Non-contact health monitoring method and system based on vision and radar dual perception

Abstract

The invention provides a non-contact health monitoring method and system based on vision and radar dual perception, and relates to the technical field of health monitoring, and the method comprises the steps: arranging a host in a first region of a target region, arranging slaves in a plurality of second regions respectively, obtaining a slave set, and completing registration and binding through a cloud platform; respectively calling a host dual perception data sequence of a radar module and a camera module of the host in a preset monitoring window for the target area and a slave function perception data sequence set of the slave set in the preset monitoring window; determining dual sensing results of the host; determining a slave function perception result set; and fusing to obtain a non-contact health monitoring result. The technical problem that health monitoring is not comprehensive and accurate enough due to the fact that health monitoring equipment in the prior art is generally single in function, depends on wearable equipment and is low in health state recognition accuracy is solved.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically to a non-contact health monitoring method and system based on dual visual and radar sensing. Background Technology

[0002] As society ages, health monitoring issues are becoming increasingly prominent, especially for high-risk groups such as the elderly, those living alone, and patients with chronic diseases. Traditional health monitoring usually relies on wearable devices or single cameras or radars, but these traditional technologies have some significant limitations.

[0003] Specifically, wearable devices require users to wear them and rely on continuous battery power, which not only causes inconvenience but also leads to inaccurate wearing or forgetting to wear them. Furthermore, the monitoring range of wearable devices is limited to specific body parts, failing to effectively monitor the user's overall health, especially for groups who cannot directly wear the devices, such as disabled patients. When used alone, cameras and radar perform unsatisfactorily in health monitoring. Cameras are ineffective in low-light environments and can easily infringe on privacy, posing privacy risks when used in sensitive locations such as restrooms, bathrooms, and bedrooms. While radar can provide spatial data, it is prone to false alarms in complex environments. Therefore, a single radar or camera cannot meet the high-precision health monitoring needs in diverse application scenarios.

[0004] Furthermore, traditional camera and radar technologies often rely on traditional algorithms or simple pattern recognition methods for processing and analyzing perceived data, lacking support from advanced AI algorithms. AI algorithms, especially deep learning and machine learning algorithms, have strong advantages in data analysis and feature extraction. However, traditional technologies often fail to fully utilize AI algorithms to intelligently process perceived data, resulting in insufficient accuracy and reliability of health monitoring results, making it difficult to meet the needs of high-precision and comprehensive health monitoring. Summary of the Invention

[0005] This application provides a non-contact health monitoring method and system based on dual vision and radar perception, aiming to solve the technical problems that existing health monitoring devices are usually single-function, rely on wearable devices, and have low accuracy in identifying health status, resulting in insufficient comprehensiveness and precision in health monitoring.

[0006] The first aspect disclosed in this application provides a non-contact health monitoring method based on dual vision and radar perception. The method includes: deploying a host in a first area of ​​a target region and deploying slave devices in multiple second areas to obtain a slave device set; registering and binding the host and slave device set through a cloud platform, wherein each slave device has a functional identifier, including an attitude monitoring identifier and a respiratory and heart rate monitoring identifier; retrieving the dual-sensing data sequence of the host in the target region from the radar module and the camera module of the host in a preset monitoring window, and retrieving the slave device set in the preset monitoring window from the slave device set in the multiple second areas; calling the multi-scale attitude sensor of the host computing unit in the computing module to analyze the dual-sensing data sequence of the host and determine the dual-sensing result of the host; calling the set of slave computing units in the computing module to analyze the set of slave device functional sensing data sequences and determine the set of slave device functional sensing results; and fusing the dual-sensing result of the host and the set of slave device functional sensing results to obtain a non-contact health monitoring result.

[0007] The second aspect of this application discloses a non-contact health monitoring system based on dual vision and radar perception. This system is used in the aforementioned non-contact health monitoring method based on dual vision and radar perception. The system includes: a host deployment module, used to deploy a host in a first area of ​​the target area and slave devices in multiple second areas to obtain a slave device set; and to complete the registration and binding of the host and slave device set through a cloud platform. Each slave device has a functional identifier, including an attitude monitoring identifier and a respiratory and heart rate monitoring identifier. A perception data retrieval module is also provided, used to retrieve data from the radar module and camera module of the host within a preset monitoring window for the target area. The system comprises: a host dual-sensing data sequence and a set of slave functional sensing data sequences for the multiple second areas within a preset monitoring window; a host data analysis module, used to call the multi-scale attitude sensor of the host computing unit in the computing module to analyze the host dual-sensing data sequence and determine the host dual-sensing result; a slave data analysis module, used to call the set of slave computing units in the computing module to analyze the set of slave functional sensing data sequences and determine the set of slave functional sensing results; and a monitoring result acquisition module, used to fuse the host dual-sensing result and the set of slave functional sensing results to obtain a non-contact health monitoring result.

[0008] One or more technical solutions provided in this application have at least the following beneficial effects: By deploying the main unit in the first area of ​​the target region and slave units in multiple second areas, the system ensures coverage of all key locations within the monitored area. The slave unit deployment allows each unit to focus on its specific function, achieving comprehensive spatial monitoring. The main unit simultaneously collects dual-sensor data via radar and camera modules. Radar signals provide spatial information, while camera modules provide image data to assist in human identification. This dual-sensor approach enhances data accuracy and reduces potential misjudgments or blind spots from a single sensor. The slave units collect functional sensing data in different areas, each focusing on a small monitoring area, thus improving the system's monitoring efficiency and accuracy across different regions. The main unit analyzes the dual-sensor data using a multi-scale attitude sensor. It can identify human posture from multiple angles. This multi-scale processing can improve the system's ability to recognize human posture at different distances and angles. Through comprehensive analysis of radar and visual data, it can more accurately detect posture changes, especially effective in identifying falls and abnormal behaviors, reducing false alarm rates. The slave unit focuses on collecting and analyzing data for specific functions. This focused design can accurately determine the health status of the area where the slave unit is located. Different slave unit computing units perform directional analysis based on the slave unit's functional identifiers, improving monitoring accuracy. By fusing the dual sensing results of the host unit with the functional sensing results of the slave unit, the resulting non-contact health monitoring results greatly reduce the probability of false alarms and missed alarms, and can provide more accurate and comprehensive status warnings.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the non-contact health monitoring method based on dual vision and radar perception provided in the embodiments of this application.

[0011] Figure 2 This is a schematic diagram of the host structure in the non-contact health monitoring method based on dual vision and radar perception provided in the embodiments of this application.

[0012] Figure 3 This is a schematic diagram of the structure of a non-contact health monitoring system based on dual vision and radar perception, provided in an embodiment of this application.

[0013] Explanation of reference numerals in the attached diagram: 10 for host deployment module, 20 for sensing data retrieval module, 30 for host data analysis module, 40 for slave data analysis module, 50 for monitoring result acquisition module, 1 for housing base, 2 for first motor, 3 for bracket, 4 for radar motherboard, 5 for second motor, 6 for camera, 7 for speaker, 8 for charging motherboard, and 9 for camera motherboard. Detailed Implementation

[0014] This application provides a non-contact health monitoring method and system based on dual vision and radar perception. It solves the technical problem that existing health monitoring devices are usually single-function and rely on wearable devices, resulting in low accuracy in identifying health status and thus insufficient comprehensiveness and precision in health monitoring. This application is only used for health monitoring and status assessment, and aims to provide non-contact health data collection and monitoring services, without involving specific interventions on health problems.

[0015] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0016] Example 1, as Figure 1 As shown, this application provides a non-contact health monitoring method based on dual vision and radar perception, the method comprising: The host is deployed in the first area of ​​the target area, and slave devices are deployed in multiple second areas to obtain a slave device set. The host and slave device set are registered and bound through the cloud platform. Each slave device has a functional identifier, which includes an attitude monitoring identifier and a respiratory and heart rate monitoring identifier.

[0017] The main unit is deployed in the first area of ​​the target region, which is the area requiring maximum monitoring coverage, such as the living room or other main living areas. The main unit should be placed in a location that maximizes the monitoring range, ensuring coverage by both radar and camera modules. The main unit uses dual-sensing technology of vision and radar to perform real-time health monitoring in this area. Multiple slave units are deployed in several second areas, such as bathrooms, kitchens, and other fragmented or special activity areas. Each slave unit has a functional identifier, including a posture monitoring identifier to monitor changes in the user's posture, such as whether they have fallen, and a heart rate monitoring identifier to monitor the user's breathing and heart rate. To protect user privacy and avoid the risk of privacy violations from cameras in privacy-sensitive areas (such as bathrooms, showers, and bedrooms), the slave units are not equipped with cameras in these areas, but rather with radar modules. These radar modules can monitor the user's health status, including changes in breathing, heart rate, and posture, without contact, thus avoiding the privacy leaks that cameras might cause.

[0018] The cloud platform serves as the central hub of the system, managing the status of hosts and slaves, data transmission, device registration and updates, and other operations. Each device (host and slave) registers with the cloud platform to ensure that the system can effectively monitor and manage the status of each device.

[0019] The radar module and camera module of the host are respectively retrieved in the preset monitoring window to obtain the host dual-sensing data sequence of the target area, and the slave set is retrieved in the preset monitoring window to obtain the slave functional sensing data sequence set of the multiple second areas.

[0020] The host system collects dual-sensor data of the target area in real time through its radar and camera modules. Specifically, the radar module detects human posture and physiological signals, such as heartbeat and respiration. Radar signals can penetrate obstacles, have strong anti-interference capabilities, and are unaffected by changes in ambient light, thus enabling stable operation around the clock. The camera module collects image and video data of the target area, particularly for posture recognition and body movement monitoring, such as falls, walking, and stillness. The camera uses image recognition technology to assist the radar signal, providing more visual information and enhancing the system's perception capabilities. The host system collects radar and video data in real time within a preset monitoring window and generates a dual-sensor data sequence containing health status information within the target area.

[0021] Each slave device collects data according to its function identifier. The radar module of the slave device detects attitude changes in the target area in real time and transmits data sequences to the host. The radar module monitors the user's heart rate and breathing status. These slave device data sequences contain physiological state data of the human body in the target area. These slave device data sequences are also collected within a preset monitoring window and form a data set as a slave device function perception data sequence set. The host receives and processes the slave device function perception data sequence set from multiple slave devices.

[0022] The multi-scale attitude sensor of the host computing unit in the computing module is invoked to analyze the host dual-sensing data sequence and determine the host dual-sensing result.

[0023] The host computer's computing module includes a multi-scale posture sensor for analyzing dual-sensor data sequences. "Multi-scale" refers to the sensor's ability to analyze data at multiple levels and scales, processing data across different time dimensions to perform real-time analysis and judgment of human movement changes. For example, if the host computer's radar module detects a person's movement has stopped, and the camera has not captured any other motion information, the data from both is combined to determine whether the person has fallen or is in an abnormal posture. Based on the multi-scale posture sensor analysis, the host computer's computing unit combines the data from both (radar and vision) to ultimately generate the host computer's dual-sensor results.

[0024] The set of slave computing units in the computing module is called to analyze the set of slave function perception data sequences respectively, and the set of slave function perception results is determined.

[0025] The slave computing unit analyzes the slave functional perception data sequence set. Specifically, for attitude monitoring slaves, the radar data transmitted by the slaves is analyzed to determine if there are abnormal posture changes such as falls. The computing unit judges whether the slave's radar signal shows a sudden change (e.g., instantaneous fall) and makes a comprehensive judgment by combining data from other slaves (e.g., abnormal heartbeat). For respiratory and heart rate monitoring slaves, the physiological data transmitted by the slaves (e.g., heart rate, respiratory rate) is further processed by the host computing unit. Abnormal fluctuations in the data (e.g., fast or slow heartbeat, respiratory arrest, etc.) are identified to determine if there is a health risk. If a slave detects an abnormality, a corresponding warning message is generated. Finally, the analysis results of all slaves are integrated into a slave functional perception result set.

[0026] By integrating the dual sensing results of the host and the functional sensing results of the slave, a non-contact health monitoring result is obtained.

[0027] The sensing results from the master and slave devices are merged according to certain rules. Specifically, if the master's posture recognition system detects a fall while the slave's heartbeat monitoring system shows cardiac arrest, the combined data confirms that the situation requires emergency handling. Similarly, if the master does not detect a fall, but the slave shows respiratory arrest or abnormal heartbeat, an alert will also be triggered. The fused data from the master and slave devices provides comprehensive contactless health monitoring results, which trigger corresponding alert mechanisms to notify the user or their guardian of relevant health issues.

[0028] Furthermore, the multi-scale attitude sensor of the host computing unit in the computing module is invoked to analyze the host dual-sensing data sequence and determine the host dual-sensing result, including: The initial attitude frame from the multi-scale attitude sensor is retrieved; the host radar perception data subsequence and the host visual perception data subsequence from the host dual-sensing data sequence are respectively input into the initial attitude frame to obtain the host radar perception attitude frame subsequence and the host visual perception attitude frame subsequence; the host radar perception attitude frame subsequence is traversed to perform attitude iterative analysis to determine the host radar perception result; the host visual perception attitude frame subsequence is traversed to perform attitude iterative analysis to determine the host visual perception result; the host radar perception result and the host visual perception result are cross-confirmed to obtain the host dual-sensing result.

[0029] The initial pose framework is a data modeling framework that performs preliminary structuring and parsing of pose data. The role of this initial pose framework is to transform the input data into a standardized format, enabling the system to effectively perform subsequent analysis. This framework includes various parameters, including key human pose points (such as head, limbs, torso, etc.) and related actions (such as standing, sitting, walking, falling, etc.).

[0030] The host radar perception data subsequence is information such as human distance, speed, and direction acquired through radar sensors; the host visual perception data subsequence is image data acquired through cameras, including information such as human appearance and posture changes. Both the radar perception data subsequence and the visual perception data subsequence are input into the initial posture framework for processing. The radar perception data subsequence, after preliminary analysis, yields information containing different posture changes, i.e., the host radar perception posture framework subsequence. These data points include changes in human position, timestamps of posture transitions, and motion trajectories. The host visual perception data subsequence generates a subsequence containing visual features, i.e., the host visual perception posture framework subsequence, such as key points of posture changes (e.g., shoulders, knees, ankles) and judgment results for different postures (e.g., standing, sitting, bending over).

[0031] The purpose of traversing the host radar's attitude perception framework subsequences is to determine the specific posture of the human body by analyzing these data points. Iterative analysis refers to improving recognition accuracy by continuously adjusting and optimizing the analysis model. For radar perception data, time series analysis methods can be used to analyze the data's changing trends, speed, and direction to identify the specific posture. During the analysis process, the data is progressively iteratively processed to eliminate noise, correct errors, and enhance the effective parts of the signal. Through attitude iterative analysis, posture changes in the host radar's attitude perception framework subsequences are classified and identified. For example, if a person suddenly changes from standing to lying down, it is identified as a fall; if it is a slow sitting motion, it is identified as sitting down, etc. The final output is the host radar perception result, that is, the specific posture and action inferred from the radar data.

[0032] Similar to the host radar's attitude perception frame subsequence, the host visual perception attitude perception frame subsequence is traversed to analyze attitude changes and movement patterns. For visual data, the analysis process relies on image processing and deep learning algorithms, especially human pose recognition models. For example, key features in images are extracted through models such as convolutional neural networks to analyze attitude changes. The visual data is also iteratively adjusted to optimize the accuracy of action recognition. After iterative analysis, the host visual perception result is determined, that is, the attitude inferred from the visual data. This is similar to the output of radar data, but visual data usually provides more attitude details and changes.

[0033] Cross-verification improves the accuracy of posture recognition by comparing data from two different sources (radar and vision). The radar perception results and the vision perception results of the host computer are compared. If the postures determined by the two are consistent, such as both determining that the user is standing, then the result is confirmed to be correct. If there is a contradiction between the radar and vision data, such as the radar determining that the user is standing while the vision determines that the user has fallen, reasonable adjustments are made according to the algorithm's preset priority. For example, vision is given priority in certain situations. Through cross-verification, the host computer obtains a dual perception result, which includes a higher confidence posture determination, thus improving recognition accuracy and reliability.

[0034] Furthermore, the attitude iterative analysis is performed by traversing the subsequences of the host radar's attitude perception framework to determine the host radar's perception results, including: Extract a first host radar attitude perception framework subsequence and a second host radar attitude perception framework subsequence from the host radar attitude perception framework subsequence; perform attitude iterative enhancement on the first host radar attitude perception framework subsequence and the second host radar attitude perception framework subsequence to obtain a first iteratively enhanced host radar attitude perception framework subsequence; use the first iteratively enhanced host radar attitude perception framework subsequence to perform attitude iterative enhancement on the third host radar attitude perception framework subsequence in the host radar attitude perception framework subsequence, and so on, to obtain a target iteratively enhanced host radar attitude perception framework subsequence; perform health perception on the target iteratively enhanced host radar attitude perception framework subsequence to obtain the host radar perception result.

[0035] The first host radar attitude perception frame subsequence and the second host radar attitude perception frame subsequence are separated from the original host radar attitude perception frame subsequence. Each subsequence contains radar data over a period of time, representing human motion or attitude changes. The extraction of these subsequences is based on the sliding window technique, which selects specific intervals in the data as independent subsequences, such as continuous intervals of 1 second, 3 seconds, 5 seconds, etc.

[0036] The pose iterative enhancement aims to improve the accuracy of pose recognition by analyzing the dynamic changes and detailed information of each subsequence. Specifically, it uses the cosine similarity function to calculate the node similarity between the first host radar perceived pose frame subsequence and the second host radar perceived pose frame subsequence, and constructs the first pose iterative enhancement adjacency matrix. Pose iterative enhancement is then performed to obtain the first iteratively enhanced host radar perceived pose frame subsequence, which represents more detailed and stable features of human pose changes.

[0037] Using the already iteratively enhanced first-iteration enhanced host radar attitude perception framework subsequence, a similar enhancement process is applied to the third host radar attitude perception framework subsequence. The key to this process is the recursive application of the already optimized subsequences. By applying the results of previous enhancements to subsequent subsequences, a synergistic closed loop is formed. Each new subsequence is optimized based on the previous enhancement result, continuously improving the accuracy and stability of the data. Each iteration enhances the radar data recognition capability by one level, ultimately resulting in a high-quality target iteratively enhanced host radar attitude perception framework subsequence.

[0038] Health perception refers to the system's analysis of human health status based on the enhanced target iterative enhancement of the host radar's posture perception framework subsequence. Health perception relies on AI algorithms, particularly machine learning algorithms that process posture data, such as classifiers and regression models. Based on training data, AI algorithms automatically identify abnormal postures or movements, thereby inferring potential health problems. For example, when analyzing radar data using AI algorithms, if a specific posture, such as a rapid fall, is detected, the machine learning model can automatically identify it as a potential health risk and quickly issue an alert. After health perception, the host radar's perception results include identified movements and health assessment results, such as health risks and fall warnings.

[0039] Furthermore, attitude iterative enhancement is performed on the first host radar attitude perception frame subsequence and the second host radar attitude perception frame subsequence to obtain a first iteratively enhanced host radar attitude perception frame subsequence, including: The node similarity between the first host radar attitude perception frame subsequence and the second host radar attitude perception frame subsequence is calculated using the cosine similarity function to obtain a node similarity set; a first attitude iterative enhancement adjacency matrix is ​​constructed based on the node similarity set; the attitude iterative enhancement of the second host radar attitude perception frame subsequence is performed using the first attitude iterative enhancement adjacency matrix to obtain a first iteratively enhanced host radar attitude perception frame subsequence.

[0040] The cosine similarity function is used to measure the similarity between two vectors. In this step, the cosine similarity function is used to calculate the similarity of each node between the two subsequences, where a node refers to a time point in the data. For the first host radar attitude perception frame subsequence and the second host radar attitude perception frame subsequence, the data is first converted into vector representations. For example, the attitude frame data (such as the position, velocity, and angle of various parts of the human body) is extracted into a high-dimensional feature vector. For each pair of corresponding nodes, such as the attitude vector at a certain time point, their similarity is calculated using the cosine similarity function. The similarity value is in the range [0,1], where 1 represents complete similarity and 0 represents complete dissimilarity. After calculating for all node pairs, a set of similarity values ​​is obtained, forming a node similarity set. This set reflects the similarity between each time point between the two subsequences.

[0041] The first-pose iteratively enhanced adjacency matrix is ​​used to represent the connection relationships between nodes. Here, nodes represent different time points in the radar data, and the similarity value determines the connection strength between them. The size of the adjacency matrix is ​​n×n, where n is the number of nodes. Each element Aij in the matrix represents the similarity between node i and node j. If the cosine similarity between node i and node j is high, for example, greater than a preset threshold, a larger value is filled in the corresponding position in the first-pose iteratively enhanced adjacency matrix, indicating that the two nodes are relatively similar. Conversely, if the similarity is low, the value is small or zero. Using the calculated set of node similarities, a connection weight is created for each pair of nodes. If the similarity value is high, the connection weight is large, and vice versa. This matrix represents the relationship between nodes in the subsequence and can be used for further graph processing algorithms, such as iterative enhancement.

[0042] The first attitude iterative enhancement adjacency matrix is ​​used to perform attitude iterative enhancement on the second host radar perception attitude framework subsequence. By enhancing the weight information in the first attitude iterative enhancement adjacency matrix, the influence of nodes with high similarity (i.e., similar attitude data) is amplified, thereby improving the overall attitude recognition accuracy. Iterative enhancement adjusts the data by using the weighted information of the first attitude iterative enhancement adjacency matrix. Specifically, techniques such as graph convolutional networks are applied to adjust the features of each node by propagating the similarity information between nodes. After iterative enhancement, the resulting first iteratively enhanced host radar perception attitude framework subsequence contains optimized attitude data, which is more accurate than the original data and can improve the accuracy of subsequent health monitoring.

[0043] Furthermore, the set of slave computing units in the computing module is invoked to analyze the set of slave function perception data sequences to determine the set of slave function perception results, including: Extract the data sequence with the function identifier of respiratory and heart rate monitoring from the slave function perception data sequence set to obtain the first slave function perception sequence set; call the anomaly identifier in the first slave computing unit set with the function identifier of respiratory and heart rate monitoring in the slave computing unit set respectively to analyze the first slave function perception sequence set to obtain the first slave function perception result set, and add the first slave function perception result set into the slave function perception result set.

[0044] The slave-level functional sensing data sequence set contains multiple data sequences with different functions, such as fall detection data and respiratory and heart rate monitoring data. Each data sequence has a function identifier to distinguish the monitoring target of the data. By filtering the data, only the data sequences with the function identifier of respiratory and heart rate monitoring are extracted. These data sequences come from specialized respiratory and heart rate monitoring devices, such as wearable sensors and chest straps. After extraction, the first slave-level functional sensing sequence set is obtained.

[0045] An anomaly detector, identified by the respiratory and heart rate monitoring identifier, is used to detect and analyze abnormal patterns in respiratory and heart rate data. It can identify abnormal health signals such as sudden changes in respiratory rate, apnea, and excessively fast or slow breathing. For the first slave functional perception sequence set, it calls the corresponding anomaly detector in the slave computing unit for data analysis. The anomaly detector typically uses time series analysis or machine learning models to determine the regularity and anomalies of the data. For example, it uses threshold judgment methods to identify deviations from the normal range, such as apnea or excessively fast breathing. After analysis by the anomaly detector, the resulting first slave functional perception result set contains the data analysis results for respiratory and heart rate monitoring, including the time point of anomaly identification and the type of anomaly (e.g., apnea or abnormal respiratory rate). The analyzed first slave functional perception result set is added to the total slave functional perception result set, forming a large set containing the analysis results of all slaves.

[0046] Furthermore, this includes: Obtain the historical host dual-sensing result set and the historical slave functional sensing result set; respectively count the frequency of anomalies in the historical host dual-sensing result set and the historical slave functional sensing result set to obtain the historical host anomaly frequency and the historical slave anomaly frequency set; based on the historical host anomaly frequency and the historical slave anomaly frequency set, dynamically update and configure the computing module resources to obtain the host computing unit and slave computing unit set.

[0047] The historical host dual-sensing result set refers to the health monitoring results previously obtained by the host under dual-sensing (radar and vision) conditions; the historical slave functional sensing result set includes the sensing results obtained by all slave devices under specific functions (such as respiratory and heart rate monitoring, fall detection, etc.). This historical data can be extracted from databases, log files, or cloud platform storage systems.

[0048] For the historical master dual-sensing result set, the frequency of abnormal situations is statistically analyzed by examining the health monitoring results from radar and visual data. For example, the number of abnormal events in health monitoring, such as abnormal posture or abnormal breathing, is counted. Similarly, for the historical slave functional sensing result set, the frequency of abnormal situations detected by each slave is counted, such as abnormalities in respiratory and heart rate monitoring (e.g., apnea) or fall detection (e.g., fall events). The frequency is obtained by statistically analyzing the abnormal detection rate of all masters and slaves, for example, the proportion of abnormal events to all detected events. After statistical analysis, the historical master abnormality frequency and historical slave abnormality frequency sets are obtained, containing specific frequency data of abnormalities occurring in past monitoring by the masters and slaves, which can provide a basis for subsequent resource allocation.

[0049] Based on the statistically analyzed historical host and slave anomaly frequencies, the computing resources of the hosts and slaves are dynamically adjusted. For example, if certain hosts or slaves frequently experience anomalies in the historical data, it means that these devices require more computing resources to process more data or perform more complex analyses. Conversely, if some devices have a lower anomaly frequency, it indicates that their computing resource requirements are lower, and resource allocation can be reduced. Based on this analysis, the computing module reconfigures computing resources to ensure that each device (host or slave) receives sufficient computing support according to its actual needs. In this way, abnormal data can be processed more effectively, improving overall monitoring and response capabilities.

[0050] Furthermore, such as Figure 2As shown, the host includes a housing base 1 for fixing the host to a wall; a first motor 2 fixed to the lower surface of the housing base 1 for driving the camera 6 to rotate; a bracket 3 fixedly connected to the first motor 2, the radar motherboard, and the camera motherboard 9; a radar motherboard 4 connected to the bracket and the radar module for fixing the radar module; a second motor 5 connected to the radar motherboard 4 for driving the radar module; a camera 6 fixed to the camera motherboard 9 for image acquisition; a speaker 7 mounted on the housing base 1 for alerting to health abnormalities; a charging motherboard 8 fixed to the base for providing power to the host; and a camera motherboard 9 fixedly connected to the bracket 3 and the camera for fixing the camera 6.

[0051] The base 1 provides a fixed support for the main unit, allowing it to be securely mounted on the wall, which helps ensure the stability of the main unit and the coverage of the monitoring area. The first motor 2 is fixed to the lower surface of the base 1, driving the camera 6 to rotate. This allows the camera 6 to rotate omnidirectionally within a preset area, improving monitoring coverage and flexibility. The bracket 3 connects the first motor 2, the radar mainboard, and the camera mainboard 9, serving as a fixed support structure for these components. The bracket 3 ensures stable connections between the components, allowing the camera 6 and the radar module to rotate and adjust as needed. The radar mainboard 4 connects to the bracket 3 and the radar module, fixing and supporting the radar module. The radar mainboard 4 provides power and data transmission interfaces for the radar module, ensuring normal radar operation. The second motor 5 connects to the radar mainboard 4, driving the radar module to rotate. Similar to the first motor 2, the second motor 5 allows the radar module to adjust its orientation, increasing the radar's monitoring range and flexibility. The camera 6, fixed to the camera motherboard 9, is responsible for image acquisition. The camera 6 can capture images of the target area in real time, assisting the radar module in comprehensive data analysis. The speaker 7, mounted on the housing base 1, is used to issue an alarm when a health abnormality is detected. The speaker 7 can provide an audible alert to the user or those nearby, enabling timely intervention. The charging motherboard 8, fixed to the base, provides power to the main unit, ensuring continuous operation. The charging motherboard 8 is typically connected to a battery or charging port, providing necessary power support to the main unit. The camera motherboard 9 is fixedly connected to the bracket 3 and the camera 6, ensuring the installation and positioning of the camera 6. The camera motherboard 9 provides electrical connections and data transmission interfaces between the camera 6 and other devices.

[0052] Furthermore, the camera includes a binocular camera and a 270-degree rotating base.

[0053] Binocular cameras simultaneously capture images of a target area using two cameras, leveraging stereoscopic vision technology for more accurate depth perception. Compared to ordinary monocular cameras, binocular cameras can measure distances and identify the shape and position of objects. This allows the camera to more accurately acquire three-dimensional spatial information of the target area, improving the accuracy of health monitoring. The 270-degree rotating base allows the binocular camera to rotate within a 270-degree range, enhancing the coverage of the monitored area. It also allows for more flexible adjustment of the camera's viewing angle, acquiring real-time data of the target area from multiple angles. The rotating base enables the camera to cover a wider area, reducing blind spots and improving overall monitoring effectiveness.

[0054] Furthermore, the host also includes a computing module, an external network communication module, and an internal network communication module. The computing module is used for data processing, the external network communication module supports 4G or 5G communication, and the internal network communication module supports Wi-Fi, Bluetooth, or StarFlash protocols.

[0055] The computing module handles the host's data processing tasks, performing real-time analysis of all data. It can process data from cameras, radar modules, and other sensors, execute complex calculations and algorithms such as posture perception and anomaly detection, and provide health monitoring results. The external network communication module enables the host to communicate with external cloud platforms in real time via 4G or 5G networks. It supports functions such as uploading data, receiving remote commands, and performing system updates, enhancing the host's remote management and real-time monitoring capabilities. The internal network communication module supports data communication within the local area network and supports wireless communication technologies such as Wi-Fi, Bluetooth, or StarScan protocols for wireless data transmission between the host and other devices (such as slave devices, smart terminals, and mobile devices), ensuring smooth data synchronization and communication between systems.

[0056] In summary, the non-contact health monitoring method based on dual vision and radar perception provided in this application has the following technical effects: By deploying the main unit in the first area of ​​the target region and slave units in multiple second areas, the system ensures coverage of all key locations within the monitored area. The slave unit deployment allows each unit to focus on its specific function, achieving comprehensive spatial monitoring. The main unit simultaneously collects dual-sensor data via radar and camera modules. Radar signals provide spatial information, while camera modules provide image data to assist in human identification. This dual-sensor approach enhances data accuracy and reduces potential misjudgments or blind spots from a single sensor. The slave units collect functional sensing data in different areas, each focusing on a small monitoring area, thus improving the system's monitoring efficiency and accuracy across different regions. The main unit analyzes the dual-sensor data using a multi-scale attitude sensor. It can identify human posture from multiple angles. This multi-scale processing can improve the system's ability to recognize human posture at different distances and angles. Through comprehensive analysis of radar and visual data, it can more accurately detect posture changes, especially effective in identifying falls and abnormal behaviors, reducing false alarm rates. The slave unit focuses on collecting and analyzing data for specific functions. This focused design can accurately determine the health status of the area where the slave unit is located. Different slave unit computing units perform directional analysis based on the slave unit's functional identifiers, improving monitoring accuracy. By fusing the dual sensing results of the host unit with the functional sensing results of the slave unit, the resulting non-contact health monitoring results greatly reduce the probability of false alarms and missed alarms, and can provide more accurate and comprehensive status warnings.

[0057] Example 2, based on the same inventive concept as the non-contact health monitoring method based on dual vision and radar perception in the previous examples, such as... Figure 3 As shown in the figure, this application provides a non-contact health monitoring system based on dual vision and radar perception. The system includes: The host deployment module 10 is used to deploy a host in a first area of ​​the target area and deploy slave devices in multiple second areas to obtain a slave device set. Registration and binding of the host and slave device set are completed through a cloud platform. Each slave device has a functional identifier, including an attitude monitoring identifier and a respiratory and heart rate monitoring identifier. The sensing data retrieval module 20 is used to retrieve the host's dual-sensing data sequence in the target area from the radar module and camera module within a preset monitoring window, as well as the slave device set's functional sensing data sequence set in the multiple second areas within the preset monitoring window. The host data analysis module 30 is used to call the multi-scale attitude sensor of the host computing unit in the computing module to analyze the host's dual-sensing data sequence and determine the host's dual-sensing results. The slave data analysis module 40 is used to call the slave computing unit set in the computing module to analyze the slave device functional sensing data sequence set and determine the slave device functional sensing result set. The monitoring result acquisition module 50 is used to fuse the host's dual-sensing results and the slave device functional sensing result set to obtain a non-contact health monitoring result.

[0058] Furthermore, the host data analysis module 30 is used to perform the following operation steps: The initial attitude frame from the multi-scale attitude sensor is retrieved; the host radar perception data subsequence and the host visual perception data subsequence from the host dual-sensing data sequence are respectively input into the initial attitude frame to obtain the host radar perception attitude frame subsequence and the host visual perception attitude frame subsequence; the host radar perception attitude frame subsequence is traversed to perform attitude iterative analysis to determine the host radar perception result; the host visual perception attitude frame subsequence is traversed to perform attitude iterative analysis to determine the host visual perception result; the host radar perception result and the host visual perception result are cross-confirmed to obtain the host dual-sensing result.

[0059] Furthermore, the host data analysis module 30 is used to perform the following operation steps: Extract a first host radar attitude perception framework subsequence and a second host radar attitude perception framework subsequence from the host radar attitude perception framework subsequence; perform attitude iterative enhancement on the first host radar attitude perception framework subsequence and the second host radar attitude perception framework subsequence to obtain a first iteratively enhanced host radar attitude perception framework subsequence; use the first iteratively enhanced host radar attitude perception framework subsequence to perform attitude iterative enhancement on the third host radar attitude perception framework subsequence in the host radar attitude perception framework subsequence, and so on, to obtain a target iteratively enhanced host radar attitude perception framework subsequence; perform health perception on the target iteratively enhanced host radar attitude perception framework subsequence to obtain the host radar perception result.

[0060] Furthermore, the host data analysis module 30 is used to perform the following operation steps: The node similarity between the first host radar attitude perception frame subsequence and the second host radar attitude perception frame subsequence is calculated using the cosine similarity function to obtain a node similarity set; a first attitude iterative enhancement adjacency matrix is ​​constructed based on the node similarity set; the attitude iterative enhancement of the second host radar attitude perception frame subsequence is performed using the first attitude iterative enhancement adjacency matrix to obtain a first iteratively enhanced host radar attitude perception frame subsequence.

[0061] Furthermore, the slave data analysis module 40 is used to perform the following operation steps: Extract the data sequence with the function identifier of respiratory and heart rate monitoring from the slave function perception data sequence set to obtain the first slave function perception sequence set; call the anomaly identifier in the first slave computing unit set with the function identifier of respiratory and heart rate monitoring in the slave computing unit set respectively to analyze the first slave function perception sequence set to obtain the first slave function perception result set, and add the first slave function perception result set into the slave function perception result set.

[0062] Furthermore, the host data analysis module 30 is used to perform the following operation steps: Obtain the historical host dual-sensing result set and the historical slave functional sensing result set; respectively count the frequency of anomalies in the historical host dual-sensing result set and the historical slave functional sensing result set to obtain the historical host anomaly frequency and the historical slave anomaly frequency set; based on the historical host anomaly frequency and the historical slave anomaly frequency set, dynamically update and configure the computing module resources to obtain the host computing unit and slave computing unit set.

[0063] Furthermore, the host includes a housing base 1 for fixing the host to a wall; a first motor 2 fixed to the lower surface of the housing base 1 for driving the camera 6 to rotate; a bracket 3 fixedly connected to the first motor 2, the radar motherboard, and the camera motherboard 9; a radar motherboard 4 connected to the bracket and the radar module for fixing the radar module; a second motor 5 connected to the radar motherboard 4 for driving the radar module; a camera 6 fixed to the camera motherboard 9 for image acquisition; a speaker 7 mounted on the housing base 1 for alerting to health abnormalities; a charging motherboard 8 fixed to the base for providing power to the host; and a camera motherboard 9 fixedly connected to the bracket 3 and the camera for fixing the camera 6.

[0064] Furthermore, the camera includes a binocular camera and a 270-degree rotating base.

[0065] Furthermore, the host also includes a computing module, an external network communication module, and an internal network communication module. The computing module is used for data processing, the external network communication module supports 4G or 5G communication, and the internal network communication module supports Wi-Fi, Bluetooth, or StarFlash protocols.

[0066] Through the foregoing detailed description of the non-contact health monitoring method based on dual vision and radar perception, those skilled in the art can clearly understand the non-contact health monitoring system based on dual vision and radar perception in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to in the method section.

[0067] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A non-contact health monitoring method based on dual perception of vision and radar, characterized in that, The method comprises: deploying a master in a first region of a target region and deploying slaves in a plurality of second regions respectively to obtain a slave set, and completing registration and binding of the master and the slave set through a cloud platform, wherein each slave has a function identifier, and the function identifier comprises a posture monitoring identifier and a breathing and heart rate monitoring identifier; respectively calling radar modules and camera modules of the master to perform double perception data sequence of the master in the target region in a preset monitoring window, and calling the slave set to perform a slave function perception data sequence set of the slaves in the plurality of second regions in the preset monitoring window; calling a multi-scale posture perceiver of a master computing unit in a computing module to analyze the double perception data sequence of the master to determine a master double perception result; calling a slave computing unit set in the computing module to analyze the slave function perception data sequence set respectively to determine a slave function perception result set; fusing the master double perception result and the slave function perception result set to obtain a non-contact health monitoring result. 2.The visual and radar dual-sensing based non-contact health monitoring method of claim 1, wherein, calling a multi-scale posture perceiver of a master computing unit in a computing module to analyze the double perception data sequence of the master to determine a master double perception result, comprising: calling an initial posture framework in the multi-scale posture perceiver; inputting master radar perception data subsequences and master visual perception data subsequences in the double perception data sequence of the master into the initial posture framework to obtain master radar perception posture framework subsequences and master visual perception posture framework subsequences; iteratively analyzing the master radar perception posture framework subsequences to determine a master radar perception result; iteratively analyzing the master visual perception posture framework subsequences to determine a master visual perception result; cross-identifying the master radar perception result and the master visual perception result to obtain the master double perception result. 3.The visual and radar dual-sensing based non-contact health monitoring method of claim 2, wherein, iteratively analyzing the master radar perception posture framework subsequences to determine a master radar perception result, comprising: extracting first master radar perception posture framework subsequences and second master radar perception posture framework subsequences from the master radar perception posture framework subsequences; iteratively enhancing the first master radar perception posture framework subsequences and the second master radar perception posture framework subsequences to obtain first iteratively enhanced master radar perception posture framework subsequences; iteratively enhancing third master radar perception posture framework subsequences in the master radar perception posture framework subsequences by using the first iteratively enhanced master radar perception posture framework subsequences, and iteratively enhancing the third master radar perception posture framework subsequences in the master radar perception posture framework subsequences by using the first iteratively enhanced master radar perception posture framework subsequences, and so on, to obtain target iteratively enhanced master radar perception posture framework subsequences; performing health perception on the target iteratively enhanced master radar perception posture framework subsequences to obtain the master radar perception result. 4.The visual and radar dual-sensing based non-contact health monitoring method of claim 3, wherein, iteratively enhancing the first master radar perception posture framework subsequences and the second master radar perception posture framework subsequences to obtain first iteratively enhanced master radar perception posture framework subsequences, comprising: Calculating node similarity of the first host radar perception pose framework subsequence and the second host radar perception pose framework subsequence by using a cosine similarity function to obtain a node similarity set; Building a first pose iterative enhancement adjacency matrix according to the node similarity set; Performing pose iterative enhancement on the second host radar perception pose framework subsequence by using the first pose iterative enhancement adjacency matrix to obtain a first iterative enhancement host radar perception pose framework subsequence. 5.The visual and radar dual-sensing based non-contact health monitoring method of claim 1, wherein, Calling a slave computing unit set in a computing module to analyze the slave function perception data sequence set respectively to determine a slave function perception result set, including: Extracting a data sequence with a function identifier of a respiratory heart rate monitoring identifier from the slave function perception data sequence set to obtain a first slave function perception sequence set; Calling an abnormality identifier in a first slave computing unit set with a function identifier of a respiratory heart rate monitoring identifier in the slave computing unit set to analyze the first slave function perception sequence set to obtain a first slave function perception result set, and adding the first slave function perception result set to the slave function perception result set. 6.The visual and radar dual-sensing based non-contact health monitoring method of claim 1, wherein, Including: Obtaining a historical host dual perception result set and a historical slave function perception result set; Statistically analyzing abnormal frequencies in the historical host dual perception result set and the historical slave function perception result set to obtain a historical host abnormal frequency and a historical slave abnormal frequency set; Performing resource dynamic update configuration on the computing module based on the historical host abnormal frequency and the historical slave abnormal frequency set to obtain a host computing unit and a slave computing unit set.

7. The visual and radar dual-sensing based contactless health monitoring method of claim 1, wherein, The host includes a shell base, which is used to fix the host to a wall position; A first motor is fixed to the lower surface of the shell base and is used to drive the camera to rotate; A support is fixedly connected with the first motor, the radar mainboard and the camera mainboard respectively; A radar mainboard is connected with the support and the radar module and is used to fix the radar module; A second motor is connected with the radar mainboard and is used to drive the radar module; A camera is fixed on the camera mainboard and is used to collect images; A loudspeaker is installed on the shell base and is used to remind health abnormalities; A charging mainboard is fixed on the base and is used to provide power for the host; A camera mainboard is fixedly connected with the support and the camera and is used to fix the camera. 8.The visual and radar dual-sensing based non-contact health monitoring method of claim 7, wherein, The camera includes a binocular camera and a 270-degree rotating base. 9.The visual and radar dual-sensing based non-contact health monitoring method of claim 7, wherein, The host further includes a computing module, an external network communication module and an internal network communication module, wherein the computing module is used for data processing, the external network communication module supports 4G or 5G communication, and the internal network communication module supports WIFI or Bluetooth or Starlink protocol.

10. A non-contact health monitoring system based on dual perception of vision and radar, characterized in that, The system is used for implementing the non-contact health monitoring method based on visual and radar dual perception according to any one of claims 1-9, and the system includes: The host arrangement module is used for arranging hosts in a first area of a target area and arranging slaves in a plurality of second areas respectively to obtain a slave set, and registration and binding of the hosts and the slave set are completed through a cloud platform, wherein each slave has a function identifier, and the function identifier includes a posture monitoring identifier and a breathing and heart rate monitoring identifier; The perception data calling module is used for calling a host double perception data sequence of the radar module and the camera module of the host in a preset monitoring window on the target area, and a slave function perception data sequence set of the slave set on the slaves in the plurality of second areas in the preset monitoring window; The host data analysis module is used for calling a multi-scale posture perception of a host computing unit in a computing module to analyze the host double perception data sequence and determine a host double perception result; The slave data analysis module is used for calling a slave computing unit set in the computing module to analyze the slave function perception data sequence set respectively and determine a slave function perception result set; The monitoring result acquisition module is used for fusing the host double perception result and the slave function perception result set to obtain a non-contact health monitoring result.

Images