A health monitoring data distribution method and system based on PCDN

Abstract

The application discloses a health monitoring data distribution method and system based on PCDN, and relates to the technical field of data distribution. The method comprises the following steps: S1, collecting depth image data and running state data of a peer node in real time, and performing preprocessing; S2, extracting human behavior structure features based on the depth image data, identifying abnormal events, comprehensively evaluating abnormal credibility, and judging whether evidence collection needs to be triggered; S3, when it is determined that the abnormal event is credible, uniformly packaging abnormal related evidence, evaluating priority in combination with the running state data of the peer node, and performing distribution optimization and collaborative caching; and S4, performing global scheduling on the peer node based on constraint rules, guiding distribution rhythm and access paths under different network conditions, and performing efficient access and long-term analysis of health monitoring data. The method solves the problems of easy bandwidth congestion and limited distribution efficiency in the centralized health monitoring data back-to-source distribution mode.

Description

Technical Field

[0001] This invention relates to the field of data distribution technology, specifically to a health monitoring data distribution method and system based on PCDN. Background Technology

[0002] With the increasing demand for remote monitoring of individual behavior and health status, related systems typically need to continuously collect multi-source sensing data in residential environments and transmit, store, and access it among multiple terminals and service nodes. Meanwhile, the complexity of network environments and the diversity of terminal node distribution mean that health monitoring data generation, distribution, and service must balance real-time performance, reliability, and resource utilization efficiency, thus placing higher demands on data distribution architecture and collaborative mechanisms.

[0003] For example, invention patent CN121357248A discloses a PCDN optimization method and system based on intelligent scheduling and dynamic governance, belonging to the field of distributed computing and edge network scheduling technology. The method includes: collecting hardware resource status data of nodes; generating node resource scheduling signals by combining user geographical location and content demand priority through a content demand optimization scheduling algorithm; constructing a multi-level cache priority system to allocate optimal node matching and data transmission paths based on hot content in different time periods and regions, combined with content timeliness and user access frequency; analyzing the dynamic relationship between node load and resource utilization through path cross-modeling scheduling method, and allocating independent scheduling windows; constructing a point-to-point content distribution network authorization feature library to classify and identify node traffic, and implementing a tiered suppression strategy for unauthorized traffic based on real-time network load, limiting traffic transmission during low load and blocking it during high load, depending on the network node load status.

[0004] For example, invention patent CN118488057A discloses a PCDN data distribution method, apparatus, device, and storage medium, applied in the field of data distribution technology. The method includes acquiring program data to be downloaded; generating multiple download tasks for the program data according to a data distribution strategy; using the multiple download tasks as a dataset to be distributed; and periodically sending download tasks to PCDN nodes according to the dataset, so that the PCDN nodes can download program data from the CDN or clients according to the download tasks. In this way, historical hot content and the latest content can be comprehensively considered to dynamically adjust the data to be distributed; in situations where storage is limited, user viewing behavior is scattered, and most content cannot be cached, only program data that is more effective in saving CDN bandwidth and improving sharing rates is stored, thus saving CDN costs.

[0005] However, in the scenario of health monitoring for elderly people living alone, terminals such as cameras and gateways continuously generate high-bitrate video and behavioral monitoring data. Existing technologies typically adopt a centralized data sourcing and distribution model of "terminal direct connection to the cloud." When multiple terminals in the community simultaneously perform video playback, anomaly verification, or centralized uploading, the cloud egress bandwidth and access links are prone to momentary congestion, leading to video stuttering, loading delays, or even service unavailability. At the same time, a large amount of duplicate video traffic is transmitted multiple times in the network, resulting in wasted bandwidth resources. The overall system distribution efficiency is difficult to scale linearly with the user base, making it difficult to meet the needs of continuous deployment at the community level.

[0006] Therefore, in order to address the above problems, there is an urgent need for a health monitoring data distribution method and system based on PCDN. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a health monitoring data distribution method and system based on PCDN, which solves the problems of bandwidth congestion and limited distribution efficiency that easily occur in the centralized health monitoring data back-to-source distribution mode.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a health monitoring data distribution method based on PCDN, comprising the following steps: S1, real-time acquisition of depth image data in the living environment of the monitored subject, and real-time acquisition of peer node operation status data, and preprocessing of multi-source data; S2, based on the preprocessed depth image data, extraction of human behavioral structural features, identification of abnormal events, analysis of changes in human posture and spatial displacement before and after the abnormal events, comprehensive evaluation of the credibility of the anomaly, and determination of whether to trigger evidence collection based on the credibility of the anomaly; S3, when a credible anomaly is determined, unified encapsulation of anomaly-related evidence collection, evaluation of the priority of peer nodes based on the operation status data of peer nodes, and execution of optimal distribution and collaborative caching based on the priority; S4, global scheduling of peer nodes based on constraint rules, and guidance of distribution rhythm and access path under different network conditions, enabling efficient access and long-term analysis of health monitoring data.

[0011] Furthermore, depth image data is collected in real time in the living environment of the monitored individual, and the operational status data of peer nodes is also collected in real time. The specific process of preprocessing multi-source data is as follows: RGB-D depth camera modules are deployed in the living environment of the monitored individual to collect depth image frames in real time, including depth pixel values ​​and pixel coordinates; simultaneously, in the PCDN network, peer nodes with peer connections to the current edge gateway are obtained to obtain a candidate peer node set, and the device operation data of the candidate peer nodes are collected in real time, including: CPU utilization, NPU utilization, available bandwidth, and link round trip. The system analyzes latency, packet loss rate, and storage usage; performs time alignment on the acquired depth image frames and device operation data; for depth image frames, it employs median filtering for noise suppression, fills in void pixels using neighborhood interpolation, and performs scale unification on the depth pixel values ​​in the depth image frames; it performs normalization on the device operation data; based on the preprocessed depth image frames, it uses a foreground segmentation algorithm to extract the human activity area, obtaining the target region depth image frame; and it establishes a data distribution database, writing the depth image frames, target region depth image frames, and device operation data into the data distribution database.

[0012] Furthermore, based on the preprocessed depth image data, the specific process of extracting human behavioral structural features and identifying abnormal events is as follows: Based on the depth image frame of the target region, depth pixels are statistically analyzed in layers along the vertical direction. In the upper layer, the pixel position with the largest vertical pixel coordinate is selected as the head key point; in the lower layer, the pixel position with the smallest vertical pixel coordinate is selected as the lower limb key point; and in the middle layer, the geometric center point of the pixel position is selected as the torso key point, thus obtaining a set of human key points. The depth pixel values ​​of the torso key points in the depth image frame of the target region are read, and the ground reference depth is determined based on the depth pixel values ​​of the ground region in the depth image frame of the target region. The depth pixel values ​​of the torso key points and the ground reference depth are then calculated. The difference in degrees is used to obtain the ground clearance value; based on the set of human body key points, head key points and torso key points are selected to construct the human torso direction vector, and the human posture angle is calculated based on the angle between the human torso direction vector and the vertical direction vector; the difference between adjacent time moments of the ground clearance value is calculated to obtain the change in ground clearance, and the second-order difference is calculated on the change in ground clearance to obtain the abrupt change value of the change in ground clearance; the difference between adjacent time moments of the human posture angle is calculated to obtain the change in human posture angle; the sampling moment when the absolute value of the abrupt change value of the ground clearance or the absolute value of the change in human posture angle exceeds the corresponding change threshold is selected as the abnormal moment; a sliding time window is constructed with the abnormal moment as the center point, and the time window before the abnormality and the time window after the abnormality are divided.

[0013] Furthermore, the specific process for analyzing changes in human posture and spatial displacement before and after the abnormal event, and comprehensively evaluating the credibility of the anomaly, is as follows: The median value of the ground clearance before the anomaly is taken within the time window before the anomaly to obtain the baseline value of ground clearance before the anomaly; the minimum value of the ground clearance after the anomaly is taken within the time window after the anomaly to obtain the minimum ground clearance after the anomaly; the minimum ground clearance after the anomaly is subtracted from the baseline value of ground clearance before the anomaly, and then divided by the sum of the baseline value of ground clearance before the anomaly and the minimum constant to obtain the ratio of sudden drop in human height; the median values ​​of human posture angles before and after the anomaly are taken within the time windows after and after the anomaly, respectively, to obtain the human posture angle values ​​before and after the anomaly, and the absolute value is calculated. Divide the difference by the angular scale constant to obtain the human body posture flipping amplitude; based on the sequence of abrupt changes in ground height within the sliding time window, calculate the maximum value and median absolute deviation, and divide the maximum value by the sum of the median absolute deviation and the minimum constant to obtain the impact abrupt change intensity; calculate the sum of the squares of the human body height drop ratio and the human body posture flipping amplitude, and take the square root to obtain the abnormal comprehensive amplitude value; take the natural logarithm of the sum of the impact abrupt change intensity and constant one to obtain the height change impact intensity value; multiply the abnormal comprehensive amplitude value and the height change impact intensity value and take the opposite number as the exponent, perform natural exponentiation, and subtract the result of the natural exponentiation from constant one to obtain the anomaly evidence trigger credibility value.

[0014] Furthermore, the specific process for determining whether to trigger evidence collection based on the credibility of the anomaly is as follows: The credibility value for anomaly evidence collection is compared with the trigger threshold. When the credibility value is less than the trigger threshold, the visible light camera module remains in standby mode, continuously performing privacy monitoring based solely on depth image frames, without outputting visible light evidence collection data. When the credibility value is greater than or equal to the trigger threshold, the current anomaly is determined to be a credible anomaly. For credible anomalies, the edge gateway sends a GPIO wake-up control signal to the visible light camera module, controlling the module to switch from standby to working mode. Visible light image frames covering the time of the anomaly are extracted from the circular buffer maintained by the visible light camera module, and visible light acquisition after the anomaly is performed, forming an anomaly evidence collection image frame sequence. The anomaly evidence collection image frame sequence is then associated with the credibility value for anomaly evidence collection, the timestamp of the anomaly time, and the depth image frames of the target area and written into the data distribution database.

[0015] Furthermore, when an anomaly is determined to be credible, the specific process of uniformly encapsulating the evidence related to the anomaly is as follows: When an anomaly is determined to be credible, the edge gateway uniformly encapsulates the data corresponding to the anomaly event: it associates and binds the corresponding anomaly evidence image frame sequence, the anomaly time timestamp, and the target area depth image frame to form a health monitoring data distribution object, and generates a unique identifier for the health monitoring data distribution object.

[0016] Furthermore, combining the operational status data of peer nodes, the specific process for evaluating the priority of peer nodes is as follows: For each candidate peer node, read the preprocessed device operation data, synchronously determine whether the candidate peer node has cached the health monitoring data distribution object; if it has cached, assign a cache hit status value of 1, otherwise assign a value of 0; add the assigned cache hit status value to a constant to obtain the cache gain value; subtract the packet loss rate from the constant to obtain the link stability value; take the reciprocal of the sum of the link round-trip delay and the minimum constant to obtain the link latency penalty value; add the CPU utilization, NPU utilization, and storage occupancy rate together to calculate the total value. The average value is used to obtain the comprehensive load value. The reciprocal of the sum of the comprehensive load value and a minimum constant is used to obtain the node load penalty value. Based on the peer node connection relationship maintained in the PCDN network, the number of peer forwarding nodes required to reach the candidate peer node is calculated using the shortest path search method to obtain the logical hop count. The reciprocal of the sum of the logical hop count and a constant is used to obtain the topology distance penalty value. The fourth root of the product of the available bandwidth, link delay penalty value, link stability value and node load penalty value is used to obtain the node carrying capacity value. The optimal peer node distribution value is obtained by multiplying the buffer gain value, node carrying capacity value and topology distance penalty value.

[0017] Furthermore, the specific process of priority-based distribution optimization and collaborative caching is as follows: Candidate peer nodes are sorted in descending order according to the peer node distribution optimization value, and the top N candidate peer nodes are selected as collaborative distribution nodes. The edge gateway sends the health monitoring data distribution object to the uncached collaborative distribution node based on the peer transmission mechanism in the PCDN network. After receiving the health monitoring data distribution object, the collaborative distribution node writes it into its local cache space and updates the cache hit status of the corresponding object to cached. When the collaborative distribution node experiences link interruption, load abnormality, or cache failure during the distribution process, the edge gateway selects candidate peer nodes to redistribute the data based on the descending order sorting result.

[0018] Furthermore, the specific process for efficient access and long-term analysis of health monitoring data, based on the global scheduling of peer nodes according to constraint rules and guiding the distribution rhythm and access path under different network conditions, is as follows: When it is detected that the current conditions are not under network load period constraints or outbound bandwidth constraints, a supplementary distribution instruction is issued to the edge gateway to re-evaluate the peer node distribution preference value and control the edge gateway to perform supplementary distribution to the health monitoring data distribution object; when it is detected that the current conditions are under network load period constraints or outbound bandwidth constraints, delayed distribution is performed on the health monitoring data distribution object, and only the cache availability of the already distributed collaborative distribution nodes is maintained; when an authorized terminal initiates an access request to the health monitoring data distribution object, the cache distribution status in the PCDN network is queried based on the unique identifier, and access services are provided from the cached peer nodes according to the peer node distribution preference value order; at the same time, based on the continuously monitored ground height value, human posture angle, and health monitoring data distribution object, health status analysis is performed on the monitored object, and a health data monitoring report is generated.

[0019] The second aspect of this invention provides a PCDN-based health monitoring data distribution system, comprising: a multi-source data acquisition and processing module, used to acquire depth image data in real time in the living environment of the monitored object, and to acquire the operating status data of peer nodes in real time, and to preprocess the multi-source data; an anomaly identification and evidence collection triggering module, used to extract human behavioral structural features based on the preprocessed depth image data, identify abnormal events, analyze changes in human posture and spatial displacement before and after the abnormal event, comprehensively evaluate the credibility of the anomaly, and determine whether to trigger evidence collection based on the credibility of the anomaly; a PCDN collaborative distribution module, used to uniformly encapsulate the evidence related to the anomaly when it is determined to be a credible anomaly, evaluate the priority of the peer nodes in combination with the operating status data of peer nodes, and perform optimal distribution and collaborative caching based on the priority; and a cloud scheduling service module, used to perform global scheduling of peer nodes based on constraint rules, and guide the distribution rhythm and access path under different network conditions, so as to achieve efficient access and long-term analysis of health monitoring data.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) This invention extracts human behavioral structural features through continuous monitoring based on depth images. Visible light evidence collection is only initiated when the credibility of the anomaly reaches the triggering condition. Combined with short-term cyclic buffering to cover the abnormal moment, the frequency of visible light collection in non-abnormal states is effectively reduced, ensuring the integrity of abnormal evidence collection while also taking into account the need for privacy protection.

[0023] (2) This invention integrates the characteristics of changes in human body height above the ground, posture changes and their temporal mutations to comprehensively evaluate the spatial displacement and posture changes of the human body before and after abnormal events, avoiding the reliance on a single threshold or instantaneous changes for judgment, thereby reducing the probability of false triggering and improving the stability and reliability of abnormal identification results.

[0024] (3) In this invention, after a trusted anomaly occurs, the distribution node is optimized and collaborative caching is performed by combining the operating status, link conditions, cache status and topology of the peer node. This reduces the pressure of repeated back-to-source and centralized transmission of health monitoring data in the network, and enables the data distribution process to better adapt to multi-node concurrent access scenarios.

[0025] (4) The present invention dynamically guides the distribution rhythm and access path under different network load and bandwidth constraints through the cloud scheduling module, so that health monitoring data can be supplemented and distributed when network conditions permit, and can maintain cache availability under limited conditions, thereby supporting continuous access to health monitoring data and subsequent health status analysis.

[0026] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0027] Figure 1 Here is a flowchart of a PCDN-based health monitoring data distribution system method;

[0028] Figure 2 This is a structural diagram of a PCDN-based health monitoring data distribution system;

[0029] Figure 3 Distribute the preferred value distribution sorting diagram to peer nodes;

[0030] Figure 4 This is a schematic diagram of the RGB-D depth camera module structure;

[0031] Figure 5 This is a schematic diagram of the layered architecture of a PCDN-based health monitoring system. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1-5This invention provides a technical solution: a health monitoring data distribution method and system based on PCDN, such as... Figure 1 As shown, the process includes the following steps: S1, real-time acquisition of depth image data in the living environment of the monitored subject, and real-time acquisition of peer node operation status data, and preprocessing of multi-source data; S2, based on the preprocessed depth image data, extraction of human behavioral structure features, identification of abnormal events, analysis of changes in human posture and spatial displacement before and after the abnormal events, comprehensive evaluation of the credibility of the anomaly, and determination of whether to trigger evidence collection based on the credibility of the anomaly; S3, when a credible anomaly is determined, unified encapsulation of the anomaly-related evidence, evaluation of the priority of the peer nodes based on the operation status data of the peer nodes, and execution of distribution optimization and collaborative caching based on the priority; S4, global scheduling of peer nodes based on constraint rules, and guidance of distribution rhythm and access path under different network conditions, to achieve efficient access and long-term analysis of health monitoring data.

[0034] Specifically, depth image data is collected in real time in the living environment of the monitored object, and the operational status data of peer nodes is also collected in real time. The specific process of preprocessing multi-source data is as follows: by deploying an RGB-D depth camera module in the living environment of the monitored object, depth image frames are collected in real time, including depth pixel values ​​and pixel coordinates; among them, the depth pixel value is the pixel-level depth measurement result directly output by the RGB-D depth camera module based on the infrared structured light ranging principle, and the pixel coordinates are the two-dimensional row and column coordinates of the corresponding depth pixel in the image plane, which are used to characterize the spatial distribution of depth information. All data are generated synchronously by the camera module during single-frame acquisition. Simultaneously, within the PCDN network, peer nodes with peer-to-peer connections to the current edge gateway are identified, resulting in a candidate peer node set. The edge gateway, as a special peer node in the PCDN network, possesses distribution, scheduling, and control permissions, and collects real-time device operation data from the candidate peer nodes, including CPU utilization, NPU utilization, available bandwidth, link round-trip latency, packet loss rate, and storage occupancy. Peer nodes are edge computing nodes and cache nodes that have established direct or indirect connections with the edge gateway via a P2P protocol. Device operation data is reported in real-time by the corresponding operating system monitoring module, network protocol stack statistics module, and container runtime interface of the peer node, reflecting the node's current computing power, communication quality, and cache resource occupancy. Time alignment is performed on the collected depth image frames and device operation data. This time alignment is achieved by matching the depth image frame timestamps with the device operation data sampling timestamps on the edge gateway side using a unified time base, ensuring comparability of data from different sources at the same time scale. For depth image frames, median filtering is used for noise suppression, neighborhood interpolation is used to fill in holes in pixels, and scale unification is performed on the depth pixel values ​​in the depth image frames. Specifically, median filtering suppresses isolated noise points generated during depth measurement, neighborhood interpolation fills in invalid depth pixels caused by insufficient reflection or occlusion, and scale unification maps depth pixel values ​​under different ranging ranges and resolutions to a unified numerical scale. The original depth pixel values ​​are mapped to a unified numerical range proportionally using a min-max normalization method to ensure the stability of subsequent feature analysis. Normalization is performed on equipment operation data; the min-max normalization method is preferred to eliminate differences in the units and value ranges of different equipment operation indicators, enabling unified comparison and fusion in subsequent peer node priority evaluation.Based on the preprocessed depth image frames, a foreground segmentation algorithm is used to extract the human activity area. The foreground segmentation algorithm is used to separate the dynamic human body area corresponding to the monitored object from the environmental background, so as to reduce the interference of static objects in the environment on human behavior analysis and obtain the target area depth image frame. A data distribution database is established, and the depth image frames, target area depth image frames and device operation data are written into the data distribution database to provide unified data storage and access support for subsequent anomaly identification, evidence collection triggering, peer node distribution optimization and cloud scheduling processes.

[0035] like Figure 4 The diagram shows a schematic of an RGB-D depth camera module. The module includes an infrared transmitter, an infrared receiver, and an RGB imaging lens. The infrared transmitter and receiver work together to acquire depth information of the monitored object and its surrounding environment, while the RGB imaging lens acquires corresponding visible light image data. The RGB-D depth camera module connects directly to the edge gateway motherboard via a multi-channel MIPI CSI-2 high-speed serial interface, enabling real-time transmission of raw RAW depth data acquired by the infrared sensor to the edge gateway's processing unit. Compared to traditional compressed or pre-processed depth data output methods, this structure avoids the additional latency caused by module-side encoding and reconstruction, allowing the edge gateway to directly perform human structural feature extraction and abnormal behavior analysis based on the raw depth pixel values ​​and pixel coordinates. This structural design ensures the timeliness and spatial accuracy of depth image data at the hardware level, providing a reliable data foundation for subsequent ground clearance calculations, attitude angle analysis, abnormal mutation identification, and credible anomaly triggering for evidence collection. This supports the overall technical performance of PCDN-based health monitoring data distribution in terms of real-time performance, accuracy, and stability.

[0036] In this implementation scheme, by synchronously collecting and uniformly preprocessing depth image data and peer node operational status data at the edge, the consistency and comparability of multi-source data in terms of time scale, numerical scale, and semantic level are ensured. Under the premise of ensuring privacy protection, effective depth information related to human activities is stably extracted, while accurately characterizing the computing power, communication quality, and cache status of peer nodes. This provides a reliable and feasible data foundation for subsequent anomaly identification, evidence collection triggering, and PCDN-based distribution optimization and global scheduling, thereby improving the overall monitoring accuracy, distribution efficiency, and operational stability of the system.

[0037] Specifically, the process of extracting human behavioral structural features and identifying abnormal events based on preprocessed depth image data is as follows: Based on the depth image frame of the target region, depth pixels are statistically analyzed in a hierarchical manner along the vertical direction, where the vertical direction is the direction of the depth image coordinate axis consistent with the ground normal; the hierarchical statistics are adaptively hierarchically based on the spatial distribution features of the human body in the vertical direction in the depth image frame of the target region, specifically: firstly, in the depth image frame of the target region, the vertical coordinate distribution of all depth pixels is statistically analyzed along the vertical direction to obtain the pixel height range of the current human body in the vertical direction; according to the pixel height range, the target region is proportionally divided into at least three continuous height intervals in the vertical direction; preferably, the boundaries of the height intervals are optimized. The height range is determined based on the proportion of the vertical height range of the human body in the current target area depth image frame, which can adapt to the spatial distribution characteristics of the human body in different heights and postures. The height range includes: the upper height range at the top, the lower height range at the bottom, and the middle height range in between. Among them, the lower height range preferably accounts for 20% to 30% of the pixel height range, the middle height range preferably accounts for 40% to 60% of the pixel height range, and the upper height range preferably accounts for 20% to 30% of the pixel height range. The upper height range corresponds to the height range where the human head may appear, the lower height range corresponds to the height range where the human lower limbs may appear, and the middle height range corresponds to the height range where the human torso is mainly distributed. In the upper layer, the pixel with the largest vertical pixel coordinate is selected as the head keypoint; in the lower layer, the pixel with the smallest vertical pixel coordinate is selected as the lower limb keypoint; and in the middle layer, the geometric center of the pixel position is selected as the torso keypoint. This results in a set of human keypoints. The head, torso, and lower limb keypoints are all spatial feature points directly calculated from single-frame depth images, independent of human model training and prior pose libraries, and can be stably and in real-time obtained under edge computing conditions. The depth pixel values ​​of the torso keypoints in the target region depth image frame are read, and the ground reference depth is determined based on the depth pixel values ​​of the ground region in the target region depth image frame. The ground region is preferably determined by depth plane fitting, which is preferably performed within the window before an anomaly and remains unchanged within the anomaly window. This depth plane fitting is used to characterize the ground height benchmark in the environment of the monitored object. The difference between the depth pixel values ​​of the torso keypoints and the ground reference depth is calculated to obtain the ground height value. The ground height value is used to characterize the actual spatial height of the human torso relative to the ground and is an important basic physical quantity for subsequent behavioral change analysis and anomaly determination.Based on a set of human body key points, a human torso direction vector is constructed by selecting head and torso key points. The human posture angle is calculated based on the angle between the human torso direction vector and the vertical direction vector. The vertical direction vector is a reference vector aligned with the direction of gravity, used to unify the posture judgment benchmark of camera module data collected at different installation angles, thus ensuring the comparability of posture angles in different deployment environments. The change in ground height is obtained by calculating the difference between adjacent time intervals of the ground height value, and a second-order difference is calculated on the change in ground height to obtain the abrupt change value of the ground height change. The difference between adjacent time intervals is used to characterize the rate of change of ground height, and the second-order difference is used to characterize the acceleration characteristics during the change of ground height, thereby enhancing the sensitivity to sudden behaviors such as falls and rapid squatting. The change in human posture angle is obtained by calculating the difference between adjacent time points. This change reflects the rotational amplitude and directional changes in human posture over a short period. The sampling time when the absolute value of the abrupt change in ground height or the absolute value of the change in human posture angle exceeds the corresponding change threshold is selected as the abnormal moment. This threshold is an adaptive threshold derived from historical data statistics, used to distinguish between slow, routine behavioral changes and sudden abnormal behaviors, avoiding false triggering of normal activities. The time point at which the anomaly first occurs between the abrupt change in ground height and the change in human posture angle is selected as the abnormal moment. A sliding time window is constructed with the abnormal moment as the center point, divided into a pre-anomaly time window and a post-anomaly time window. The optimal duration of both the pre-anomaly and post-anomaly time windows is 1 to 3 seconds, to simultaneously cover the stable state before the anomaly and the behavioral outcome after the anomaly, providing complete temporal context information for subsequent anomaly confidence assessment.

[0038] In this implementation scheme, key spatial features of the human body are constructed directly from depth image data at the edge side, achieving a stable characterization of human posture and spatial height changes without relying on human model training and prior pose libraries. Through temporal analysis of changes in ground height and posture angle, the structural features of sudden behaviors are effectively captured. Combined with adaptive thresholds and sliding time windows, the timing of anomalies and their evolution before and after are accurately located. This ensures real-time performance and feasibility while reducing the risk of false triggering, providing a reliable, continuous, and physically meaningful behavioral analysis basis for subsequent anomaly credibility assessment and evidence collection.

[0039] Specifically, the process of analyzing changes in human posture and spatial displacement before and after an abnormal event, and comprehensively evaluating the credibility of the anomaly, is as follows: The median value of the ground clearance before the anomaly is taken within the time window before the anomaly to obtain the baseline value of the ground clearance before the anomaly. Using the median value instead of the average value is used to reduce the impact of occasional posture fluctuations within the pre-anomaly window on the baseline height calculation results, thereby improving the stability of the height baseline. The minimum value of the ground clearance after the anomaly is taken within the time window after the anomaly to obtain the minimum ground clearance after the anomaly. This minimum ground clearance after the anomaly reflects the minimum spatial distance between the human body and the ground after the anomaly occurs, thus characterizing the sudden drop in height caused by actions such as falling or rapid squatting. The minimum ground clearance after the anomaly is subtracted from the baseline value of ground clearance before the anomaly and divided by the sum of the baseline value of ground clearance before the anomaly and the minimum constant to obtain the human height drop ratio. The minimum constant is used to avoid numerical instability caused by the baseline value of ground clearance before the anomaly approaching zero, and the human height drop ratio is used to normalize the height changes of individuals of different heights. The median of human posture angles is taken in the time windows before and after the anomaly to obtain the post-anomaly and pre-anomaly human posture angle values. The absolute difference is calculated and divided by the angle scale constant to obtain the human posture rollover amplitude. The angle scale constant is pi, used to map posture angle changes to a uniform scale, thus ensuring the comparability of posture rollover amplitudes under different anomaly scenarios. Based on the sequence of abrupt changes in ground height within the sliding time window, the maximum value and median absolute deviation are statistically analyzed. The maximum value is divided by the sum of the median absolute deviation and the minimum constant to determine the impact abrupt change intensity. The median absolute deviation characterizes the overall fluctuation level of the abrupt change in ground height, and the maximum value characterizes the peak impact intensity at the moment of the anomaly. The combination of the two can distinguish between slow height changes and sudden impact-type changes. The sum of squares of the human height drop ratio and the human posture rollover amplitude is calculated, and the square root is taken to obtain the comprehensive anomaly amplitude value. The method of using the sum of squares and taking the square root is used to integrate the two orthogonal physical quantities of height change and posture change, avoiding the dominance of a single feature in the anomaly assessment results. The natural logarithm of the sum of the shock mutation intensity and the constant is used to obtain the height-varying shock intensity value. The natural logarithm operation is used to suppress the numerical amplification effect of the shock mutation intensity under extreme conditions, so that the shock intensity value remains smooth under different abnormal scenarios. The anomaly comprehensive amplitude value is multiplied by the height-varying shock intensity value and the opposite number is taken as the exponent. Natural exponentiation is then performed, and the result of the natural exponentiation operation is subtracted from the constant to obtain the anomaly evidence trigger confidence value.By integrating multi-dimensional behavioral characteristics such as changes in human spatial height before and after an abnormal event, the amplitude of posture reversal, and the intensity of abrupt changes during height changes, a unified quantitative assessment of the amplitude, directionality, and impact of abnormal behavior is conducted. An exponential mapping method is used to compress the comprehensive assessment results into a bounded interval, thus forming a continuous characterization of the credibility of abnormal events. This approach can suppress interference from everyday slow behaviors while highlighting the salient characteristics of sudden, high-risk abnormal events, providing a stable, interpretable, and robust basis for determining whether to trigger evidence collection. The obtained anomaly evidence collection trigger credibility value is a continuous value between 0 and 1, used to characterize the credibility of the current abnormal event as a genuine abnormal behavior; a larger value indicates a higher credibility.

[0040] The specific formula for the trusted value triggered by abnormal evidence collection is as follows:

[0041] ;

[0042] In the formula, This represents the confidence value for triggering abnormal evidence collection. It is used to generate a confidence value between 0 and 1 near the time of the abnormality, taking into account the magnitude of the sudden drop in human height, the magnitude of the change in posture, and the suddenness of the change in height. This value is used to determine whether the current abnormality needs to trigger visible light evidence collection. It represents the baseline value of ground clearance before the abnormality, characterizing the normal height level of the human body before the abnormality occurred; This represents the lowest ground clearance value after an anomaly, indicating the minimum height of a human body above the ground after an anomaly occurs. This represents the angle value of the human body posture after the abnormality, indicating the degree of tilt of the human torso relative to the vertical direction after the abnormality occurs. This represents the human body's posture angle value before the abnormality occurs, and characterizes the normal posture angle of the human body before the abnormality occurs. This represents the angular scale constant, mapping angular changes to 0 to 1; This represents the maximum value of the abrupt change in altitude above ground, characterizing the most dramatic instantaneous change in altitude above ground. The median absolute deviation of abrupt changes in altitude describes the overall fluctuation level of the altitude abrupt change sequence; To represent a very small constant and prevent the denominator from being zero, the preferred value is [value missing]. arrive .

[0043] In this implementation plan, a multi-feature fusion assessment mechanism with physical interpretability for abnormal behavior is constructed by collaboratively analyzing changes in human spatial height, the amplitude of posture flipping, and the intensity of abrupt changes in height within the time window before and after an anomaly. This mechanism utilizes robust statistics and normalization processing to enhance its adaptability to individual differences and noise interference, and constrains the assessment results to a finite interval through exponential mapping, thereby achieving a continuous and stable characterization of anomaly credibility. This effectively distinguishes between daily activities and sudden high-risk abnormal behaviors, providing a reliable, robust, and implementable basis for evidence collection and decision-making.

[0044] Specifically, the process of determining whether to trigger evidence collection based on the credibility of anomalies is as follows: The credibility value of anomaly evidence collection trigger is compared with the trigger threshold. The credibility value of anomaly evidence collection trigger is a continuous value obtained by comprehensively evaluating behavioral characteristics such as changes in human spatial height, posture rotation amplitude, and intensity of height change abrupt changes before and after the abnormal event. The trigger threshold is used to distinguish between general behavioral changes and high-risk abnormal behaviors and can be adaptively updated based on the distribution of historical anomaly evidence collection trigger credibility values. When the credibility value of anomaly evidence collection trigger is less than the trigger threshold, the visible light camera module is kept in standby mode. In standby mode, the visible light camera module disables the active evidence collection output function, only maintains basic power supply and control interface listening, does not perform image output operations for storage and transmission, and only continuously performs privacy monitoring based on depth image frames, without outputting visible light evidence collection data. By relying solely on depth image frames for behavior monitoring, unnecessary visible light image collection can be avoided while ensuring the continuity of anomaly detection, thereby reducing the risk of privacy leakage and system computing power and bandwidth consumption. When the credibility value for anomaly evidence collection is greater than or equal to the trigger threshold, the current anomaly is determined to be a credible anomaly. A credible anomaly indicates that the current behavior meets the anomaly judgment conditions in terms of spatial displacement amplitude, attitude change degree, and sudden impact characteristics, and has a high risk and necessity for evidence collection. For credible anomalies, the edge gateway sends a GPIO wake-up control signal to the visible light camera module. The GPIO wake-up control signal is directly output by the edge gateway through the hardware control interface and is used to trigger the working state switch of the visible light camera module within a millisecond time scale, controlling the visible light camera module to switch from standby state to working state. After switching to working state, the visible light camera module begins execution. The document outlines a process for controlling the acquisition of evidence. It extracts visible light image frames covering the time of anomalies from a short-term circular buffer maintained by the visible light camera module. This short-term circular buffer is a ring-shaped cache continuously maintained by the visible light camera module in standby mode with low storage overhead. It temporarily stores visible light image data acquired within a recent period. The buffer duration is preferably 1 to 3 seconds to ensure that the image content corresponding to the time of the anomaly can still be retrieved after the anomaly determination is completed, and visible light acquisition after the anomaly is performed to form an anomaly evidence image frame sequence. This post-anomaly visible light acquisition is used to supplement the behavioral evolution process after the anomaly occurs, providing continuous evidence information. The anomaly evidence image frame sequence is associated with the anomaly evidence trigger credibility value, the anomaly time timestamp, and the target area depth image frames and written into the data distribution database. By uniformly associating and storing visible light evidence data, anomaly credibility indicators, time information, and corresponding depth image data, a structured anomaly evidence data object is formed. This facilitates unified management, rapid location, and efficient access during subsequent PCDN peer-to-peer distribution, collaborative caching, and cloud scheduling.

[0045] This implementation combines multidimensional feature analysis of abnormal human behavior with an evidence-triggering mechanism to achieve accurate identification and graded response to high-risk abnormal events based on continuous privacy monitoring of deep images. Visible light evidence collection is triggered only when the credibility of the anomaly meets the judgment criteria, effectively avoiding redundant collection and privacy leakage risks in irrelevant scenarios. At the same time, a short-term cyclic buffer mechanism is used to achieve complete evidence backtracking at the moment of anomaly occurrence, ensuring the continuity and integrity of evidence data. This method improves the accuracy and real-time performance of anomaly evidence collection while reducing system computing power and bandwidth consumption, enhancing the overall security, interpretability, and engineering feasibility of the monitoring.

[0046] Specifically, after an anomaly is determined to be credible, the unified encapsulation process for anomaly-related evidence is as follows: When an anomaly is determined to be credible, the edge gateway uniformly encapsulates the data corresponding to the anomaly event: it associates and binds the corresponding anomaly evidence image frame sequence, the anomaly time timestamp, and the target area depth image frame to form a health monitoring data distribution object, and generates a unique identifier for the health monitoring data distribution object. The anomaly evidence image frame sequence consists of continuous image data acquired by the visible light camera module before and after the anomaly triggering time, used to completely record the occurrence process and outcome status of the anomaly event; the anomaly time timestamp is an anomaly occurrence time marker determined on the edge gateway side based on a unified time reference, used to establish a precise time correspondence between multi-source data; the target area depth image frame is the depth image data corresponding to the anomaly occurrence time, used to characterize the spatial positional relationship between the human body and the environment when the anomaly event occurs, thereby providing three-dimensional structural reference information for subsequent analysis. The association and binding are preferably achieved by generating a unified data index structure for the same anomaly event on the edge gateway side, logically associating the anomaly evidence image frame sequence, the anomaly time timestamp, and the target area depth image frame with the same anomaly event identifier, avoiding data confusion between different anomalies. The unique identifier is preferably a globally unique identifier generated on the edge gateway side. It is used to uniquely locate and control the distribution objects of health monitoring data during PCDN network distribution, cache collaboration and cloud scheduling, thereby ensuring the consistency, traceability and integrity of abnormal evidence data in the subsequent distribution, storage and access process.

[0047] In this implementation scheme, multi-source forensic data related to trusted anomalies is uniformly encapsulated and associated at the edge, enabling anomaly forensic images, time information, and spatial depth information to form a structurally complete and semantically consistent health monitoring data distribution object. A unique identifier is used to achieve precise positioning and traceable management, effectively avoiding the fragmentation and confusion of anomaly data. This approach ensures the integrity and spatiotemporal consistency of anomaly event forensic information and provides a clear and feasible data foundation for efficient distribution, collaborative caching, and cloud scheduling in the PCDN network, thereby improving the overall reliability, scalability, and engineering feasibility of the system.

[0048] Specifically, the process of evaluating the priority of peer nodes based on their operational status data is as follows: For each candidate peer node, preprocessed device operational data is read, and it is simultaneously determined whether the candidate peer node has cached the health monitoring data distribution object. Cache status is determined by querying the peer node's local cache index table, which reflects whether the node has the ability to directly provide the target data. If cached, a cache hit status value of 1 is assigned; otherwise, a value of 0 is assigned. The cache hit status value is added to a constant to obtain a cache gain value. This cache hit status value characterizes whether the peer node has a local copy of the target data. By explicitly increasing the gain of existing cache nodes, it enables them to obtain the target data in subsequent services. Higher priority ensures that cached data is not repeatedly transmitted during subsequent access services and diffusion scheduling phases; the link stability value is obtained by subtracting the packet loss rate from a constant, with the packet loss rate calculated in real-time by the edge gateway based on the number of sent and received data packets counted by the network protocol stack, used to characterize the data reliability of the current peer link; the link latency penalty value is obtained by taking the reciprocal of the sum of the link round-trip latency and a minimal constant, using the reciprocal form to give low-latency links a greater positive contribution, while using a minimal constant to avoid numerical instability caused by latency approaching zero; the comprehensive load value is obtained by adding and averaging the CPU utilization, NPU utilization, and storage utilization, with each utilization indicator derived from the local system monitoring and capacity of the peer nodes. The runtime interface comprehensively reflects the current computing and storage resource usage of a node. The reciprocal of the sum of the comprehensive load value and a minimal constant is used to obtain a node load penalty value. This ensures that nodes with lower loads and sufficient resource margins receive higher weight during the selection process, preventing the distribution of data to high-load nodes and avoiding performance degradation. Based on the peer-to-peer node connection relationships maintained in the PCDN network, the number of peer-to-peer forwarding nodes required to reach a candidate peer node is calculated using a shortest path search method to obtain the logical hop count. The logical hop count characterizes the forwarding complexity and potential latency risk required for data transmission in the peer-to-peer network. The reciprocal of the sum of the logical hop count and a constant is used to obtain a topology distance penalty value, which further optimizes the topology location and forwarding speed. Nodes with shorter paths have an advantage in distribution optimization. The node carrying capacity value is obtained by multiplying the available bandwidth, link latency penalty value, link stability value, and node load penalty value and taking the fourth root. By multiplying multiple factors and taking the higher root, communication capacity and computing capacity are balanced and integrated, avoiding the excessive amplification of a single indicator on the overall evaluation result. The peer node distribution optimization value is obtained by multiplying the cache gain value, node carrying capacity value, and topology distance penalty value. The peer node distribution optimization value is used to uniformly quantify and rank candidate peer nodes, which is the direct basis for subsequent collaborative distribution node selection and dynamic redistribution decisions, thereby ensuring that the distribution process achieves a balance between real-time performance, stability, and resource utilization efficiency.By explicitly increasing the capacity of existing cache nodes, existing data replicas are prioritized to reduce redundant transmissions. Simultaneously, considering available bandwidth, link latency, and packet loss rate, data is preferentially distributed to nodes with stable communication conditions and high transmission efficiency. Furthermore, by combining the current computing and storage load of nodes with the topology hop count, nodes with limited resources or long paths are naturally suppressed. This optimal value, expressed as a continuous numerical value, characterizes the overall carrying capacity and service capabilities of peer nodes, providing a stable, interpretable, and implementable ranking basis for subsequent collaborative distribution and dynamic scheduling.

[0049] The specific formula for the preferred value of peer node distribution is as follows:

[0050] ;

[0051] In the formula, This represents the preferred value for peer node distribution, comprehensively evaluating the overall distribution suitability of each candidate peer node in terms of bandwidth capacity, link quality, load status, topology location, and cache hit rate, and guiding the priority distribution and collaborative caching selection of health monitoring data distribution objects. This indicates the cache hit status value, and existing cache nodes are explicitly prioritized for amplification to reduce network load. This represents the available bandwidth, characterizing the node's current transmission capacity for data distribution. The higher the bandwidth, the greater the optimal distribution value. This represents the link round-trip time, characterizing the transmission delay of data from the edge gateway to the peer node; This represents the packet loss rate, which characterizes the stability of the link. This represents the overall load value, which is obtained by combining CPU utilization, NPU utilization, and storage utilization, to avoid distributing data to nodes that have no idle capacity. This represents the logical hop count, reflecting the minimum number of peer forwarding nodes traversed between the edge gateway and the peer node. It is used for nearest distribution to reduce cross-node forwarding costs. To represent a very small constant and prevent the denominator from being zero, the preferred value is [value missing]. arrive .

[0052] In this embodiment, Table 1 is a data table of preferred values ​​for peer node distribution. The minimum constant is... The table details the cache hit status, available bandwidth, link round-trip time (RTT), packet loss rate, overall load, logical hop count, and peer distribution optimization value for five candidate peer nodes. Specifically, node 1 has a cache hit status of 1, available bandwidth of 0.90, RTT of 0.20, packet loss rate of 0.02, overall load of 0.30, logical hop count of 1, and peer distribution optimization value of 1.9577; node 2 has a cache hit status of 0, available bandwidth of 0.85, RTT of 0.35, packet loss rate of 0.05, overall load of 0.45, logical hop count of 2, and peer distribution optimization value of 0.5015; node 3 has a cache hit status of 0, available bandwidth of 0.70, RTT of 0.50, packet loss rate of 0.02, overall load of 0.30, logical hop count of 1, and peer distribution optimization value of 1.9577; node 2 has a cache hit status of 0, available bandwidth of 0.70, RTT of 0.50, packet loss rate of 0.02, overall load of 0.30, logical hop count of 1, and peer distribution optimization value of 1.9577; node 3 has a cache hit status of 0, available bandwidth of 0.70, RTT of 0.50, packet loss rate of 0.02, overall load of 0.45, logical hop count of 2, and peer distribution optimization value of 0.5015; and node 3 has a cache hit status of 0, available bandwidth of 0.70, RTT of 0.50, packet loss rate of 0.02, overall load of 0.30, logical hop count of 1, and peer distribution optimization value of 1.9577. The packet loss rate is 0.08, the overall load value is 0.60, the logical hop count is 3, and the preferred value for peer distribution is 0.3026; the cache hit status value of node 4 is 0, the available bandwidth is 0.60, the link round-trip latency is 0.80, the packet loss rate is 0.12, the overall load value is 0.70, the logical hop count is 2, and the preferred value for peer distribution is 0.3284; the cache hit status value of node 5 is 1, the available bandwidth is 0.50, the link round-trip latency is 1.00, the packet loss rate is 0.15, the overall load value is 0.80, the logical hop count is 4, and the preferred value for peer distribution is 0.3415.

[0053] Table 1. Peer Node Distribution Optimal Value Data Table

[0054]

[0055] like Figure 3 The figure shows the distribution ranking of peer node distribution preference values. In the figure, the horizontal axis represents the node number, and the vertical axis represents the corresponding peer node distribution preference value. Larger bar values ​​indicate that the node is more suitable for collaborative distribution and caching of health monitoring data. The figures are sorted in descending order of peer node distribution preference values. (Referring to Table 1 and...) Figure 3 It can be seen that Node 1 has a significantly higher peer distribution priority value than the other candidate nodes, indicating that it has comprehensive advantages in terms of cache hit rate, link conditions, load level, and network topology distance, making it suitable as a priority distribution node. Nodes 2, 4, and 5 are in the middle range, possessing certain distribution capabilities and can be used as redundant cache nodes. Node 3 has a relatively low peer distribution priority value, mainly limited by its large link latency, high overall load, and large number of logical hops, making its distribution suitability relatively weak. Overall, this effectively distinguishes the comprehensive adaptability of different peer nodes in the health monitoring data distribution scenario, providing a clear basis for subsequent collaborative distribution decisions based on priority ranking.

[0056] This implementation scheme achieves a unified quantitative evaluation of the comprehensive carrying capacity and service capabilities of candidate peer nodes by integrating the cache status, link quality, resource load, and network topology location of peer nodes, thus enabling the rational selection of collaborative distribution nodes. On the one hand, by providing explicit gains to cached nodes, existing data copies are reused first to avoid duplicate transmissions and reduce redundant network traffic. On the other hand, by comprehensively considering bandwidth, latency, packet loss rate, and node load, data is preferentially distributed to nodes with stable communication, sufficient resources, and close topological distance, thereby improving distribution efficiency and system stability. Overall, this method can balance resource utilization and network overhead while ensuring real-time performance, providing a stable, interpretable, and implementable decision-making basis for the subsequent collaborative distribution and dynamic scheduling of health monitoring data.

[0057] Specifically, the process of priority-based distribution optimization and collaborative caching is as follows: Candidate peer nodes are sorted in descending order based on their peer node distribution optimization values. The top N candidate peer nodes are selected as collaborative distribution nodes. The descending order ensures that, under current network and resource conditions, nodes with stronger overall carrying capacity and higher service reliability are prioritized for collaborative distribution. N is determined based on the current available candidate peer node scale and abnormal data access needs, preferably 2 to 5, to avoid single-point cache failures and prevent excessive waste of network and storage resources caused by excessive duplicate distribution. The system determines whether the collaborative distribution node has cached the corresponding health monitoring data distribution object. The edge gateway, based on the peer-to-peer transmission mechanism in the PCDN network, sends the health monitoring data distribution object to the uncached collaborative distribution node. The uncached collaborative distribution node refers to a node that has not yet stored the target health monitoring data distribution object in its local cache space. By only transmitting actual data to uncached nodes, duplicate data transmission to nodes that already have data copies can be avoided, thereby reducing network redundancy traffic and improving distribution efficiency. The peer-to-peer transmission mechanism is preferably implemented based on a P2P data transmission protocol to support direct and multi-hop data transmission between nodes. After receiving a health monitoring data distribution object, the collaborative distribution node writes it to its local cache space and updates the cache hit status of the corresponding object to "cached". The local cache space is preferably a semi-persistent storage area pre-allocated by the peer node to store health monitoring data distribution objects within a certain period. Updating the cache hit status is achieved by maintaining the node's local cache index table to quickly respond to data access requests from authorized terminals and other nodes. When a collaborative distribution node experiences link interruption, abnormal load, or cache failure during distribution, the edge gateway selects candidate peer nodes for redistribution based on descending order. The determination of link interruption is preferably achieved through network connectivity detection and link quality monitoring. Specifically, when the edge gateway fails to receive a heartbeat response from the target peer node for multiple consecutive detection periods, or detects that the link round-trip latency continuously exceeds a latency threshold (e.g., 500 milliseconds to 2000 milliseconds) for a set duration preferably 3 to 10 seconds, a link interruption is determined to have occurred at the peer node. The determination of abnormal load is preferably based on a comprehensive judgment of the peer node's operating status data. When any one of the peer node's CPU utilization, NPU utilization, or storage occupancy rate continuously exceeds the corresponding load threshold, preferably 80% to 95%, and continues for a set duration, preferably 5 to 30 seconds, the peer node is determined to be in an abnormal load state and is not suitable to continue participating in the current health monitoring data distribution process. The determination of cache failure is preferably achieved through the feedback of the cache write result after the peer node receives the health monitoring data distribution object. When the peer node returns a cache write failure status code and a verification failure identifier, a cache failure is determined to have occurred. The status code and confirmation information are fed back to the edge gateway by the peer node's local cache management module through the peer transmission protocol.When any of the above-mentioned abnormal conditions are detected, the edge gateway will mark the corresponding peer node as currently unavailable for distribution. Based on the descending sorting result of the peer node distribution preference value, it will automatically select the next candidate peer node to perform alternative distribution, thereby ensuring the continuity, reliability, and fault tolerance of the health monitoring data distribution process. This approach effectively controls the number of distributions and network overhead while ensuring data distribution coverage, giving the collaborative caching process good scalability and fault tolerance, and enabling it to stably support the efficient distribution and access of health monitoring data in complex network environments.

[0058] In this implementation scheme, by ranking the preferred distribution values ​​of peer nodes, a small number of nodes with strong carrying capacity and high service reliability are dynamically selected to participate in collaborative distribution. Actual data transmission is only performed to nodes that are not cached, effectively avoiding network redundancy caused by repeated distribution. Simultaneously, in the event of node anomalies or link fluctuations, alternative distribution can be quickly completed based on the preferred results, ensuring the continuity and reliability of data distribution. By controlling the number of collaborative distribution nodes and combining it with a caching collaboration mechanism, the availability and access efficiency of anomaly evidence data are improved while significantly reducing network bandwidth and storage resource consumption. This gives the system good scalability, fault tolerance, and engineering feasibility, making it suitable for health monitoring data distribution scenarios in complex network environments.

[0059] Specifically, the process of performing global scheduling of peer nodes based on constraint rules and guiding the distribution rhythm and access path under different network conditions to achieve efficient access and long-term analysis of health monitoring data is as follows: When it is detected that the current conditions are not under network load constraints and outbound bandwidth constraints, a supplementary distribution instruction is issued to the edge gateway to re-evaluate the preferred distribution value of peer nodes and control the edge gateway to perform supplementary distribution on the health monitoring data distribution objects; the supplementary distribution recalculates the preferred distribution value of each candidate peer node and selects peer nodes that have not been cached or whose cached replicas are lower than the target redundancy threshold to perform actual data distribution, in order to improve the replica coverage and access reachability of health monitoring data distribution objects in the PCDN network; the target redundancy threshold is preferably 2 to 5 replicas, in order to improve fault tolerance while avoiding invalid diffusion. When the current conditions are detected to be under network load constraints or outbound bandwidth limitations, delayed distribution is performed on the health monitoring data distribution objects, maintaining only the cache availability of the already distributed collaborative distribution nodes. Delayed distribution means pausing the issuance of new data copies, only keeping the cached state of the existing collaborative distribution nodes for the health monitoring data distribution objects intact. This method avoids introducing additional transmission load under network congestion and bandwidth constraints, thereby ensuring the stable operation of deployed services. Specifically, network load constraints are determined by counting the number of access requests per unit time within the PCDN network on the cloud side, based on a sliding time window preferably 1 to 5 minutes. When the number of requests exceeds the access load threshold, it is determined to be under network load. Outbound bandwidth limitations are determined by real-time collection of regional network outbound bandwidth utilization. When the bandwidth utilization exceeds the utilization threshold preferably 70% to 85% and continues for a set duration preferably 30 to 60 seconds, it is determined to be an outbound bandwidth limited state. When neither of the above two constraints is triggered, the current network is determined to be in a state of supplementary distribution, with real-time transmission margin for data copy diffusion. When an authorized terminal initiates an access request to a health monitoring data distribution object, the cache distribution status in the PCDN network is queried based on the unique identifier. Access services are then provided from the cached peer nodes according to the peer node distribution preference order. The unique identifier is used to accurately locate the corresponding health monitoring data distribution object in the PCDN network, and the cache distribution status is obtained by querying the peer node cache index table. Providing access services according to the peer node distribution preference order can prioritize guiding access requests to nodes with better communication quality, lower load, and closer topological distance, thereby reducing access latency and reducing network overhead caused by cross-node forwarding.Simultaneously, based on continuously monitored ground clearance, human posture angles, and health monitoring data distribution targets, health status analysis is performed on the monitored subjects, and health data monitoring reports are generated. Among them, continuously monitored ground clearance and human posture angles are used to depict the behavioral change trends of the monitored subjects over time, while abnormal evidence data from the health monitoring data distribution targets provides event-level reference for health analysis. The combination of the two can support comprehensive analysis of health indicators such as fall frequency, abnormal duration, and recovery status. Health data monitoring reports are generated periodically to provide readable health status assessment results to authorized terminals and management platforms.

[0060] like Figure 5 The diagram illustrates the layered architecture of a PCDN-based health monitoring system. It shows the layered collaborative architecture used in the PCDN-based health monitoring data distribution system, comprising, from bottom to top, a perception layer, an edge layer, a transmission layer, and a cloud layer. The perception layer includes multimodal perception terminals deployed within the monitored individual's living environment, preferably RGB-D depth camera modules, used to continuously collect human depth image data and provide basic perception information for abnormal behavior identification. The edge layer includes intelligent gateways directly connected to the perception terminals. These gateways possess local computing and control capabilities, used for preprocessing the collected depth image data, extracting human behavioral features, and identifying abnormal events. When a credible abnormality is identified, they trigger evidence collection and data encapsulation, and also participate in data distribution and caching collaboration as edge nodes in the PCDN network. The transmission layer is a PCDN distributed network composed of multiple peer nodes. These peer nodes form a multi-path collaborative structure through peer-to-peer connections, used to carry collaborative caching and proximity access for health monitoring data distribution objects, and dynamically adjust data distribution and access paths based on node operating status and topology. The cloud layer comprises a cloud server cluster, used for global scheduling of distribution status and access load within the PCDN network, guiding distribution rhythm under different network conditions, and performing health status analysis and trend assessment based on long-term accumulated health monitoring data, providing authorized terminals with health data monitoring reports and service access support. This layered architecture enables collaborative work among sensing, computing, distribution, and analysis, effectively reducing centralized back-to-origin bandwidth pressure while ensuring privacy protection and real-time response, and improving the distribution efficiency and overall scalability of health monitoring data.

[0061] This implementation scheme introduces a global scheduling mechanism with dual constraints of network load and egress bandwidth. When network conditions are favorable, it adaptively distributes supplementary health monitoring data, improving data copy coverage and accessibility. When network congestion or bandwidth limitations occur, it automatically switches to delayed distribution, maintaining only the availability of existing caches, thereby effectively suppressing redundant transmissions and ensuring stable system operation. Simultaneously, it guides access paths using peer-to-peer distribution optimization values, ensuring authorized terminals prioritize data acquisition from nodes with high communication quality, low load, and close topological distance, significantly reducing access latency and network overhead. Furthermore, it integrates long-term behavioral monitoring data and anomaly evidence information to achieve continuous analysis and periodic assessment of the health status of monitored objects, balancing scalability, service reliability, and practical value in health management.

[0062] Reference Figure 2 As shown, the second aspect of this invention provides a PCDN-based health monitoring data distribution system, applied to the aforementioned PCDN-based health monitoring data distribution method, comprising: a multi-source data acquisition and processing module, used to acquire depth image data in real time in the living environment of the monitored object, and to acquire the operational status data of peer nodes in real time, and to preprocess the multi-source data; an anomaly identification and evidence collection triggering module, used to extract human behavioral structural features based on the preprocessed depth image data, identify abnormal events, analyze changes in human posture and spatial displacement before and after the abnormal event, comprehensively evaluate the credibility of the anomaly, and determine whether evidence collection needs to be triggered based on the credibility of the anomaly; a PCDN collaborative distribution module, used to uniformly encapsulate the evidence related to the anomaly after it is determined to be a credible anomaly, combine the operational status data of peer nodes to evaluate the priority of peer nodes, and perform distribution optimization and collaborative caching according to the priority; and a cloud scheduling service module, used to perform global scheduling of peer nodes based on constraint rules, and guide the distribution rhythm and access path under different network conditions, to achieve efficient access and long-term analysis of health monitoring data.

[0063] This implementation plan organically integrates depth-sensing monitoring, anomaly credibility assessment, and PCDN collaborative distribution mechanism to optimize the entire process of health monitoring data from collection and identification to distribution and access. On the one hand, relying on the extraction of human structural features and anomaly credibility determination from depth images, it enables accurate triggering and evidence collection of high-risk abnormal events while ensuring privacy. On the other hand, it optimizes distribution and performs collaborative caching based on the running status of peer nodes, and dynamically adjusts the distribution rhythm and access path under cloud-constrained scheduling. This effectively reduces the pressure of centralized back-to-source and redundant transmission overhead, improves the timeliness of data access and the overall scalability of the system, thereby providing reliable support for community-level, long-term stable deployment of health monitoring services.

[0064] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0065] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for distributing health monitoring data based on PCDN, characterized in that, Includes the following steps: S1 collects depth image data in real time in the living environment of the monitored object, and collects the operating status data of peer nodes in real time, and preprocesses the multi-source data. S2, based on preprocessed depth image data, extracts human behavioral structural features, identifies abnormal events, analyzes changes in human posture and spatial displacement before and after abnormal events, comprehensively evaluates the credibility of the anomaly, and determines whether evidence collection needs to be triggered based on the credibility of the anomaly. S3, once a trusted anomaly is identified, the evidence related to the anomaly is uniformly encapsulated, and the priority of the peer node is evaluated by combining the running status data of the peer node. Based on the priority, the distribution optimization and collaborative caching are performed. S4 performs global scheduling on peer nodes based on constraint rules and guides the distribution rhythm and access path under different network conditions to enable efficient access and long-term analysis of health monitoring data.

2. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of real-time acquisition of depth image data in the living environment of the monitored object, real-time acquisition of operational status data of peer nodes, and preprocessing of multi-source data is as follows: By deploying RGB-D depth camera modules in the living environment of the monitored object, depth image frames are acquired in real time, including depth pixel values ​​and pixel coordinates; at the same time, in the PCDN network, peer nodes with peer connections to the current edge gateway are obtained to obtain a set of candidate peer nodes, and the device operation data of the candidate peer nodes are collected in real time, including: CPU utilization, NPU utilization, available bandwidth, link round-trip latency, packet loss rate and storage occupancy rate. The acquired depth image frames and device operation data are time-aligned; for the depth image frames, median filtering is used for noise suppression, and hole pixels are filled based on neighborhood interpolation; the depth pixel values ​​in the depth image frames are scaled uniformly; the device operation data is normalized; based on the preprocessed depth image frames, a foreground segmentation algorithm is used to extract the human activity area to obtain the target area depth image frame; a data distribution database is established, and the depth image frames, target area depth image frames, and device operation data are written into the data distribution database.

3. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of extracting human behavioral structural features and identifying abnormal events based on preprocessed depth image data is as follows: Based on the depth image frame of the target region, the depth pixels are statistically analyzed in layers along the vertical direction. In the upper layer, the pixel position with the largest vertical pixel coordinate is selected as the head key point. In the lower layer, the pixel position with the smallest vertical pixel coordinate is selected as the lower limb key point. In the middle layer, the geometric center point of the pixel position is selected as the torso key point. The set of human body key points is obtained by statistics. Read the depth pixel values ​​of the torso key points in the depth image frame of the target area, determine the ground reference depth based on the depth pixel values ​​of the ground area in the depth image frame of the target area, and calculate the difference between the depth pixel values ​​of the torso key points and the ground reference depth to obtain the height above the ground. Based on the set of human body key points, head key points and torso key points are selected to construct the human torso direction vector, and the human posture angle is calculated based on the angle between the human torso direction vector and the vertical direction vector. The change in ground height is obtained by calculating the difference between adjacent time intervals of the ground height value, and the second-order difference is calculated on the change in ground height to obtain the abrupt change value of the ground height change; the change in human posture angle is obtained by calculating the difference between adjacent time intervals of the human posture angle; the sampling time when the absolute value of the abrupt change value of the ground height change or the absolute value of the change in human posture angle exceeds the corresponding change threshold is selected as the abnormal time; a sliding time window is constructed with the abnormal time as the center point, and the time window before the abnormality and the time window after the abnormality are divided.

4. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process for analyzing changes in human posture and spatial displacement before and after an abnormal event, and comprehensively assessing the credibility of the anomaly, is as follows: The median value of the ground clearance in the time window before the anomaly is taken to obtain the ground clearance baseline value before the anomaly; the minimum value of the ground clearance in the time window after the anomaly is taken to obtain the ground clearance minimum value after the anomaly; the ground clearance baseline value before the anomaly is subtracted from the ground clearance minimum value after the anomaly, and then divided by the sum of the ground clearance baseline value before the anomaly and the minimum constant, to obtain the sudden drop ratio of human height. The median of the human posture angles in the time window after the anomaly and the time window before the anomaly are taken respectively to obtain the human posture angle values ​​after the anomaly and before the anomaly. The absolute difference is calculated and then divided by the angle scale constant to obtain the human posture flipping amplitude. Based on the sequence of abrupt changes in ground clearance within a sliding time window, the maximum value and median absolute deviation are statistically analyzed, and the maximum value is divided by the sum of the median absolute deviation and the minimum constant to determine the impact abrupt change intensity. The sum of the squares of the sudden drop in human height and the amplitude of human posture rotation is calculated, and the square root is taken to obtain the abnormal comprehensive amplitude value; the natural logarithm of the sum of the impact mutation intensity and constant 1 is taken to obtain the height change impact intensity value; the abnormal comprehensive amplitude value and the height change impact intensity value are multiplied and the opposite number is taken as the exponent for natural exponent calculation, and the result of the natural exponent calculation is subtracted from constant 1 to obtain the credibility value of the abnormal evidence trigger.

5. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process for determining whether to trigger evidence collection based on the credibility of the anomaly is as follows: Compare the confidence value for anomaly evidence collection with the trigger threshold: When the confidence value for abnormal evidence collection is less than the trigger threshold, the visible light camera module is kept in standby mode, and privacy monitoring is continuously performed based only on depth image frames, without outputting visible light evidence collection data. When the trusted trigger value for anomaly evidence collection is greater than or equal to the trigger threshold, the current anomaly is determined to be a trusted anomaly. For trusted anomalies, the edge gateway sends a GPIO wake-up control signal to the visible light camera module to switch the visible light camera module from standby to working state. It extracts the visible light image frame covering the time of the anomaly from the circular buffer maintained by the visible light camera module and performs visible light acquisition after the anomaly to form an anomaly evidence collection image frame sequence. The anomaly evidence collection image frame sequence is associated with the trusted trigger value for anomaly evidence collection, the timestamp of the time of the anomaly, and the depth image frame of the target area and written into the data distribution database.

6. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of uniformly encapsulating the evidence related to the anomaly after it has been determined to be a credible anomaly is as follows: When an anomaly is determined to be a credible anomaly, the edge gateway performs unified encapsulation of the data corresponding to the anomaly event: it associates and binds the corresponding anomaly evidence image frame sequence, the anomaly time timestamp, and the target area depth image frame to form a health monitoring data distribution object, and generates a unique identifier for the health monitoring data distribution object.

7. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of evaluating the priority of peer nodes by combining their operational status data is as follows: For each candidate peer node, read the preprocessed device operation data, and synchronously determine whether the candidate peer node has cached the health monitoring data distribution object. If it has been cached, assign a cache hit status value of 1; otherwise, assign a value of 0. Add the assigned cache hit status value to a constant to obtain the cache gain value. The link stability value is obtained by subtracting the packet loss rate from the constant, and the link delay penalty value is obtained by taking the reciprocal of the sum of the link round-trip delay and the minimum constant. The overall load value is obtained by adding up the CPU utilization, NPU utilization, and storage utilization and calculating the average value. The node load penalty value is obtained by taking the reciprocal of the sum of the overall load value and a very small constant. Based on the peer node connection relationship maintained in the PCDN network, the number of peer forwarding nodes required to reach a candidate peer node is calculated using the shortest path search method to obtain the logical hop count. The reciprocal of the sum of the logical hop count and a constant is taken to obtain the topology distance penalty value. The node carrying capacity value is obtained by multiplying the available bandwidth, link latency penalty value, link stability value and node load penalty value and taking the fourth root. The optimal value for peer node distribution is obtained by multiplying the cache gain value, node carrying capacity value and topology distance penalty value.

8. The method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of performing priority-based distribution optimization and collaborative caching is as follows: Candidate peer nodes are sorted in descending order based on the peer node distribution preference value, and the top N candidate peer nodes are selected as collaborative distribution nodes. The edge gateway sends the health monitoring data distribution object to the uncached collaborative distribution node based on the peer transmission mechanism in the PCDN network. After receiving the health monitoring data distribution object, the collaborative distribution node writes it to the local cache space and updates the cache hit status of the corresponding object to be cached. When a collaborative distribution node experiences link interruption, abnormal load, or cache failure during the distribution process, the edge gateway selects candidate peer nodes for redistribution based on the descending order of the distribution results.

9. A method for distributing health monitoring data based on PCDN according to claim 1, characterized in that, The specific process of performing global scheduling of peer nodes based on constraint rules, guiding the distribution rhythm and access path under different network conditions, and enabling efficient access and long-term analysis of health monitoring data is as follows: When it is detected that the current conditions are not under network load period constraints and outbound bandwidth constraints, a supplementary distribution instruction is issued to the edge gateway to re-evaluate the peer node distribution preference value and control the edge gateway to perform supplementary distribution on the health monitoring data distribution object; When it is detected that the current conditions are under network load constraints or outbound bandwidth constraints, delayed distribution is performed on the health monitoring data distribution objects, and only the cache availability of the already distributed collaborative distribution nodes is maintained; When an authorized terminal initiates an access request to the health monitoring data distribution object, the cache distribution status in the PCDN network is queried based on the unique identifier, and access services are provided from the cached peer nodes according to the peer node distribution preference order. Meanwhile, based on continuously monitored ground height, human posture angles, and health monitoring data distribution targets, health status analysis is performed on the monitored individuals, and health data monitoring reports are generated.

10. A health monitoring data distribution system based on PCDN, characterized in that, include: The multi-source data acquisition and processing module is used to acquire depth image data in real time in the living environment of the monitored object, and to acquire the operating status data of peer nodes in real time, and to preprocess the multi-source data. The anomaly identification and evidence collection triggering module is used to extract human behavioral structural features based on preprocessed depth image data, identify abnormal events, analyze changes in human posture and spatial displacement before and after the abnormal event, comprehensively evaluate the credibility of the anomaly, and determine whether evidence collection needs to be triggered based on the credibility of the anomaly. The PCDN collaborative distribution module is used to uniformly encapsulate the evidence related to the anomaly after it is determined to be a trusted anomaly. It combines the running status data of peer nodes to evaluate the priority of peer nodes and performs distribution optimization and collaborative caching according to the priority. The cloud-based scheduling service module is used to perform global scheduling on peer nodes based on constraint rules, and guide the distribution rhythm and access path under different network conditions to enable efficient access and long-term analysis of health monitoring data.

Images