Fall early warning system based on multi-dimensional feature fusion and hierarchical trigger mechanism

Abstract

The invention discloses a tumble early warning system based on multi-dimensional feature fusion and a hierarchical trigger mechanism, and the system integrates the data of seven dimensions: traditional Chinese medicine diagnosis features (face image, eye diagnosis, tongue image, ear image and pulse condition), western medicine clinical indexes (electronic medical record and past medical history), posture behaviors, environment monitoring, inquiry semantics, voice voiceprint and psychological state. And accurate fall risk assessment is realized. The system adopts a three-stage trigger mechanism: a first-stage trigger module independently scores traditional Chinese medicine and western medicine features, and preliminarily quantifies the falling risk; the secondary trigger module independently scores the seven-dimensional features and outputs a multi-modal risk assessment result; and the third-level trigger module performs weighted fusion on the scores of all dimensions based on medical priori knowledge to generate low-risk, medium-risk and high-risk third-level early warning signals. Through multi-source data collaborative analysis and hierarchical decision making, the accuracy and clinical applicability of fall early warning are remarkably improved, and the method is suitable for old people health monitoring and medical aid decision making.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically to a fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism, and more specifically to a fall warning method implemented through a weighted scoring mechanism. Background Technology

[0002] According to the World Health Organization, approximately 28% to 35% of people aged 65 and over experience at least one fall each year, with severe cases resulting in fractures, traumatic brain injury, and even death. Falls not only cause personal suffering and a decline in quality of life but also impose a heavy medical burden and social care costs. Therefore, how to achieve early identification and intervention of fall risks is one of the core technological challenges in the field of smart healthcare and elderly care.

[0003] Patent document CN120048072A (application number: 202510271936.6) provides a fall prevention early warning system and method based on multi-dimensional monitoring. This system integrates and deploys sensors of different monitoring categories on a wearable device through a sensor configuration module. Based on the integration and deployment results, it monitors the user's walking status in real time, obtaining multi-category sensor data. A fall probability prediction module constructs a fall risk model and performs correlation analysis on the multi-category sensor data based on the fall risk model to predict the user's fall probability. An early warning notification module triggers an early warning mechanism based on the fall probability and responds to the user with a fall warning based on the triggering result. However, this system still has certain limitations, such as neglecting soft indicators like traditional Chinese medicine and psychology, making it difficult to fully reflect the complex mechanisms of fall formation, and lacking effective hierarchical early warning capabilities. Therefore, there is an urgent need to construct a multimodal, multi-level, and interpretable fall risk identification mechanism to improve the overall early warning level. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism. The system integrates multiple information dimensions such as traditional Chinese medicine diagnostic features (tongue appearance, facial appearance, pulse appearance, etc.), Western medicine indicators, posture and behavior, environmental monitoring, consultation semantics, voiceprint, and psychological state, and constructs a feature weighting mechanism. Through intra-module scoring and global weighted synthesis, it achieves early warning of falls.

[0005] A fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism provided by the present invention includes:

[0006] The data processing module collects data features from traditional Chinese medicine, Western medicine, posture, environment, consultation, voiceprint, and psychology through video, wearable devices, and question-and-answer methods, and filters out features that are involved in fall warning.

[0007] The first-level trigger module calculates risk scores for each feature based on five categories of characteristic parameters from traditional Chinese medicine (TCM): facial features, eye diagnosis, tongue appearance, ear appearance, and pulse. It also uses two categories of clinical indicators from Western medicine: electronic medical records and past medical history. This allows for a preliminary quantitative assessment of fall risk. The TCM features focus on analyzing parameters such as facial color, tongue shape, and pulse changes, while the Western medicine indicators emphasize the weight of chronic diseases such as diabetes and minor stroke on fall risk. The module outputs independent warning scores for each feature.

[0008] The secondary trigger module implements multimodal risk assessment across seven dimensions (Traditional Chinese Medicine, Western Medicine, posture, environment, consultation, voiceprint, and psychology): First, it integrates weighted scores from five sub-dimensions of Traditional Chinese Medicine (TCM) facial features, eye diagnosis, tongue appearance, ear appearance, and pulse diagnosis, and calculates a comprehensive TCM early warning value using adaptive weighting coefficients. Simultaneously, it generates a Western Medicine risk score based on electronic medical record data and past medical history (including key indicators such as diabetes and minor stroke). In addition, it independently calculates risk values ​​for five specific dimensions: consultation semantic features, posture parameters, psychological assessment, environmental factors, and voiceprint features, outputting risk assessment results across seven dimensions to provide multi-dimensional decision-making support for the tertiary trigger module.

[0009] The three-level trigger module employs a weighted fusion algorithm based on prior medical knowledge to scientifically integrate warning scores from seven dimensions: Traditional Chinese Medicine diagnostic features (facial appearance, eye diagnosis, tongue appearance, ear appearance, pulse appearance), Western medicine clinical indicators (electronic medical records, past medical history), posture and behavior, environmental monitoring, consultation semantics, voiceprint, and psychological state. Based on clinical medical experience and evidence-based medicine, weight coefficients for each dimension are preset, and a comprehensive fall risk score is calculated through weighted averaging. The system, based on preset clinical grading standards, classifies the risk score into three warning levels: low risk, medium risk, and high risk, and automatically triggers the corresponding warning mechanisms, providing a scientific basis for medical decision-making that aligns with clinical practice.

[0010] Preferably, the data processing module adopts a multi-source heterogeneous data acquisition method, including:

[0011] The system acquires facial, eye, tongue, and ear features from TCM diagnostic data using video capture equipment, and collects pulse data using a pulse diagnosis instrument; it integrates medical history records and physical examination indicators from the electronic medical record system as Western medicine clinical data; it collects behavioral and posture data, including gait, balance, and other movement parameters, using wearable devices; it acquires environmental monitoring data, such as ground humidity and light intensity, using IoT devices; it conducts psychological assessments using standardized questionnaires combined with voice emotion analysis technology; and it collects voiceprint feature parameters using audio equipment to achieve multi-dimensional data acquisition and processing.

[0012] Feature indicators highly correlated with fall risk are screened from multi-source data, including but not limited to: traditional Chinese medicine features such as facial color, tongue slenderness index, and the degree of wiryness and slipperiness of the pulse; Western medicine indicators such as history of diabetes, history of minor stroke, degree of Alzheimer's disease, and frequency of hypotension; behavioral parameters such as gait asymmetry, frequency of dragging steps, duration of standing instability, and angle of forward / backward tilt; environmental factors such as ground friction coefficient and light intensity; psychological indicators such as anxiety and depression; and voiceprint features such as the frequency of voice tremor, to provide accurate data support for subsequent graded early warning.

[0013] Preferably, the primary triggering module independently scores and calculates the traditional Chinese medicine characteristics (facial features, eye diagnosis, tongue appearance, ear appearance, pulse) and Western medicine indicators (electronic medical records, past medical history), including:

[0014] Based on the selected facial fall warning features, such as whether the complexion is dull, the overall color of the face, whether there are facial twitches, and whether the face is symmetrical, when a feature triggers the warning condition, the overall fall risk score A1 of the facial image is calculated, as shown in the following expression: Where, n 1 S represents the number of facial warning features. 1 This represents the set of all facial features that can trigger an alert. Indicates the weight of facial warning features. This indicates whether facial features triggered an alert; 1 indicates triggered, and 0 indicates not triggered.

[0015] Based on the selected visual fall warning features, such as a blank stare and eyelid swelling, when a feature triggers a warning condition, the overall visual fall risk score A2 is calculated, as shown in the following expression: Where, n 2 S represents the number of eye warning features. 2 This represents the set of all visual diagnostic features that can trigger an alert. Indicates the weight of eye warning features. This indicates whether the eye feature triggers a warning; 1 indicates a trigger and 0 indicates no trigger.

[0016] Based on the selected fall warning features of the tongue image, such as whether the tongue is deviated and the color of the tongue coating, when a feature triggers the warning condition, the overall fall risk score A3 of the tongue image is calculated, as shown in the following expression: Where, n 3 S represents the number of tongue appearance warning features. 3 This represents the set of all tongue features that can trigger an alert. Indicates the weight of tongue appearance warning features. This indicates whether the tongue appearance features trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0017] Based on the selected fall warning features of the ear image, such as redness of the ear and auricle, when a feature triggers the warning condition, the overall fall risk score A4 of the ear image is calculated, as shown in the following expression: Where, n 4 S represents the number of ear-image warning features. 4 This represents the set of all ear features that can trigger an alert. Indicates the weight of ear-image warning features. This indicates whether the ear feature triggers an alert; 1 indicates triggering and 0 indicates not triggering.

[0018] Based on the selected pulse fall warning features, such as wiry pulse and slippery pulse, when a feature triggers the warning condition, the overall pulse fall risk score A5 is calculated, as shown in the following expression: Where, n 5 S represents the number of pulse warning features. 5 This represents the set of all pulse characteristics that can trigger an early warning. Indicates the weight of pulse warning features. This indicates whether the pulse characteristics trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0019] Preferably, the primary triggering module independently scores and calculates the traditional Chinese medicine characteristics (facial features, eye diagnosis, tongue appearance, ear appearance, pulse) and Western medicine indicators (electronic medical records, past medical history), including:

[0020] Based on the selected fall warning indicators from electronic medical records, such as indicators of hypotension, when an indicator triggers a warning condition, the overall fall risk score B1 of the electronic medical record is calculated, as shown in the following expression: Where, m 1 T represents the number of early warning indicators for electronic medical records. 1 This represents the set of all electronic medical record indicators that can trigger alerts. Indicates the weight of the early warning indicators for electronic medical records. This indicates whether an electronic medical record indicator has triggered a warning; 1 indicates a triggered warning, and 0 indicates no triggered warning.

[0021] Based on the selected fall risk indicators from past medical records, such as indicators for diabetes, mini-stroke, and Alzheimer's disease, when an indicator triggers a warning condition, the overall fall risk score B2 of the past medical history is calculated, as shown in the following expression: Where, m 2 Indicates the number of warning indicators based on past medical history, T 2 This represents the set of all past medical history indicators that can trigger an alert. This indicates the weight of the warning indicator based on past medical history. This indicates whether the past medical history indicators have triggered an alert; 1 indicates a triggered alert, and 0 indicates no triggered alert.

[0022] Preferably, the secondary triggering module includes:

[0023] The weighted average of the scores from five feature categories—facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse appearance—in the first-level trigger module is used to generate a comprehensive TCM fall risk assessment value P1, as shown in the following expression: in, The weighting of the fall risk score based on five aspects: facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse.

[0024] Based on the weighted average of the scores of electronic medical records and past medical history in the first-level trigger module, a comprehensive Western medicine fall risk assessment value P2 is generated, as shown in the following expression: in, Weighting of fall risk scores based on electronic medical records and past medical history

[0025] Preferably, the secondary triggering module includes:

[0026] Based on the selected posture-related fall warning behaviors, such as gait symmetry, dragging frequency, duration of instability, and forward / backward tilt angles, when a posture behavior triggers a warning condition, the overall fall risk score P3 for the posture behavior is calculated, as shown in the following expression: Where z represents the number of warning behaviors in the pose, R represents the set of all pose behaviors that can trigger warnings, and W R Z represents the weight of the attitude warning behavior. i This indicates whether the behavior in the posture triggers a warning; 1 indicates a warning and 0 indicates no warning.

[0027] Based on the selected environmental monitoring fall warning features, such as whether the ground is slippery and the level of ambient light, when an environmental feature triggers a warning condition, the overall fall risk score P4 for the environmental feature is calculated, as shown in the following expression: Where v represents the number of early warning features in environmental monitoring, L represents the set of all environmental features that can trigger early warnings, and W L V represents the weight of environmental early warning features. i This indicates whether an environmental feature triggers a warning; 1 indicates a trigger, and 0 indicates no trigger.

[0028] Based on the selected physiological features for fall risk warning in the consultation semantics, such as dizziness, weakness, and slow reaction time, when a physiological feature triggers a warning condition, the overall fall risk score P5 of the consultation semantics is calculated, as shown in the following expression: Where d represents the number of warning physiological features in the consultation semantics, γ represents the set of all physiological features that can trigger warnings, and W γ D represents the weight of the early warning physiological features in the semantics of medical history taking. i This indicates whether the physiological characteristics in the consultation semantics trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0029] Based on the selected voiceprint fall warning features, such as unclear pronunciation and slurred enunciation, when a voiceprint feature triggers a warning condition, the overall fall risk score P6 of the voiceprint is calculated, as shown in the following expression: Where q represents the number of warning voiceprints in the speech, δ represents the set of all voiceprints that can trigger warnings, and W δ Indicates the voiceprint weight for voice warnings, Q i This indicates whether the voiceprint triggers an alert; 1 indicates triggered, and 0 indicates not triggered.

[0030] Based on the selected psychological fall warning states, such as depression and anxiety, when a psychological state triggers a warning condition, the overall fall risk score P7 for that psychological state is calculated, as shown in the following expression: Where k represents the number of psychological warning states, β represents the set of all psychological states that can trigger a warning, and W K K represents the weight of the psychological warning state. i This indicates whether a psychological state triggers a warning; 1 indicates a trigger and 0 indicates no trigger.

[0031] Preferably, the three-level triggering module includes:

[0032] A multi-dimensional feature fusion algorithm was used to deeply integrate the scoring results of seven dimensions, including TCM comprehensive score, Western medicine comprehensive score, posture behavior, environmental monitoring, medical history analysis, voiceprint features, and psychological assessment. A weighted average of the scores from each dimension was calculated using a clinically validated weighting scheme to generate the total fall risk score, Score, as shown in the following expression: Among them, W i Represents the weight coefficients of each dimension (∑W) i =1), P i For the standardized score, multi-level feature fusion is used to retain the specific contribution of each dimension. At the same time, a comprehensive quantitative assessment of fall risk is carried out to output a standardized risk index. The system automatically generates three-level warnings based on preset thresholds: low risk, medium risk and high risk, completing the entire process assessment from feature extraction to graded warning. Beneficial effects

[0033] This invention achieves accurate early warning of fall risk through multi-dimensional feature fusion and a hierarchical triggering mechanism. Its beneficial effects are mainly reflected in the following aspects:

[0034] 1. The system integrates multi-dimensional data such as TCM diagnostic features (e.g., tongue appearance, facial appearance, pulse appearance), Western medicine indicators (e.g., electronic medical records, past medical history), posture and behavior, natural environment, consultation semantics, voiceprints, and psychological state. It comprehensively analyzes fall risk from multiple perspectives, including physiology, behavior, psychology, and environment, significantly improving the comprehensiveness and accuracy of early warning.

[0035] 2. A modular weighted scoring mechanism is adopted to calculate fall warning scores for various dimensions, including Traditional Chinese Medicine and Western Medicine, layer by layer, and generate tiered warning signals. This mechanism not only improves the system's sensitivity to subtle changes in risk but also makes the warning results more interpretable, facilitating medical staff or family members to understand the source of risk and take targeted measures.

[0036] 3. By introducing psychological state assessment (such as anxiety and depression) and voiceprint analysis (such as voice tremor characteristics), the system can capture potential risks that are difficult to detect by traditional physiological indicators, such as gait instability or distraction caused by emotional fluctuations, thereby achieving early warning of pre-fall states.

[0037] 4. The system is compatible with ordinary cameras, wearable devices, and mobile terminals, requiring no complex hardware support and can be flexibly deployed in communities, homes, or elderly care institutions. Its modular design also allows for adjustments to data collection dimensions according to actual needs, balancing cost and efficiency, making it suitable for widespread application in large-scale health screening scenarios.

[0038] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Attached Figure Description

[0039] Figure 1 This is a framework diagram of a fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism. Specific Implementation

[0040] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0041] Example 1

[0042] According to the present invention, a fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism is provided, such as... Figure 1 As shown, it includes:

[0043] The data processing module collects data features from traditional Chinese medicine, Western medicine, posture, environment, consultation, voiceprint, and psychology through video, wearable devices, and question-and-answer methods, and filters out features that are involved in fall warning.

[0044] The first-level trigger module calculates risk scores for each feature based on five categories of characteristic parameters from traditional Chinese medicine (TCM): facial features, eye diagnosis, tongue appearance, ear appearance, and pulse. It also uses two categories of clinical indicators from Western medicine: electronic medical records and past medical history. This allows for a preliminary quantitative assessment of fall risk. The TCM features focus on analyzing parameters such as facial color, tongue shape, and pulse changes, while the Western medicine indicators emphasize the weight of chronic diseases such as diabetes and minor stroke on fall risk. The module outputs independent warning scores for each feature.

[0045] The secondary trigger module implements multimodal risk assessment across seven dimensions (Traditional Chinese Medicine, Western Medicine, posture, environment, consultation, voiceprint, and psychology): First, it integrates weighted scores from five sub-dimensions of Traditional Chinese Medicine (TCM) facial features, eye diagnosis, tongue appearance, ear appearance, and pulse diagnosis, and calculates a comprehensive TCM early warning value using adaptive weighting coefficients. Simultaneously, it generates a Western Medicine risk score based on electronic medical record data and past medical history (including key indicators such as diabetes and minor stroke). In addition, it independently calculates risk values ​​for five specific dimensions: consultation semantic features, posture parameters, psychological assessment, environmental factors, and voiceprint features, outputting risk assessment results across seven dimensions to provide multi-dimensional decision-making support for the tertiary trigger module.

[0046] The three-level trigger module employs a weighted fusion algorithm based on prior medical knowledge to scientifically integrate warning scores from seven dimensions: Traditional Chinese Medicine diagnostic features (facial appearance, eye diagnosis, tongue appearance, ear appearance, pulse appearance), Western medicine clinical indicators (electronic medical records, past medical history), posture and behavior, environmental monitoring, consultation semantics, voiceprint, and psychological state. Based on clinical medical experience and evidence-based medicine, weight coefficients for each dimension are preset, and a comprehensive fall risk score is calculated through weighted averaging. The system, based on preset clinical grading standards, classifies the risk score into three warning levels: low risk, medium risk, and high risk, and automatically triggers the corresponding warning mechanisms, providing a scientific basis for medical decision-making that aligns with clinical practice.

[0047] Specifically, the data processing module adopts a multi-source heterogeneous data acquisition method, including:

[0048] The system acquires facial, eye, tongue, and ear features from TCM diagnostic data using video capture equipment, and collects pulse data using a pulse diagnosis instrument; it integrates medical history records and physical examination indicators from the electronic medical record system as Western medicine clinical data; it collects behavioral and posture data, including gait, balance, and other movement parameters, using wearable devices; it acquires environmental monitoring data, such as ground humidity and light intensity, using IoT devices; it conducts psychological assessments using standardized questionnaires combined with voice emotion analysis technology; and it collects voiceprint feature parameters using audio equipment to achieve multi-dimensional data acquisition and processing.

[0049] Feature indicators highly correlated with fall risk are screened from multi-source data, including but not limited to: traditional Chinese medicine features such as facial color, tongue slenderness index, and the degree of wiryness and slipperiness of the pulse; Western medicine indicators such as history of diabetes, history of minor stroke, degree of Alzheimer's disease, and frequency of hypotension; behavioral parameters such as gait asymmetry, frequency of dragging steps, duration of standing instability, and angle of forward / backward tilt; environmental factors such as ground friction coefficient and light intensity; psychological indicators such as anxiety and depression; and voiceprint features such as the frequency of voice tremor, to provide accurate data support for subsequent graded early warning.

[0050] Specifically, the primary triggering module independently scores and calculates the traditional Chinese medicine characteristics (facial features, eye diagnosis, tongue appearance, ear appearance, and pulse) and Western medicine indicators (electronic medical records and past medical history), including:

[0051] Based on the selected facial fall warning features, such as whether the complexion is dull, the overall color of the face, whether there are facial twitches, and whether the face is symmetrical, when a feature triggers the warning condition, the overall fall risk score A1 of the facial image is calculated, as shown in the following expression: Where, n 1 S represents the number of facial warning features. 1 This represents the set of all facial features that can trigger an alert. Indicates the weight of facial warning features. This indicates whether facial features triggered an alert; 1 indicates triggered, and 0 indicates not triggered.

[0052] Based on the selected visual fall warning features, such as a blank stare and eyelid swelling, when a feature triggers a warning condition, the overall visual fall risk score A2 is calculated, as shown in the following expression: Where, n 2 S represents the number of eye warning features. 2 This represents the set of all visual diagnostic features that can trigger an alert. Indicates the weight of eye warning features. This indicates whether the eye feature triggers a warning; 1 indicates a trigger and 0 indicates no trigger.

[0053] Based on the selected fall warning features of the tongue image, such as whether the tongue is deviated and the color of the tongue coating, when a feature triggers the warning condition, the overall fall risk score A3 of the tongue image is calculated, as shown in the following expression: Where, n 3 S represents the number of tongue appearance warning features. 3 This represents the set of all tongue features that can trigger an alert. Indicates the weight of tongue appearance warning features. This indicates whether the tongue appearance features trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0054] Based on the selected fall warning features of the ear image, such as redness of the ear and auricle, when a feature triggers the warning condition, the overall fall risk score A4 of the ear image is calculated, as shown in the following expression: Where, n 4 S represents the number of ear-image warning features. 4 This represents the set of all ear features that can trigger an alert. Indicates the weight of ear-image warning features. This indicates whether the ear feature triggers an alert; 1 indicates triggering and 0 indicates not triggering.

[0055] Based on the selected pulse fall warning features, such as wiry pulse and slippery pulse, when a feature triggers the warning condition, the overall pulse fall risk score A5 is calculated, as shown in the following expression: Where, n 5 S represents the number of pulse warning features. 5 This represents the set of all pulse characteristics that can trigger an early warning. Indicates the weight of pulse warning features. This indicates whether the pulse characteristics trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0056] Specifically, the primary triggering module independently scores and calculates the traditional Chinese medicine characteristics (facial features, eye diagnosis, tongue appearance, ear appearance, and pulse) and Western medicine indicators (electronic medical records and past medical history), including:

[0057] Based on the selected fall warning indicators from electronic medical records, such as indicators of hypotension, when an indicator triggers a warning condition, the overall fall risk score B1 of the electronic medical record is calculated, as shown in the following expression: Where, m 1 T represents the number of early warning indicators for electronic medical records. 1 This represents the set of all electronic medical record indicators that can trigger alerts. Indicates the weight of the early warning indicators for electronic medical records. This indicates whether an electronic medical record indicator has triggered a warning; 1 indicates a triggered warning, and 0 indicates no triggered warning.

[0058] Based on the selected fall risk indicators from past medical records, such as indicators for diabetes, mini-stroke, and Alzheimer's disease, when an indicator triggers a warning condition, the overall fall risk score B2 of the past medical history is calculated, as shown in the following expression: Where, m 2 Indicates the number of warning indicators based on past medical history, T 2 This represents the set of all past medical history indicators that can trigger an alert. This indicates the weight of the warning indicator based on past medical history. This indicates whether the past medical history indicators have triggered an alert; 1 indicates a triggered alert, and 0 indicates no triggered alert.

[0059] Specifically, the secondary triggering module includes:

[0060] The weighted average of the scores from five feature categories—facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse appearance—in the first-level trigger module is used to generate a comprehensive TCM fall risk assessment value P1, as shown in the following expression: in, The weighting of the fall risk score based on five aspects: facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse.

[0061] More specifically, in the TCM-based fall risk assessment, the weights of each feature are allocated as follows: Tongue appearance accounts for 35%, as its changes can comprehensively reflect imbalances in Qi and blood, and abnormalities in organ function. For example, a pale tongue indicates Yang deficiency, a red tongue indicates internal heat, and a swollen tongue indicates dampness, all of which are directly related to high-risk factors for falls such as physical weakness and insufficient blood supply to the brain; Facial appearance accounts for 30%, as facial twitching and asymmetry are highly correlated with conditions such as stroke and facial paralysis, and are important indicators for predicting falls; Pulse appearance accounts for 20%, as a thin and weak pulse indicates deficiency of both Qi and blood, and a choppy pulse reflects poor blood circulation, both of which are related to decreased muscle strength; Eye examination accounts for 10%, as a dull gaze indicates mental abnormalities, and eyelid edema is associated with kidney deficiency and water retention, indicating physical decline; Ear appearance accounts for 5%, as red ears can indicate liver Yang hyperactivity, but have low specificity for predicting falls.

[0062] Specifically, the secondary triggering module includes:

[0063] Based on the weighted average of the scores of electronic medical records and past medical history in the first-level trigger module, a comprehensive Western medicine fall risk assessment value P2 is generated, as shown in the following expression: in, Weighting of fall risk scores based on electronic medical records and past medical history

[0064] More specifically, in the fall risk assessment of Western medicine indicators, the weights of each feature are allocated as follows: past medical history accounts for 55%, because it contains long-term records of key fall risk factors such as mini-stroke, Alzheimer's disease, and diabetes, and has important predictive value; electronic medical records account for 45%, reflecting the patient's muscle strength, balance ability, and potential risk factors through real-time updated functional assessment data.

[0065] Specifically, the secondary triggering module includes:

[0066] Based on the selected posture-related fall warning behaviors, such as gait symmetry, dragging frequency, duration of instability, and forward / backward tilt angles, when a posture behavior triggers a warning condition, the overall fall risk score P3 for the posture behavior is calculated, as shown in the following expression: Where z represents the number of warning behaviors in the pose, R represents the set of all pose behaviors that can trigger warnings, and W R Z represents the weight of the attitude warning behavior. i This indicates whether the behavior in the posture triggers a warning; 1 indicates a warning and 0 indicates no warning.

[0067] Based on the selected environmental monitoring fall warning features, such as whether the ground is slippery and the level of ambient light, when an environmental feature triggers a warning condition, the overall fall risk score P4 for the environmental feature is calculated, as shown in the following expression: Where v represents the number of early warning features in environmental monitoring, L represents the set of all environmental features that can trigger early warnings, and W L V represents the weight of environmental early warning features. i This indicates whether an environmental feature triggers a warning; 1 indicates a trigger, and 0 indicates no trigger.

[0068] Based on the selected physiological features for fall risk warning in the consultation semantics, such as dizziness, weakness, and slow reaction time, when a physiological feature triggers a warning condition, the overall fall risk score P5 of the consultation semantics is calculated, as shown in the following expression: Where d represents the number of warning physiological features in the consultation semantics, γ represents the set of all physiological features that can trigger warnings, and W γ D represents the weight of the early warning physiological features in the semantics of medical history taking. i This indicates whether the physiological characteristics in the consultation semantics trigger an alert; 1 indicates triggering and 0 indicates not triggering.

[0069] Based on the selected voiceprint fall warning features, such as unclear pronunciation and slurred enunciation, when a voiceprint feature triggers a warning condition, the overall fall risk score P6 of the voiceprint is calculated, as shown in the following expression: Where q represents the number of warning voiceprints in the speech, δ represents the set of all voiceprints that can trigger warnings, and W δ Indicates the voiceprint weight for voice warnings, Q i This indicates whether the voiceprint triggers an alert; 1 indicates triggered, and 0 indicates not triggered.

[0070] Based on the selected psychological fall warning states, such as depression and anxiety, when a psychological state triggers a warning condition, the overall fall risk score P7 for that psychological state is calculated, as shown in the following expression: Where k represents the number of psychological warning states, β represents the set of all psychological states that can trigger a warning, and W K K represents the weight of the psychological warning state. i This indicates whether a psychological state triggers a warning; 1 indicates a trigger and 0 indicates no trigger.

[0071] Specifically, the three-level triggering module includes:

[0072] A multi-dimensional feature fusion algorithm was used to deeply integrate the scoring results of seven dimensions, including TCM comprehensive score, Western medicine comprehensive score, posture behavior, environmental monitoring, medical history analysis, voiceprint features, and psychological assessment. A weighted average of the scores from each dimension was calculated using a clinically validated weighting scheme to generate the total fall risk score, Score, as shown in the following expression: Among them, W i Represents the weight coefficients of each dimension (∑W) i =1), P i This is the standardized score.

[0073] More specifically, in the overall fall risk assessment, the weights of each characteristic are allocated as follows: Postural behavior accounts for 30%, as it directly reflects fall risk; abnormal gait, turning too quickly, and unsteady standing are the most direct warning signs of a fall; Western medicine comprehensive weight accounts for 20%, as clinical conditions such as low blood pressure, diabetes, and mini-stroke are important medical risk factors for falls; The semantic weight of the consultation accounts for 15%, as chief symptoms such as dizziness, palpitations, and fatigue directly reflect an unstable state of bodily function; Environmental monitoring accounts for 12%, as environmental hazards such as insufficient light and slippery ground, especially during nighttime activities, significantly increase the probability of falls; Traditional Chinese medicine comprehensive weight accounts for 10%, as analysis of tongue and facial features can help determine constitutional risks such as insufficient qi and blood and internal liver wind; Voice and voiceprint weight accounts for 8%, as changes such as slow speech and unclear pronunciation can indirectly indicate abnormalities in the nervous system; Psychological state weight accounts for 5%, as psychological factors such as depression and anxiety, although not direct causes, can indirectly affect fall risk by reducing reaction ability and attention.

[0074] By employing a multi-level feature fusion method, the system comprehensively quantifies and assesses fall risk while preserving the specific contributions of each dimension. The final output is a standardized risk index of 0-100 points. Based on clinically validated preset thresholds, the system automatically generates three levels of warning signals: 0-39 points for low risk, 40-59 points for medium risk, and 60-100 points for high risk. This achieves a complete assessment process from multi-dimensional feature extraction to intelligent graded warning.

[0075] The present invention also provides a fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism. The fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism can be implemented by executing the process steps of the fall warning method based on multi-dimensional feature fusion and hierarchical triggering mechanism. That is, those skilled in the art can understand the fall warning method based on multi-dimensional feature fusion and hierarchical triggering mechanism as a preferred embodiment of the fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism.

[0076] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0077] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism, characterized in that, include: The data processing module collects data features from traditional Chinese medicine, Western medicine, posture, environment, consultation, voiceprint, and psychology through video, wearable devices, and question-and-answer methods, and filters out features that are involved in fall warning. The first-level trigger module calculates risk scores for each feature based on five categories of characteristic parameters from traditional Chinese medicine (TCM): facial features, eye diagnosis, tongue appearance, ear appearance, and pulse. It also uses two categories of clinical indicators from Western medicine: electronic medical records and past medical history. This allows for a preliminary quantitative assessment of fall risk. The TCM features focus on analyzing parameters such as facial color, tongue shape, and pulse changes, while the Western medicine indicators emphasize the weight of chronic diseases such as diabetes and minor stroke on fall risk. The module outputs independent warning scores for each feature. The secondary trigger module implements multimodal risk assessment across seven dimensions (Traditional Chinese Medicine, Western Medicine, posture, environment, consultation, voiceprint, and psychology): First, it integrates weighted scores from five sub-dimensions of Traditional Chinese Medicine (TCM) facial features, eye diagnosis, tongue appearance, ear appearance, and pulse diagnosis, and calculates a comprehensive TCM early warning value using adaptive weighting coefficients. Simultaneously, it generates a Western Medicine risk score based on electronic medical record data and past medical history (including key indicators such as diabetes and minor stroke). In addition, it independently calculates risk values ​​for five specific dimensions: consultation semantic features, posture parameters, psychological assessment, environmental factors, and voiceprint features, outputting risk assessment results across seven dimensions to provide multi-dimensional decision-making support for the tertiary trigger module. The three-level trigger module employs a weighted fusion algorithm based on prior medical knowledge to scientifically integrate warning scores from seven dimensions: Traditional Chinese Medicine diagnostic features (facial appearance, eye diagnosis, tongue appearance, ear appearance, pulse appearance), Western medicine clinical indicators (electronic medical records, past medical history), posture and behavior, environmental monitoring, consultation semantics, voiceprint, and psychological state. Based on clinical medical experience and evidence-based medicine, weight coefficients for each dimension are preset, and a comprehensive fall risk score is calculated through weighted averaging. The system, based on preset clinical grading standards, classifies the risk score into three warning levels: low risk, medium risk, and high risk, and automatically triggers the corresponding warning mechanisms, providing a scientific basis for medical decision-making that aligns with clinical practice.

2. The fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism according to claim 1, characterized in that, The data processing module includes: The system acquires facial, eye, tongue, and ear features from TCM diagnostic data using video capture equipment, and collects pulse data using a pulse diagnosis instrument; it integrates medical history records and physical examination indicators from the electronic medical record system as Western medicine clinical data; it collects behavioral and posture data, including gait, balance, and other movement parameters, using wearable devices; it acquires environmental monitoring data, such as ground humidity and light intensity, using IoT devices; it conducts psychological assessments using standardized questionnaires combined with voice emotion analysis technology; and it collects voiceprint feature parameters using audio equipment to achieve multi-dimensional data acquisition and processing. Feature indicators highly correlated with fall risk are screened from multi-source data, including but not limited to: traditional Chinese medicine features such as facial color, tongue slenderness index, and the degree of wiryness and slipperiness of the pulse; Western medicine indicators such as history of diabetes, history of minor stroke, degree of Alzheimer's disease, and frequency of hypotension; behavioral parameters such as gait asymmetry, frequency of dragging steps, duration of standing instability, and angle of forward / backward tilt; environmental factors such as ground friction coefficient and light intensity; psychological indicators such as anxiety and depression; and voiceprint features such as the frequency of voice tremor, to provide accurate data support for subsequent graded early warning.

3. The fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism according to claim 1, characterized in that, The primary triggering module independently scores and calculates TCM characteristics (facial features, eye diagnosis, tongue appearance, ear appearance, pulse) and Western medicine indicators (electronic medical records, past medical history), including: Based on the selected facial fall warning features, such as whether the complexion is dull, the overall color of the face, whether there are facial twitches, and whether the face is symmetrical, when a feature triggers the warning condition, the overall fall risk score A1 of the facial image is calculated, as shown in the following expression: Where, n 1 S represents the number of facial warning features. 1 This represents the set of all facial features that can trigger an alert. Indicates the weight of facial warning features. This indicates whether facial features triggered an alert; 1 indicates triggered, and 0 indicates not triggered. Based on the selected visual fall warning features, such as a blank stare and eyelid swelling, when a feature triggers a warning condition, the overall visual fall risk score A2 is calculated, as shown in the following expression: Where, n 2 S represents the number of eye warning features. 2 This represents the set of all visual diagnostic features that can trigger an alert. Indicates the weight of eye warning features. This indicates whether the eye feature triggers a warning; 1 indicates a trigger and 0 indicates no trigger. Based on the selected fall warning features of the tongue image, such as whether the tongue is deviated and the color of the tongue coating, when a feature triggers the warning condition, the overall fall risk score A3 of the tongue image is calculated, as shown in the following expression: Where, n 3 S represents the number of tongue appearance warning features. 3 This represents the set of all tongue features that can trigger an alert. Indicates the weight of tongue appearance warning features. This indicates whether the tongue appearance features trigger an alert; 1 indicates triggering and 0 indicates not triggering. Based on the selected fall warning features of the ear image, such as redness of the ear and auricle, when a feature triggers the warning condition, the overall fall risk score A4 of the ear image is calculated, as shown in the following expression: Where, n 4 S represents the number of ear-image warning features. 4 This represents the set of all ear features that can trigger an alert. Indicates the weight of ear-image warning features. This indicates whether the ear feature triggers an alert; 1 indicates triggering and 0 indicates not triggering. Based on the selected pulse fall warning features, such as wiry pulse and slippery pulse, when a feature triggers the warning condition, the overall pulse fall risk score A5 is calculated, as shown in the following expression: Where, n 5 S represents the number of pulse warning features. 5 This represents the set of all pulse characteristics that can trigger an early warning. Indicates the weight of pulse warning features. This indicates whether the pulse characteristics trigger an alert; 1 indicates triggering and 0 indicates not triggering. Based on the selected fall warning indicators from electronic medical records, such as indicators of hypotension, when an indicator triggers a warning condition, the overall fall risk score B1 of the electronic medical record is calculated, as shown in the following expression: Where, m 1 T represents the number of early warning indicators for electronic medical records. 1 This represents the set of all electronic medical record indicators that can trigger alerts. Indicates the weight of the early warning indicators for electronic medical records. This indicates whether an electronic medical record indicator has triggered a warning; 1 indicates a triggered warning, and 0 indicates no triggered warning. Based on the selected fall risk indicators from past medical records, such as indicators for diabetes, mini-stroke, and Alzheimer's disease, when an indicator triggers a warning condition, the overall fall risk score B2 of the past medical history is calculated, as shown in the following expression: Where, m 2 Indicates the number of warning indicators based on past medical history, T 2 This represents the set of all past medical history indicators that can trigger an alert. This indicates the weight of the warning indicator based on past medical history. This indicates whether the past medical history indicators have triggered an alert; 1 indicates a triggered alert, and 0 indicates no triggered alert.

4. A fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism according to claim 1, characterized in that, The secondary triggering module includes: The weighted average of the scores from five feature categories—facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse appearance—in the first-level trigger module is used to generate a comprehensive TCM fall risk assessment value P1, as shown in the following expression: in, The weighting of the fall risk score based on five aspects: facial appearance, eye diagnosis, tongue appearance, ear appearance, and pulse. Based on the weighted average of the scores of electronic medical records and past medical history in the first-level trigger module, a comprehensive Western medicine fall risk assessment value P2 is generated, as shown in the following expression: in, Weighting of fall risk scores based on electronic medical records and past medical history Based on the selected posture-related fall warning behaviors, such as gait symmetry, dragging frequency, duration of instability, and forward / backward tilt angles, when a posture behavior triggers a warning condition, the overall fall risk score P3 for the posture behavior is calculated, as shown in the following expression: Where z represents the number of warning behaviors in the pose, R represents the set of all pose behaviors that can trigger warnings, and W R Z represents the weight of the attitude warning behavior. i This indicates whether the behavior in the posture triggers a warning; 1 indicates a warning and 0 indicates no warning. Based on the selected environmental monitoring fall warning features, such as whether the ground is slippery and the level of ambient light, when an environmental feature triggers a warning condition, the overall fall risk score P4 for the environmental feature is calculated, as shown in the following expression: Where v represents the number of early warning features in environmental monitoring, L represents the set of all environmental features that can trigger early warnings, and W L V represents the weight of environmental early warning features. i This indicates whether an environmental feature triggers a warning; 1 indicates a trigger, and 0 indicates no trigger. Based on the selected physiological features for fall risk warning in the consultation semantics, such as dizziness, weakness, and slow reaction time, when a physiological feature triggers a warning condition, the overall fall risk score P5 of the consultation semantics is calculated, as shown in the following expression: Where d represents the number of warning physiological features in the consultation semantics, γ represents the set of all physiological features that can trigger warnings, and W γ D represents the weight of the early warning physiological features in the semantics of medical history taking. i This indicates whether the physiological characteristics in the consultation semantics trigger an alert; 1 indicates triggering and 0 indicates not triggering. Based on the selected voiceprint fall warning features, such as unclear pronunciation and slurred enunciation, when a voiceprint feature triggers a warning condition, the overall fall risk score P6 of the voiceprint is calculated, as shown in the following expression: Where q represents the number of warning voiceprints in the speech, δ represents the set of all voiceprints that can trigger warnings, and W δ Indicates the voiceprint weight for voice warnings, Q i This indicates whether the voiceprint triggers an alert; 1 indicates triggered, and 0 indicates not triggered. Based on the selected psychological fall warning states, such as depression and anxiety, when a psychological state triggers a warning condition, the overall fall risk score P7 for that psychological state is calculated, as shown in the following expression: Where k represents the number of psychological warning states, β represents the set of all psychological states that can trigger a warning, and W K K represents the weight of the psychological warning state. i This indicates whether a psychological state triggers a warning; 1 indicates a trigger and 0 indicates no trigger.

5. A fall warning system based on multi-dimensional feature fusion and hierarchical triggering mechanism according to claim 1, characterized in that, The secondary triggering module includes: A multi-dimensional feature fusion algorithm was used to deeply integrate the scoring results of seven dimensions, including TCM comprehensive score, Western medicine comprehensive score, posture behavior, environmental monitoring, medical history analysis, voiceprint features, and psychological assessment. A weighted average of the scores from each dimension was calculated using a clinically validated weighting scheme to generate the total fall risk score, Score, as shown in the following expression: Among them, W i Represents the weight coefficients of each dimension (∑W) i =1), P i For the standardized score, multi-level feature fusion is used to retain the specific contribution of each dimension. At the same time, a comprehensive quantitative assessment of fall risk is carried out to output a standardized risk index. The system automatically generates three-level warnings based on preset thresholds: low risk, medium risk and high risk, completing the entire process assessment from feature extraction to graded warning.

Images