-modal fusion radar dynamic attitude recognition method for sleep scenarios in Embodiment 1.

[0085] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.

[0086] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0087] Example 3:

[0088] The computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the dual-modal fusion radar dynamic attitude recognition method for sleep scenarios in Embodiment 1.

[0089] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0090] Those skilled in the art will understand that the content disclosed in the embodiments can be provided as a method, system, or computer program product. Therefore, this solution can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this solution can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage) containing computer-usable program code.

[0091] This solution is described with reference to flowchart illustrations and / or block diagrams of methods and computer program products according to embodiments of this solution. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0095] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the specific embodiments described above. The specific embodiments and descriptions in the specification are merely for further illustrating the principles of the invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the claims and their equivalents.

Claims

1. A dual-modal fusion radar dynamic attitude recognition method for sleep scenarios, characterized in that, Includes the following steps: S1. Acquisition and demodulation of raw radar data; Millimeter-wave radar is used to continuously acquire I / Q echo signals during sleep at a fixed frame period. The I / Q echo signals are demodulated and deinterleaved to obtain a multi-channel raw data sequence arranged in frames. The human body principal reflection range unit and its neighborhood are determined based on the range dimension energy distribution of the first frame or multiple frames. S2. Construction of dual-window features for velocity and structural modes; construction of short-time-window velocity modes and long-time-window structural modes respectively; S3. Velocity-guided bimodal temporal alignment and feature reweighting; Based on the location of velocity peaks and energy abrupt changes in the short-time-window velocity modes, multiple time anchors are determined; and the structural segment sequences corresponding to each time anchor are located on a unified time axis. Based on the velocity change amplitude, the energy of the structural segment and the time distance between them, the reweighting coefficient of each structural segment is calculated, and the structural segments are selectively enhanced or suppressed to obtain the time-aligned structural enhancement features. The structural enhancement features and velocity modal features are then residually fused to obtain dual-modal enhancement features for subsequent sleeping posture classification. During the fusion process, the velocity peak detected in the velocity mode is used as the time anchor point to align and weight the time position of the structural segment sequence; S4. Dynamic Sleep Posture Classification Output: The dual-modal enhanced features are input into the classification network, and the probability of dynamic sleep posture is output for continuous frames through a sliding window method, so as to realize the recognition of continuous posture changes throughout the night.

2. The method as described in claim 1, characterized in that, In S1, the millimeter-wave radar is positioned 0.2-3 meters from the side of the bed to cover the dynamic transition area during attitude switching.

3. The method as described in claim 1, characterized in that, S2 includes the following specific steps: S21. Short-time-window velocity mode construction: A weighting factor is constructed for the velocity signal in the neighborhood of the main reflection range cell according to the combination of the mean amplitude, energy trend factor and distance Gaussian weight, and the neighborhood velocity signal is weighted and fused. The fusion results are sequentially subjected to DC removal, trend suppression and robust filtering, and a short-time Fourier transform with a short time window and high overlap rate is used to generate velocity mode features. The short time window is used to capture micro-motion velocity changes with a duration of tens of milliseconds. S22. Long-term window structure modality construction: The radar amplitude vectors of consecutive frames are flattened to form a structure vector sequence. Based on the statistical characteristics of the duration of sleep turning over (1.8–2.5 seconds), the structure vectors are sliced ​​according to a long time window covering 50–80 frames, with adjacent slices sliding with an overlap rate of 80%–90%. Each structural slice is input into a multilayer perceptron for encoding, resulting in a structural segment label with a unified dimension; a long time window is used to represent the slow attitude change phase.

4. The method as described in claim 3, characterized in that, The weighting factors in S21 include at least two combinations of the mean amplitude, energy trend factor, and distance Gaussian weight.

5. The method according to claim 4, characterized in that, In S22, the length of the structural slice is set to cover 50 to 80 frames based on the duration of the sleep turning action, so as to fully cover a turning phase lasting about 1.8 to 2.5 seconds, so that each structural slice sequence can represent a stable posture phase.

6. The method according to claim 5, characterized in that, The overlap rate of adjacent structural slices is set to 80% to 90% to maintain the continuity of structural modes in the time dimension and reduce attitude phase jumps.

7. The method according to claim 6, characterized in that, Step S3 employs a velocity-information-guided bimodal time alignment and feature reweighting mechanism. By using the peak velocity change as the time anchor point, structural features in different time slices are aligned and weights are redistributed to achieve correspondence and semantic consistency between velocity modes and structural modes in multiple subspaces.

8. The method according to claim 7, characterized in that, The sliding window in step S4 is used to output attitude prediction for consecutive frames, and probability smoothing is performed between different windows to improve the stability of monitoring throughout the night.

9. A dual-modal fusion radar dynamic attitude recognition device for sleep scenarios, characterized in that: The device is used to perform the steps in the dual-modal fusion radar dynamic attitude recognition method for sleep scenarios as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps in the dual-modal fusion radar dynamic attitude recognition method for sleep scenarios as described in any one of claims 1 to 8.

Images