Fall prevention monitoring method and system based on smart bed

Abstract

The application discloses a fall-prevention monitoring method and system based on an intelligent bed, relates to the technical field of intelligent beds with behavior feature recognition, and through the construction of a three-dimensional space monitoring framework, the accurate division of bed body and surrounding space coordinate ranges, the real-time collection of human motion data by fusing multi-sensor technology, the dynamic tracking of the changes of the center of mass trajectory, the calculation of the center of mass horizontal deviation rate and the center of mass deviation direction vector, the deep analysis of the limb suspension proportion and the posture inclination angle, the determination of the trend of the increase of the limb suspension duration, and the further evaluation of the fall risk probability value. The application realizes the establishment of an accurate three-dimensional space monitoring range around the bed body by the combination of the dynamic trajectory analysis of the center of mass and the multi-level monitoring grid, and realizes the real-time calculation of the dynamic relationship change rate of the center of mass and the bed body coordinates, and realizes the early positioning of the fall occurrence space.

Description

Technical Field

[0001] This invention relates to the field of smart bed technology based on behavioral feature recognition, and more particularly to a fall prevention monitoring method and system based on smart beds. Background Technology

[0002] With the increasing aging of the population, the risk of falls in bed for the elderly and patients with limited mobility is becoming increasingly prominent. This issue has become a pressing safety hazard that needs to be addressed in the field of medical care, directly affecting life, health, and the burden on families. Current fall prevention monitoring methods mostly rely on wearable devices or fixed cameras in the room. However, these methods are easily affected by changes in lighting, privacy restrictions, and poor patient compliance, resulting in insufficient monitoring continuity, especially at night or when patients are alone in bed, making it difficult to achieve reliable coverage around the clock.

[0003] In the field of smart bed fall detection, accurately delineating the space around the bed and effectively managing risk areas presents significant challenges. The space around the bed is complex and variable, encompassing various areas such as the bed surface, bedside, and the ground. The system needs to differentiate these areas hierarchically; otherwise, it's difficult to accurately capture the moment a body part enters a high-risk position. However, simply delineating the space is insufficient to address practical problems because a person's posture and center of gravity constantly change during the process of getting out of the bed, and the relative positional relationship between body parts and the bed boundaries adjusts continuously. This makes it difficult for the monitoring system to grasp the dynamic space occupancy status in real time.

[0004] This combination of spatial division and dynamic positional relationships further complicates the prediction of fall locations. When a patient's legs gradually dangle off the bed or their upper body's center of gravity slowly shifts beyond the bed edge, if the system cannot quickly calculate the coordinate changes of body parts in three-dimensional space and the rate of center of gravity shift, it will lose the optimal intervention opportunity before protective measures are activated. This results in the brief process from the intention to leave the bed to the actual fall not being effectively intercepted. The real-time capture of these dynamic spatial changes and the early judgment of the fall location are interconnected, directly limiting the accurate triggering of protective measures.

[0005] Therefore, how to establish an accurate three-dimensional spatial monitoring range around the bed and calculate the dynamic relationship between the human body's center of mass and the bed's coordinates in real time, so as to realize the early location of the fall-prone spatial area, has become a key issue in the research of fall prevention monitoring methods and systems based on intelligent beds. Summary of the Invention

[0006] This invention provides a fall prevention monitoring method and system based on a smart bed, which establishes a precise three-dimensional spatial monitoring range around the bed and calculates in real time the dynamic relationship change rate between the human body's center of mass and the bed's coordinates, thereby enabling early location of the fall-prone spatial area.

[0007] This invention provides a fall prevention monitoring method based on a smart bed, executed by a computer, comprising:

[0008] Based on the pre-established 3D model of the bed, the space around the bed is divided to obtain the coordinate range of the bed surface, bed edge and ground. Based on the coordinate range of the bed surface, bed edge and ground, a multi-level monitoring grid is constructed in the divided space area to obtain the initial 3D spatial monitoring framework.

[0009] Based on the initial three-dimensional spatial monitoring framework, multi-sensor fusion technology is used to collect human motion data in real time, and based on the human motion data, the spatial coordinate distribution of various parts of the human body is determined, wherein the spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass.

[0010] By continuously tracking the spatial coordinate distribution of various parts of the human body, the dynamic trajectory of the human body's center of mass in three-dimensional space is calculated to obtain the horizontal offset rate and the offset direction vector of the center of mass.

[0011] If the horizontal displacement rate of the center of mass exceeds the preset threshold range, and the displacement direction vector of the center of mass points to the coordinate range of the bedside, then a deep analysis of the dynamic change trajectory is performed to calculate the proportion of limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension.

[0012] Based on the increasing trend of the duration of limb suspension, the results of human posture change analysis are analyzed in real time, and based on the results of human posture change analysis, the curvature change of the center of mass trajectory is calculated to determine the correlation data between the acceleration component of the center of mass and the change of trajectory curvature.

[0013] Based on the associated data, the centroid offset direction vector and the bed coordinate offset are calculated, and the fall risk prediction probability value is determined based on the centroid offset direction vector and the bed coordinate offset.

[0014] According to the fall prevention monitoring method based on a smart bed provided by the present invention, the method involves using multi-sensor fusion technology to collect human motion data in real time based on the initial three-dimensional spatial monitoring framework, and determining the spatial coordinate distribution of various parts of the human body based on the human motion data, including:

[0015] Based on the initial three-dimensional spatial monitoring framework, human motion data is acquired in real time through multi-sensor fusion technology, and the human motion data is denoised to obtain cleaned motion signal data.

[0016] Based on the cleaned motion signal data, the position information is initially screened using preset filtering rules to remove outliers and determine the preliminary spatial coordinates of various parts of the human body.

[0017] Based on the preliminary spatial coordinates, the relative positional relationships of various parts of the human body are calculated using geometric transformation methods, so as to determine the suspended state of the human limbs based on the relative positional relationships.

[0018] Based on the suspended state of the human limbs, the initial three-dimensional spatial coordinates of the centroid are calculated using a weighted average method to obtain a preliminary estimate of the centroid's position.

[0019] Based on the preliminary estimate of the centroid position, the Kalman filter algorithm is used to dynamically correct the centroid position to obtain the three-dimensional coordinates of the centroid.

[0020] Based on the three-dimensional coordinates of the centroid and the suspended state of the human limbs, a human posture space model is constructed, and the spatial coordinate distribution is determined based on the human posture space model.

[0021] According to the fall prevention monitoring method based on a smart bed provided by the present invention, the method involves continuously tracking the spatial coordinate distribution of various parts of the human body and calculating the dynamic trajectory of the human body's center of mass in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the center of mass, including:

[0022] By continuously tracking the spatial coordinate distribution of various parts of the human body, a real-time coordinate sequence is obtained;

[0023] Based on the real-time coordinate sequence, the change in the position of the human body's centroid in the current frame is calculated to obtain the dynamic trajectory of the human body's centroid in three-dimensional space.

[0024] Based on the dynamic trajectory of the centroid, Kalman filtering is used for smoothing to determine the smoothed centroid trajectory.

[0025] Based on the smoothed centroid trajectory, the horizontal and vertical displacements of the centroid between adjacent frames are calculated to obtain the horizontal offset rate and the vertical sinking rate of the centroid.

[0026] The centroid offset direction vector is synthesized based on the horizontal offset rate of the centroid and the vertical sinking rate of the centroid.

[0027] According to the fall prevention monitoring method based on a smart bed provided by the present invention, if the horizontal offset rate of the center of gravity exceeds a preset threshold range, and the offset direction vector of the center of gravity points to the coordinate range of the bedside, then a depth analysis is performed on the dynamic trajectory to calculate the proportion of limbs suspended above the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension, including:

[0028] If the horizontal offset rate of the centroid exceeds a preset threshold range, then based on the angle between the offset direction vector of the centroid and the coordinate range of the bedside, it is determined that the offset direction vector of the centroid points to the coordinate range of the bedside.

[0029] When the centroid offset direction vector points to the bedside coordinate range, the dynamic trajectory is processed by limb segmentation to identify limb parts located outside the bedside coordinate range and to calculate the proportion of limbs suspended at the bedside.

[0030] The attitude tilt angle is calculated based on the shortest distance between the centroid projection point and the edge of the bed surface in the dynamic trajectory.

[0031] Based on the proportion of limbs suspended at the bedside and the tilt angle of the posture, a linear regression model is used to fit the slope of the change of the proportion of limbs suspended at the bedside over time to determine the trend of increasing duration of limb suspension.

[0032] If the slope of the change is positive, it indicates that the duration of limb suspension is increasing.

[0033] According to the fall prevention monitoring method based on a smart bed provided by the present invention, the method involves analyzing the human posture change results in real time based on the increasing trend of the duration of limb suspension, and calculating the curvature change of the center of mass trajectory based on the human posture change analysis results to determine the correlation data between the acceleration component of the center of mass and the change in trajectory curvature, including:

[0034] Based on the increasing trend of the duration of limb suspension, the three-dimensional coordinate changes of key points of the human body are calculated in real time to determine the results of the human posture change analysis.

[0035] Based on the analysis results of the human posture change, the limb tilt angle is extracted, and based on the limb tilt angle, the mapping parameters between the tilt angle threshold and the bed coordinates are dynamically adjusted.

[0036] Based on the adjusted mapping parameters, the depth of the limb part entering the high-risk area is calculated to obtain a depth change sequence.

[0037] Based on the depth change sequence, the posture adjustment frequency is determined, and based on the posture adjustment frequency and the human posture change analysis results, the centroid position sequence of the human body is determined.

[0038] The curvature change of the center of mass trajectory is calculated based on the center of mass position sequence, and a random forest regression model is used to determine the correlation data between the center of mass acceleration component and the trajectory curvature change based on the center of mass position sequence and the curvature change.

[0039] According to the fall prevention monitoring method based on a smart bed provided by the present invention, the step of calculating the centroid offset direction vector and the bed coordinate offset based on the associated data, and determining the fall risk prediction probability value based on the centroid offset direction vector and the bed coordinate offset, includes:

[0040] Based on the aforementioned correlation data, real-time centroid acceleration data is collected to obtain a centroid acceleration sequence;

[0041] Based on the centroid acceleration sequence, the trajectory curvature change of the centroid is calculated to obtain the trajectory curvature change sequence;

[0042] Based on the trajectory curvature change sequence, the centroid offset direction is extracted to obtain the centroid offset direction vector;

[0043] Based on the centroid offset direction vector, a sliding window statistical method is used to determine the degree of consistency of the offset direction, and a direction consistency index is obtained.

[0044] Based on the directional consistency index and the coordinate range of the bed surface, bedside and ground, calculate the bed coordinate offset and determine the coordinate offset value;

[0045] The coordinate offset value is input into the support vector machine model to obtain the fall risk prediction probability output by the support vector machine model.

[0046] The fall prevention monitoring method based on a smart bed provided by the present invention further includes:

[0047] If the predicted probability value of the fall risk reaches a preset alarm threshold, a protective trigger signal is generated, wherein the protective trigger signal is used to determine the timing of the activation of protective measures;

[0048] Based on the protection trigger signal, the posture and position data of the bedside are collected, and based on the posture and position data, a random forest classifier is used to classify the posture features and determine the classification result of the posture features.

[0049] If the classification result has a triggering condition, then the signal type information and severity information are extracted from the protection trigger signal;

[0050] Based on the signal type information and the severity information, the signal priority is determined, and based on the signal priority, multiple concurrent signals of bedside protection actions are sorted to obtain a sorted signal priority sequence.

[0051] The intelligent bed control module receives the sorted signal priority sequence and generates target execution instructions for corresponding bedside protection actions.

[0052] Based on the target execution command, the smart bed is driven to perform bedside lifting or bed tilting actions to intercept the risk of falling from the bedside in real time.

[0053] The present invention also provides a fall prevention monitoring system based on a smart bed, comprising:

[0054] The monitoring framework generation module is used to divide the space around the bed based on a pre-established three-dimensional model of the bed, obtain the coordinate range of the bed surface, bed edge and ground, and construct a multi-level monitoring grid for the divided space area based on the coordinate range of the bed surface, bed edge and ground to obtain an initial three-dimensional spatial monitoring framework.

[0055] The spatial coordinate distribution generation module is used to collect human motion data in real time using multi-sensor fusion technology based on the initial three-dimensional spatial monitoring framework, and to determine the spatial coordinate distribution of various parts of the human body based on the human motion data. The spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass.

[0056] The centroid trajectory determination module is used to continuously track the spatial coordinate distribution of various parts of the human body and calculate the dynamic trajectory of the human centroid in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the centroid.

[0057] The trend determination module is used to perform in-depth analysis on the dynamic change trajectory if the horizontal offset rate of the center of mass exceeds a preset threshold range and the offset direction vector of the center of mass points to the coordinate range of the bedside, and calculate the proportion of the limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension.

[0058] The associated data generation module is used to analyze the human posture change analysis results in real time based on the increasing trend of the duration of the limb suspension, and to calculate the curvature change of the center of mass trajectory based on the human posture change analysis results, so as to determine the associated data between the acceleration component of the center of mass and the change of trajectory curvature.

[0059] The fall risk generation module is used to calculate the centroid offset direction vector and bed coordinate offset based on the associated data, and to determine the fall risk prediction probability value based on the centroid offset direction vector and bed coordinate offset.

[0060] This invention provides a fall prevention monitoring method and system based on a smart bed. Addressing the business scenario of predicting fall risks by analyzing dynamic changes in human posture in the space surrounding a smart bed, it constructs a three-dimensional spatial monitoring framework to precisely delineate the coordinate range of the bed and its surrounding space. It integrates multi-sensor technology to collect human motion data in real time, dynamically tracks changes in the center of mass trajectory, calculates the horizontal offset rate and direction vector of the center of mass, deeply analyzes the proportion of limbs suspended in the air and the tilt angle of the posture, determines the increasing trend of limb suspension duration, and thus assesses the probability of a fall. This invention, through the combination of dynamic trajectory analysis of the center of mass and a multi-layered monitoring grid, significantly improves the accuracy and timeliness of fall risk prediction, providing a solution for bedside safety protection. It effectively reduces the probability of accidental falls, ensures user safety, establishes a precise three-dimensional spatial monitoring range around the bed, and calculates the dynamic relationship between the human center of mass and the bed coordinates in real time, enabling early location of the fall-prone spatial area. Attached Figure Description

[0061] Figure 1 This is one of the flowcharts of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0062] Figure 2 This is the second flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0063] Figure 3 This is the third flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0064] Figure 4 This is the fourth flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0065] Figure 5 This is the fifth flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0066] Figure 6 This is the sixth flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention;

[0067] Figure 7 This is the seventh flowchart of the fall prevention monitoring method based on a smart bed provided in this embodiment of the invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Reference Figure 1 This invention provides a fall prevention monitoring method based on a smart bed, comprising the following steps:

[0070] Step 100: Based on the pre-established three-dimensional model of the bed, the space around the bed is divided to obtain the coordinate range of the bed surface, bed edge and ground. Based on the coordinate range of the bed surface, bed edge and ground, a multi-level monitoring grid is constructed for the divided space area to obtain the initial three-dimensional spatial monitoring framework.

[0071] The core of this step lies in constructing a structured three-dimensional monitoring spatial framework. First, based on the actual physical structure and dimensions of the bed, a corresponding three-dimensional model of the bed is pre-established. This three-dimensional model represents the shape and spatial occupancy of the bed. Based on this three-dimensional model, the key spaces around the bed are logically divided, defining the three-dimensional coordinate ranges of the bed surface area, the adjacent area next to the bed, and the ground area. This discretizes the continuous physical space into logical subspaces with different functions and risk levels.

[0072] After defining the three-dimensional coordinate ranges of the bed surface area, the adjacent area, and the ground area, a multi-layered monitoring grid is constructed for each of the divided spatial areas. This monitoring grid can be adaptively configured according to the characteristics and monitoring needs of different areas. For example, a denser grid can be used in the higher-risk bedside area to improve monitoring sensitivity, while the grid density can be appropriately reduced in other relatively safe areas to optimize computing resources. Through this hierarchical grid construction, a structured initial three-dimensional spatial monitoring framework is formed.

[0073] Step 200: Based on the initial three-dimensional spatial monitoring framework, multi-sensor fusion technology is used to collect human motion data in real time, and based on the human motion data, the spatial coordinate distribution of each part of the human body is determined, wherein the spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass.

[0074] After obtaining the initial three-dimensional spatial monitoring framework, dynamic monitoring and data analysis are performed. Multi-sensor fusion technology is used to collect multi-dimensional data reflecting human posture and movement in real time. Human movement data can be mapped to this initial three-dimensional spatial monitoring framework in real time to ensure a synchronous correspondence between human movement and the digital model. Then, based on the collected human movement data, the detailed configuration of the human body in three-dimensional space is analyzed. Specifically, posture estimation algorithms and models can be used to determine the spatial coordinate distribution of major human body parts, such as the head, torso, and limbs, at the current moment. This spatial coordinate distribution is used to determine whether a specific limb (e.g., a leg or arm) is suspended in the air or in contact with a solid support surface such as the bed or ground; these are important features for judging key actions such as getting out of bed, sitting up, or loss of balance.

[0075] Subsequently, by weighting and calculating the spatial coordinates of various parts of the human body, the initial three-dimensional spatial coordinates of the body's center of mass can be calculated in real time. These center of mass coordinates can be used to measure the overall balance of the human body, enabling continuous tracking of the center of mass's trajectory, velocity, and positional relationship relative to safety boundaries such as the bedside and the ground. This step transforms the raw sensor signals into spatial coordinate parameters with clear physical meaning, including information on limb suspension representing the local state and the spatial position of the center of mass representing the overall state.

[0076] Step 300: By continuously tracking the spatial coordinate distribution of various parts of the human body, calculate the dynamic trajectory of the human body's center of mass in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the center of mass.

[0077] After obtaining the initial three-dimensional spatial coordinates of the human body's center of mass, the spatial coordinate distribution of various parts of the body is sampled and updated at high frequency to achieve millisecond-level continuous tracking of the overall center of mass, thereby constructing a dynamic trajectory describing the evolution of the center of mass in three-dimensional space over time. This dynamic trajectory uses time as the horizontal axis to represent the position of the center of mass at each successive moment. Based on this continuous and smooth trajectory, dynamic parameters are extracted. By calculating the ratio of the distance traveled by the projection point of the center of mass on a horizontal plane (usually a plane parallel to the ground) to the corresponding time, the horizontal offset rate of the center of mass is calculated in real time. This horizontal offset rate describes how fast the body's center of mass moves laterally. A slow, controllable offset usually represents a normal sitting up or turning, while a sudden, high-speed offset is very likely a direct signal of slipping, misstepping, or the onset of imbalance.

[0078] Simultaneously, by analyzing the vector relationships formed by continuous points on the dynamic trajectory of the center of mass in three-dimensional space, the direction vector of the center of mass offset is calculated. This direction vector contains information about the magnitude of the offset and indicates the direction of the center of mass's movement in three-dimensional space. For example, this direction vector can clearly distinguish whether the center of mass is moving towards the outside of the bed, towards the ground, or sliding along the edge of the bed. Furthermore, the horizontal offset rate of the center of mass (the intensity of the movement) can be combined with the direction vector of the center of mass offset (the spatial tendency of the movement) to obtain a set of core characteristic quantities that can comprehensively characterize the current overall motion state and risk of imbalance of the human body.

[0079] Step 400: If the horizontal offset rate of the center of mass exceeds the preset threshold range, and the offset direction vector of the center of mass points to the coordinate range of the bedside, then perform in-depth analysis on the dynamic change trajectory, calculate the proportion of the limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension.

[0080] When the detected rate of centroid horizontal displacement exceeds a preset safety threshold based on normal human activity patterns, and simultaneously the centroid displacement direction vector points towards a predefined bedside coordinate range, it signifies a potential high-risk situation. This combination of "high speed" and "outward orientation" strongly suggests that the human body may not be engaging in controlled, slow bedside activity, but rather is in a potentially unbalanced state of rapid movement away from the bed. In this case, a deep analysis mode is employed to further quantify and assess the fine posture of the human body in the local area around the bed. First, the proportion of limbs suspended above the bedside is calculated. This involves calculating the volume or key node proportion of body parts (such as lower limbs) located in the bedside area whose spatial coordinates show no contact with any support surface, thus obtaining the proportion of limbs suspended above the bedside. This proportion reflects the extent to which the body has deviated from stable support.

[0081] Simultaneously, by combining a human skeletal model, the body's tilt angle is calculated. This tilt angle measures the degree of deviation of the torso's central axis from the direction of vertical gravity. A continuously increasing tilt angle indicates that the body is losing its upright or sitting balance and is developing towards a falling posture, leaning to the side or forward. In-depth analysis of the dynamic trajectory essentially involves continuous time-series tracking of two key indicators: the proportion of limbs suspended above the bedside and the tilt angle, to capture the dynamic evolution trend of the center of mass. Therefore, by analyzing whether the proportion of limbs suspended above the bed continuously increases and whether the tilt angle continuously worsens within a short time window, it is possible to determine whether this dangerous state of limb suspension is continuously worsening and at what rate, thus enabling the assessment of whether it is a gradual, rather than instantaneous, fall or slippage process.

[0082] Step 500: Based on the increasing trend of the duration of the limb being suspended in the air, analyze the results of the human posture change analysis in real time, and calculate the curvature change of the center of mass trajectory based on the results of the human posture change analysis, so as to determine the correlation data between the acceleration component of the center of mass and the change of trajectory curvature.

[0083] By analyzing the trend of increasing limb suspension duration in real time, the results of human posture change analysis are analyzed. When the analysis indicates that the limb suspension state is continuously worsening, it suggests that a potential imbalance event is evolving, rather than a momentary fluctuation. The real-time human posture change analysis results are then further analyzed. These results include static limb parameters such as limb angles and torso orientation, as well as dynamic patterns showing how these parameters change over time. By analyzing the dynamic patterns of the limb parameters in real time, the human intention or loss of control corresponding to the current posture change can be determined, such as whether it is an active leaning forward or a passive slip.

[0084] Subsequently, the changes in the trajectory curvature were calculated in real time by analyzing the three-dimensional centroid trajectory obtained through continuous tracking. This curvature characterizes the degree of bending of the trajectory; a sudden or drastic change indicates a sharp shift in the direction of motion. This is typically a key geometric feature in the process of imbalance, where the body's movement path changes from controllable to uncontrollable after losing support. For example, a trajectory that abruptly changes from horizontal movement to vertical descent will exhibit a peak in curvature at the turning point.

[0085] Furthermore, based on the curvature changes of the obtained center-of-mass trajectory, the intrinsic correlation data between the acceleration components of the center of mass and the changes in trajectory curvature is explored in depth. A dynamic mapping relationship between the two is established by performing correlation analysis between the acceleration vector describing the rate of change of velocity and the curvature parameter describing the rate of change of path direction. The correlation data is used to characterize the physical nature of the imbalance process: for example, a large normal acceleration accompanied by a sharp increase in curvature may correspond to the instant when the body is tripped and rapidly turns and falls; while the abrupt change in tangential acceleration and the asynchronous change in curvature may characterize another instability mode.

[0086] Step 600: Based on the associated data, calculate the centroid offset direction vector and the bed coordinate offset, and determine the fall risk prediction probability value based on the centroid offset direction vector and the bed coordinate offset.

[0087] Based on the analysis of the correlation data between the center of mass acceleration and trajectory curvature, a comprehensive risk assessment and quantification is performed. This involves integrating the obtained multi-dimensional dynamic parameters into the assessment model to calculate two key risk determination factors. First, the consistency of the center of mass offset direction is assessed by analyzing the sequence of center of mass offset direction vectors over a continuous period to determine whether it exhibits stable and continuous directional characteristics, such as consistently and clearly pointing away from the safe area outside the bed. High consistency indicates that the human body is being driven away from the safe position by a continuous trend, rather than random swaying. Second, the bed coordinate offset is calculated, which quantifies the change in spatial position of the human body's center of mass (or key torso positions) relative to a pre-defined safe area (such as the center of the bed surface) in the three-dimensional model of the bed.

[0088] Finally, an intelligent algorithm model can be used to fuse and weight the two core factors—the centroid offset direction vector and the bed coordinate offset—with multi-source, heterogeneous information such as the acquired limb suspension trend, posture tilt angle, and trajectory curvature correlation data. This intelligent algorithm model is used to simulate the comprehensive decision-making process. By analyzing the intensity, persistence, and combination patterns of various risk indicators, a quantified fall risk prediction probability value is obtained. This fall risk prediction probability value is a continuous spectral scale from 0% to 100% to reflect the immediate likelihood of a fall occurring at the current moment.

[0089] This invention provides a fall prevention monitoring method and system based on a smart bed. Addressing the business scenario of predicting fall risks by analyzing dynamic changes in human posture in the space surrounding a smart bed, it constructs a three-dimensional spatial monitoring framework to precisely delineate the coordinate range of the bed and its surrounding space. It integrates multi-sensor technology to collect human motion data in real time, dynamically tracks changes in the center of mass trajectory, calculates the horizontal offset rate and direction vector of the center of mass, deeply analyzes the proportion of limbs suspended in the air and the tilt angle of posture, determines the increasing trend of limb suspension duration, and thus assesses the probability of a fall. This invention, through the combination of dynamic trajectory analysis of the center of mass and a multi-layered monitoring grid, significantly improves the accuracy and timeliness of fall risk prediction, providing an innovative solution for bedside safety protection. It effectively reduces the probability of accidental falls, ensures user safety, establishes a precise three-dimensional spatial monitoring range around the bed, and calculates the dynamic relationship between the human center of mass and the bed coordinates in real time, enabling early location of the fall-prone spatial area.

[0090] In one embodiment, please refer to Figure 2 The process, based on the initial three-dimensional spatial monitoring framework, employs multi-sensor fusion technology to collect human motion data in real time, and determines the spatial coordinate distribution of various parts of the human body based on the human motion data, including:

[0091] Step 201: Based on the initial three-dimensional spatial monitoring framework, human motion data is acquired in real time through multi-sensor fusion technology, and the human motion data is denoised to obtain cleaned motion signal data.

[0092] Step 202: Based on the cleaned motion signal data, the position information is initially screened using preset filtering rules to remove outliers and determine the preliminary spatial coordinates of various parts of the human body.

[0093] Step 203: Based on the preliminary spatial coordinates, apply a geometric transformation method to calculate the relative positional relationship of various parts of the human body, so as to determine the suspended state of the human limbs based on the relative positional relationship;

[0094] Step 204: Based on the suspended state of the human limbs, the initial three-dimensional spatial coordinates of the center of mass are calculated by weighted average method to obtain a preliminary estimate of the position of the center of mass.

[0095] Step 205: Based on the preliminary estimate of the centroid position, the Kalman filter algorithm is used to dynamically correct the centroid position to obtain the three-dimensional coordinates of the centroid.

[0096] Step 206: Based on the three-dimensional coordinates of the centroid and the suspended state of the human limbs, construct a human posture space model, and determine the spatial coordinate distribution based on the human posture space model.

[0097] Based on the established initial 3D spatial monitoring framework, raw human motion data is first acquired in real time by fusing data from multiple sensors deployed within the framework. This motion data is then processed using a denoising algorithm to filter out environmental interference and inherent device noise, resulting in cleaned motion signal data. Following this, pre-defined intelligent filtering rules are used to initially screen the positional information in the cleaned motion signal data, effectively identifying and eliminating outliers caused by brief occlusions or signal jumps, thereby obtaining stable and reliable preliminary spatial coordinates for various parts of the human body. Subsequently, geometric transformation methods are applied to analyze the relative positional relationships between different body parts. This analysis based on relative relationships can determine whether a limb is truly in a suspended state without contact with any supporting surface.

[0098] After determining the limb's suspended state, a weighted average method based on a human anatomical model is used to calculate the initial three-dimensional spatial coordinates of the human center of mass, obtaining a preliminary estimate. To further improve the accuracy and real-time smoothness of this key parameter, a Kalman filter algorithm is used to dynamically track and correct this preliminary estimate. Specifically, this Kalman filter algorithm is used in the motion model based on the center of mass, fusing current observations with historical predictions to output stable and disturbance-resistant three-dimensional coordinates of the center of mass. Finally, the three-dimensional coordinates of the center of mass are fused with previously determined information such as the limb's suspended state to construct a human posture spatial model that comprehensively reflects the human body in three-dimensional space. This human posture spatial model, as the culmination of all spatial information, accurately determines the final spatial coordinate distribution of various parts of the human body.

[0099] For example, in the initial three-dimensional spatial monitoring framework, multiple depth sensors and inertial sensors are first deployed for multi-sensor fusion. The sensors are arranged around the room at 2.0-meter intervals, using a total of 6 depth sensors and 17 inertial sensors worn by each person to achieve real-time acquisition of human motion data. The sampling frequency is uniformly 60Hz. After time stamp alignment, extended Kalman filtering is used to fuse depth point cloud and acceleration / angular velocity data, reducing the noise standard deviation to within 0.01 meters. Next, the collected raw location information was initially filtered using a statistical outlier removal algorithm. With a neighborhood number of 20 and a standard deviation threshold of 3.0, significantly deviating noise points were removed. Simultaneously, voxel mesh downsampling (0.05-meter voxel size) was applied to reduce point cloud density and improve subsequent processing efficiency, thereby accurately determining the spatial coordinate distribution of various parts of the human body. A joint detection model pre-trained based on a random forest classifier identified the 3D coordinates of 18 key joints, including the head, shoulders, elbows, wrists, hips, knees, and ankles, with an average positioning error controlled within 0.03 meters. Based on this, the suspended state of human limbs was calculated. First, the angle between each limb segment and the ground normal vector (0,1,0) was determined using the vector cross product. If the angle was greater than 30°, it was marked as suspended. Simultaneously, the acceleration magnitude of the inertial sensor was used to analyze static and dynamic support, eliminating false ground contact judgments, ultimately obtaining the proportion and distribution of suspended limbs. By further utilizing the identified key point coordinates, the initial three-dimensional spatial coordinates of the centroid are calculated using a weighted average method. The weights are based on the standard physiological proportions of the human body (e.g., the head accounts for 8% and the torso accounts for 50%). Through iterative optimization, the centroid projection falls within the supporting polygon, and the initial coordinate error is reduced to 0.02 meters.

[0100] In this embodiment, the accuracy and robustness of the suspension state judgment and centroid positioning are improved through multi-level processing of geometric relationship analysis and dynamic filtering correction. Furthermore, the discrete coordinate points are transformed into a human posture space model, which enhances the logical consistency and interpretability of the system analysis.

[0101] In one embodiment, please refer to Figure 3 The method involves continuously tracking the spatial coordinate distribution of various parts of the human body and calculating the dynamic trajectory of the human body's center of mass in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the center of mass, including:

[0102] Step 301: Obtain a real-time coordinate sequence by continuously tracking the spatial coordinate distribution of various parts of the human body;

[0103] Step 302: Based on the real-time coordinate sequence, calculate the change in the position of the human body's center of mass in the current frame to obtain the dynamic trajectory of the human body's center of mass in three-dimensional space;

[0104] Step 303: Based on the dynamic trajectory of the centroid, Kalman filtering is used for smoothing to determine the smoothed centroid trajectory;

[0105] Step 304: Based on the smoothed centroid trajectory, calculate the horizontal and vertical displacements of the centroid between adjacent frames to obtain the horizontal offset rate and the vertical sinking rate of the centroid.

[0106] Step 305: Based on the horizontal offset rate of the centroid and the vertical sinking rate of the centroid, synthesize the centroid offset direction vector.

[0107] After determining the spatial coordinate distribution based on the human posture space model, dynamic tracking and quantitative analysis of the motion state are performed. First, the spatial coordinates of various parts of the human body in the human posture space model are sampled continuously at high frequency to obtain a real-time coordinate sequence that evolves over time. Based on this real-time coordinate sequence, the instantaneous spatial position of the human center of mass is calculated frame by frame, and these continuous position points are connected to construct a dynamic trajectory of the center of mass depicting its movement path in three-dimensional space. To overcome the interference of minor sensor jitter and random errors on the trajectory, a Kalman filter algorithm is further used to smooth the dynamic trajectory of the center of mass, resulting in a smoothed trajectory that more realistically reflects the trend of human motion.

[0108] Next, dynamic parameters are extracted based on the smoothed centroid trajectory. By calculating the displacement of the centroid point in the horizontal and vertical directions between adjacent time frames on the smoothed trajectory, the horizontal offset rate and vertical sinking rate of the centroid are calculated. The horizontal offset rate quantifies the speed of lateral movement of the human body and can be used to detect lateral displacement due to imbalance; the vertical sinking rate quantifies the speed of descent of the human body and can be used to identify falls or sitting down. Finally, based on the horizontal offset rate and the vertical sinking rate, these two physically orthogonal rate components are vector-synthesized to obtain the centroid offset direction vector, which characterizes the instantaneous direction and trend of the centroid's motion. This centroid offset direction vector indicates the direction of motion (e.g., towards the outside of the bed or obliquely towards the ground), and its magnitude also comprehensively reflects the intensity of the motion.

[0109] For example, a depth camera or multi-camera system can be used to continuously capture the 3D spatial coordinates of 17 key points on the human body. For instance, OpenPose or MediaPipe pose estimation algorithms can be used to acquire the key point location data for each frame in real time, with the coordinate unit being meters. A sampling frequency of 30Hz is ensured to guarantee smooth trajectory. Based on a human anatomical mass distribution model, the human body is divided into the head (8.1% mass), trunk (50.0%), upper arm (2.8%), forearm (1.6%), hand (0.6%), thigh (14.2%), lower leg (4.7%), and foot (1.4%). A corresponding mass weight is assigned to each key point, and the centroid position of each segment is calculated as a weighting point. The overall centroid coordinates of the human body in the current frame are then calculated. For example, in one frame, the centroid might be located at (0.50, 1.05, 0.10) meters. Subsequently, the dynamic trajectory is calculated using the centroid displacement between adjacent frames. For example, if the centroid moves from (0.50, 1.05, 0.10) to (0.51, 1.04, 0.11) between two frames, the displacement vector is (0.01, -0.01, 0.01) meters, and the time interval is 0.033 seconds, then the horizontal offset rate is calculated using the horizontal component norm. ≈ 0.43 m / s, the absolute value of the negative z-component of the vertical sinking velocity. The offset direction vector is the normalized horizontal component (0.707, -0.707, 0). Furthermore, after applying a low-pass filter to remove noise from the multi-frame centroid trajectory sequence, the trajectory continuity deviation is calculated as the standard deviation of the Euclidean distance between adjacent points. For example, a sequence deviation of 0.015 meters indicates a stable trajectory. The centroid height variation is the difference between the maximum and minimum z-coordinates; for example, the variation from 1.05 meters to 1.00 meters is 0.05 meters. This variation is combined with the sinking rate to assess the risk of falling.

[0110] In this embodiment, by independently calculating and synthesizing horizontal and vertical velocities, a synthetic direction vector that better reveals complex unbalanced motion is constructed. Simultaneously, applying Kalman filtering to the real-time smoothing of the centroid trajectory not only suppresses noise but also enables short-term trajectory prediction based on the motion model. This results in higher stability and immediacy of the calculated velocity and direction vectors, effectively reducing false alarms caused by data jitter.

[0111] In one embodiment, please refer to Figure 4 If the horizontal displacement rate of the center of mass exceeds a preset threshold range, and the displacement direction vector of the center of mass points to the bedside coordinate range, then a depth analysis is performed on the dynamic trajectory to calculate the proportion of limbs suspended at the bedside and the tilt angle of the posture, in order to determine the trend of increasing duration of limb suspension, including:

[0112] Step 401: If the horizontal offset rate of the centroid exceeds a preset threshold range, then based on the angle between the offset direction vector of the centroid and the coordinate range of the bedside, determine that the offset direction vector of the centroid points to the coordinate range of the bedside.

[0113] Step 402: When the centroid offset direction vector points to the bedside coordinate range, perform limb segmentation processing on the dynamic change trajectory, identify the limb parts located outside the bedside coordinate range, and calculate the proportion of limbs suspended at the bedside.

[0114] Step 403: Calculate the attitude tilt angle based on the shortest distance between the centroid projection point and the edge of the bed surface in the dynamic trajectory.

[0115] Step 404: Based on the proportion of limbs suspended at the bedside and the tilt angle of the posture, a linear regression model is used to fit the slope of the change of the proportion of limbs suspended at the bedside over time to determine the trend of increasing duration of limb suspension.

[0116] If the slope of the change is positive, it indicates that the duration of limb suspension is increasing.

[0117] After calculating the horizontal offset rate and offset direction vector of the center of gravity in real time, when the horizontal offset rate exceeds the safety threshold range set according to the normal human activity pattern, subsequent in-depth analysis is initiated. By calculating the spatial angle between the offset direction vector of the center of gravity and the predefined boundary of the bedside coordinate range, it is determined whether the vector is clearly pointing towards the bedside area. This dual condition judgment of "high speed" and "outward orientation" constitutes a filter for high-risk event identification, effectively screening out safe activities that, although fast, are oriented towards the inside of the bed or sliding along the bedside.

[0118] After establishing the risk conditions, the local state of the human body in the risk area (bedside) is analyzed. First, continuous motion trajectory data is processed by limb segmentation to identify which body parts have actually exceeded the bedside coordinate range, including limbs located outside the bedside coordinate range and calculating the proportion of limbs suspended above the bedside. This proportion of limbs suspended above the bedside is used to quantify the range of limbs in an unsupported risk state and is a key spatial indicator for assessing the severity of imbalance. Simultaneously, the system combines the geometric relationship between the center of mass projection point and the edge of the bed surface to calculate the body's tilt angle. This angle is not a simple absolute angle but reflects the degree of deviation of the body's center of gravity relative to its supporting foundation (bed edge), sensitively capturing unbalanced postures such as forward or sideways tilting.

[0119] Next, the data sequence of the proportion of limbs dangling from the bedside over time was input into a linear regression model for fitting analysis. By calculating the slope of this proportion over time, it was determined whether the limb dangling state was worsening, stabilizing, or improving. If the slope was positive, it confirmed that the duration of limb dangling was increasing, enabling the differentiation between momentary, self-correctable dangerous actions and continuous, worsening processes of falling from the bed, such as quickly reaching for something and then retracting it, thus achieving the prediction of dangerous progression.

[0120] For example, when monitoring and analyzing the horizontal offset rate of the center of mass, assuming that the user's posture data on the bed is collected in real time by sensors, the horizontal offset rate of the center of mass is calculated from two consecutive frames of data. In the current frame, the center of mass position is (1.2, 0.8) meters, and in the previous frame it was (1.0, 0.8) meters, with a time interval of 0.5 seconds. Therefore, the offset rate is approximately 0.4 meters per second. If the preset threshold range is 0.1 to 0.3 meters per second, this is clearly outside the range. Next, the offset direction vector of the center of mass is calculated. The current vector is (0.2, 0.0). Assuming the x-axis range of the bedside is 1.5 to 2.0 meters, by comparing it with the bedside coordinate range, it is determined that the vector points towards the bedside. Then, the deep analysis stage begins. A human body model is constructed using 3D posture data, and the proportion of limbs suspended above the bedside is calculated. Assuming that point cloud data analysis reveals that some points on the right arm and right leg extend beyond the bedside range, accounting for approximately 25%, this exceeds the safe proportion threshold of 20%. Simultaneously, the distance of the centroid projection from the bed surface was calculated. Assuming the bed height is 0.5 meters, the centroid z-axis height is 0.7 meters, and the deviation distance is 0.2 meters, combined with historical data analysis, the deviation distance increased from 0.1 meters to 0.2 meters in the past 10 minutes, showing a continuous increasing trend. To further determine the increasing trend of limb suspension duration, the suspension time was recorded. The current suspension duration is 5 minutes. Historical data sampling per minute shows that the duration increased from 2 minutes to 5 minutes, a significant increase, triggering the early warning mechanism.

[0121] This embodiment improves the predictive ability for high-risk scenarios that develop slowly but continue to deteriorate (such as the gradual slippage of elderly people in a state of drowsiness) by using a time-series trend criterion based on linear regression. Furthermore, the analysis method that combines limb segmentation and proportional calculation of the centroid trajectory with the geometric relationship of the bed improves the accuracy and reliability of risk assessment in complex situations.

[0122] In one embodiment, please refer to Figure 5 The method involves analyzing the human posture change results in real time based on the increasing trend of the limb suspension duration, and calculating the curvature change of the center of mass trajectory based on the human posture change analysis results to determine the correlation data between the center of mass acceleration component and the trajectory curvature change, including:

[0123] Step 501: Based on the increasing trend of the duration of the limb being suspended in the air, calculate the changes in the three-dimensional coordinates of key points of the human body in real time and determine the results of the human posture change analysis.

[0124] Step 502: Extract the limb tilt angle based on the human posture change analysis results, and dynamically adjust the mapping parameters between the tilt angle threshold and the bed coordinates based on the limb tilt angle.

[0125] Step 503: Based on the adjusted mapping parameters, calculate the depth at which the limb part enters the high-risk area to obtain a depth change sequence;

[0126] Step 504: Based on the depth change sequence, determine the posture adjustment frequency, and based on the posture adjustment frequency and the human posture change analysis results, determine the centroid position sequence of the human body's centroid.

[0127] Step 505: Calculate the curvature change of the center of mass trajectory based on the center of mass position sequence, and use a random forest regression model based on the center of mass position sequence and the curvature change to determine the correlation data between the center of mass acceleration component and the trajectory curvature change.

[0128] After determining that the number of limbs suspended in the air is continuously increasing, the three-dimensional coordinates of key points representing the human skeleton are first tracked and calculated. Subtle changes in human posture are calculated in real time, yielding dynamic analysis results of human posture changes. These results describe the static geometric characteristics of the posture and reveal its dynamic evolution patterns.

[0129] Based on the analysis of human posture changes, the tilt angles of key limbs such as the torso and thighs are further extracted. These real-time angles are then used to dynamically adjust the mapping parameters between the tilt angle threshold and the bed coordinate space. This adaptive mechanism allows for personalized assessment of the degree of tilt that constitutes a real risk in the current position and posture, significantly improving the contextual relevance of the assessment. Subsequently, using the adjusted mapping parameters, the depth of limb intrusion into a predefined high-risk area at the bedside can be accurately calculated, forming a continuous depth change sequence. This depth change sequence records the evolution of the risk exposure level.

[0130] Furthermore, by analyzing this depth change sequence, the subject's posture adjustment frequency, i.e., the rate at which they attempt to regain balance or change posture, is calculated. Combining this posture adjustment frequency with the posture change analysis results allows for a more accurate distinction between voluntary movements and passive imbalances, thereby screening and determining the human center of mass position sequence. Finally, based on this center of mass position sequence, the geometric characteristics of its trajectory, i.e., curvature change, are calculated. Using curvature change and center of mass position information as input, a random forest regression model—a machine learning method—is employed to mine and determine the complex correlation between the center of mass acceleration component and the trajectory curvature change. This random forest regression model is able to capture the nonlinear, high-dimensional interaction patterns between the two.

[0131] In this embodiment, dynamic parameter mapping and attitude adjustment frequency analysis enhance the ability to perceive the context and distinguish intentions, effectively differentiating between voluntary getting out of bed and passive slippage, and greatly reducing false alarms. Secondly, the random forest regression model is applied to the correlation analysis of acceleration and curvature, two deep dynamic features, to achieve a deeper understanding of the underlying physical mechanisms of different types of imbalance events and more reliable predictions.

[0132] In one embodiment, please refer to Figure 6 The step of calculating the centroid offset direction vector and bed coordinate offset based on the associated data, and determining the fall risk prediction probability value based on the centroid offset direction vector and bed coordinate offset, includes:

[0133] Step 601: Based on the associated data, collect real-time centroid acceleration data to obtain a centroid acceleration sequence;

[0134] Step 602: Based on the centroid acceleration sequence, calculate the trajectory curvature change of the centroid to obtain the trajectory curvature change sequence;

[0135] Step 603: Based on the trajectory curvature change sequence, extract the centroid offset direction to obtain the centroid offset direction vector;

[0136] Step 604: Based on the centroid offset direction vector, the sliding window statistical method is used to determine the degree of consistency of the offset direction and obtain the direction consistency index.

[0137] Step 605: Based on the directional consistency index and the coordinate range of the bed surface, bed edge and ground, calculate the bed coordinate offset and determine the coordinate offset value;

[0138] Step 606: Input the coordinate offset value into the support vector machine model to obtain the fall risk prediction probability output by the support vector machine model.

[0139] After establishing the correlation data between the center of mass acceleration and trajectory curvature, high-frequency center of mass acceleration data is first collected in real time based on this correlation data, forming a sequence reflecting the dynamic state of motion, namely the center of mass acceleration sequence. Then, the geometric curvature characteristics of the motion path are dynamically calculated based on this acceleration sequence, generating a synchronous trajectory curvature change sequence. The center of mass offset direction vector at each moment is then calculated and extracted from the continuous curvature changes, ensuring that the direction information originates from a deep derivation of the overall motion geometry, thus improving the robustness of the direction determination.

[0140] To assess movement trends, a sliding window statistical method was used to analyze the extracted direction vector sequence. This method calculates a direction consistency index by examining the concentration and dispersion of direction vectors within a specific time window, thereby quantifying whether the body's center of mass is continuously moving in a stable direction or oscillating erratically. This direction consistency index can effectively distinguish between purposeful bed-leaning movements and panicked struggles before imbalance. Simultaneously, combining the direction consistency index with established coordinate ranges for the bed surface, bedside, and ground, the body's coordinate offset relative to these safety reference areas was calculated. This offset characterizes the spatial distance and velocity of the body leaving a stable support position.

[0141] Finally, the orientation consistency index and coordinate offset values ​​are used as a set of highly refined features and input into a pre-trained support vector machine (SVM) model. This SVM model is able to find the optimal classification boundary in the high-dimensional feature space that distinguishes between safe states and different levels of risk states, and ultimately outputs the predicted probability value of fall risk.

[0142] In this embodiment, a semantic understanding of the risk situation is achieved through directional consistency analysis based on sliding window statistics; furthermore, the use of a support vector machine model for probabilistic output can comprehensively weigh complex features with multiple dimensions and time-series correlations, and make nonlinear comprehensive judgments.

[0143] In one embodiment, please refer to Figure 7 It also includes:

[0144] Step 700: If the predicted probability value of the fall risk reaches a preset alarm threshold, a protection trigger signal is generated, wherein the protection trigger signal is used to determine the timing of the activation of the protection measures.

[0145] Step 800: Based on the protection trigger signal, collect the posture and position data at the bedside, and based on the posture and position data, use a random forest classifier to classify the posture features and determine the classification result of the posture features.

[0146] Step 900: If the classification result has a triggering condition, then extract the signal type information and severity information from the protection trigger signal;

[0147] Step 1000: Based on the signal type information and the severity information, determine the signal priority, and based on the signal priority, sort multiple concurrent signals of bedside protection actions to obtain a sorted signal priority sequence.

[0148] Step 1100: Receive the sorted signal priority sequence through the intelligent bed control module and generate the target execution command for the corresponding bedside protection action;

[0149] Step 1200: Based on the target execution command, drive the smart bed to perform bedside lifting or bed tilting actions to intercept the risk of falling off the bedside in real time.

[0150] When the predicted probability of a fall exceeds a preset alarm threshold, a high-risk event is deemed imminent, and a structured protective trigger signal is generated. This signal includes a comprehensive assessment of the risk status, used to accurately determine the optimal time to activate protective measures. Based on this signal, high-frequency monitoring is activated to collect detailed posture and position data of the person at the bedside. Key features are extracted from the posture and position data at the bedside and input into a random forest classifier for rapid and reliable classification of these real-time postures, resulting in a posture feature classification. This classification result is used to determine the specific pattern of the current risk, such as whether it is a lateral slip or a forward roll.

[0151] If the classification results meet the preset triggering conditions, key signal type information (such as risk category) and severity information (such as probability value) are parsed from the initial protection trigger signals. Then, based on the signal type and severity information, the signal priority of each protection action to be executed is dynamically determined. When multiple protection requests need to be executed concurrently, such as simultaneously raising the guardrail and adjusting the bed tilt, all signals are sorted according to this priority, generating a sorted signal priority sequence. This ensures that in the most urgent situations, resources are prioritized for the most critical protection actions.

[0152] Subsequently, the smart bed control module receives and parses the sorted signal priority sequence, transforming it into a series of specific, executable target execution instructions. Finally, based on the target execution instructions, it drives the smart bed's mechanical mechanism to perform physical protective actions such as raising the bedside rails or adjusting the bed's tilt angle. This proactive and timely physical intervention and spatial limitation of the body's tendency to lose balance allows for real-time interception in the very short time before a fall occurs.

[0153] In this embodiment, by using a pose reconfirmation and signal parsing mechanism based on a random forest classifier, the accuracy of the intervention can be ensured by adding this highly reliable verification and refinement step, thus avoiding unnecessary interference caused by misjudgment.

[0154] The fall prevention monitoring system based on a smart bed provided by the present invention will be described below. The fall prevention monitoring system based on a smart bed described below can be referred to in correspondence with the fall prevention monitoring method based on a smart bed described above.

[0155] The present invention also provides a fall prevention monitoring system based on a smart bed, comprising:

[0156] The monitoring framework generation module is used to divide the space around the bed based on a pre-established three-dimensional model of the bed, obtain the coordinate range of the bed surface, bed edge and ground, and construct a multi-level monitoring grid for the divided space area based on the coordinate range of the bed surface, bed edge and ground to obtain an initial three-dimensional spatial monitoring framework.

[0157] The spatial coordinate distribution generation module is used to collect human motion data in real time using multi-sensor fusion technology based on the initial three-dimensional spatial monitoring framework, and to determine the spatial coordinate distribution of various parts of the human body based on the human motion data. The spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass.

[0158] The centroid trajectory determination module is used to continuously track the spatial coordinate distribution of various parts of the human body and calculate the dynamic trajectory of the human centroid in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the centroid.

[0159] The trend determination module is used to perform in-depth analysis on the dynamic change trajectory if the horizontal offset rate of the center of mass exceeds a preset threshold range and the offset direction vector of the center of mass points to the coordinate range of the bedside, and calculate the proportion of the limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension.

[0160] The associated data generation module is used to analyze the human posture change analysis results in real time based on the increasing trend of the duration of the limb suspension, and to calculate the curvature change of the center of mass trajectory based on the human posture change analysis results, so as to determine the associated data between the acceleration component of the center of mass and the change of trajectory curvature.

[0161] The fall risk generation module is used to calculate the centroid offset direction vector and bed coordinate offset based on the associated data, and to determine the fall risk prediction probability value based on the centroid offset direction vector and bed coordinate offset.

[0162] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fall prevention monitoring method based on a smart bed, characterized in that, Executed by a computer, including: Based on the pre-established three-dimensional model of the bed, the space around the bed is divided to obtain the coordinate range of the bed surface, bed edge and ground. Based on the coordinate range of the bed surface, bed edge and ground, a multi-level monitoring grid is constructed for the divided space area to obtain the initial three-dimensional spatial monitoring framework. Based on the initial three-dimensional spatial monitoring framework, multi-sensor fusion technology is used to collect human motion data in real time, and based on the human motion data, the spatial coordinate distribution of various parts of the human body is determined, wherein the spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass. By continuously tracking the spatial coordinate distribution of various parts of the human body, the dynamic trajectory of the human centroid in three-dimensional space is calculated to obtain the horizontal offset rate and the offset direction vector of the centroid. Specifically, this includes: continuously tracking the spatial coordinate distribution of various parts of the human body to obtain a real-time coordinate sequence; calculating the change in the position of the human centroid in the current frame based on the real-time coordinate sequence to obtain the dynamic trajectory of the human centroid in three-dimensional space; smoothing the dynamic trajectory of the centroid using Kalman filtering to determine the smoothed centroid trajectory; calculating the horizontal and vertical displacements of the centroid between adjacent frames based on the smoothed centroid trajectory to obtain the horizontal offset rate and the vertical sinking rate of the centroid; and synthesizing the offset direction vector of the centroid based on the horizontal offset rate and the vertical sinking rate of the centroid. If the horizontal displacement rate of the center of mass exceeds the preset threshold range, and the displacement direction vector of the center of mass points to the coordinate range of the bedside, then a depth analysis is performed on the dynamic change trajectory to calculate the proportion of limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension. Based on the increasing trend of the duration of limb suspension, the results of human posture change analysis are analyzed in real time, and based on the results of human posture change analysis, the curvature change of the center of mass trajectory is calculated to determine the correlation data between the acceleration component of the center of mass and the change of trajectory curvature. Based on the associated data, the centroid offset direction vector and bed coordinate offset are calculated. A fall risk prediction probability value is then determined based on the centroid offset direction vector and the bed coordinate offset. Specifically, this includes: collecting real-time centroid acceleration data based on the associated data to obtain a centroid acceleration sequence; calculating the trajectory curvature change of the centroid based on the centroid acceleration sequence to obtain a trajectory curvature change sequence; extracting the centroid offset direction based on the trajectory curvature change sequence to obtain a centroid offset direction vector; using a sliding window statistical method to determine the consistency of the offset direction based on the centroid offset direction vector to obtain a direction consistency index; calculating the bed coordinate offset based on the direction consistency index and the coordinate range of the bed surface, bed edge, and ground to determine the coordinate offset value; and inputting the coordinate offset value into a support vector machine model to obtain the fall risk prediction probability output by the support vector machine model.

2. The fall prevention monitoring method based on a smart bed according to claim 1, characterized in that, The method, based on the initial three-dimensional spatial monitoring framework, employs multi-sensor fusion technology to collect human motion data in real time, and determines the spatial coordinate distribution of various parts of the human body based on the human motion data, including: Based on the initial three-dimensional spatial monitoring framework, human motion data is acquired in real time through multi-sensor fusion technology, and the human motion data is denoised to obtain cleaned motion signal data. Based on the cleaned motion signal data, the position information is initially screened using preset filtering rules to remove outliers and determine the preliminary spatial coordinates of various parts of the human body. Based on the preliminary spatial coordinates, the relative positional relationships of various parts of the human body are calculated using geometric transformation methods, so as to determine the suspended state of the human limbs based on the relative positional relationships. Based on the suspended state of the human limbs, the initial three-dimensional spatial coordinates of the centroid are calculated using a weighted average method to obtain a preliminary estimate of the centroid's position. Based on the preliminary estimate of the centroid position, the Kalman filter algorithm is used to dynamically correct the centroid position to obtain the three-dimensional coordinates of the centroid. Based on the three-dimensional coordinates of the centroid and the suspended state of the human limbs, a human posture space model is constructed, and the spatial coordinate distribution is determined based on the human posture space model.

3. The fall prevention monitoring method based on a smart bed according to claim 1, characterized in that, If the horizontal displacement rate of the center of gravity exceeds a preset threshold range, and the displacement direction vector of the center of gravity points to the bedside coordinate range, then a depth analysis is performed on the dynamic trajectory to calculate the proportion of limbs suspended above the bedside and the tilt angle of their posture, in order to determine the increasing trend of the duration of limb suspension, including: If the horizontal offset rate of the centroid exceeds a preset threshold range, then based on the angle between the offset direction vector of the centroid and the coordinate range of the bedside, it is determined that the offset direction vector of the centroid points to the coordinate range of the bedside. When the centroid offset direction vector points to the bedside coordinate range, the dynamic trajectory is processed by limb segmentation to identify limb parts located outside the bedside coordinate range and to calculate the proportion of limbs suspended at the bedside. The attitude tilt angle is calculated based on the shortest distance between the centroid projection point and the edge of the bed surface in the dynamic trajectory. Based on the proportion of limbs suspended at the bedside and the tilt angle of the posture, a linear regression model is used to fit the slope of the change of the proportion of limbs suspended at the bedside over time to determine the trend of increasing duration of limb suspension. If the slope of the change is positive, it indicates that the duration of limb suspension is increasing.

4. The fall prevention monitoring method based on a smart bed according to claim 1, characterized in that, The method involves analyzing the changes in human posture in real time based on the increasing trend of the duration of limb suspension, and calculating the curvature change of the center of mass trajectory based on the results of the human posture change analysis to determine the correlation data between the acceleration component of the center of mass and the change in trajectory curvature, including: Based on the increasing trend of the duration of limb suspension, the three-dimensional coordinate changes of key points of the human body are calculated in real time to determine the results of the human posture change analysis. Based on the analysis results of the human posture change, the limb tilt angle is extracted, and based on the limb tilt angle, the mapping parameters between the tilt angle threshold and the bed coordinates are dynamically adjusted. Based on the adjusted mapping parameters, the depth of the limb part entering the high-risk area is calculated to obtain a depth change sequence. Based on the depth change sequence, the posture adjustment frequency is determined, and based on the posture adjustment frequency and the human posture change analysis results, the centroid position sequence of the human body is determined. The curvature change of the center of mass trajectory is calculated based on the center of mass position sequence, and a random forest regression model is used to determine the correlation data between the center of mass acceleration component and the trajectory curvature change based on the center of mass position sequence and the curvature change.

5. The fall prevention monitoring method based on a smart bed according to claim 1, characterized in that, Also includes: If the predicted probability value of the fall risk reaches a preset alarm threshold, a protective trigger signal is generated, wherein the protective trigger signal is used to determine the timing of the activation of protective measures; Based on the protection trigger signal, the posture and position data of the bedside are collected, and based on the posture and position data, a random forest classifier is used to classify the posture features and determine the classification result of the posture features. If the classification result has a triggering condition, then the signal type information and severity information are extracted from the protection trigger signal; Based on the signal type information and the severity information, the signal priority is determined, and based on the signal priority, multiple concurrent signals of bedside protection actions are sorted to obtain a sorted signal priority sequence. The intelligent bed control module receives the sorted signal priority sequence and generates target execution instructions for corresponding bedside protection actions. Based on the target execution command, the smart bed is driven to perform bedside lifting or bed tilting actions to intercept the risk of falling from the bedside in real time.

6. A fall prevention monitoring system based on a smart bed, employing the fall prevention monitoring method based on a smart bed as described in any one of claims 1-5, characterized in that, include: The monitoring framework generation module is used to divide the space around the bed based on a pre-established three-dimensional model of the bed, obtain the coordinate range of the bed surface, bed edge and ground, and construct a multi-level monitoring grid for the divided space area based on the coordinate range of the bed surface, bed edge and ground to obtain an initial three-dimensional spatial monitoring framework. The spatial coordinate distribution generation module is used to collect human motion data in real time using multi-sensor fusion technology based on the initial three-dimensional spatial monitoring framework, and to determine the spatial coordinate distribution of various parts of the human body based on the human motion data. The spatial coordinate distribution includes the suspended state of the human limbs and the initial three-dimensional spatial coordinates of the center of mass. The centroid trajectory determination module is used to continuously track the spatial coordinate distribution of various parts of the human body and calculate the dynamic trajectory of the human centroid in three-dimensional space to obtain the horizontal offset rate and the offset direction vector of the centroid. The trend determination module is used to perform in-depth analysis on the dynamic change trajectory if the horizontal offset rate of the center of mass exceeds a preset threshold range and the offset direction vector of the center of mass points to the coordinate range of the bedside, and calculate the proportion of the limbs suspended at the bedside and the tilt angle of the posture, so as to determine the trend of increasing duration of limb suspension. The associated data generation module is used to analyze the human posture change analysis results in real time based on the increasing trend of the duration of the limb suspension, and to calculate the curvature change of the center of mass trajectory based on the human posture change analysis results, so as to determine the associated data between the acceleration component of the center of mass and the change of trajectory curvature. The fall risk generation module is used to calculate the centroid offset direction vector and bed coordinate offset based on the associated data, and to determine the fall risk prediction probability value based on the centroid offset direction vector and bed coordinate offset.

Images