Personal health service system based on smart watch digital human

Abstract

The invention relates to the technical field of intelligent wearable equipment, and provides a personal health service system based on an intelligent watch digital human, the system comprises an intelligent data analysis platform module comprising an intelligent watch end and a cloud end, the intelligent watch end comprises a digital human interaction module and a multi-source data acquisition module, the digital human interaction module constructs a multi-mode interaction mechanism based on a virtual digital human, obtains user health related information, receives a user instruction and outputs a health analysis result; the multi-source data acquisition module acquires physiological feature data and behavior activity data of a user in real time, and performs dynamic alignment and consistency verification of multi-dimensional data through a space-time association algorithm; and the intelligent data analysis platform module adopts a hierarchical collaborative architecture of edge computing and cloud deep mining to intelligently analyze and process the collected data and generate a personalized health assessment report and a dynamic intervention suggestion, so that the digital human interaction experience can be optimized, and the personalization, intelligence and naturalization of health management can be realized.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of intelligent wearable devices, in particular to a personal health service system based on a smart watch digital person. BACKGROUND

[0002] In recent years, with the popularity of intelligent wearable devices, smart watches have gradually become an important carrier for user health management due to their portability. Currently, the health management functions of smart watches rely on sensors to collect basic data such as heart rate and movement trajectory, and feedback the results to users in the form of text or simple charts. The interaction mode is single and lacks intuitiveness, making it difficult to meet the user's demand for in-depth understanding and natural interaction of health information.

[0003] At the same time, the data analysis of existing health management systems focuses on the statistics of single-dimensional data, and fails to effectively integrate user static health baseline (such as age, resting heart rate) and dynamic physiological and behavioral data (such as heart rate fluctuation and step frequency change during exercise), resulting in a lack of personalization in health assessment and insufficient targeting of intervention suggestions or warning information, making it difficult to adapt to the health status and behavior habits of different users.

[0004] In addition, the virtual digital person technology in the existing scheme is mostly used as a display carrier, without deep integration of multi-modal interaction capabilities and intelligent data analysis functions, and cannot realize the closed-loop service of "demand interaction-data collection-analysis feedback". Moreover, due to the hardware performance of smart watches, the digital person model often faces problems such as rendering lag and interaction delay, which restricts the improvement of user experience.

[0005] Therefore, there is an urgent need for an AI digital person health management scheme based on a smart watch. SUMMARY

[0006] The present disclosure provides a personal health service system based on a smart watch digital person, aiming to optimize the digital person interaction experience, strengthen the multi-source data fusion analysis capability, and realize the personalization, intelligentization and naturalization of health management, in order to solve the problems of single interaction, lack of targeting in evaluation and insufficient depth of digital person application in the prior art.

[0007] According to a first aspect of the present disclosure, a personal health service system based on a smart watch digital person is provided, comprising: An intelligent data analysis platform module including a smart watch end and a cloud end, the smart watch end including a digital person interaction module and a multi-source data collection module, the smart watch end and the cloud end realizing bidirectional collaboration through a dynamic encryption link, wherein, The digital person interaction module constructs a multi-modal interaction mechanism based on a virtual digital person, acquires user health-related information through multi-modal perception fusion technology, receives user instructions, and outputs health analysis results in a multi-modal manner; The multi-source data acquisition module cooperates with multiple types of sensors to collect user physiological feature data and behavior activity data in real time, and realizes dynamic alignment and consistency verification of multi-dimensional data through a space-time correlation algorithm. The intelligent data analysis platform module adopts a layered collaborative architecture of edge computing and cloud deep mining to construct a dynamically updated personalized health assessment model, intelligently analyze and process the collected data, and generate a personalized health assessment report and dynamic intervention suggestions.

[0008] As a preferred embodiment, the virtual digital person in the digital person interaction module adopts a parameterized dynamic modeling mechanism and a dynamic image adaptation mechanism, wherein, The parameterized dynamic modeling mechanism constructs a basic 3D model based on the SMPL framework, generates a high-precision initial model through a camera array to collect multi-angle photos of the user, and performs topological simplification and PBR material compression processing on the high-precision initial model to generate a lightweight format model. The dynamic image adaptation mechanism supports automatic adjustment of image style and interaction tone based on user portrait features and use scenarios, and supports user personalized configuration of the appearance of the virtual digital person through a self-defined template.

[0009] As a preferred embodiment, the multi-modal interaction mechanism of the digital person interaction module includes: Voice interaction link: voice recognition, semantic analysis, and voice synthesis are used to realize instruction response and synchronously drive virtual digital person lip movement; Touch and gesture interaction link: pre-set actions are triggered through touch-sensitive operations and bound to the digital person skeleton, while supporting cross-device gesture capture and real-time interaction; Context-aware adjustment: the user's voice instructions, touch operations, and real-time physiological data are analyzed, and when it is detected that the user is in a motion state, the system automatically switches to a simple interaction mode, and when the user is stationary, it switches to a detailed interaction mode.

[0010] As a preferred embodiment, the interface layout of the digital person interaction module includes: Core interaction area: shows the real-time state of the virtual digital person, adapts to real-time rendering requirements through dynamic resolution adjustment technology, optimizes efficiency by combining pre-computed light maps, and ensures complete display of digital person actions; Data display area: adaptively switches the visualization dimension according to the current health theme, and synchronizes content updates with digital person voice feedback; Function navigation area: dynamically sorts function entries based on user historical operation frequency, and supports sliding gesture triggered area switching.

[0011] In a preferred embodiment, the multi-source data acquisition module includes a physiological sensing unit and a behavioral sensing unit, wherein the physiological sensing unit and the behavioral sensing unit adopt a trigger-based collaborative acquisition strategy. When the behavior sensing unit detects that the user has entered a state of motion, it automatically wakes up the physiological sensing unit to increase the sampling frequency; when it detects that the user has entered a resting state, it controls the physiological sensing unit to switch to intermittent sampling mode to reduce power consumption and extend battery life.

[0012] In a preferred embodiment, the multi-source data acquisition module performs spatiotemporal matching analysis on the collected physiological feature data and behavioral activity data through a cross-modal data association verification mechanism. When a logical conflict occurs between the physiological feature data and the behavioral activity data, it automatically triggers an abnormal data marker and triggers a virtual digital human to perform interactive confirmation.

[0013] As a preferred embodiment, the intelligent data analysis platform module adopts a layered collaborative architecture of edge computing and cloud-based deep mining, specifically including: Deploy lightweight algorithm models at the edge computing layer to perform outlier detection and key feature extraction on real-time collected data, and generate an instant health status profile; By integrating the results of edge computing layer processing with the user's historical health data in the cloud-based deep mining layer, a comprehensive health profile is generated through a multi-dimensional health assessment model. At the decision output layer, personalized intervention plans are matched based on the comprehensive health profile, and the plans are transformed into a set of interactive instructions that can be executed by the virtual digital human.

[0014] As a preferred implementation, the intelligent data analysis platform module is also equipped with an adaptive noise reduction mechanism, which uses differentiated preprocessing algorithms for different types of sensor data and achieves feature normalization through statistical standardization methods.

[0015] As a preferred implementation, the dynamic encrypted link between the smartwatch and the cloud adopts a session key dynamic update mechanism. Before each data transmission, a temporary encryption key is generated through device authentication. Furthermore, physiologically sensitive data is desensitized at the edge before transmission, retaining only health-related information.

[0016] In a preferred embodiment, the digital human interaction module and the multi-source data acquisition module form a closed-loop interaction, which can dynamically adjust the digital human interaction strategy according to the real-time collected physiological data, and automatically trigger the early warning interaction process when the user has an abnormal physiological state.

[0017] Compared with the prior art, this disclosure achieves the following beneficial effects: (1) This disclosure uses a rounded square screen to adapt to the virtual digital human for display, which fully presents the digital human's facial expressions and body movements, avoiding the problem of screen cropping; at the same time, it combines multimodal interaction with voice, touch and gesture and gamified interface design, which breaks through the traditional text button interaction mode, enhances the naturalness and immersion of user communication with the system, and supports personalized settings of digital human image, which can meet the user's customization needs.

[0018] (2) This disclosure integrates a nine-axis attitude sensor, an optical sensor, a multi-mode GPS and other multi-source acquisition devices to collect dynamic physiological and behavioral data in a targeted manner. It combines Kalman filtering, wavelet transform and other preprocessing techniques to remove noise interference and ensure data quality. Through sensor fusion and spatiotemporal alignment technology, it further improves the accuracy of motion intensity assessment and fatigue calculation, and provides a reliable data foundation for subsequent analysis.

[0019] (3) This disclosure constructs a fusion analysis architecture of "static data + dynamic data", combines the LSTM model to extract time series features, and generates a personalized health assessment including quantitative results such as fatigue index; based on baseline data such as user age and exercise habits, the analysis conclusions are corrected so that intervention suggestions (such as exercise intensity adjustment) and early warning information (such as abnormal physiological indicators) are more in line with individual needs and avoid the limitations of generalized analysis.

[0020] (4) This disclosure forms a complete service chain of “user needs - data collection - cloud analysis - feedback interaction”. The digital human module runs through the entire process to realize the integration of demand response and result feedback. A new closed-loop optimization step is added. By recording the user’s execution status and subsequent physiological changes, the algorithm model and interaction strategy are iteratively optimized to continuously improve the service accuracy and user adaptability.

[0021] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0022] Figure 1 This diagram illustrates the overall architecture design of a personal health service system based on a smartwatch digital human, according to an embodiment of this disclosure. Figure 2 A reference diagram of a watch with a rounded square screen, according to an embodiment of the present disclosure, is shown. Figure 3 This illustration shows a schematic diagram of the interface development of a personal health service system based on a smartwatch digital human, according to an embodiment of this disclosure. Figure 4 A fatigue index display diagram of a personal health service system based on a smartwatch digital human according to an embodiment of this disclosure is shown. Figure 5 This invention discloses an architecture diagram of an intelligent data analysis platform for a personal health service system based on a smartwatch digital human, according to an embodiment of the present disclosure. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0024] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0025] The following is in conjunction with the appendix Figure 1 - Appendix Figure 5 Taking the typical application scenario of a user querying and obtaining a sports fatigue assessment through a smartwatch as an example, this paper details the specific implementation process of this disclosure.

[0026] like Figure 1 The diagram shown is a schematic of the overall architecture of a personal health service system based on a smartwatch digital human according to an embodiment of this disclosure. The core of the system consists of a smartwatch terminal and a cloud-based intelligent data analysis platform module. The two are connected through a dynamic encrypted link to achieve bidirectional collaboration. The smartwatch terminal further includes a digital human interaction module and a multi-source data acquisition module, forming a closed-loop service link of "interaction-acquisition-analysis-feedback".

[0027] Among them, the digital human interaction module serves as the core interaction entry point between users and the system. It constructs a multimodal interaction mechanism based on virtual digital humans, obtains users' health needs through multimodal perception fusion technology, analyzes users' command intentions, and outputs health analysis results in a multimodal manner. The multi-source data acquisition module relies on multiple types of sensors in the smartwatch hardware layer to collect user physiological characteristics and behavioral activity data in real time, and achieves dynamic alignment and consistency verification of multi-dimensional data through spatiotemporal correlation algorithms. The intelligent data analysis platform module adopts a layered collaborative architecture of "edge computing + cloud deep mining". By building a dynamically updated personalized health assessment model, it intelligently analyzes and processes the collected data to generate personalized health assessment reports and dynamic intervention suggestions.

[0028] In some embodiments, the core of the digital human interaction module is to achieve natural and immersive interaction through virtual digital humans, which includes three parts: interface adaptation, virtual digital human modeling and driving, and the construction of a multimodal interaction link. Interface adaptation includes screen selection and layout design: like Figure 2 As shown, this embodiment uses a rounded square screen as the interactive interface carrier for the smartwatch. The aspect ratio of this screen is close to that of a smartphone screen, and mature mobile UI components can be directly reused, reducing the workload of adaptation. At the same time, its rectangular display area can fully present the facial expressions (such as smiles and neutral expressions) and body movements (such as waving and nodding) of the virtual digital human, avoiding the image edge cropping problem that may occur with circular screens.

[0029] The overall interface layout adopts a "functional partitioning + dynamic adaptation" design, such as... Figure 3 As shown, it is specifically divided into three main areas: Core Interaction Area: Occupying the largest area of ​​the screen, this area displays the user's personalized virtual avatar. Users can set their avatar in two ways: first, by directly selecting from the system's default avatar library; second, by scanning themselves or a designated object using the watch's camera to generate a personalized 3D avatar. This core interaction area also supports dynamic resolution adjustment technology and pre-calculated lightmap optimization, ensuring smooth avatar movements and clear details at a rendering frame rate of 720p@30fps. Data Display Area: Located to the left of the digital human avatar, the visualization content adapts and switches according to the current health theme. For example, in a fatigue assessment scenario, it displays the fatigue index and corresponding suggestions (such as "adjust your breathing and reduce your pace by 10%) in real time), and the content updates are synchronized with the digital human's voice feedback to avoid information gaps. Function navigation area: Located at the top of the screen, it adopts a horizontal navigation bar design and includes four major function entrances: "Default Interface", "Exercise Report", "Health Check Report" and "Settings |". The system will dynamically sort the entrance positions based on the user's historical operation frequency. For example, for users who frequently use "Exercise Report", this entrance will be placed at the top. It also supports swipe gestures to trigger area switching, improving the convenience of operation.

[0030] Virtual digital human modeling and driving: A lightweight modeling and multimodal driving scheme is adopted to ensure smooth interaction under the limited hardware resources of smartwatches. Specifically, Based on the SMPL (Skinned Multi-Person Linear Model) framework, a basic 3D model is built. Users can take at least 30 photos from multiple angles (covering the whole body and facial details) with their mobile phones, upload them to the cloud, and generate an initial model with sub-millimeter precision using camera array scanning technology. Then, the initial model is topologically simplified (the number of vertices is controlled below 5000) and PBR material is compressed (texture resolution is compressed to 1024x1024). Finally, it is exported as a lightweight model in .glb format and transmitted to a smartwatch via Bluetooth or Wi-Fi. Employing a dual-drive mode of "voice + touch," it integrates a TTS (text-to-speech) API to convert text commands into digital human actions, while also supporting touchscreen-triggered preset actions (such as tapping the screen to trigger a "smile" from the digital human). Specifically, it uses Unity's lightweight animation system to bind common interactive actions such as waving, nodding, and lip movements to the digital human's skeleton; for gesture interaction, it can capture user gestures using the phone's camera combined with OpenCV, and transmit them to the watch via WebSocket, achieving real-time cross-device driving.

[0031] Multimodal interaction chain operation: Taking the user's "voice query fatigue index" as an example, the user issues the voice command "query current fatigue index", the multimodal input module on the smartwatch collects the voice signal and transmits it to the ASR (Automatic Speech Recognition) module; The ASR module converts the speech into text, "Query current fatigue index," and transmits it to the NLP (Natural Language Processing) module. The NLP module parses the text intent, determines that the user's need is "fatigue assessment", and simultaneously generates the response text "Collecting data and assessing fatigue index for you", which is then transmitted to the TTS module and the motion parameter extraction module respectively. The TTS module synthesizes the reply text into speech and transmits it to the audio-visual fusion module; at the same time, the motion parameter extraction module extracts lip shape parameters and facial expression parameters (such as "neutral expression") from the reply text and speech and sends them to the graphics module. The graphics module generates digital human lip movements and facial expression animations based on parameters and transmits them to the audio-visual fusion module; The audio-visual fusion module synchronously overlays voice and animation to generate digital human interactive audio-visual data, which is then presented to the user through the multimodal output module to fulfill the user's needs.

[0032] In addition, the module also has context-aware adjustment capabilities: by associating user voice commands, touch operations and real-time physiological data (such as heart rate and accelerometer), if it detects that the user is in motion (such as accelerometer data showing a step frequency ≥ 120 steps / minute), it automatically switches to a simple interaction mode (only voice broadcast of core results + short text); if the user is stationary (accelerometer data shows no significant fluctuations), it switches to a detailed interaction mode (displaying a complete evaluation report + suggestions for a dynamic demonstration of the digital human).

[0033] In some embodiments, the multi-source data acquisition module relies on the sensor combination built into the smartwatch and adopts a triggered collaborative acquisition + cross-modal verification strategy to ensure accurate and efficient data acquisition. To acquire motion-related dynamic data, a six-axis or nine-axis attitude sensor can be selected. This embodiment uses a nine-axis attitude sensor, and the combination and division of labor are shown in Table 1 below: Table 1

[0034] The nine-axis attitude sensor (IMU) includes a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer. The accelerometer collects 3-axis motion acceleration at a frequency of ≥100Hz for analyzing step frequency and stride length; the gyroscope collects 3-axis angular velocity at a frequency of ≥50Hz for detecting arm stability; and the magnetometer is used for direction sensing to improve the accuracy of motion trajectory. Optical sensor: Includes an optical heart rate monitor and a blood oxygen sensor, based on PPG (photoplethysmography) technology - the transmitting LED emits green light (or red / infrared light), which penetrates the skin, and the receiving photodiode captures the light reflected from the capillaries to calculate the real-time heart rate (collection frequency ≥1Hz) and blood oxygen saturation (intermittent measurement). Multi-mode GPS module: Supports GPS+GLONASS+BeiDou multi-system positioning, collects location data at a frequency of ≥1Hz, converts it into travel distance using the Havesing formula, and calculates real-time pace using timestamps to assess exercise intensity.

[0035] When the nine-axis attitude sensor detects that the user has entered a state of motion (such as the 3-axis acceleration fluctuation frequency matching the running characteristics), it automatically wakes up the optical sensor, increases the heart rate acquisition frequency from "intermittent" to ≥1Hz, and at the same time starts the multi-mode GPS module to continuously collect location data. When the sensor detects that the user has entered a resting state (e.g., acceleration fluctuation ≤ threshold, GPS position unchanged for 5 minutes), it controls the optical sensor to switch back to intermittent sampling mode (heart rate is collected once every 5 minutes) and turns off the GPS module to reduce power consumption.

[0036] Furthermore, to ensure the validity of the collected data, the raw data needs to undergo cross-modal correlation verification and noise reduction processing. Specifically, Cross-modal verification: The collected physiological data and behavioral data are matched and analyzed through spatiotemporal correlation algorithms. For example, if the GPS shows that the user is running (pace 6 min / km), but the heart rate collected by the optical sensor is only 60 bpm (below the normal heart rate range during exercise), the system will determine that there is a logical conflict in the data, automatically mark the abnormal data, and trigger digital human interaction confirmation (such as the digital human asking "Are you in an exercise state?"). Data denoising: Kalman filtering is used to denoise IMU motion data (acceleration, angular velocity) through a "prediction-update" loop (based on state prediction equation, observation equation, and Kalman gain calculation) to eliminate sensor electronic noise and interference from arm tremors; wavelet transform is used to denoise PPG heart rate signals by performing multi-level decomposition of the signal through a mother wavelet function (such as db6), and reconstructing the signal after soft thresholding of high-frequency noise coefficients to ensure stable heart rate data.

[0037] In some embodiments, the cloud-based intelligent data analysis platform module includes an edge computing layer, a cloud-based deep mining layer, and a decision output layer. These three layers work together to complete fatigue assessment. The specific implementation process is as follows: First, data preprocessing is performed: The smartwatch transmits the preprocessed dynamic data (60 seconds of sensor data, including acceleration, angular velocity, heart rate, GPS pace, etc., in a 60×8 matrix format) and the user's static data (age, gender, resting heart rate, VO2max, running experience, such as [30,0,58,45,5]) to the cloud via a dynamic encrypted link. The dynamic encrypted link uses a session key dynamic update mechanism. Before each transmission, a temporary key is generated through device authentication. Physiologically sensitive data (such as the original heart rate waveform) is desensitized at the edge, retaining only health characteristic information such as heart rate value and heart rate variability. Secondly, the preprocessed data is standardized: static data is directly standardized using the formula " ( The original value, The mean, Standardization is performed for the standard deviation, and dynamic data is also standardized after denoising to ensure that the feature scales of static and dynamic data are consistent.

[0038] Next, the edge computing layer performs preliminary analysis on the real-time collected dynamic data (such as heart rate and cadence). If a heart rate ≥180 bpm (exceeding the safety threshold) is detected, an instant health profile of "abnormal heart rate" is immediately generated, triggering an alert through the digital human without waiting for cloud analysis. The cloud deep mining layer receives the processing results from the edge computing layer, integrates the user's historical health data (such as changes in fatigue index and exercise load over the past 7 days), and extracts the temporal features of the dynamic data through an LSTM model. First, a 64-unit LSTM layer captures short-term patterns (such as heart rate fluctuations within 10 seconds), and then a 32-unit LSTM layer extracts high-level temporal features (such as cadence change trends within 60 seconds), outputting a 32-dimensional feature vector. This vector is then concatenated with standardized static data (5 dimensions) to form a 37-dimensional fused feature. The decision output layer inputs the aforementioned fused features into the fully connected layer, outputs a probability value of 0-1 through the sigmoid activation function, and then calculates the fatigue index (FI) using the formula "FI = 10 × probability value". For example, when the fully connected layer outputs 0.63, FI = 6.3, corresponding to a "moderate to high fatigue" level. At the same time, based on the user's static data (such as age 30 and running experience of 5 years), suggestions are revised to generate a personalized intervention plan of "immediately slow down and replenish water", and the plan is converted into interactive commands (such as voice text and facial expression parameters) that can be executed by the digital human.

[0039] Finally, the cloud transmits the fatigue index (FI=6.3), assessment level (moderate to high fatigue), and intervention suggestions to the smartwatch, such as... Figure 4 As shown, the digital human interaction module provides feedback in a multimodal form - the voice broadcast says "Current fatigue index 6.3, moderate to high fatigue, it is recommended to slow down immediately and replenish water", the data display area displays the FI value and the suggested text, and the digital human presents a "slightly frowning" expression to enhance the information delivery; To achieve closed-loop optimization and improve service accuracy, the system further records the user's execution of suggestions (such as whether they slowed down) and changes in physiological state over the following 30 minutes (such as whether heart rate decreased or FI decreased after slowing down). The system feeds back the correlation data between "execution behavior and physiological changes" to the cloud. Based on this data, the cloud iteratively optimizes the feature weights of the LSTM model (such as increasing the feature importance of "heart rate drift rate") and adjusts the digital human interaction strategy (such as increasing the frequency of voice reminders for users who have not executed suggestions in the next feedback), continuously improving service accuracy.

[0040] Example 2: To further illustrate the universality and adaptability of the technical solution disclosed herein, Example 2 is provided below, taking the scenario of "users obtaining sleep health assessments through smartwatches" as an example, to elaborate on the specific implementation process of the system in non-sports scenarios.

[0041] This embodiment aims to demonstrate the system's entire process of interaction, data collection, analysis, and feedback in a static (sleep) scenario, with the user's goal of "checking their sleep health status over the past 7 days via a smartwatch," further verifying the system's adaptability to multiple health scenarios.

[0042] The core user requirement is to "obtain the sleep structure (deep sleep / light sleep / REM sleep duration), sleep quality score and improvement suggestions for the past 7 days". The system response logic follows the path of "receiving the request - retrieving historical data - multi-dimensional analysis - personalized feedback". The digital human interaction module leads the interaction process, the multi-source data collection module focuses on organizing historical data, and the cloud-based intelligent data analysis platform module focuses on mining sleep characteristics. The three work together to achieve sleep health assessment.

[0043] First, the user triggers a request via touch interaction on the smartwatch, tapping the dialog box button at the bottom of the screen and entering "request for sleep report of the past 7 days" in the text input box, or inputting the same command via voice (multimodal interaction is optional). The processing flow of the digital human interaction module is as follows: If the input is text, it is directly transmitted to the NLP module; if the input is voice, it is first converted into text "Query sleep report for the past 7 days" by the ASR module. The NLP module parses the text intent, determines the request type as sleep health assessment, confirms the need to retrieve the historical sleep data of the past 7 days, generates the response text "Retrieving your sleep data of the past 7 days and generating a sleep report for you", and simultaneously transmits it to the TTS module and the motion parameter extraction module; Based on the above response text, the motion parameter extraction module extracts the virtual digital human's interaction parameters (such as "gentle expression" and "slow nodding action"), and drives the virtual digital human to respond with natural posture, avoiding stiff interaction caused by the static attributes of the "sleep assessment" scenario.

[0044] Since the user is in a static scenario (typically in a non-moving state during queries), the digital human interaction module automatically switches to a detailed interaction mode, and the interface layout is optimized and adapted to the sleep scenario based on the framework of Implementation Example 1: Core Interaction Area: The virtual digital human maintains a "sitting posture + gentle expression" and avoids large dynamic body movements, which is in line with the calm tone of the sleep scene; at the same time, through pre-calculated light mapping technology, the screen brightness is automatically reduced in low-light environments (such as nighttime queries) to reduce visual interference; Data display area: Adaptively switches to the "Sleep Structure Visualization" interface, using a pie chart to show the percentage of deep sleep (e.g., 25%), light sleep (e.g., 55%), and REM sleep (e.g., 20%) over the past 7 days, and a line chart to show the daily sleep quality score (1-10 points) trend. Data updates are synchronized with digital human voice narration (e.g., "Your average deep sleep duration over the past 7 days was 2.1 hours, an increase of 0.3 hours compared to last week"). Function Navigation Area: A temporary "Sleep Details" entry has been added. Users can click on it to view the specific duration of each sleep stage each day (e.g., "2.3 hours of deep sleep and 5.1 hours of light sleep on October 1st"). The entry is dynamically sorted based on "relevance to the current scenario" and will automatically revert to the default sorting after use.

[0045] The multi-source data acquisition module relies on the following sensors built into the smartwatch to automatically collect data during the user's sleep, with specific parameters adapted to sleep scenario requirements: Nine-axis attitude sensor: Reduce the acquisition frequency to 10Hz (no need for high-frequency acquisition in motion scenarios), capture the user's nighttime turning movements through a 3-axis accelerometer (to determine sleep depth, such as turning ≤1 time / hour for deep sleep, ≥5 times / hour for light sleep), and use a 3-axis gyroscope to assist in recognizing sleeping posture (supine / side-lying / prone). Optical sensor: It adopts an "intermittent + trigger" sampling mode, collecting heart rate once every 30 minutes (heart rate is usually low during sleep, so high-frequency monitoring is not required). When abnormal heart rate fluctuations are detected (such as a sudden increase of ≥20 bpm), the sampling frequency is temporarily increased to 1 Hz, and the fluctuation time point is recorded (correlated with whether it is an awake state). Ambient light sensor (supplementary sensor, adapted for sleep scenarios): Collects ambient light intensity at night to determine if there is a "light interference with sleep" problem (e.g., light intensity ≥50 lux for 10 minutes may affect REM sleep).

[0046] Next, to ensure the validity of the data used for analysis, the multi-source data acquisition module performs cross-modal verification and data integration on the collected raw sleep data. Specifically, By matching "posture data - heart rate data" using a spatiotemporal correlation algorithm, if the nine-axis posture sensor shows that the user has not turned over for 1 hour (determined to be in deep sleep), but the heart rate fluctuation collected by the optical sensor is ≥15 bpm (not in line with the heart rate stability during deep sleep), the system marks the data for that period as "abnormal" and provides a prompt through the digital human's subsequent feedback: "The data from 2-3 am on October 3rd is interfered with. It is recommended to maintain a stable sleep environment." Data integration: Sensor data from the past 7 days is categorized by "date-sleep period" and core sleep features (such as total sleep duration, deep sleep percentage, turning frequency, heart rate variability, and ambient light duration) are extracted to form a "daily sleep feature matrix" (formatted as 7×5, corresponding to 7 days × 5 feature categories). After edge-end desensitization (removing the precise timestamps of specific sleep periods and retaining only time period labels such as "22:00-6:00 the next day"), the data is uploaded to the cloud via a dynamically encrypted link.

[0047] The cloud-based intelligent data analysis platform module receives the "daily sleep feature matrix" and user static data uploaded from the smartwatch. It uses "statistical standardization" on the sleep feature data (such as converting the total sleep duration into "deviation rate relative to the user's historical mean") and "neighboring day mean imputation" on outlier data (such as replacing the marked "data from the early morning of October 3rd" with the mean of data from the same period on October 2nd and October 4th) to avoid outliers affecting the analysis results.

[0048] A "sleep stage determination rule base" is deployed at the edge computing layer of the intelligent data analysis platform module. Based on the "turning over frequency - heart rate" threshold, the duration of each sleep stage is initially determined each day, and a "basic sleep profile" is generated. In the cloud-based deep mining layer of the intelligent data analysis platform module, basic sleep profiles and user static data are integrated to calculate core evaluation indicators, including sleep quality score (calculated based on total sleep duration, deep sleep percentage, and number of awakenings, with a maximum score of 10), sleep disturbance factors (such as ambient light duration ≥1 hour, turning over frequency ≥6 times / hour), and sleep trend (the slope of the change in the percentage of deep sleep in the past 7 days to determine whether there is improvement). Finally, based on the above analysis results, a "sleep intervention suggestion library" was matched. For example, for "low percentage of deep sleep + interference from ambient light", the suggestion was "keep the bedroom dark and turn off electronic devices 1 hour before bedtime to increase the percentage of deep sleep"; for "high frequency of turning over (7 times / hour)", the suggestion was "adjust the sleeping position to side lying and use a buckwheat pillow to support the neck".

[0049] The cloud transmits the obtained sleep quality score, deep sleep percentage, interfering factors, and improvement suggestions to the smartwatch. The digital human interaction module provides feedback in the following ways: voice broadcast of core conclusions (e.g., "Your sleep quality has been good for the past 7 days, and the percentage of deep sleep is slightly lower than historically, mainly due to light interference"), data display area showing pie charts and trend charts, and digital human demonstrating the "side sleeping posture" to enhance the intuitiveness of the suggestions. To achieve closed-loop optimization of the system and improve service accuracy, the system also records the user's implementation of the suggestions (such as detecting "whether to reduce light exposure 1 hour before bedtime" through an ambient light sensor) and changes in sleep data over the next 7 days (such as whether the percentage of deep sleep has increased to 27%). The system feeds back the data linking "implementation behavior - sleep improvement effect" to the cloud, iteratively optimizing the "sleep intervention suggestion library" (such as prioritizing similar suggestions for users whose deep sleep improves after implementing "reduce light exposure"). At the same time, the system adjusts the digital human feedback strategy (such as adding an explanation of "the dangers of light interference" to the next feedback for users who have not implemented the suggestions).

[0050] According to the above embodiments of this disclosure, the following technical effects are achieved: (1) This disclosure uses a rounded square screen to adapt to the virtual digital human for display, which fully presents the digital human's facial expressions and body movements, avoiding the problem of screen cropping; at the same time, it combines multimodal interaction with voice, touch and gesture and gamified interface design, which breaks through the traditional text button interaction mode, enhances the naturalness and immersion of user communication with the system, and supports personalized settings of digital human image, which can meet the user's customization needs.

[0051] (2) This disclosure integrates a nine-axis attitude sensor, an optical sensor, a multi-mode GPS and other multi-source acquisition devices to collect dynamic physiological and behavioral data in a targeted manner. It combines Kalman filtering, wavelet transform and other preprocessing techniques to remove noise interference and ensure data quality. Through sensor fusion and spatiotemporal alignment technology, it further improves the accuracy of motion intensity assessment and fatigue calculation, and provides a reliable data foundation for subsequent analysis.

[0052] (3) This disclosure constructs a fusion analysis architecture of "static data + dynamic data", combines the LSTM model to extract time series features, and generates a personalized health assessment including quantitative results such as fatigue index; based on baseline data such as user age and exercise habits, the analysis conclusions are corrected so that intervention suggestions (such as exercise intensity adjustment) and early warning information (such as abnormal physiological indicators) are more in line with individual needs and avoid the limitations of generalized analysis.

[0053] (4) This disclosure forms a complete service chain of “user needs - data collection - cloud analysis - feedback interaction”. The digital human module runs through the entire process to realize the integration of demand response and result feedback. A new closed-loop optimization step is added. By recording the user’s execution status and subsequent physiological changes, the algorithm model and interaction strategy are iteratively optimized to continuously improve the service accuracy and user adaptability.

[0054] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0055] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0056] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0057] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A personal health service system based on a smartwatch digital human, characterized in that, The system includes a smart watch app and a cloud-based intelligent data analysis platform module. The smart watch app includes a digital human interaction module and a multi-source data acquisition module. The smart watch app and the cloud achieve bidirectional collaboration through a dynamically encrypted link. The digital human interaction module is based on a virtual digital human to build a multimodal interaction mechanism. It acquires user health-related information through multimodal perception fusion technology, receives user commands, and outputs health analysis results in a multimodal manner. The multi-source data acquisition module works in collaboration with multiple types of sensors to collect user physiological characteristic data and behavioral activity data in real time, and uses a spatiotemporal correlation algorithm to achieve dynamic alignment and consistency verification of multi-dimensional data. The intelligent data analysis platform module adopts a layered collaborative architecture of edge computing and cloud-based deep mining to build a dynamically updated personalized health assessment model, intelligently analyze and process the collected data, and generate personalized health assessment reports and dynamic intervention suggestions.

2. The system according to claim 1, characterized in that, The virtual digital human in the digital human interaction module adopts a parametric dynamic modeling mechanism and a dynamic image adaptation mechanism, wherein, The parametric dynamic modeling mechanism is based on the SMPL framework to build a basic 3D model. It generates a high-precision initial model by acquiring multi-angle photos of the user through a camera array, and performs topology simplification and PBR material compression on the high-precision initial model to generate a lightweight format model. The dynamic image adaptation mechanism supports automatically adjusting the image style and interactive tone based on user profile characteristics and usage scenarios, and also allows users to personalize the appearance of the virtual digital human through custom templates.

3. The system according to claim 2, characterized in that, The multimodal interaction mechanism of the digital human interaction module includes: Voice interaction link: Command response is achieved through speech recognition, semantic parsing and speech synthesis, and the virtual digital human's lip movements are driven synchronously; Touch and gesture interaction link: Preset actions are triggered by touch-sensitive operations and bound to the digital human skeleton, while supporting cross-device gesture capture and real-time interaction; Context-aware mode: It correlates and analyzes the user's voice commands, touch operations and real-time physiological data, and automatically switches to a simple interaction mode when the user is in motion, and switches to a detailed interaction mode when the user is stationary.

4. The system according to claim 3, characterized in that, The interface layout of the digital human interaction module includes: Core interactive area: Displays the real-time status of the virtual digital human, adapts to real-time rendering requirements through dynamic resolution adjustment technology, and optimizes efficiency by combining pre-calculated light maps to ensure complete display of the digital human's movements; Data display area: The visualization dimensions are automatically switched according to the current health theme, and the content updates are synchronized with the digital human's voice feedback; Function navigation area: Function entries are dynamically sorted based on the user's historical operation frequency, and support switching of areas triggered by swipe gestures.

5. The system according to claim 1, characterized in that, The multi-source data acquisition module includes a physiological sensing unit and a behavioral sensing unit, which employ a trigger-based collaborative acquisition strategy. When the behavior sensing unit detects that the user has entered a state of motion, it automatically wakes up the physiological sensing unit to increase the sampling frequency; when it detects that the user has entered a resting state, it controls the physiological sensing unit to switch to intermittent sampling mode to reduce power consumption and extend battery life.

6. The system according to claim 5, characterized in that, The multi-source data acquisition module performs spatiotemporal matching analysis on the collected physiological feature data and behavioral activity data through a cross-modal data association verification mechanism. When a logical conflict occurs between the physiological feature data and the behavioral activity data, it automatically triggers an abnormal data marker and triggers the virtual digital human to perform interactive confirmation.

7. The system according to claim 6, characterized in that, The intelligent data analysis platform module adopts a layered collaborative architecture of edge computing and cloud-based deep mining, specifically including: Deploy lightweight algorithm models at the edge computing layer to perform outlier detection and key feature extraction on real-time collected data, and generate an instant health status profile; By integrating the processing results of the edge computing layer with the user's historical health data in the cloud-based deep mining layer, a comprehensive health profile is generated through a multi-dimensional health assessment model. At the decision output layer, personalized intervention plans are matched based on the comprehensive health profile, and the plans are transformed into a set of interactive instructions that can be executed by the virtual digital human.

8. The system according to claim 7, characterized in that, The intelligent data analysis platform module is also equipped with an adaptive noise reduction mechanism, which uses differentiated preprocessing algorithms for different types of sensor data and achieves feature normalization through statistical standardization methods.

9. The system according to claim 1, characterized in that, The dynamic encrypted link between the smartwatch and the cloud adopts a session key dynamic update mechanism. Before each data transmission, a temporary encryption key is generated through device authentication. Furthermore, physiologically sensitive data is desensitized at the edge before transmission, retaining only health-related information.

10. The system according to claim 1, characterized in that, The digital human interaction module and the multi-source data acquisition module form a closed-loop interaction, which can dynamically adjust the digital human interaction strategy according to the real-time collected physiological data, and automatically trigger the early warning interaction process when the user has an abnormal physiological state.

Images