Smart pillow regulation method and system for sleep posture recognition and neck support optimization

Abstract

The application relates to an intelligent pillow control method and system for sleep posture recognition and neck support optimization. The control method comprises the following steps: collecting distributed pressure array multi-region pressure data and head relative pillow posture data output by an inertial measurement unit; aligning the two types of data in time sequence and jointly estimating in a pillow coordinate system to obtain head rigid body and cervical spine reference segment relative posture parameters; inputting the posture parameters and pillow surface partition pressure feature vectors into a cervical spine biomechanics model to identify supine, lateral and prone sleeping postures; combining the sleeping posture and the posture to generate target pressure instructions for at least three independent air bag cavities, the air bags are corresponding to the pressure collection partition space, and the air bags are cooperatively controlled to maintain the cervical spine sagittal plane curvature in a physiological interval; and driving the independent inflation and deflation of the air bags. The application fuses multi-source data to accurately recognize head and neck postures, and multiple air bags are cooperatively adjusted to reduce support mutations, thereby improving the continuity, stability and comfort of neck support during the sleeping process.

Description

Technical Field

[0001] This application relates to the field of smart bedding and sleep monitoring technology, and in particular to a smart pillow control method and smart pillow control system for sleep posture recognition and neck support optimization. Background Technology

[0002] With the continued growth in demand for sleep health and cervical spine health management, smart pillows, which use sensors on or inside the pillow surface to sense the contact state between the head and the pillow, and adjust their internal support structure accordingly to provide dynamic support for the head and neck during sleep, have been applied in the consumer market and rehabilitation care scenarios. One common technological approach involves arranging a distributed pressure sensor array on the pillow surface to obtain multi-regional pressure distribution and inferring the user's sleeping posture based on changes in pressure distribution over time.

[0003] In related technologies, sleeping positions are typically categorized into supine, lateral, or prone positions based on the morphological characteristics of pressure distribution images or whether several pressure channels exceed fixed thresholds. There are also solutions that roughly determine the direction of head rotation based on the area where the pressure peak is located. Regarding support adjustment, some products use a single air chamber or a few air chambers that are uniformly inflated and deflated to change pillow height or overall firmness. While some products have multiple air chambers, the pressure of each air chamber often switches according to preset levels, resulting in a weak correlation between the pressure and the real-time sleeping position and cervical spine biomechanics, making it difficult to maintain continuous support during the transition phase of turning over.

[0004] However, classifying sleeping postures solely based on pressure distribution images or pressure thresholds is insufficient to fully reflect the spatial posture information of the head relative to the pillow and cervical spine. This can easily lead to misjudgments when transitioning between side-lying and supine positions, when there are minor head and neck movements, or when pressure distribution is partially absent. Furthermore, if the adjustment of the multi-cavity air bladders does not establish a stable closed-loop connection with physiological goals such as the sagittal curvature of the cervical spine, the independent adjustment of each cavity can easily result in sudden changes in support force or mismatch with the sleeping posture, thereby affecting sleep continuity and neck comfort.

[0005] Therefore, there is a need for an intelligent pillow control solution that can integrate pressure and inertial measurement information in the pillow coordinate system, coordinate the adjustment of no less than three independent air bladder cavities in conjunction with the biomechanical constraints of the human cervical spine, and maintain the sagittal curvature of the cervical spine within a preset physiological target range. Summary of the Invention

[0006] The main objective of this application is to provide a smart pillow control method and system for sleep posture recognition and neck support optimization. This system can reliably estimate the sleeping posture and the relative posture of the head and neck, and maintain the sagittal curvature of the cervical spine within a preset physiological target range under the synergistic effect of multi-cavity airbags, thereby improving the stability, continuity and comfort of neck support during sleep.

[0007] To achieve the above objectives, embodiments of the present invention provide a smart pillow control method for sleep posture recognition and neck support optimization, the method comprising the following steps: Acquire multi-region pressure distribution data output by the distributed pressure array in the pillow support area, and attitude data of the head relative to the pillow coordinate system output by at least one inertial measurement unit. Align the multi-region pressure distribution data and the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system to obtain the relative posture parameters between the head rigid body and the cervical spine reference segment; wherein, the joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds; The relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data are input into the preset human cervical spine biomechanical model to obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone. Based on the sleeping posture type determination result and the relative posture parameters, a target support pressure command is generated for no less than three independent airbag cavities in the multi-cavity airbag system. Each airbag cavity and the corresponding acquisition partition of the distributed pressure array have a preset correspondence in spatial arrangement. The target support pressure command is used to make the support force of each airbag cavity work together and maintain the sagittal curvature of the cervical spine within the preset physiological target range. The multi-chamber airbag system is driven to independently inflate and deflate each airbag chamber according to the target support pressure command.

[0008] This invention also provides an intelligent pillow control system for sleep posture recognition and neck support optimization. The system includes: a distributed pressure sensing module for collecting pressure distribution data of multiple areas on the pillow surface through a distributed pressure array; An inertial measurement module is used to acquire attitude data representing the head relative to the occipital coordinate system through at least one inertial measurement unit; The posture reconstruction module is used to align the multi-region pressure distribution data with the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system, and output the relative posture parameters between the head rigid body and the cervical spine reference segment. The joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds. The sleeping posture determination module is used to input the relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data into a preset human cervical spine biomechanical model and obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone. A support decision module is used to generate target support pressure commands for at least three independent airbag cavities in the multi-cavity airbag system based on the sleeping posture type determination result and the relative posture parameters. Each airbag cavity has a preset spatial correspondence with the corresponding acquisition partition of the distributed pressure array, and the target support pressure commands are used to ensure that the support forces of each airbag cavity work synergistically to maintain the sagittal curvature of the cervical spine within the preset physiological target range. The airbag drive module is used to perform independent inflation and deflation regulation of the multi-chamber airbag system according to the target support pressure command.

[0009] In summary, this application integrates pillow surface pressure and inertial measurements in the pillow coordinate system and obtains the relative posture parameters between the head rigid body and the cervical spine reference segment through joint estimation, which is different from sleeping posture classification that only relies on pressure image morphology or fixed thresholds. By combining the human cervical spine biomechanical model and the coordinated pressure planning of no less than three air sac cavities, the sagittal curvature of the cervical spine is limited to a preset physiological target range, and the inflation and deflation of the cavities is performed, which can reduce support abrupt changes and improve the support continuity during the sleeping posture transition process, thereby improving the stability, continuity and comfort of neck support during sleep. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram illustrating an application scenario of an intelligent pillow control system provided in an embodiment of this application; Figure 2 A flowchart illustrating a smart pillow control method provided in an embodiment of this application; Figure 3 This application provides a schematic diagram of a data acquisition and synchronous analysis data window formation process in an embodiment of the present application. Figure 4 A schematic diagram illustrating a joint estimation of relative attitude parameters provided in an embodiment of this application; Figure 5 A schematic diagram of a sleep posture determination and support decision-making process based on a biomechanical model provided in this application embodiment; Figure 6 A schematic diagram illustrating a progressive inflation / deflation and closed-loop correction process during sleep position switching, provided as an embodiment of this application; Figure 7A schematic diagram of an instruction update process provided in an embodiment of this application; Figure 8 A schematic diagram illustrating a process for determining individualized initial values, provided for an embodiment of this application; Figure 9 A flowchart illustrating instruction iteration correction provided in an embodiment of this application; Figure 10 A schematic diagram illustrating an anomaly recovery process provided in an embodiment of this application; Figure 11 This application provides a structural schematic of an intelligent pillow system according to an embodiment. Figure 12 This is a schematic diagram of the structure of a smart pillow with built-in control and drive circuit provided in an embodiment of this application. Detailed Implementation

[0012] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings, so that the technical solution, corresponding hardware structure, and beneficial effects of this application will be clearer and more complete. It should be understood that the embodiments described herein are only for explaining this application and do not constitute a limitation on the scope of protection of this application.

[0013] It should be noted that before and during the collection of pressure distribution data and inertial measurement data related to user sleep, this application may explain the purpose, scope, storage method, and deletion method of data collection to the user through prompt interfaces, pop-ups, or voice prompts. The corresponding data acquisition steps will only be executed after obtaining the user's confirmation of the above prompts; if user confirmation is not obtained, data acquisition may be suspended or not performed. The user data processing activities involved in this application shall comply with the laws and regulations of relevant countries and regions, as well as the requirements regarding personal information protection and data security. Data that can be used to identify specific natural persons should preferably be de-identified or anonymized before being used for non-real-time control purposes such as model training or statistical analysis. Data necessary for real-time control should be processed in a terminal device or user-controlled environment and its retention period should be limited.

[0014] In this embodiment, the pillow coordinate system refers to a right-handed rectangular coordinate system fixed to the pillow body structure of the smart pillow. Its origin can be selected as the projection point of the geometric center of the pillow body or the geometric center of the pressure array on the bottom surface of the pillow body. The coordinate axis directions can be defined by the manufacturer during factory calibration and stored in the non-volatile memory of the device. The acceleration, angular velocity and attitude angle measured by the inertial measurement unit are transformed to the pillow coordinate system through the known installation attitude matrix.

[0015] Cervical reference segments refer to the reference segments used in human cervical spine biomechanical models to characterize the relative posture relationship between the head and neck. For example, one or more segments selected from the second to the seventh cervical vertebrae can be used as references. The model parameters may include the equivalent stiffness, equivalent damping, and geometric constraints of the reference segment, which are used to establish a calculable relationship between head posture and cervical curvature index.

[0016] The preset physiological target range refers to the acceptable numerical range set for the sagittal curvature of the cervical spine or its monotonically related equivalent surrogate indicators. Its upper and lower limits can be determined by the range recommended by medical guidelines, the quantiles of statistical populations, or the calibration results of individual users, and are stored as configuration parameters.

[0017] Synchronous analysis data windows refer to continuous time segments extracted from pressure and attitude data streams under the condition of uniform sampling period and timestamp alignment, which are used to complete joint estimation, sleeping posture update and support decision input construction within an analysis period.

[0018] For ease of explanation, the following uses only necessary symbols when dealing with matrix operations, filtering estimation, or pressure optimization solutions, and explains the technical meaning of each symbol by its Chinese name upon its first appearance. For example, pressure frame P_Frame represents a frame of pressure data output by the pressure array at a sampling time, and inertial attitude data IMU_Pose represents the attitude and motion data output by the inertial measurement unit. For non-formulaic objects such as sleeping posture type, support requirement level, synchronous analysis data window, target support pressure command, and anomaly threshold, Chinese names are used as much as possible in the following descriptions. These names are not limited to variable naming methods, data storage formats, or communication message formats in the software.

[0019] In related technologies, sleeping posture recognition based on pressure distribution images or thresholds is easily affected by pillowcases and temperature drift, leading to deviations from the actual head and neck posture. Threshold boundaries are also prone to inducing classification jitter. Multi-chamber airbags, if they only switch pressure levels without coordination with sleeping posture and cervical curvature targets, are prone to support mismatch or localized compression. To address these shortcomings, this application integrates pressure and inertial information for joint estimation in the pillow coordinate system, and combines a human cervical spine biomechanical model with coordinated pressure commands from at least three independent airbag cavities to maintain the sagittal curvature of the cervical spine within a preset physiological target range. The following description, in conjunction with accompanying drawings and embodiments, illustrates this process.

[0020] See Figure 1The application scenarios of this application embodiment may include a smart pillow 110, a user's head and neck, and a computing unit 130 that communicates with or is built into the pillow body. A distributed pressure array 114 is arranged on the surface of the smart pillow 110, and a multi-cavity airbag system 112 and air path components are arranged inside the pillow body. The computing unit 130 may include a microcontroller, digital signal processor, or application processor, etc., for executing the method steps of this application embodiment. In some embodiments, the computing unit 130 is located in a circuit compartment inside the pillow body to shorten the length of the sensor sampling harness and reduce external electromagnetic interference; in other embodiments, the computing unit 130 may be partially or entirely located in a mobile phone, gateway, or edge computing device connected to the smart pillow 110 via wired or wireless connection, which is not limited in this application embodiment. The smart pillow 110 may also include a power management unit and a charging interface for providing stable power to the distributed pressure array, inertial measurement unit, micro air pump, and solenoid valve; and may include vibration damping pads, sound-absorbing cavities, or flexible covering structures to reduce the transmission of air pump and valve operation noise to the pillow surface and reduce interference with the sleep sound environment.

[0021] In terms of specific structure, the smart pillow 110 may also include a pillow surface layer 111, a cushioning layer, a multi-chamber airbag system 112, and a base 113. The pillow surface layer may adopt a composite structure of breathable fabric and slow-rebound sponge or gel thin layer to ensure a stable transmission relationship between the contact load measured by the pressure array and the user's perceived firmness, and reduce the impact of sweat accumulation on the pressure baseline. The cushioning layer is located between the pressure array 114 and the multi-chamber airbag system 112, and is used to homogenize the rib feel of the airbags and couple the deformation of each cavity into a continuous support surface. Its thickness and Shore hardness can be designed in conjunction with the geometric lift of the neck support area to make the cervical spine reference segment closer to the model's preset sagittal alignment trend in the supine position.

[0022] The distributed pressure array 114 divides the occipital support area into multiple pressure acquisition zones, exemplarily including at least three zones from the top of the head, back of the head, neck support, and lateral restraint zone; each zone may correspond to a set of pressure sensing units or a sub-region of a flexible pressure sensing surface. The scanning circuit and flexible printed circuit can be serpentinely routed along the width of the occipital surface to reduce bending stress concentration; each zone pin corresponds one-to-one with the analog-to-digital converter front end or the line scanning controller channel within the circuit compartment, thereby fixing the mapping relationship between zones, channels, and storage indexes at the hardware level, providing a definite boundary for the extraction of pressure feature vectors from the occipital surface zones.

[0023] The multi-chamber airbag system 112 includes at least three independent airbag chambers. For example, the first chamber corresponds to the main support of the occipital bone, the second chamber corresponds to the neck support lifting, the third chamber corresponds to lateral or top head restraint, and the fourth chamber can serve as an auxiliary balancing chamber. Each airbag chamber can be connected to an inflation valve, a deflation valve, and a miniature air pump. The air path can use silicone tubing connected to the air distribution block for convergence. The tubing route should avoid being directly below the pressure array sensitive nodes to reduce pressure crosstalk caused by compression. In terms of spatial arrangement, there is a preset correspondence between each airbag chamber and the pressure acquisition zone. This correspondence is stored in the form of a calibration table and can record the coupling coefficient between the air pressure sensor readings of each chamber and the total pressure of the zone, which is used for rapid feedforward correction before the target support pressure command is issued. The inertial measurement unit can be embedded in the side of the headrest or the forehead support to make it closer to the head's center of mass movement trend under typical usage posture. The installation rotation matrix between its sensitive axis and the headrest coordinate system is calibrated at the factory and can be supplemented with heat conduction and grounding measures to reduce the impact of digital scanning noise on gyroscope zero bias estimation.

[0024] Furthermore, in this embodiment, multi-region pressure distribution data can be obtained by scanning and sampling from the distributed pressure array 114, and attitude data can be sent to the computing unit by the inertial measurement unit via serial bus interrupt or direct memory access; joint estimation is completed by reading synchronous samples in the annular buffer within a fixed control cycle; the biomechanical model parameters of the human cervical spine and the target support curve template can be loaded into static random access memory or serial flash memory; after the airbag drive module drives the valve pump according to the instruction, the pressure and air pressure feedback enter the buffer again, forming an intraoccipital closed loop of sensing, estimation, decision-making, execution to re-sensing.

[0025] In one embodiment, reference Figure 2 The smart pillow control method for sleep posture recognition and neck support optimization may specifically include steps S30 to S34, which will be described in detail below: S30: Acquire multi-region pressure distribution data output by the distributed pressure array in the pillow support area, and attitude data of the head relative to the pillow coordinate system output by at least one inertial measurement unit.

[0026] In this embodiment, step S30 is used to establish the original observations required for subsequent alignment and estimation; the multi-region pressure distribution data can be regarded as the pressure set of discrete sampling points on the pillow surface at a single moment, and the attitude data is the kinematic quantity after being transformed to the pillow coordinate system by the installation attitude matrix. For ease of description, a frame of pressure data output by the pressure array at a sampling moment can be denoted as pressure frame P_Frame, and the elements in the pressure frame represent the pressure value of the corresponding pressure sampling point; the attitude and motion data output by the inertial measurement unit can be denoted as inertial attitude data IMU_Pose, and the inertial attitude data may include at least one of three-axis acceleration, three-axis angular velocity, attitude angle, or attitude quaternion.

[0027] For example, when the distributed pressure array scans at tens of hertz, it outputs M×N pressure values ​​per scan cycle, where M represents the number of rows in the pressure array and N represents the number of columns in the pressure array; the inertial measurement unit outputs triaxial acceleration, triaxial angular velocity and fused attitude quaternions at a higher frequency, and the firmware adds the same clock domain timestamp to the two types of data in a circular buffer.

[0028] In one embodiment, the pressure array interface writes the pressure frame to the buffer via direct memory access or interrupt mode; the inertial interface reads the inertial attitude data via a serial peripheral interface or integrated circuit bus, and extracts several inertial samples with the smallest time difference from the current pressure frame within the same control cycle for averaging or interpolation, so as to pair with the pressure frame.

[0029] S31: After aligning the multi-region pressure distribution data and the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system to obtain the relative posture parameters between the head rigid body and the cervical spine reference segment; wherein, the joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds.

[0030] In this embodiment, step S31 fuses pressure space features and inertial features in the pillow-fixed coordinate system and outputs the relative attitude parameter RelPose. Time alignment refers to ensuring that the pressure sample and attitude sample have a consistent or interpolable time reference within the same synchronous analysis data window; coordinate transformation refers to mapping the inertial measurement unit sensor coordinate system measurement multiplied by the mounting attitude matrix R_IMU2Pillow to the pillow coordinate system, where the mounting attitude matrix represents the mounting attitude matrix from the inertial measurement unit coordinate system to the pillow coordinate system, used to transform the attitude data collected by the inertial measurement unit to the pillow coordinate system.

[0031] For example, the set of effective pressure points within the window can be used to calculate the pressure center of gravity and the direction of the principal axis. The gyro integral and acceleration observation input can be extended by Kalman filtering or complementary filtering to obtain the head rigid body attitude quaternion q_head. The head rigid body attitude quaternion is used to characterize the three-dimensional attitude of the head rigid body relative to the occipital coordinate system. Then, the relative flexion-extension angle, lateral flexion angle and axial deflection can be calculated by combining the pressure change in the cervical support area.

[0032] In one embodiment, when using extended Kalman filtering, the prediction step updates the head rigid body attitude quaternion using gyro integrals, and the update step corrects pitch and roll using tilt observations derived from pressure. When using complementary filtering, the attitude estimate obtained from angular velocity integrals and the attitude observations derived from pressure distribution characteristics can be weighted and fused according to preset fusion weights to obtain a fused head attitude estimate. The fusion weights are used to adjust the proportion of inertial measurement results and pressure observation results in the attitude estimation, so as to maintain tracking of slowly varying contact changes while suppressing high-frequency noise.

[0033] S32: Input the relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data into the preset human cervical spine biomechanical model to obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone positions.

[0034] In this embodiment, step S32 can map the mechanical features and relative posture parameters together to the sleeping posture category and the selectable support requirement level; the pillow surface partition pressure feature vector can be obtained by splicing the load ratio, peak value and gradient statistics of each pressure acquisition partition in a fixed order; the human cervical spine biomechanical model outputs intermediate quantities such as cervical extension torque, lateral load degree and contact stability index, and then obtains the sleeping posture type through a threshold or classifier.

[0035] For example, the pressure feature vector of the occipital region can be concatenated with the cervical spine posture feature group to form the model input vector, which can be forward calculated by the state space equation or a feedforward network with limited layers to obtain a mechanical index vector including posterior extension moment of the neck, degree of lateral load and contact stability index.

[0036] In one embodiment, a hysteresis comparator is set for each component in the mechanical index vector: when the lateral load is higher than a preset lateral load threshold and the contact stability index is lower than a preset lateral stability threshold, the current sleeping position can be determined to be lateral; when the neck extension torque is higher than a preset supine extension torque threshold and the lateral load is lower than a preset supine load threshold, the current sleeping position can be determined to be supine, thereby reducing boundary posture jitter.

[0037] S33: Based on the sleeping posture type determination result and the relative posture parameters, generate a target support pressure command for no less than three independent airbag cavities in the multi-cavity airbag system. Each airbag cavity and the corresponding acquisition partition of the distributed pressure array have a preset correspondence in spatial arrangement. The target support pressure command is used to make the support force of each airbag cavity work together to maintain the sagittal curvature of the cervical spine within the preset physiological target range.

[0038] In this embodiment, step S33 solves for the collaborative target pressure vector on each independent airbag cavity of the multi-cavity airbag system, making the predicted zone load after spatial correspondence matrix mapping close to the target load, while ensuring that the cervical spine sagittal curvature correspondence index CurvIndex falls within the preset physiological target interval [Curv_min, Curv_max]. Here, the spatial correspondence matrix represents the preset spatial correspondence between the pressure acquisition zone and the airbag cavity, used to characterize the influence of pressure changes in each airbag cavity on the support load of each pressure acquisition zone; Curv_min represents the lower limit of the preset physiological target interval, and Curv_max represents the upper limit of the preset physiological target interval; when using a single reference value for control, the target reference value within the interval can be recorded as the cervical spine sagittal curvature reference value Curv_ref. The target support curve template can be called according to the current sleeping posture type and cervical spine support requirement level, and provides the baseline pressure trajectory of each airbag cavity; the relative posture parameters are used to calculate the compensation increment.

[0039] For example, the reference pressure vector of each air bladder cavity at the starting point of the current analysis cycle can be obtained by looking up a table. Then, based on the deviation between the corresponding index of cervical spine sagittal curvature and the reference value of cervical spine sagittal curvature, the pressure compensation increment of the air bladder corresponding to the cervical support area can be calculated. The pressure compensation increment of the air bladder corresponding to the cervical support area can be determined by both proportional and differential terms. The proportional term is used to reflect the current curvature deviation, and the differential term is used to reflect the trend of curvature deviation. The occipital cavity and lateral cavity are given a smaller gain to obtain the candidate pressure vector.

[0040] In one embodiment, the objective function can be constructed using the pressure vector p of each airbag cavity as the optimization variable: F(p)=||Mapping·p-f_load||^2+λ||p-P_personal||^2.

[0041] The constraint is: p_min≤p≤p_max.

[0042] Where F(p) represents the objective function to be minimized, Mapping represents the spatial correspondence matrix, Mapping·p represents the partitioned support load vector obtained from the pressure prediction of each airbag cavity, f_load represents the target partitioned load vector, P_personal represents the individualized initial value vector, λ represents the regularization coefficient, p_min represents the lower limit vector of allowable pressure for each airbag cavity, and p_max represents the upper limit vector of allowable pressure for each airbag cavity; p_min≤p≤p_max means that each airbag pressure component in p lies between the corresponding lower and upper limit of allowable pressure; ||·||^2 represents the sum of squares of the components of the corresponding vector. By solving for p that minimizes F(p) under the above constraints, the cooperative target pressure vector can be obtained, and the target support pressure command can be formed after further applying the pressure change rate constraint.

[0043] S34: Drive the multi-chamber airbag system to perform independent inflation and deflation regulation of each airbag cavity according to the target support pressure command.

[0044] In this embodiment, step S34 can convert the target support pressure command into the duty cycle or switching sequence of the inflation valve, deflation valve and micro air pump of each cavity, so that the pressure of each cavity tracks the target; each cavity can be independently inflated, deflated or pressure maintained, and the actual pressure of each airbag cavity is updated without affecting each other.

[0045] For example, a proportional valve or two-position valve is driven by pulse width modulation, the pump supplies air within the upper limit of the duty cycle, and the air pressure sensor of each chamber provides feedback for inner loop stabilization or feedforward correction.

[0046] In one embodiment, when a change in sleeping position is detected and the confidence level is higher than a threshold, the target support pressure command is first segmented and interpolated within a preset transition time window before being issued for execution. The preset transition time window represents the length of time taken for the smooth transition of target pressure during the change in sleeping position, so as to avoid pressure step. After the transition is completed, the endpoint is written to the current target register and a short-term anti-shake timer is started.

[0047] In one embodiment, reference Figure 3 Step S30 may specifically include steps S301 to S304, which will be described in detail below: S301: Divide the distributed pressure array into at least three pressure acquisition zones among the top of the head, the back of the head, the neck brace, and the lateral restraint zone.

[0048] In this embodiment of the application, step S301 can establish a one-to-one mapping between pressure acquisition partitions and array row and column indices at the software mask or hardware channel grouping level, so that subsequent partition statistics have definite boundaries; the partition labels can be stored as a two-dimensional mask table, and the element values ​​of the two-dimensional mask table are partition enumerations.

[0049] For example, the top of the head, the back of the head, the neck support, and the lateral restraint area can be coded as 1 to 4 respectively. The two-dimensional mask table is aligned with the flexible printed circuit wiring partition and the airbag hot-pressing dividing line in the top view projection, which facilitates the production measurement and positioning of broken lines or crosstalk partitions.

[0050] In one embodiment, the factory calibration writes a two-dimensional mask table into the configuration sector with verification and binds it to the analog-to-digital conversion channel number; during runtime, the pressure frames are reduced and summed or weighted averaged according to the two-dimensional mask table to obtain the composite pressure of each pressure acquisition partition; each composite pressure corresponds to a pressure acquisition partition, which can be used to calculate the load ratio, pressure centroid and partition pressure characteristics in subsequent calculations.

[0051] S302: Extract the effective support area based on the contact start time of each pressure acquisition zone, and remove invalid pressure points below the contact threshold.

[0052] In this embodiment, the contact start time refers to the first moment when the synthetic pressure of a certain zone rises steadily from below the contact threshold to above the threshold, which is used to define the reference time when the head enters the zone for support; the effective support area is the set of measurement points above the threshold within the window, which is used to calculate the pressure center of gravity and load ratio, and points below the threshold are marked as invalid and have zero weight in the feature calculation.

[0053] For example, a first-order hysteresis comparator is used for the composite pressure in the occipital region. When the pressure rises from below 0.8 kPa to above 1.2 kPa and remains there for 50 milliseconds, the contact start time is latched. From this contact start time, single points below 0.5 kPa are considered invalid pressure points and are discarded.

[0054] In one embodiment, independent upper and lower hysteresis thresholds are maintained for each partition to prevent oscillations near the threshold from causing frequent jumps in the effective support area; the effective support area can be cached in bitmap form for reuse in steps S31 and S32.

[0055] S303: Use the gravity component in the static state to perform zero-bias calibration on the at least one inertial measurement unit to obtain an initial attitude reference relative to the pillow coordinate system.

[0056] In this embodiment, the static state refers to the state when the macroscopic motion of the head relative to the pillow body is below the threshold and the accelerometer mainly senses gravity; the zero bias calibration estimates the zero-point offset of the gyroscope and the accelerometer, and the initial attitude reference is the reference quaternion or Euler angle written to the register before entering the joint estimation.

[0057] For example, when the user presses and holds the calibration button and the three-axis acceleration variance is below the threshold for 2 seconds, the installation error and zero bias are estimated and fixed by using the difference between the expected direction and the actual direction of gravity in the pillow coordinate system.

[0058] In one embodiment, a combination of six-position flip calibration and on-site single-position gravity alignment is used: coarse calibration parameters are written into the production line, and users only need to update the gyroscope zero bias, accelerometer zero bias and small attitude correction during calibration to reduce the burden of on-site operation.

[0059] S304: The pressure data stream and attitude data stream are timestamped according to a unified sampling period to form a synchronous analysis data window. The synchronous analysis data window serves as the input data window for aligning the multi-region pressure distribution data and the attitude data on the time axis.

[0060] In this embodiment, the unified sampling period is the clock cycle of the controller scheduling pressure scanning and inertial readings; timestamp alignment pairs the pressure frame time with the inertial sample set to form a synchronous analysis data window of a preset window length, which is used as the input for step S31.

[0061] For example, when the pressure frame rate is 20 Hz and the inertial sampling is 200 Hz, the average of the 10 inertial samples with the smallest time difference at each pressure frame moment is taken as the paired sample; the window length is about 1 second, which contains about 20 pressure frames.

[0062] In one embodiment, half-window overlap or recursive updates are used at the boundaries of the synchronous analysis data window to ensure the continuity of the filter state in step S31; for lost stress frames, the previous frame can be used to hold the frame briefly and mark the quality flag for subsequent abnormal degradation control.

[0063] In one embodiment, reference Figure 4 Step S31 may specifically include steps S311 to S314, which will be described in detail below: S311: Based on the pressure distribution of the effective support area within the synchronous analysis data window, determine the pressure center of gravity, the direction of the main pressure axis, and the difference between the left and right loads.

[0064] In this embodiment, the pressure centroid is a weighted average of the coordinates of the sensing points with pressure values ​​as weights; the pressure principal axis direction is obtained by the eigenvalue decomposition of the second-order central moment matrix of the pressure distribution; the difference between the left and right loads is the difference between the combined pressures of the left and right sides of the pillow coordinate system, which is used to characterize the asymmetry of the lateral loads.

[0065] For example, on the set of effective points, the pressure centroid can be obtained by weighted averaging of the pressure values ​​and coordinates of each effective pressure point; the difference between the left and right loads can be obtained by subtracting the combined effective pressure value on the left and the combined effective pressure value on the right from the combined effective pressure value on the right side of the pillow coordinate system.

[0066] In one embodiment, the second-order moment matrix is ​​decomposed into eigenvalues, and the projection angle of the eigenvector corresponding to the largest eigenvalue on the pillow surface is taken as the direction of the pressure principal axis. The pressure centroid, the direction of the pressure principal axis, and the difference between the left and right loads are averaged to reduce single-window noise.

[0067] S312: Decompose the attitude data into pitch angle, roll angle and angular velocity components, and transform them to the pillow coordinate system.

[0068] In this embodiment of the application, step S312 decomposes the output of the inertial unit into pitch angle, roll angle and angular velocity vectors, and transforms them to the bolster coordinate system by installing the attitude matrix, thereby obtaining the pitch angle, roll angle and angular velocity vectors in the bolster coordinate system.

[0069] For example, when the inertial unit outputs the raw attitude quaternion q_raw, it can be converted to Euler angles before applying the installation attitude matrix; alternatively, the raw attitude quaternion can be transformed into quaternion coordinates based on the installation attitude quaternion q_install to obtain the attitude quaternion q_pillow transformed to the pillow coordinate system. In one implementation, this transformation can be expressed as q_pillow = q_install. q_raw, where This represents quaternion multiplication.

[0070] In one embodiment, the pitch and roll angles in the pillow coordinate system are subjected to limiting and low-pass filtering, and the angular velocity vector in the pillow coordinate system is zero-bias compensation and then input into the filter prediction step of step S313.

[0071] S313: Using the pressure center of gravity, the pressure axis direction, the left and right load difference, the pitch angle, the roll angle, and the angular velocity components as joint observations, the extended Kalman filter or complementary filter is used to estimate the rigid body attitude of the head.

[0072] In this embodiment, the joint observations are simultaneously entered into the observation equation; the extended Kalman filter updates the head attitude quaternion using angular velocity vector integration in the prediction step, and corrects the attitude using tilt angle derived from pressure in the update step; the complementary filter fuses the high-frequency information from the gyroscope and the low-frequency information from pressure according to the frequency domain division of labor.

[0073] For example, the extended Kalman filter state can be a nine-dimensional or seven-dimensional minimal set, and the observations are the deviations between the pressure tilt angle and the pressure center of gravity relative to the nominal position; complementary filtering can perform weighted fusion of attitude angles or attitude quaternions. When using attitude quaternion representation, the predicted attitude can first be obtained from the integral of angular velocity, then the attitude correction can be obtained from the pressure distribution characteristics, and the predicted attitude and attitude correction can be fused according to the complementary filtering weight coefficient α. After fusion, normalization is performed to obtain the head attitude estimation result. Here, α is used to adjust the fusion ratio of the inertial prediction result and the pressure observation result.

[0074] In one embodiment, when the pressure missing rate within the synchronous analysis data window increases, the proportion of inertial prediction results in the complementary filter weight coefficients is automatically increased, making attitude estimation more dependent on inertia; when inertial drift detection is triggered, the proportion of pressure observation results is temporarily increased, and pressure tilt angle is used to pull it back.

[0075] S314: Using the head rigid body posture and the pressure change in the cervical collar area, calculate the relative flexion-extension angle, lateral flexion angle, and axial deflection between the head rigid body and the cervical reference segment. The relative flexion-extension angle, lateral flexion angle, and axial deflection are used to characterize the relative posture parameters.

[0076] In this embodiment, the change in neck brace pressure is the combined neck brace pressure difference between the current synchronous analysis data window and the reference window; the relative flexion-extension angle, lateral flexion angle, and axial deflection are obtained by mapping the head rigid body posture angle, the change in neck brace pressure, and the difference in left and right loads through the posture calibration matrix K_pose, and the three constitute the relative posture parameters; wherein, the posture calibration matrix represents a mapping matrix obtained according to user calibration or factory calibration, used to convert the head posture amount and the change in neck brace pressure into relative posture parameters.

[0077] For example, the relative flexion-extension angle can be determined by the pitch angle in the pillow coordinate system and the pressure change in the neck support area, the lateral flexion angle can be calculated by the roll angle in the pillow coordinate system and the difference between the left and right loads using the least squares method, and the axial deflection can be obtained by the difference between the pressure principal axis direction and the model reference direction.

[0078] In one embodiment, physiologically feasible domain constraints are applied to the relative flexion-extension angle, lateral flexion angle, and axial deflection and output to step S32; when the change in pressure in the cervical collar area exceeds the reasonable range, the weight of the change in pressure in the cervical collar area in the solution of relative posture parameters is reduced to prevent abnormal contact from causing a jump in relative posture parameters.

[0079] In one embodiment, reference Figure 5 Step S32 may specifically include steps S321 to S324, which will be described in detail below: S321: Construct a cervical spine posture feature group based on the relative flexion-extension angle, lateral flexion angle, and axial deflection.

[0080] In this embodiment, the cervical spine posture feature group includes relative flexion-extension angle, lateral flexion angle, axial deflection and its derived quantities such as first-order difference and sliding window variance, which are used to characterize the relative movement trend of the head and neck. The relative flexion-extension angle, lateral flexion angle and axial deflection can be expressed in degrees or radians, and the units should remain consistent within the same control cycle.

[0081] For example, the cervical spine posture characteristics group may include relative flexion-extension angle, lateral flexion angle, axial deflection, rate of change of relative flexion-extension angle over time, and variance of lateral flexion angle within the synchronous analysis data window. Among them, the rate of change of relative flexion-extension angle over time is used to reflect the trend of head and neck flexion-extension changes, and the variance of lateral flexion angle is used to reflect the degree of lateral posture stability.

[0082] In one embodiment, the components of the cervical spine posture feature group are standardized, or normalized according to the mean and variance stored by the user's individual calibration, in order to improve the numerical stability of the model.

[0083] S322: Construct a pressure feature vector for the pillow surface region based on the load ratio, contact area, and pressure gradient of each pressure acquisition zone.

[0084] In this embodiment, the load ratio is the ratio of the synthesized pressure of the i-th pressure acquisition zone to the total effective synthesized pressure of the entire pillow, the contact area is the number of points exceeding the threshold or the equivalent area within the i-th pressure acquisition zone, and the pressure gradient is the pressure spatial gradient norm in the neighborhood of the i-th pressure acquisition zone, where i represents the pressure acquisition zone number; the pressure feature vector of the pillow surface zone can be obtained by splicing the load ratio, contact area, and pressure gradient of each zone according to a preset feature order.

[0085] For example, when dividing into four regions, 12 original features are first obtained, and then the dimensionality is reduced to 6 dimensions through principal component analysis or linear layer as part of the model input.

[0086] In one embodiment, the load percentage is normalized and untouched partitions are eliminated to avoid division by zero; the pressure gradient is calculated using finite difference within the partition boundaries.

[0087] S323: Input the cervical spine posture feature group and the occipital surface zone pressure feature vector into the human cervical spine biomechanical model to calculate the posterior extension torque, lateral load degree and contact stability index.

[0088] In this embodiment, the human cervical spine biomechanical model maps the model input vector to a mechanical index vector including cervical extension torque, lateral load degree, and contact stability index; the model can be an analytical rigid body chain, a reduced-order finite element proxy, or a small feedforward network.

[0089] For example, a linear model can take the form of "mechanical index vector = weight matrix × model input vector + bias vector", where the weight matrix represents the mapping weight of each feature in the model input vector to the mechanical index, and the bias vector represents the baseline offset of the model output; a nonlinear model can take two fully connected layers plus rectified linear unit activation.

[0090] In one embodiment, model parameters are grouped and stored in flash memory according to age, gender, and body type. After the user selects a type on the terminal, the corresponding weight matrix and bias vector are loaded.

[0091] S324: Determine the current sleeping posture type and the corresponding cervical spine support requirement level based on the posterior neck extension torque, the lateral load degree, and the contact stability index.

[0092] In this embodiment, the sleeping posture type and cervical support requirement level are determined by the posterior neck extension torque, lateral load degree, and contact stability index and threshold surface or classification hyperplane; the cervical support requirement level is used in step S33 to select template strength and change rate constraints.

[0093] For example, when the lateral load is higher than the lateral load threshold and the contact stability index is lower than the lateral stability threshold, the current sleeping position is determined to be lateral and a higher cervical support requirement level is determined; when the posterior extension torque is higher than the supine extension torque threshold and the lateral load is lower than the supine load threshold, the current sleeping position is determined to be supine and a medium cervical support requirement level is determined.

[0094] In one embodiment, a consistent two-window hysteresis rule is applied to the determination result to avoid light sleep micro-movements causing changes in sleep posture type; the sleep posture type, cervical spine support requirement level, and relative posture parameters are output to step S33.

[0095] In one embodiment, reference Figure 6 Step S33 can specifically include steps S331 to S334, which will be described in detail below: S331: According to the sleeping posture type and the cervical spine support requirement level, call the target support curve template corresponding to the top of the head, the back of the head, and the neck support area.

[0096] In this embodiment, the target support curve template is obtained by looking up a table based on the sleeping posture type and the cervical spine support requirement level; the template stores the reference pressure nodes of each air bladder cavity over time or phase; the top of the head, the back of the head, the neck support area and the corresponding cavity or cavity combination are pre-bound in engineering.

[0097] For example, the high demand in lateral lying corresponds to a higher reference node for the lateral restraint cavity and the cervical support cavity, while the demand in supine lying corresponds to a node sequence dominated by the posterior occipital cavity and moderately oriented for the cervical support cavity.

[0098] In one embodiment, the template distinguishes between daytime nap and nighttime sleep modes, with the nighttime mode having a longer transition time and stricter change rate constraints; the template version number is written into the communication message capability field to ensure terminal compatibility during distribution.

[0099] S332: Calculate the pressure compensation increment of each airbag cavity based on the relative flexion-extension angle, lateral flexion angle and axial deflection, wherein the compensation increment of the airbag corresponding to the cervical collar area is used to preferentially correct the sagittal curvature deviation of the cervical spine.

[0100] In this embodiment, each component of the pressure compensation increment vector is calculated by the relative flexion-extension angle, lateral flexion angle, axial deflection, and curvature deviation through a gain matrix; the compensation gain corresponding to the cavity of the cervical collar is the largest, which is used to preferentially pull the cervical spine sagittal curvature corresponding index back to the preset physiological target range.

[0101] For example, when calculating the pressure compensation increment of the airbag corresponding to the neck brace area using the proportional-derivative compensation method, it can be expressed as: The pressure compensation increment of the neck brace airbag = k_p × e_Curv + k_d × d(e_Curv) / dt.

[0102] Wherein, e_Curv represents the deviation between the cervical spine sagittal curvature index CurvIndex and the cervical spine sagittal curvature reference value Curv_ref. The sign of e_Curv indicates the direction of deviation of the current curvature index relative to the reference curvature index. k_p represents the proportional compensation coefficient, k_d represents the differential compensation coefficient, and d(e_Curv) / dt represents the rate of change of this deviation over time. The increments in the posterior occipital and lateral cavities can use smaller proportional compensation coefficients, differential compensation coefficients, or only proportional terms. The positive and negative directions of the pressure compensation increment can be determined according to the airbag arrangement and factory calibration relationship, and are used to indicate the pressure increase or decompression trend of the corresponding airbag cavity.

[0103] In one embodiment, when the absolute value of the curvature deviation is less than a preset dead zone, the pressure compensation increment is not updated to reduce steady-state oscillations; when the absolute value of the curvature deviation exceeds a safety threshold, the amplitude of the pressure compensation increment is limited and an alarm flag is triggered.

[0104] S333: Based on the preset spatial correspondence between each pressure acquisition zone and each airbag cavity, solve for the cooperative target pressure value of no less than three independent airbag cavities.

[0105] In this embodiment, the preset spatial correspondence can be represented by the spatial correspondence matrix Mapping. For each airbag cavity pressure vector p to be solved, Mapping·p represents the partition support load vector predicted by the pressure vector; by solving the airbag cavity pressure vector p, the spatial correspondence matrix Mapping·p is made as close as possible to the target partition load vector f_load, thereby obtaining the cooperative target pressure value.

[0106] In one embodiment, the number of rows in the spatial correspondence matrix Mapping corresponds to the number of pressure acquisition zones, and the number of columns corresponds to the number of independent airbag cavities. When the spatial correspondence matrix Mapping is a full-rank square matrix, the collaborative target pressure value can be solved by matrix inversion; when the spatial correspondence matrix Mapping is a non-square matrix or does not meet the invertibility condition, it can be solved by non-negative least squares, pseudo-inverse, or least squares with regularization. The regularization term can be expressed as λ||p-P_personal||^2, where P_personal represents the individualized initial value vector, and λ represents the regularization coefficient, used to limit the deviation of the collaborative target pressure value from the individualized initial value. When lying on one's side, different upper and lower pressure limits can be applied to the lateral airbag cavities according to the lateral support requirements corresponding to the left or right lateral lying position, to adapt to the difference in lateral support caused by shoulder elevation.

[0107] S334: Apply upper and lower pressure limits and change rate constraints to the cooperative target pressure value to form the target support pressure command.

[0108] In this embodiment, the collaborative target pressure value is component-wise trimmed to between the lower and upper limits of the allowable pressure, and then the variation range between the collaborative target pressure value and the actual pressure in the previous control cycle is limited within adjacent control cycles to obtain the target support pressure command; the rate of change constraint is used to prevent sudden changes in the user's sensation.

[0109] For example, the lower and upper limits of allowable pressure are calibrated according to the safety and comfort of the cavity, and the rate of change is constrained to take a smaller value in the long sleep mode at night.

[0110] In one embodiment, if the pressure upper and lower limit constraints cause the index corresponding to the sagittal curvature of the cervical spine to still exceed the limit, the allocation ratio of the cervical support cavity change rate budget is increased in the next cycle, but not exceeding the upper limit of the hardware pump valve capacity.

[0111] In one embodiment, reference Figure 7 Step S34 may specifically include steps S341 to S344, which will be described in detail below: S341: Detect whether the sleeping posture type changes within two adjacent synchronous analysis data windows, and calculate the switching confidence level.

[0112] In this embodiment, two adjacent synchronous analysis data windows are consecutive or partially overlapping windows in time; sleep posture switching refers to changes in sleep posture type category; the switching confidence can be obtained by the product of the classification probabilities of the consecutive windows, Bayesian posterior, or evidence theory.

[0113] For example, if the probability of a patient being in a lateral position in the first window is 0.85 and the probability of a patient being in a supine position in the second window is 0.82, the switching confidence level can be calculated based on these two probabilities to characterize the reliability of switching from a lateral position to a supine position.

[0114] In one embodiment, a switch is only considered if two or more consecutive synchronous analysis data windows point to the new sleeping position and the switch confidence level is higher than a high threshold, in order to reduce false alarms; if the threshold is not reached, the original sleeping position type and target support pressure trajectory are maintained.

[0115] S342: When the switching confidence level is higher than a preset threshold, the collaborative target pressure value before and after the switching is segmented and interpolated within a preset transition time window.

[0116] In this embodiment, the preset threshold is the threshold for enabling smooth transition; the collaborative target pressure values ​​before and after the switch are respectively used as the interpolation start point and the interpolation end point; a transition target pressure sequence is generated within the preset transition time window, which can be piecewise linear or cubic spline.

[0117] For example, the preset transition time window can be 8 seconds. The first 25% of the time maintains the collaborative target pressure value before the switch, the middle 50% of the time is linearly interpolated, and the last 25% of the time approximates the collaborative target pressure value after the switch.

[0118] In one embodiment, the pressure components of each airbag cavity in the transition target pressure sequence are interpolated independently but with synchronized start and end times to avoid asynchrony between cavities leading to torsional discomfort; the interpolation results are written into the target queue for consumption by the valve pump scheduler.

[0119] S343: Control the inflation valve, deflation valve and micro air pump of each airbag cavity to perform progressive inflation and deflation according to the segmented interpolation results.

[0120] In this embodiment, progressive inflation and deflation are achieved by limiting the duty cycle ramp of the valve, the upper limit of the pump pulse width modulation, or staged pressurization and depressurization; the inflation valve, deflation valve, and micro pump corresponding to the i-th airbag cavity are controlled respectively, where i represents the airbag cavity number, and closed-loop adjustment is performed according to the deviation between the transition target pressure sequence and the actual pressure of each airbag cavity.

[0121] For example, the maximum duty cycle of the valve is gradually increased from 20% to the target duty cycle per cycle to prevent excessive airflow impact noise.

[0122] In one embodiment, when the actual pressure of a certain cavity lags behind the transition target pressure sequence by more than the tolerance, the upper limit of the duty cycle of that cavity is temporarily increased. However, due to the constraint of the total power of the pump, the remaining duty cycle budget needs to be allocated according to the priority of the neck brace, the backrest, and the side.

[0123] S344: After the transition time window ends, lock the updated target support pressure command to suppress sleep disturbances caused by sudden changes in support force.

[0124] In this embodiment of the application, locking refers to rejecting large target rewriting that conflicts with the current transition before the transition is completed, or writing the interpolation endpoint into the target support pressure register and starting the anti-shake timer; the anti-shake timer is used to suppress short-term repeated sleeping position switching triggers within a preset anti-shake duration after the transition is completed.

[0125] For example, the preset stabilization duration can be 3 seconds; during the lock-up period, subsequent closed-loop small-step corrections are still allowed, and if they do not conflict with the transition, they are arbitrated by the state machine.

[0126] In one embodiment, after the transition is completed, the switching event code and the sleeping posture type before and after the switch are recorded in the log for operation and maintenance diagnosis and user application display of sleeping posture statistics.

[0127] In one embodiment, before performing step S331 of calling the target support curve template, reference is made to... Figure 8 The individualized calibration steps may specifically include steps S335 to S338, which will be described in detail below: S335: Collects baseline pressure distribution and baseline posture data while the user remains in a resting supine position.

[0128] In this embodiment, the resting supine position is the calibrated posture of the user lying naturally with the head and neck relaxed; the reference pressure distribution is the result of multi-frame pressure frame fusion; and the reference attitude data is the median or mean of the inertial attitude angle or attitude quaternion within the same acquisition time window.

[0129] For example, prompt the user to lie flat for 15 seconds, collect 30 frames of pressure and take the average to obtain the baseline pressure distribution, and simultaneously collect a quaternion sequence and take the median posture to obtain the baseline posture data.

[0130] In one embodiment, the body motion variance is detected in real time during the data acquisition process. If the variance exceeds the limit, the acquisition time is extended or a recalibration prompt is given to ensure that the reference pressure distribution and reference posture data represent the steady-state supine position.

[0131] S336: Determine the individualized spacing parameters between the main support point of the occipital bone and the starting point of the neck brace based on the aforementioned reference pressure distribution.

[0132] In this embodiment, the main support point of the back of the head is the position where the back of the user's head mainly presses against the pillow surface, which can be determined by the position of the maximum pressure or the position of the pressure center in the back of the head section; the starting point of the neck support is the position where the pillow surface begins to form obvious support for the user's neck, which can be determined by the position in the neck support section where the pressure exceeds a preset threshold; the individualized spacing parameter is the distance between the above two positions along the front-back direction of the pillow or the direction of the pillow surface arc.

[0133] For example, if the main support point of the occipital bone and the starting point of the neck brace are 6 sampling intervals apart on the pressure array, and the adjacent sampling interval is 8 mm, then the individualized spacing parameter is 48 mm.

[0134] In one embodiment, the ratio of the individualized spacing parameter to the pre-statistically obtained group average spacing parameter is used as the template vertical scaling factor and written into the user configuration file.

[0135] S337: Correct the initial target pressure value in the target support curve template based on the reference attitude data and the individualized spacing parameters.

[0136] In this embodiment, the modification is to add an offset or multiply a scaling factor to the reference pressure of each cavity of the template curve at the initial moment or in the steady state segment; the deviation of the reference attitude data relative to the typical pitch angle of the group is mapped to a fine adjustment of the opposite sign of the occipital cavity and the top-of-the-head limiting cavity.

[0137] For example, when the pitch angle is 3 degrees greater than the group average, the initial target pressure in the occipital cavity increases by 0.3 kPa and the initial target pressure in the top of the head decreases by 0.2 kPa.

[0138] In one embodiment, the correction amount, after being trimmed by comfort and safety boundaries, is written to random access memory for use during the current sleep cycle, and simultaneously mirrored to flash memory as an individualized initial value.

[0139] S338: The corrected initial target pressure value is determined as the individualized initial value, which is used as a reference parameter in the calculation when solving the collaborative target pressure value and maintaining the cervical spine sagittal curvature in the preset physiological target range.

[0140] In this embodiment, the individualized initial value is used in the regularization term of step S333 and the projection of the initial value in step S33; it serves as a safe default cavity pressure reference during abnormal degradation recovery.

[0141] For example, the individualized initial values ​​of the three cavities are written to the user configuration sector and the checksum is calculated; the checksum is loaded at startup, and if it fails, it falls back to the model's default table.

[0142] In one embodiment, the user can reset the individualized calibration on the terminal, the device clears the old individualized initial values ​​and prompts the user to re-execute steps S335 to S338.

[0143] In one embodiment, reference Figure 9 After step S34, the iterative correction steps may specifically include steps S345 to S348, which will be described in detail below: S345: Reacquire the adjusted multi-region pressure distribution data and attitude data.

[0144] In this embodiment of the application, after the valve pump executes the target support pressure command, step S345 reads the pressure frame and inertial attitude data again after a preset waiting time to obtain a sample reflecting the latest contact and attitude; the preset waiting time must be greater than the main part of the gas path response time constant.

[0145] For example, the preset waiting time can be 500 milliseconds, and a whole frame of pressure and 200 Hz inertial short window samples can be collected simultaneously.

[0146] In one embodiment, if a body movement event is detected within a preset waiting period, the waiting time can be extended or the sample can be discarded to avoid misjudgment based on the transition state in the closed loop.

[0147] S346: Calculate the deviation between the adjusted sagittal curvature index and the preset physiological target interval based on the re-collected data.

[0148] In this embodiment, the index corresponding to the sagittal curvature of the cervical spine is consistent with the biomechanical model or regressor; the deviation is the signed distance of the index corresponding to the sagittal curvature of the cervical spine from the preset physiological target interval, or the error norm to the target reference value.

[0149] For example, the normalized target range for the cervical spine sagittal curvature index is [0.85, 0.95]. When the current value is 0.80, it is lower than the lower limit of the range of 0.85. The deviation can be 0.05, which is used to indicate that the curvature index needs to be corrected to the target range.

[0150] In one embodiment, the deviation is simultaneously low-pass filtered to avoid closed-loop oscillation caused by single noise; when the deviation exceeds the limit, data quality monitoring is triggered to accelerate the evaluation.

[0151] S347: Decompose the deviation into occipital support deviation, neck brace support deviation and lateral limiting deviation, and map them into secondary compensation amounts for the corresponding airbag cavities.

[0152] In this embodiment, the scalar or low-dimensional deviation is mapped to the functional basis vector through weighted least squares or pseudo-inverse, and then multiplied by the cavity mapping to obtain the secondary compensation quantity; the decomposition makes the three types of functional deviations of the back of the head, neck brace and side correspond to the physical cavity.

[0153] For example, when solving for the secondary compensation amount, the secondary compensation vector ΔP2 can be used as the optimization variable to construct the following secondary compensation objective function: F2(ΔP2)=||A·ΔP2-e_vec||^2.

[0154] The constraint is: -ΔP2_max≤ΔP2≤ΔP2_max.

[0155] Where F2(ΔP2) represents the objective function of the quadratic compensation to be minimized, A represents the sensitivity matrix between curvature deviation and the secondary compensation amount of the airbag, A·ΔP2 represents the curvature deviation correction amount predicted by the secondary compensation amount, e_vec represents the error vector constructed from the curvature deviation amount, ΔP2 represents the secondary compensation amount vector of each airbag cavity, ΔP2_max represents the maximum allowable amplitude of the secondary compensation amount, and -ΔP2_max≤ΔP2≤ΔP2_max indicates that each compensation component in ΔP2 is within the corresponding allowable range. By solving for ΔP2 that minimizes F2(ΔP2) under the above constraints, the secondary compensation amount of each airbag cavity can be obtained.

[0156] In one embodiment, the neck support component is allowed a larger upper limit in the secondary compensation amount to prioritize the correction of curvature, and the upper limit of the lateral component is relaxed when sleeping on the side and tightened when sleeping on the back, depending on the sleeping position.

[0157] S348: The target support pressure command is iteratively corrected based on the secondary compensation amount until the deviation is lower than the preset tolerance threshold.

[0158] In this embodiment, the iterative correction loop executes steps S345 to S347, where the target support pressure command is superimposed with a secondary compensation amount in each round and then issued again; the preset tolerance threshold is the stopping condition; the maximum number of iterations N_max represents the preset maximum number of rounds that the closed-loop correction is allowed to execute, which is used to avoid continuous iteration under abnormal conditions.

[0159] For example, the preset tolerance threshold can be set to 2% relative error of the index corresponding to the sagittal curvature of the cervical spine, and the maximum number of iterations can be set to 5; when a sudden change in sleeping posture or abnormal degradation control is triggered, the loop is interrupted and the process switches to a sleeping posture transition process.

[0160] In one embodiment, a minimum time interval is inserted between iterations to wait for the gas path to stabilize; if the deviation increases for two consecutive rounds, the step size of the secondary compensation is reduced or the process is stopped and a divergence flag is recorded.

[0161] In one embodiment, reference Figure 10 During the execution of steps S345 to S348, the anomaly degradation control steps may specifically include steps S349 to S3412; wherein, the data quality judgment acts on the iterative correction process, and the stress level adopted by the safe default mode corresponds to the individualized initial value. This will be described in detail below: S349: Calculate data quality indicators for the re-acquired multi-region pressure distribution data and attitude data, wherein the data quality indicators include at least pressure missing rate, attitude drift amount and contact stability.

[0162] In this embodiment, the pressure loss rate is the number of invalid or saturated measurement points divided by the total number of pressure measurement points; the attitude drift is the norm of the online estimated zero bias relative to the factory nominal zero bias; and the contact stability is negatively correlated with the total pressure variation coefficient within the synchronous analysis data window. A higher pressure loss rate and attitude drift indicate a higher degree of anomaly, while lower contact stability indicates poorer contact stability.

[0163] For example, the pressure loss rate can be determined by the ratio of the number of invalid or saturated pressure points to the total number of pressure points; the attitude drift can be determined by the norm of the difference between the online estimated zero bias and the nominal zero bias; and the contact stability can be determined based on the coefficient of variation of the total pressure within the synchronous analysis data window. The larger the coefficient of variation, the lower the contact stability.

[0164] In one embodiment, the hardware channel saturation count and row scan failure rate are incorporated into the overall quality score Q, where Q represents a normalized index used to comprehensively evaluate data quality. In another implementation, a larger Q indicates worse data quality, and when the overall quality threshold is exceeded, abnormal degradation control is triggered.

[0165] S3410: When any of the data quality indicators exceeds the preset abnormal threshold, the iterative correction is paused and the current airbag cavity pressure is maintained.

[0166] In this embodiment, pausing iterative correction means stopping subsequent rounds of step S348, maintaining the target support pressure command and valve / pump status unchanged, or only maintaining inner loop pressure stabilization; and recording alarm codes for terminal display.

[0167] For example, when the stress loss rate is greater than 30%, the loop is frozen and only events are recorded.

[0168] In one embodiment, the sleeping posture transition initiated in step S341 is still allowed to complete during the pause to avoid discomfort caused by pressure suspension midway, with the priority defined by the state machine.

[0169] S3411: If the abnormality persists for more than the preset duration, each airbag cavity will be switched to the safe default support mode, and the pressure setting of the safe default support mode corresponds to the individualized initial value.

[0170] In this embodiment, the preset duration represents the time threshold for allowing data quality anomalies to persist; when the anomaly duration exceeds the preset duration, the current acquisition or execution link is determined to be in a continuous abnormal state. The security default support mode sets the target support pressure command to an individualized initial value or a conservative value slightly lower than the individualized initial value to ensure predictable support.

[0171] For example, the preset duration can be 90 seconds, and if the timeout occurs, the three-chamber target will be set as the individualized initial value vector.

[0172] In one embodiment, steps S345 to S348 are disabled when switching the default safety support mode until the user confirms maintenance or the pressure loss rate is restored, to prevent continued integral correction in a fault state.

[0173] S3412: After the data quality indicators are restored to the normal range, resume the determination of sleeping posture type, the generation of target support pressure command and the closed-loop correction.

[0174] In this embodiment, the normal range is the acceptable range where each indicator is below the recovery threshold; after recovery, steps S32 to S34 and steps S345 to S348 are allowed to run according to the original logic again.

[0175] For example, if the stress loss rate is below 5% for 10 consecutive synchronous analysis data windows and the attitude drift returns to within the window, then the downgrade is lifted.

[0176] In one embodiment, during recovery, an individualized calibration quick check is performed first, or the user is prompted to confirm that the pillowcase is flat, before continuing nighttime adjustment.

[0177] In one embodiment, reference Figure 11 The intelligent pillow control system of this application embodiment may specifically include: The distributed pressure sensing module 120 is used to collect pressure distribution data of multiple regions on the pillow surface through a distributed pressure array.

[0178] In this embodiment, the distributed pressure sensing module acquires data related to the pressure frame P_Frame, including a thin-film piezoresistive array, a multiplexer, an analog-to-digital converter, and scanning firmware; the output is reduced to the composite pressure of each partition by a partition mask table for use by subsequent modules.

[0179] For example, the distributed pressure sensing module 120 scans at 20 Hz to 50 Hz and writes the original frame to a double buffer; it supports temperature-compensated lookup tables.

[0180] In one embodiment, the self-test of the distributed pressure sensing module 120 includes open-circuit and short-circuit detection and noise floor test. If the test fails, normal control is prohibited during the startup phase.

[0181] The inertial measurement module 121 is used to acquire attitude data representing the head relative to the occipital coordinate system through at least one inertial measurement unit.

[0182] In this embodiment, the inertial measurement module acquires inertial attitude data IMU_Pose and outputs the attitude and angular velocity of the bolster system via the installed attitude matrix R_IMU2Pillow; the communication interface can be a serial peripheral interface or an integrated circuit bus.

[0183] For example, the six-axis microelectromechanical system unit outputs at 200 Hz, and the firmware performs installation matrix multiplication and zero-bias compensation.

[0184] In one embodiment, the module supports switching between low-noise mode and wide-range mode, with low-noise mode as the default during sleep.

[0185] The posture reconstruction module 122 is used to align the multi-region pressure distribution data and the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system, and output the relative posture parameters between the head rigid body and the cervical spine reference segment. The joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds.

[0186] In this embodiment, the attitude reconstruction module receives the pressure frame P_Frame and the inertial attitude data IMU_Pose, forms a synchronous analysis data window, and extracts joint observations such as the pressure center of gravity, pressure principal axis direction, left and right load difference, pitch angle, roll angle and angular velocity components in the pillow coordinate system, and outputs the relative attitude parameter RelPose; internally, extended Kalman filtering or complementary filtering and pressure feature extraction can be performed.

[0187] For example, the posture reconstruction module reads the synchronous analysis data window and outputs the relative posture parameters to the sleeping posture determination module, which takes less time than the duration of a single window to ensure real-time performance.

[0188] In one embodiment, the attitude reconstruction module performs filtering and matrix operations with a fixed number of points when resources are limited, thereby reducing power consumption.

[0189] The sleeping posture determination module 123 is used to input the relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data into a preset human cervical spine biomechanical model and obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone positions.

[0190] In this embodiment, the sleeping posture determination module receives the relative posture parameters output by the posture reconstruction module and the zoned pressure features output by the distributed pressure sensing module, constructs the cervical spine posture feature group and the pillow surface zoned pressure feature vector, inputs the two into the human cervical spine biomechanical model, calculates the posterior extension torque, lateral load degree and contact stability index, and outputs the sleeping posture type and the corresponding cervical spine support requirement level.

[0191] The support decision module 124 is used to generate target support pressure commands for no less than three independent airbag cavities in the multi-cavity airbag system based on the sleeping posture type determination result and the relative posture parameters. Each airbag cavity and the corresponding acquisition partition of the distributed pressure array have a preset correspondence in spatial arrangement. The target support pressure command is used to make the support forces of each airbag cavity work together to maintain the sagittal curvature of the cervical spine within the preset physiological target range.

[0192] In this embodiment, the support decision module receives the sleeping posture type, cervical spine support requirement level and relative posture parameters, and combines the cervical spine sagittal plane curvature deviation, individualized initial values ​​and spatial correspondence to generate the collaborative target pressure value and target support pressure command for each airbag cavity.

[0193] The airbag drive module 125 is used to perform independent inflation and deflation regulation of the multi-chamber airbag system according to the target support pressure command.

[0194] In this embodiment, the airbag driving module receives the target support pressure command and converts it into the driving timing of the inflation valve, deflation valve and micro air pump corresponding to each airbag cavity; when the sleeping position changes, the airbag driving module can perform progressive inflation and deflation according to the transition target pressure sequence.

[0195] See Figure 12 In this embodiment, the control and drive circuit 800 built into or integrated with the smart pillow may include a processor 801, a memory 802, a communication interface 803, a pressure array interface 804, an inertial measurement interface 805, and a valve pump drive interface 806. The memory 802 stores firmware and calibration parameters. When the processor 801 executes the firmware, it implements the control method of this embodiment. The pressure array interface 804 is connected to the analog-to-digital conversion front-end or line scan controller of the distributed pressure array. The inertial measurement interface 805 is connected to the digital interface of the inertial measurement unit. The valve pump drive interface 806 is connected to a micro air pump and a solenoid valve via a power stage. The communication interface 803 can be used for firmware upgrades, debugging, or synchronizing summary data with an external terminal. In the printed circuit board layout, the valve pump drive power devices and the processor should maintain a complete planar reference and be spaced apart. The pressure array simulation traces should be kept away from the high-current loop of the pump to reduce electromagnetic interference coupling during joint estimation and control closed-loop execution.

[0196] In some embodiments, the software architecture can be divided into a driver layer, a middleware layer, and an application layer. The driver layer encapsulates pressure scanning, serial bus access to the inertial measurement unit, general input / output control of valves and pumps, and analog-to-digital conversion for reading pressure sensors, etc.; the middleware layer provides time synchronization services, a circular buffer, non-volatile parameter storage, and logs; the application layer implements the logic corresponding to the distributed pressure sensing module, inertial measurement module, attitude reconstruction module, sleeping posture determination module, support decision module, and airbag driving module, and runs a state machine, which includes at least initialization, individualized calibration, normal control, closed-loop correction, abnormal degradation, and low-power shutdown states. The factory calibration process may include applying known surface loads to each pressure acquisition zone on a standard load stage to fit linear coefficients and temperature compensation coefficients; recording the output of the inertial measurement unit and the true attitude on a standard tilt fixture to solve for the installation matrix and zero bias; and sequentially inflating the multi-chamber airbag to multiple calibration points to fit the elements in the spatial corresponding matrix. Calibration data can be written to configuration sectors with verification or signature and bound to hardware channel numbers to reduce the risk of miscontrol due to incorrect gas path connection or incorrect partition mapping.

[0197] In some embodiments, the parameters of the human cervical spine biomechanical model can be stored in groups according to age, gender, and body type. After the user selects a body type on the mobile terminal, the controller loads the corresponding parameter set, ensuring that the preset physiological target interval and the default value of the template curve are consistent with the group statistics, and then converges to the user's own value through individualized calibration. The target support curve template can also be divided into a daytime nap mode and a nighttime long sleep mode: the former has a shorter transition time window and a wider rate of change constraint; the latter has a longer transition time window and a stricter rate of change constraint. Communication between the smart pillow and external devices can be achieved using Bluetooth Low Energy, Wi-Fi, or Near Field Communication; in asynchronous scenarios, periodic broadcast summaries can be used, and a long connection can be established only when historical window data needs to be uploaded.

[0198] In some embodiments, the miniature air pump can be arranged in the base air duct or anechoic chamber to facilitate heat dissipation and noise reduction; the sensing grid of the distributed pressure array can be a rectangular or a variant layout conforming to the neck curve, and the mapping relationship between each grid unit and the partition mask can be uniquely determined in the factory calibration file. When the mobile terminal synchronizes individualized calibration results, preset physiological target intervals, or sleep posture statistical summaries to the pillow, the message should preferably be transmitted in encrypted form and accompanied by a device-side random number to resist replay; in the local real-time control mode only, the communication interface can be mainly used for maintenance diagnosis and firmware upgrades.

[0199] In some embodiments, the production line can pressurize each chamber on a pressure-holding fixture and record the pressurization time constant and steady-state leakage rate. Defective products can have a locking code written into a non-volatile tag to prevent them from entering normal control mode. Maintenance personnel can export statistical summaries and state machine jump records from several recent synchronous analysis data windows through the service interface to differentiate between pressure sampling link anomalies, gas path micro-blockages, and model parameter mismatches. The power supply side can be configured with overcharge / over-discharge, temperature, and short-circuit protection, and the firmware monitors the pump's continuous operating time and valve opening / closing frequency, initiating cooling or flow-limiting strategies when thresholds are exceeded.

[0200] It should be noted that the numerical ranges, device selections, formula symbols, matrix forms, threshold names, airbag quantity and structural examples in the above embodiments are only used to help those skilled in the art understand the technical solutions of this application, and do not constitute an implicit limitation on the scope of protection of this application; the symbols such as pressure frame P_Frame, inertial attitude data IMU_Pose, spatial correspondence matrix Mapping, and cervical spine sagittal curvature correspondence index CurvIndex are only used to illustrate the data processing or control calculation process, and do not limit their variable names, data structures or storage formats in specific products. Multiple equivalent expressions may appear for the same technical feature without contradicting the solutions already described in this application; the interpretation should be based on the technical solutions described in the application documents.

[0201] When the device is first started or a major function is updated, it can display a summary of data processing rules to the user and obtain explicit consent. In applicable jurisdictions, it can provide data portability and deletion interfaces as required by law. Deletion operations should simultaneously clear individualized initial values ​​and recoverable diagnostic log summaries in non-volatile memory, and retain anonymized statistical information.

[0202] In industrial applications, this application can be used for smart pillows for home use, neck pillows for rehabilitation and nursing, and smart bedding in hotels. By linking with mobile applications, users can view sleeping posture statistics and cervical curvature trends, and adjust preset physiological target ranges with compliance prompts. The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A smart pillow control method for sleep posture recognition and neck support optimization, characterized in that, Includes the following steps: Acquire multi-region pressure distribution data output by the distributed pressure array in the pillow support area, and attitude data of the head relative to the pillow coordinate system output by at least one inertial measurement unit. Align the multi-region pressure distribution data and the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system to obtain the relative posture parameters between the head rigid body and the cervical spine reference segment; wherein, the joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds; The relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data are input into the preset human cervical spine biomechanical model to obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone. Based on the sleeping posture type determination result and the relative posture parameters, a target support pressure command is generated for no less than three independent airbag cavities in the multi-cavity airbag system. Each airbag cavity and the corresponding acquisition partition of the distributed pressure array have a preset correspondence in spatial arrangement. The target support pressure command is used to make the support force of each airbag cavity work together and maintain the sagittal curvature of the cervical spine within the preset physiological target range. as well as The multi-chamber airbag system is driven to independently inflate and deflate each airbag chamber according to the target support pressure command.

2. The intelligent pillow control method according to claim 1, characterized in that, The acquisition of the multi-region pressure distribution data and the attitude data includes the following steps: The distributed pressure array is divided into at least three pressure acquisition zones: the top of the head, the back of the head, the neck brace, and the lateral restraint zone. The effective support area is extracted based on the contact start time of each pressure acquisition zone, and invalid pressure points below the contact threshold are removed. The at least one inertial measurement unit is zero-biased calibrated using the gravity component in a static state to obtain an initial attitude reference relative to the pillow coordinate system. The pressure data stream and attitude data stream are timestamped according to a unified sampling period to form a synchronous analysis data window. The synchronous analysis data window serves as the input data window for aligning the multi-region pressure distribution data and the attitude data on the time axis.

3. The intelligent pillow control method according to claim 2, characterized in that, The process of aligning the multi-region pressure distribution data and the attitude data on the time axis, performing coordinate transformation and joint estimation in the pillow coordinate system to obtain the relative attitude parameters includes the following steps: Based on the pressure distribution of the effective support area within the synchronous analysis data window, the pressure center of gravity, the direction of the main pressure axis, and the difference between the left and right loads are obtained. The attitude data is decomposed into pitch angle, roll angle and angular velocity components, and then transformed into the pillow coordinate system; Using the pressure center of gravity, the pressure axis direction, the left and right load difference, the pitch angle, the roll angle, and the angular velocity components as joint observations, extended Kalman filtering or complementary filtering is used to estimate the rigid body attitude of the head. By combining the head rigid body posture with the pressure change in the cervical collar area, the relative flexion-extension angle, lateral flexion angle, and axial deflection between the head rigid body and the cervical reference segment are calculated. The relative flexion-extension angle, lateral flexion angle, and axial deflection are used to characterize the relative posture parameters.

4. The intelligent pillow control method according to claim 3, characterized in that, The step of inputting the relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data into a preset human cervical spine biomechanical model to obtain the current sleeping posture type determination result includes the following steps: A set of cervical spine posture characteristics is constructed based on the relative flexion-extension angle, lateral flexion angle, and axial deflection. Construct a pressure feature vector for each pressure acquisition zone based on the load percentage, contact area, and pressure gradient of each zone. The cervical spine posture feature group and the occipital surface zone pressure feature vector are input into the human cervical spine biomechanical model to calculate the posterior extension torque, lateral load degree and contact stability index. The current sleeping posture type and the corresponding cervical spine support requirement level are determined based on the posterior neck extension torque, the degree of lateral load, and the contact stability index.

5. The intelligent pillow control method according to claim 4, characterized in that, The step of generating a target support pressure command for at least three independent airbag cavities in the multi-cavity airbag system based on the sleeping posture type determination result and the relative posture parameters includes the following steps: Based on the sleeping posture type and the cervical spine support requirement level, call the target support curve template corresponding to the top of the head area, the back of the head area, and the neck support area; The pressure compensation increment of each air bladder cavity is calculated based on the relative flexion-extension angle, lateral flexion angle and axial deflection, wherein the compensation increment of the air bladder corresponding to the cervical collar area is used to preferentially correct the sagittal curvature deviation of the cervical spine. Based on the preset spatial correspondence between each pressure acquisition zone and each airbag cavity, the cooperative target pressure value of no less than three independent airbag cavities is solved. Apply upper and lower pressure limits and change rate constraints to the coordinated target pressure value to form the target support pressure command.

6. The intelligent pillow control method according to claim 5, characterized in that, The step of driving the multi-chamber airbag system to independently inflate and deflate each airbag chamber according to the target support pressure command includes the following steps: Detect whether the sleeping posture type changes within two adjacent synchronous analysis data windows, and calculate the confidence level of the change; When the switching confidence level is higher than a preset threshold, the collaborative target pressure values ​​before and after the switching are interpolated in segments within a preset transition time window; Based on the segmented interpolation results, the inflation valve, deflation valve, and micro air pump of each airbag cavity are controlled to perform progressive inflation and deflation. After the transition time window ends, the updated target support pressure command is locked to suppress sleep disturbances caused by sudden changes in support force.

7. The intelligent pillow control method according to claim 5, characterized in that, Before calling the target support curve template corresponding to the top of the head, back of the head, and neck brace area according to the sleeping posture type and the cervical spine support requirement level, an individualized calibration step is performed; the correction result output by the individualized calibration step is used as the reference for calling the target support curve template and subsequent adjustment, and includes the following steps: Baseline pressure distribution and baseline posture data were collected while the user remained in a resting supine position. The individualized spacing parameters between the main support point of the occipital bone and the starting point of the neck brace are determined based on the baseline pressure distribution. The initial target pressure value in the target support curve template is corrected based on the baseline attitude data and the individualized spacing parameters. The corrected initial target pressure value is determined as the individualized initial value, which is used as a reference parameter in the calculation when solving the collaborative target pressure value and maintaining the cervical spine sagittal curvature within the preset physiological target range.

8. The intelligent pillow control method according to claim 6, characterized in that, After the multi-chamber airbag system is driven to independently inflate and deflate each airbag cavity according to the target support pressure command, a command correction step is further included. This command correction step is used to provide feedback correction to the target support pressure command based on the preset physiological target range, and includes the following steps: The adjusted multi-region pressure distribution data and attitude data were reacquired. The deviation between the adjusted sagittal curvature index of the cervical spine and the preset physiological target interval is calculated based on the re-collected data. The deviation is decomposed into occipital support deviation, neck brace support deviation, and lateral limiting deviation, and mapped to the secondary compensation amount of the corresponding airbag cavity respectively. The target support pressure command is iteratively corrected based on the secondary compensation amount until the deviation is lower than the preset tolerance threshold.

9. The intelligent pillow control method according to claim 7 or 8, characterized in that, The execution of the instruction correction step also includes an exception degradation control step, which includes the following steps: Calculate data quality indicators for the re-acquired multi-region pressure distribution data and attitude data. The data quality indicators include at least pressure missing rate, attitude drift, and contact stability. When any of the data quality indicators exceeds a preset anomaly threshold, the iterative correction is paused and the current airbag cavity pressure is maintained. If the abnormality persists for more than the preset duration, each of the airbag cavities will be switched to the safe default support mode, and the pressure setting of the safe default support mode corresponds to the individualized initial value. After the data quality indicators are restored to the normal range, the sleep posture type determination, the generation of the target support pressure command, and the closed-loop correction are resumed.

10. A smart pillow control system for sleep posture recognition and neck support optimization, characterized in that, include: The distributed pressure sensing module is used to collect pressure distribution data of multiple regions on the pillow surface through a distributed pressure array; An inertial measurement module is used to acquire attitude data representing the head relative to the occipital coordinate system through at least one inertial measurement unit; The posture reconstruction module is used to align the multi-region pressure distribution data with the posture data on the time axis, perform coordinate transformation and joint estimation in the pillow coordinate system, and output the relative posture parameters between the head rigid body and the cervical spine reference segment. The joint estimation is different from sleeping posture classification based solely on pressure distribution images or pressure thresholds. The sleeping posture determination module is used to input the relative posture parameters and the pillow surface partition pressure feature vector extracted from the multi-region pressure distribution data into a preset human cervical spine biomechanical model and obtain the current sleeping posture type determination result. The sleeping posture type includes at least one of supine, lateral, and prone. The support decision module is used to generate target support pressure commands for no less than three independent airbag cavities in the multi-cavity airbag system based on the sleeping posture type determination result and the relative posture parameters. Each airbag cavity and the corresponding acquisition partition of the distributed pressure array have a preset correspondence in spatial arrangement. The target support pressure command is used to make the support forces of each airbag cavity work together to maintain the sagittal curvature of the cervical spine within the preset physiological target range. as well as The airbag drive module is used to perform independent inflation and deflation regulation of the multi-chamber airbag system according to the target support pressure command.

Images