An intelligent sleeping posture optimization method and system based on pressure sensing and human body reconstruction

Abstract

This invention provides an intelligent sleeping posture optimization method and system based on pressure sensing and human body reconstruction, belonging to the field of smart home and medical health technology. The invention collects static pressure distribution images through a matrix of pressure sensors arranged on the mattress surface and inputs them into a trained parametric human body model fitter to reconstruct a three-dimensional human body model of the user. Using the height of each motor supporting the mattress surface as optimization variables, a multi-objective optimization function integrating pressure comfort and joint posture comfort is constructed. Under the conditions of satisfying motor travel, height difference, and human body balance constraints, an optimization algorithm is used to solve for the optimal combination of motor heights. Finally, the motor array is controlled to form a support surface that matches the user's body shape and posture. This invention, employing the aforementioned intelligent sleeping posture optimization method and system based on pressure sensing and human body reconstruction, achieves a breakthrough from passive adjustment to active predictive optimization, and from single pressure control to pressure-posture collaborative optimization.

Description

Technical Field

[0001] This invention relates to the fields of smart home and medical health technology, and in particular to a smart sleeping posture optimization method and system based on pressure sensing and human body reconstruction. Background Technology

[0002] The comfort of lying posture is closely related to human health. Poor posture support can lead to excessive local pressure and abnormal spinal curvature, resulting in problems such as bedsores and lower back pain. Existing smart bed adjustment technologies mainly focus on pressure distribution and posture adaptation, but they all have limitations in terms of systematicness, proactivity, and personalization, as analyzed in detail below:

[0003] First, while threshold-triggered passive feedback systems can alleviate high-pressure points in real time, they cannot achieve preventative adjustment and only aim to reduce local pressure, neglecting overall posture and spinal health. Second, while matching systems based on predefined posture patterns respond quickly, their classification is coarse and their support schemes are rigid, making it difficult to adapt to individual body shape differences and non-standard posture variations. Third, while non-contact systems based on 3D contour reconstruction can acquire body surface shape, they cannot perceive the true posture of internal bones and joints, easily leading to inconsistencies between spinal support and physiological curvature. Fourth, while academic research based on simplified mechanical models has introduced the idea of ​​joint posture optimization, the models are overly simplified, not coupled with specific users, and rely on high-cost sensors, limiting their engineering practicality. The common limitation of these methods is that none of them have established a model that can accurately invert the coupling relationship between a user's personalized internal posture and external pressure from easily accessible data, thus failing to achieve systematic, proactive, and adaptive adjustment centered on human biomechanical comfort.

[0004] In summary, the root cause of the shortcomings of existing technologies lies in the failure to establish an accurate model that can deduce the coupling relationship between the user's personalized internal skeletal posture and body surface pressure distribution from easily accessible data at low cost. Consequently, it is also impossible to achieve synergistic optimization with the goal of whole-body systemic comfort (taking into account both pressure balance and physiological curvature). Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent sleeping posture optimization method and system based on pressure perception and human body reconstruction, so as to solve the problems of existing technologies that cannot take into account both pressure distribution and physiological posture, are passive in adjustment and lack personalized adaptation.

[0006] To achieve the above objectives, this invention provides an intelligent sleeping posture optimization method based on pressure sensing and human body reconstruction, comprising the following steps:

[0007] Step S1: Collect static pressure distribution data of the user in a static lying position by using a pressure sensor matrix arranged on the mattress surface, and generate a static pressure distribution image;

[0008] Step S2: Input the static pressure distribution image into the pre-trained parametric human body model fitter to reconstruct a three-dimensional human body model containing the user's key joint angles and body shape features.

[0009] Step S3: Using the height of each motor in the adjustable support bed as the optimization variable, construct a multi-objective optimization function that integrates pressure comfort and joint posture comfort;

[0010] Step S4: Under the premise of satisfying the preset constraints, use the optimization algorithm to solve the multi-objective optimization function and obtain the optimal combination of motor heights that minimizes the objective function value;

[0011] Step S5: Adjust the motor array according to the optimal height combination to form a support surface that matches the user's current body shape and posture.

[0012] Preferably, in step S2, the parametric human body model fitter is constructed based on the SMPL model and regresses the user's body shape parameters and posture parameters from the stress distribution image through a convolutional neural network.

[0013] Preferably, the parametric human model fitter is trained by minimizing a composite loss function, which includes body shape parameter loss, pose parameter loss, 3D joint point loss, and mesh vertex loss.

[0014] Preferably, in step S3, the multi-objective optimization function is:

[0015] ;

[0016] ;

[0017] ;

[0018] in, Represents a multi-objective optimization function. This indicates the pressure comfort level. This indicates the comfort level of joint posture. Indicates peak pressure. Indicates the pressure gradient. Indicates the pressure uniformity index. Indicates the current angle of the joint. This represents the median comfortable joint angle. Indicates the range of comfortable joint angles. , , , , All of these represent weighting coefficients.

[0019] Preferably, in step S4, the constraints include:

[0020] Height and travel constraints for each motor;

[0021] Height difference constraint between adjacent motors;

[0022] The projection of the human body's center of gravity is located inside the mattress.

[0023] Preferably, in step S4, the optimization algorithm is the particle swarm optimization algorithm.

[0024] Preferably, the parameter settings of the particle swarm optimization algorithm include: the particle encoding dimension is equal to the number of motors, the particle population size is 30, the maximum number of iterations is 100, and the inertia weight, individual cognitive weight and group social weight in the velocity update are 0.5, 1.5 and 1.5, respectively.

[0025] This invention also provides an intelligent sleeping posture optimization system based on pressure sensing and human body reconstruction, used to execute the intelligent sleeping posture optimization method based on pressure sensing and human body reconstruction as described above, including:

[0026] The sensing input layer is used to acquire static pressure distribution images of the user on the mattress;

[0027] The digital modeling layer is used to fit a parametric human body model based on stress images and output joint angles and body shape features.

[0028] The intelligent decision-making layer is used to construct and solve a multi-objective optimization problem with motor height as the optimization variable, and generate motor height instructions.

[0029] The physical execution layer is used to control the movement of the motor array according to altitude commands.

[0030] Preferably, the sensing input layer is configured as a flexible thin-film pressure sensor pad, the size of which is the effective sensing area. The spatial resolution is 36mm.

[0031] Therefore, the present invention employs the above-mentioned intelligent sleeping posture optimization method and system based on pressure sensing and human body reconstruction, and the beneficial technical effects are as follows:

[0032] (1) By reconstructing the user's personalized three-dimensional human body model in real time from a single frame static pressure map and performing multi-objective optimization prediction on this basis, the system can actively adjust the support surface before body pressure accumulation and posture deviation occur, overcoming the limitation of existing threshold-triggered systems that can only intervene after the fact.

[0033] (2) This invention incorporates pressure uniformity, pressure gradient and the comfort angle range of key joints such as spine, neck, shoulder and hip into the optimization objective function, and solves it collaboratively through a unified algorithm. This overcomes the shortcomings of existing technologies that only focus on a single indicator or cannot perceive the true posture of the internal skeleton, and achieves systematic comfort that conforms to ergonomics.

[0034] (3) While ensuring personalized adaptation accuracy, it has good engineering feasibility and practical potential. The modeling method based on widely used parametric human body models (such as SMPL) and actual pressure data is adopted. This avoids the dependence on high-cost depth cameras, easily interfered non-contact sensing or dense flexible sensor networks, and solves the problem of simplified mechanical models being divorced from real user data. This makes the system easier to deploy, more stable to operate and suitable for real sleep environments. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the architecture of an intelligent sleeping posture optimization system based on pressure sensing and human body reconstruction according to the present invention.

[0036] Figure 2 This is a flowchart of an intelligent sleeping posture optimization method based on pressure sensing and human body reconstruction according to the present invention. Detailed Implementation

[0037] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0039] Example 1

[0040] This embodiment provides an intelligent sleeping posture optimization method based on pressure sensing and human body reconstruction, the process of which is as follows: Figure 2 As shown, it specifically includes:

[0041] Step S1: Static pressure distribution data acquisition.

[0042] A flexible thin-film pressure sensor matrix, laid on the mattress surface, collects real-time data on the user's body surface pressure distribution while the user is lying still, and generates a static pressure distribution image. The overall size of the sensor is... The effective sensing area is The spatial resolution is 36 mm. The acquired pressure image resolution is... A pixel is a sensing unit in a sensor array, and its grayscale value reflects the pressure at that location.

[0043] Step S2: Reconstruction of the three-dimensional human body model.

[0044] The static pressure distribution image obtained in step S1 is input into a pre-trained parametric human body model fitter to reconstruct the user's 3D human body model. In this embodiment, the fitter is constructed based on the SMPL (Skinned Multi-PersonLinear) parametric human body model. This model defines a standard human body mesh using 6890 vertices and 13776 triangular faces, and controls the human body shape and joint posture using a 10-dimensional body shape parameter vector β and a 24-dimensional posture parameter vector θ, respectively.

[0045] The fitter employs a convolutional neural network (CNN) structure to achieve end-to-end regression from stress images to SMPL parameters. The CNN structure includes: 1) A feature encoder: composed of alternating stacks of multiple convolutional layers, batch normalization layers, and activation function layers, used to extract multi-level features from stress images. For example, four convolutional modules can be used, each containing convolution, batch normalization, and ReLU activation operations, and downsampling is performed through convolutional or pooling layers with a stride of 2, ultimately encoding the image into a one-dimensional feature vector. 2) A fully connected regressor: composed of several fully connected layers, mapping the feature vector output by the feature encoder to the target parameter space. The final output layer of the network has 82 neurons, corresponding to 10-dimensional body shape parameters and 72-dimensional pose parameters.

[0046] The training process is as follows:

[0047] 1. Data Preprocessing: To ensure accurate and consistent human feature parameters when using pressure sensors of different specifications, the image data is scaled down to a realistic scale using spatial resolution parameters (i.e., from...). Transform into ).

[0048] 2. Loss Function Design: A composite supervised loss function is adopted. This ensures that the model can accurately fit both the parameter space and the vertex space.

[0049] ;

[0050] in, and To balance the hyperparameters, they were experimentally set to 0.001 and 5.0 respectively. The definitions of each loss component are as follows:

[0051] Body size parameter loss ( ): Calculate predicted body shape parameters Compared with the true value The mean square error between them ensures that the model accurately captures the macroscopic shape of the human body.

[0052] ;

[0053] in, Index representing body shape parameters.

[0054] Attitude parameter loss ( ): Calculate the predicted attitude parameters Compared with the true value The mean square error between them.

[0055] ;

[0056] in, Indicates the index of the predicted attitude parameters.

[0057] 3D joint loss ( First, align the two point sets with the pelvic joint as the center to eliminate the influence of global translation, and then calculate the average Euclidean distance of all relevant nodes.

[0058] ;

[0059] in, and They represent the first The predicted coordinates and actual coordinates of each joint.

[0060] Mesh vertex loss ( ): Calculate the predicted 3D mesh for all vertices With the true vertex The mean absolute error between the two ensures the accuracy of the reconstructed surface.

[0061] ;

[0062] in, Represents the vertex index.

[0063] 3. Training data: Synthetic datasets (PressurePose, BodyPressureSD) and real datasets (SLP) are used together to improve the model's generalization ability and robustness on stress data from different sources.

[0064] This process outputs the following two key pieces of information:

[0065] Joint information: parameters that describe the rotational state of each joint.

[0066] Pressure distribution information: Based on the vertex data of the reconstructed 3D human body model, the system performs the following steps to establish a simplified physical model for rapid pressure estimation:

[0067] (1) Body segment division.

[0068] Based on human anatomical standards, the reconstructed model is divided into five main segments: head, chest, pelvis, thigh, and lower leg. This division is based on the spatial clustering results of the model vertices in standard poses, combined with the boundaries determined by predefined key joint positions (such as cervical vertebrae, lumbar vertebrae, hip, knee, and ankle).

[0069] (2) Segment weight estimation.

[0070] A statistical regression model based on anthropometry is used, according to the user's body shape parameters. Estimate the weight percentage of each segment. The specific calculation formula is as follows:

[0071] ;

[0072] in, For the first The quality of each segment; This is the mass percentage coefficient of this segment (obtained from the Winter human body parameter table); The overall mass is estimated based on the user's height, weight, and body shape parameters.

[0073] (3) Establishment of segment-motor mapping.

[0074] Establish a spatial mapping relationship between the bottom surface of each segment (the area in contact with the mattress) and the motor below. The mapping rules are as follows:

[0075] Project the bottom surface of each segment onto the motor array plane;

[0076] If a motor is located within the projection area of ​​a segment, it is considered that the motor participates in supporting that segment;

[0077] Each motor can support one or more segments, and its supporting force is distributed proportionally to the projected area.

[0078] (4) Simplify the construction of the spring mass model.

[0079] Each body segment is considered a rigid point (mass block), and each motor is equivalent to a linear spring, forming a spring-mass system. The physical parameters of the spring are set as follows:

[0080] Spring stiffness is directly proportional to the equivalent Young's modulus of the mattress material;

[0081] The natural length of the spring is determined by the current height of the motor;

[0082] Spring compression This is the difference between the height of the motor and the height of the bottom surface of the segment.

[0083] When the The motor supports the first When there are multiple segments, the supporting force of the motor on that segment. Calculated using Hooke's Law:

[0084] ;

[0085] in, For the first The motor supports the first Spring compression at each segment;

[0086] This force is further converted into a portion of the supporting force experienced by the segment according to the proportion of the projected area.

[0087] (5) Pressure distribution calculation.

[0088] The system obtains the final interface pressure distribution by iteratively calculating the mechanical balance of all segments and the motor. The following balance conditions are considered during the calculation:

[0089] ;

[0090] in, This is the acceleration due to gravity. Ultimately, the pressure value at each motor location can be obtained by dividing the sum of the supporting forces of that motor by its sensing area.

[0091] Step S3: Using the height of each motor under the bed surface as the optimization variable, construct the following multi-objective optimization function:

[0092] ;

[0093] ;

[0094] ;

[0095] in, Indicates peak pressure. The pressure gradient is represented by the coefficient of variation of the pressure distribution, which is used as the pressure uniformity index. ( Standard deviation, (arithmetic mean) Indicates the current angle of the joint. This represents the median comfortable joint angle. These parameters represent the range of comfortable joint angles, derived from clinical studies and biomechanical literature research on the comfortable angles or neutral positions and ranges of motion of various joints; the weights for each parameter are: .

[0096] Step S4: Under the following constraints, solve the above optimization problem using the Particle Swarm Optimization (PSO) algorithm:

[0097] 1. The height of each motor is within its travel range: ,in Indicates the first The height of each motor is a decision variable in the optimization problem. This indicates the minimum allowable height of the motor (lower limit of travel). This indicates the maximum allowable height of the motor (maximum travel limit).

[0098] 2. The height difference between adjacent motors does not exceed the set threshold: Motors a and b are adjacent. Indicates the first The height of each motor This indicates the maximum allowable height difference between adjacent motors, used to ensure a smooth transition of the bed surface.

[0099] 3. The projection of the human body's center of gravity is located inside the mattress.

[0100] The PSO solver settings are as follows:

[0101] Particle coding adopts dimensional vector directly mapped to the mattress Each motor has a height, and each dimension corresponds to the real-time height value of a single motor. In this example... Take the number of particles ;

[0102] During initialization, particle positions are randomly and uniformly generated within the allowable height range of the motor, and the velocity is initialized to a random value within 10% of the height range;

[0103] The iterative update uses a constrained dynamic PSO algorithm, which updates the velocity by combining the inertia term, individual cognition term, and group social term, with weights set to 0.5, 1.5, and 1.5, respectively. After the position update, hard constraints are applied (boundary truncation and adjacent height difference limit). In each iteration, the comfort optimizer calculates the weighted discomfort of pressure distribution and joint posture as the fitness.

[0104] Set the maximum number of iterations to 100;

[0105] In the output phase, the globally optimal particle is decoded to determine the optimal motor height. .

[0106] Step S5, The command is sent to a multi-motor array controller, which drives each motor to move synchronously to a specified height, forming a support surface that conforms to the user's body shape and posture. The system can repeat the above process at regular intervals or triggered cycles to achieve dynamic adaptation.

[0107] Example 2

[0108] This embodiment provides an intelligent sleeping posture optimization system for implementing the above method, the architecture of which is as follows: Figure 1 As shown, it includes:

[0109] The sensing input layer is used to acquire static pressure distribution images of the user on the mattress;

[0110] The digital modeling layer is used to fit a parametric human body model based on stress images and output joint angles and body shape features.

[0111] The intelligent decision-making layer is used to construct and solve a multi-objective optimization problem with motor height as the optimization variable, and generate motor height instructions.

[0112] The physical execution layer is used to control the movement of the motor array according to altitude commands.

[0113] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0114] Therefore, the present invention adopts the above-mentioned intelligent sleeping posture optimization method and system based on pressure perception and human body reconstruction. By reconstructing a personalized three-dimensional human body model in real time from a single frame static pressure map and integrating pressure distribution and joint posture comfort for collaborative optimization, the present invention solves the problems of passive response, rough adaptation and neglect of internal physiological posture in existing sleeping posture adjustment technologies. Thus, it realizes active, accurate and biomechanical intelligent optimization and support for the user's sleeping posture.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart sleeping posture optimization method based on pressure perception and human body reconstruction, characterized in that, Includes the following steps: Step S1: Collect static pressure distribution data of the user in a static lying position by using a pressure sensor matrix arranged on the mattress surface, and generate a static pressure distribution image; Step S2: Input the static pressure distribution image into the pre-trained parametric human body model fitter to reconstruct a three-dimensional human body model containing the user's key joint angles and body shape features. Based on human anatomy standards, spatial clustering results of model vertices, and preset key joint positions for the cervical spine, lumbar spine, hip, knee, and ankle, the 3D human body model is divided into five independent human body segments: head, chest, pelvis, thigh, and calf. Using anthropometry statistical regression models based on user body shape parameters, the mass and weight ratio coefficient of each human body segment are calculated. The bottom surface area of ​​each human body segment in contact with the mattress is vertically projected onto the motor array plane. The corresponding supporting motor is determined by the projection coverage area; a single motor can support multiple human body segments. The supporting force weight of each segment on the corresponding motor is allocated according to the proportion of the segment's projected area, establishing a spatial and force mapping relationship between the human body segments and the motor array. Each human body segment is equivalent to a rigid point mass block, and each motor is equivalent to a linear spring unit, constructing a simplified physical model of spring mass. Hooke's law is used to calculate the spring compression and supporting force of the motors, and combined with the overall mechanical equilibrium conditions, iterative solutions are obtained to obtain the interface pressure distribution at each motor position on the mattress. Step S3: Using the height of each motor in the adjustable support bed as the optimization variable, construct a multi-objective optimization function that integrates pressure comfort and joint posture comfort; Step S4: Under the premise of satisfying the preset constraints, use the optimization algorithm to solve the multi-objective optimization function and obtain the optimal combination of motor heights that minimizes the objective function value; Step S5: Adjust the motor array according to the optimal height combination to form a support surface that matches the user's current body shape and posture; In step S3, the multi-objective optimization function is: ； ； ； in, Represents a multi-objective optimization function. This indicates the pressure comfort level. This indicates the comfort level of joint posture. Indicates peak pressure. Indicates the pressure gradient. Indicates the pressure uniformity index. Indicates the current angle of the joint. This represents the median comfortable joint angle. Indicates the range of comfortable joint angles. , , , , All represent weighting coefficients; In step S4, the constraints include: Height and travel constraints for each motor; Height difference constraint between adjacent motors; The projection of the human body's center of gravity is located inside the mattress.

2. The intelligent sleeping posture optimization method based on pressure perception and human body reconstruction according to claim 1, characterized in that, In step S2, the parameterized human body model fitter is constructed based on the SMPL model and regresses the user's body shape and posture parameters from the stress distribution image through a convolutional neural network.

3. The intelligent sleeping posture optimization method based on pressure perception and human body reconstruction according to claim 2, characterized in that, The parametric human model fitter is trained by minimizing a composite loss function, which includes body shape parameter loss, pose parameter loss, 3D joint loss, and mesh vertex loss.

4. The intelligent sleeping posture optimization method based on pressure perception and human body reconstruction according to claim 1, characterized in that, In step S4, the optimization algorithm is the particle swarm optimization algorithm.

5. The intelligent sleeping posture optimization method based on pressure perception and human body reconstruction according to claim 1, characterized in that, The parameter settings for the particle swarm optimization algorithm include: particle encoding dimension equal to the number of motors, particle population size of 30, maximum number of iterations of 100, and inertia weight, individual cognitive weight and group social weight in velocity update of 0.5, 1.5 and 1.5 respectively.

6. An intelligent sleeping posture optimization system based on pressure sensing and human body reconstruction, characterized in that, The method for implementing the intelligent sleep posture optimization method based on pressure sensing and human body reconstruction as described in any one of claims 1-5 includes: The sensing input layer is used to acquire static pressure distribution images of the user on the mattress; The digital modeling layer is used to fit a parametric human body model based on stress images and output joint angles and body shape features. The intelligent decision-making layer is used to construct and solve a multi-objective optimization problem with motor height as the optimization variable, and generate motor height instructions. The physical execution layer is used to control the movement of the motor array according to altitude commands.

7. The intelligent sleep posture optimization system based on pressure sensing and human body reconstruction according to claim 6, characterized in that, The sensing input layer is set as a flexible thin-film pressure sensor pad, and the size of its effective sensing area is [size missing]. The spatial resolution is 36mm.

Images