Ward nursing monitoring data processing method and system based on big data

Abstract

The present application relates to the technical field of data processing, in particular to a ward nursing monitoring data processing method and system based on big data, comprising the following steps: acquiring monitoring image frames and nursing big data samples, extracting standard action sequence numbers, screening offset segments, completing occlusion contours, counting number repetition and physiological fluctuations, generating stability control data, comparing node offset to generate label mapping, constructing risk score and abnormal trajectory, and outputting abnormal focus information group. The present application realizes accurate screening of behavior dislocation paragraphs by comparing the position difference of the action node sequence in the image frames with the standard action sequence in the nursing task list, realizes focused labeling of abnormal nodes in a high-frequency update environment by extracting the time sequence features and spatial offset parameters of the action nodes, and constructing the score and trajectory segment on the time axis combined with various parameters, and enhances the structured processing capability and abnormal recognition accuracy of nursing monitoring behavior.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for processing ward nursing monitoring data based on big data. Background Technology

[0002] The field of data processing technology involves the collection, storage, analysis, and mining of multi-source data. Its core aspects include data modeling, data fusion, pattern recognition, statistical computation, and predictive analysis. This technology is widely applied in various industries such as healthcare, finance, transportation, and industrial manufacturing. It aims to extract valuable information from large-scale datasets of structured and unstructured data to support business decision-making and automated management. In the healthcare context, data processing extends to the dynamic processing and application of patient vital signs, treatment records, and nursing behaviors. By constructing intelligent data analysis mechanisms oriented towards clinical practice and nursing, it improves the efficiency and responsiveness of healthcare management.

[0003] Traditional ward nursing monitoring data processing methods refer to the analysis and processing of monitoring data generated during patient care in wards. These methods typically rely on image and audio data collected by monitoring equipment deployed within the ward. Key events are screened and recorded manually through playback, or static rules are used to trigger detection of abnormal patient behavior. Combined with basic data such as medical records and nursing plans, manual input and interpretation are essential for assessing nursing behavior, tracking nursing quality, and statistically analyzing nursing indicators. Traditional methods often rely on fixed-threshold frame difference comparison, target trajectory tracking, or image region comparison for image and audio content analysis. They lack the ability to jointly process multi-dimensional data, and the overall processing flow depends on human judgment and post-event analysis. They lack structured extraction and behavioral classification models for complex nursing events and cannot effectively address the high-frequency updates and semantic relationship modeling requirements of monitoring data in a big data environment. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a data processing method and system for ward nursing monitoring based on big data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a ward nursing monitoring data processing method based on big data, comprising the following steps:

[0006] S1: Obtain continuous frame images from ward nursing monitoring videos, extract the corresponding standard action sequence numbers from the patient nursing task list based on nursing big data samples, determine whether the cumulative difference between adjacent numbers exceeds the set sequence offset threshold, filter out the position segment sequences with offsets, and generate a set of nursing behavior misalignment segments.

[0007] S2: Call the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extract the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculate the direction deviation of the gradient values ​​of the Sobel operator inside and outside the edge of the occluded area, compare the trend of the occluded boundary path change in continuous frames, record the information of the reconstructed contour closed line segment, and generate the contour completion structure of the occluded area.

[0008] S3: Based on the outline completion structure of the occluded area, the standard deviation of the number of repeated numbers between adjacent nodes and the time interval is calculated, and the respiratory rate and heart rate variability consistent with the time segment of the key physiological indicators monitored during the nursing period are extracted. The time continuity parameters and physiological fluctuations of each action sequence are combined to generate a control data set of nursing behavior stability.

[0009] S4: Based on the occluded area contour completion structure and the set of nursing behavior misalignment segments, extract the action nodes within the contour closed area in the monitoring image, and compare the Euclidean distance between the marked position of each action node in the image frame and the task target area position of the corresponding action label, filter the node coordinates that exceed the offset range, and generate an occluded area behavior offset label mapping set.

[0010] S5: Call the nursing behavior stability control data set and the occluded area behavior offset label mapping set to construct the risk score of each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, establish a monitoring label structure corresponding to patient identifiers and nursing unit identifiers, and generate a ward nursing monitoring behavior abnormality focus information group.

[0011] The present invention improves upon the following: the set of misaligned nursing behavior segments includes execution node index numbers, action path time segments, and sequence misalignment offset values; the occluded area contour completion structure includes pixel coordinates of edge closed connection segments, occlusion boundary extension path vectors, and the length of the matching sliding window between frame sequences; the nursing behavior stability control data set includes behavior label stability coefficients, physiological index fluctuations, and action execution cycle standard deviations; the occluded area behavior offset label mapping set includes label position offset differences, coordinate deviation node indexes, and task target area offset markers; and the ward nursing monitoring behavior anomaly focusing information set includes scoring node number sequences, abnormal path segment trajectory coordinates, and trajectory risk level identifiers.

[0012] The present invention is improved in that the specific steps for obtaining the set of misaligned nursing behavior segments are as follows:

[0013] S111: Acquire continuous frame images from ward nursing monitoring video, identify the action behavior area of ​​nursing staff in the frame, extract the time label, behavior target number and behavior sequence mark from the image frame, sort all extracted data according to the time of action occurrence, establish an action time sequence list, and generate a nursing action node sequence list.

[0014] S112: Based on the behavior target number and time tag in the nursing action node sequence list, retrieve the corresponding nursing task item number and standard operation sequence number from the patient nursing task list, construct the standard action sequence and match it one-to-one with the sequence mark in the nursing action node sequence list, calculate the difference between the two numbers and accumulate the offset value as the offset judgment index, and obtain the sequence number offset comparison result set.

[0015] S113: Based on the cumulative offset value of each item in the sequence number offset comparison result set, determine whether it exceeds the preset sequence offset threshold, filter the segment index corresponding to the number whose cumulative offset value exceeds the sequence offset threshold, and organize the action node segments in the corresponding time range into a set to establish a nursing behavior misalignment segment set.

[0016] The present invention is improved in that the step of obtaining the contour completion structure of the occluded area is specifically as follows:

[0017] S211: Call the image frame corresponding to each action node in the set of misaligned nursing behavior segments, extract the image pixel matrix in the current frame, perform Canny edge detection on the edge region in the image, identify the gray-scale change region between the nursing staff and the bed, and convert the region into a binary edge map structure to generate the edge pixel structure map of the monitoring image.

[0018] S212: Based on the location coordinates of the edge region of the hospital bed and the pixel distribution of the neighboring region in the edge pixel structure diagram of the monitoring image, extract the Sobel gradient vector group inside and outside the occluded block, obtain the gradient values ​​of the x and y directions corresponding to each pixel, calculate the gradient direction difference value, and match the point pair with the smallest gradient direction difference value to generate occluded edge direction comparison information.

[0019] S213: Based on the coordinate position sequence of the matching point pairs in the occlusion edge direction comparison information, construct the pixel displacement path between adjacent frame images, construct the contour connection trajectory according to the contour change trend of continuous frames, identify the spatial changes of connected segments in the point pair path and record the closed edge position index, and generate the contour completion structure of the occlusion area.

[0020] The present invention is improved in that the formula for obtaining the gradient direction difference value is specifically as follows:

[0021] ;

[0022] in, This represents the normalized gradient vector of the pixels within the occluded region of the i-th pair of edge points. This represents the normalized gradient vector of the pixel outside the occluded region in the i-th pair of edge points. Let represent the magnitude of the normalized gradient vector within the occlusion region at the i-th pair of edge points. This represents the magnitude of the normalized gradient vector outside the occlusion in the i-th pair of edge points. This represents the normalized Euclidean distance between the coordinates of the i-th pair of edge points. Represents the spatial distance suppression constant. This represents the difference in gradient direction between the i-th pair of edge points.

[0023] The present invention is improved in that the steps for obtaining the nursing behavior stability control data group are specifically as follows:

[0024] S311: Based on the time sequence label of each segment in the contour completion structure of the occluded area, extract each nursing action node corresponding to the segment, obtain the frame index length between the start frame and the end frame of the node, calculate the corresponding action duration value in combination with the frame frequency, and extract the action number as the behavior label. After summarizing, establish an action duration and label number table.

[0025] S312: Call the action duration and the tag number sequence in the tag number table, count the cumulative number of repeated number items between adjacent nodes, calculate the standard deviation of the action duration time interval between adjacent nodes, construct an interval stability parameter set based on the repetition frequency and the degree of time fluctuation, and generate a behavior sequence stability evaluation matrix.

[0026] S313: Based on the time window of each segment in the behavioral sequence stability assessment matrix, extract the respiratory rate value and heart rate interval sequence synchronously collected in the nursing monitoring data, count the number of fluctuations of physiological signals within the segment, and bind them with the segment corresponding to the stability parameter. After fusion, generate a control structure data table and establish a nursing behavior stability control data group.

[0027] The present invention is improved in that the step of obtaining the occlusion region behavior offset label mapping set is specifically as follows:

[0028] S411: Based on the mapping relationship between the time index and node number in the set of misaligned nursing behavior segments of the occluded area contour completion structure, locate the corresponding image frame number for each mapped node, retrieve the set of pixel coordinates of the closed contour area in the image frame, identify the spatial position of the action behavior annotation point in the area, and extract the coordinate point set of the action node in the area to generate the coordinate group of the action node in the contour.

[0029] S412: Call the image frame position coordinates of each node in the action node coordinate group within the contour, calculate the Euclidean distance with the reference center coordinates of the task target area corresponding to the action label, determine whether the Euclidean distance exceeds the allowable offset threshold for task execution, filter the nodes that exceed the threshold and record the offset position and label number, and generate a node spatial offset label sequence.

[0030] S413: Based on the coordinate point information, label number and frame number index in the node spatial offset label sequence, construct a mapping structure to record the coordinates of the abnormal location points corresponding to the behavior labels, organize the index paths of the offset nodes by frame, and perform cross-identification by the corresponding number of the task target area to establish a behavior offset label mapping set for the occluded area.

[0031] The present invention is improved in that the steps for obtaining the abnormal behavior focusing information group in the ward nursing monitoring are specifically as follows:

[0032] S511: Call the action number, action duration value and spatial offset distance value corresponding to each data segment in the nursing behavior stability control data group and the occluded area behavior offset label mapping set, construct a two-dimensional mapping matrix for each behavior node with time index as the horizontal axis and stability parameter and spatial offset as the vertical axis, calculate the behavior risk score for each node, and generate a behavior node score matrix.

[0033] S512: Based on the behavior node scoring matrix, the node numbers that exceed the preset abnormal judgment threshold are filtered, and the nodes are marked in time index order. The connection relationship between continuously distributed nodes is identified, the trajectory connection path structure between nodes is constructed, and the time series index and spatial coordinate sequence of abnormal nodes are recorded to obtain the abnormal node trajectory connection set.

[0034] S513: Based on the trajectory sequence in the trajectory connection set of the abnormal nodes, the abnormal path is structurally mapped and bound to the corresponding action number, patient unique identifier and nursing unit number. A monitoring graph structure index is constructed according to the attribution relationship of each trajectory segment. High-frequency nodes in the trajectory path are extracted for focused annotation, and a ward nursing monitoring behavior abnormal focused information group is established.

[0035] A big data-based ward nursing monitoring data processing system is provided to implement the aforementioned big data-based ward nursing monitoring data processing method. The system includes:

[0036] The behavior misalignment recognition module acquires continuous frame images from ward nursing monitoring videos, extracts the corresponding standard action sequence numbers from the patient nursing task list based on nursing big data samples, determines whether the cumulative difference of the offset between adjacent numbers exceeds the set sequence offset threshold, filters out the position segment sequences with offsets, and generates a set of nursing behavior misalignment segments.

[0037] The occlusion area analysis module calls the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extracts the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculates the direction deviation of the gradient values ​​of the Sobel operator inside and outside the occlusion area edge, compares the trend of occlusion boundary path change in continuous frames, records the information of the reconstructed contour closed line segment, and generates the contour completion structure of the occlusion area.

[0038] The behavioral parameter analysis module, based on the outline completion structure of the occluded area, calculates the standard deviation of the number of repeated numbers between adjacent nodes and the time interval, and extracts the respiratory rate and heart rate variability consistent with the time segment from the key physiological indicators monitored during the nursing period. It combines and generates time continuity parameters and physiological fluctuation quantity comparison data under each action sequence to generate a nursing behavior stability comparison data group.

[0039] The behavior offset segmentation module extracts action nodes within the contour closed area in the monitoring image based on the occluded area contour completion structure and the set of nursing behavior misalignment segments. It then compares the Euclidean distance between the marked position of each action node in the image frame and the task target area position of the corresponding action label, filters out node coordinates that exceed the offset range, and generates an occluded area behavior offset label mapping set.

[0040] The abnormal information processing module calls the nursing behavior stability control data set and the occluded area behavior offset label mapping set to construct a risk score for each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, establish a monitoring label structure corresponding to patient identifiers and nursing unit identifiers, and generate a ward nursing monitoring behavior abnormality focus information set.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0042] In this invention, by comparing the positional difference between the sequence of action nodes in an image frame and the standard action sequence in the nursing task list, accurate screening of misaligned segments of behavior is achieved. Closed boundaries are constructed by using the edge direction of the occluded area and contour completion strategies, enabling structural reconstruction and continuous tracking of the occluded area in the image. By extracting the temporal sequence features and spatial offset parameters of the action nodes and fusing respiratory rate and heart rate variability data consistent with the segment time, a comparison mechanism between temporal continuity and physiological fluctuations is established, achieving stable modeling of nursing behavior. Abnormal behavior areas are marked by spatial comparison of the degree of node offset and the distance to the task target area. By combining various parameters, a score and trajectory segment are constructed on the time axis, enabling focused marking of abnormal nodes in a high-frequency update environment, enhancing the structured processing capability and anomaly recognition accuracy of nursing monitoring behavior. Attached Figure Description

[0043] Figure 1 This is a flowchart of the method of the present invention;

[0044] Figure 2 This is a flowchart illustrating the process of obtaining a set of misaligned nursing behavior segments according to the present invention.

[0045] Figure 3 This is a flowchart illustrating the process of obtaining the contour completion structure of the occluded region according to the present invention;

[0046] Figure 4 This is a flowchart illustrating the process of obtaining a control group of stable nursing behaviors according to the present invention.

[0047] Figure 5 This is a flowchart illustrating the process of obtaining the behavior offset label mapping set of the occluded region according to the present invention;

[0048] Figure 6 This is a flowchart for the present invention to obtain a group of abnormal behavior focus information in ward nursing monitoring. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0050] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0051] Please see Figure 1 This invention provides a technical solution: a method for processing ward nursing monitoring data based on big data, comprising the following steps:

[0052] S1: Obtain continuous frame images from the ward nursing monitoring video, extract the action occurrence time, operation object number and sequence number from each frame image, extract the corresponding standard action sequence number from the patient nursing task list based on the nursing big data sample, perform position difference calculation based on the two sets of sequence numbers, determine whether the cumulative difference of the offset between adjacent numbers exceeds the set sequence offset threshold, filter the position segment sequence with offset, and generate a set of nursing behavior misalignment segments.

[0053] S2: Call the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extract the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculate the direction deviation of the gradient values ​​of the Sobel operator inside and outside the edge of the occluded area, locate the extension direction between the point pairs with the smallest gradient direction difference, compare the trend of occlusion boundary path change in continuous frames, record the information of the reconstructed contour closed line segment, and generate the contour completion structure of the occluded area.

[0054] The gradient value of the Sobel operator is a commonly used edge detection metric in image processing. It uses the gradient changes of gray values ​​in the image along the x and y axes to extract edges. Points with consistent orientation are calculated from the angle between the gradient vectors and are often used to detect contour extension trends.

[0055] S3: Based on the time sequence information of each segment in the structure of the occluded area contour completion, obtain the duration and label number of each action node, count the number of repeated numbers and the standard deviation of the time interval between adjacent nodes, and extract the respiratory rate and heart rate variability that are consistent with the segment time period from the key physiological indicators monitored during the nursing period. Combine and generate time continuity parameters and physiological fluctuation quantity comparison data under each action sequence to generate nursing behavior stability comparison data group.

[0056] Heart rate variability is a commonly used parameter for measuring the variability of vital signs, reflecting the autonomic nervous system's regulatory capacity, and serves as an auxiliary indicator of the impact of response interventions in the nursing process; standard deviation is used to measure the dispersion of action execution time.

[0057] S4: Based on the mapping relationship between the outline completion structure of the occluded area and the node position in the set of misaligned nursing behavior segments, extract the action nodes within the outline closed area in the monitoring image, and compare the Euclidean distance between the marked position of each action node in the image frame and the task target area position of the corresponding action label to determine whether the distance is greater than the allowable offset range of task execution, filter the node coordinates that exceed the offset range, and generate the occluded area behavior offset label mapping set.

[0058] Euclidean distance is used to measure the spatial offset between image coordinate points and is widely used in image target detection and trajectory matching. The allowable offset range can be set according to the ward monitoring resolution and scene, and is generally set as a threshold within 5 to 20 pixels.

[0059] S5: Call the action number, duration and spatial offset value of each data segment in the nursing behavior stability control data group and the occluded area behavior offset label mapping set, construct the score value of each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, and count the frequency and location sequence of abnormal nodes, establish a monitoring label structure corresponding to patient identification and nursing unit identification, and generate a ward nursing monitoring behavior abnormal focus information group.

[0060] The set of misaligned nursing behaviors includes execution node index number, action path time segment and sequence misalignment offset value; the occluded area contour completion structure includes edge closure connection segment pixel coordinates, occluded boundary extension path vector and frame sequence matching sliding window length; the nursing behavior stability control data set includes behavior label stability coefficient, physiological index fluctuation number and action execution cycle standard deviation; the occluded area behavior offset label mapping set includes label position offset difference, coordinate deviation node index and task target area offset mark; the ward nursing monitoring behavior abnormality focusing information set includes scoring node number sequence, abnormal path segment trajectory coordinates and trajectory risk level identifier.

[0061] Please see Figure 2 The specific steps for obtaining the set of misaligned nursing behaviors are as follows:

[0062] S111: Acquire continuous frame images from ward nursing monitoring video, identify the action behavior area of ​​nursing staff in the frame, extract the time label, behavior target number and behavior sequence mark from the image frame, sort all extracted data according to the time of action occurrence, establish an action time sequence list, and generate a nursing action node sequence list.

[0063] The resolution of the camera installed in the ward is [not specified]. Frame rate The real-time monitoring video stream is broken down into a sequence of consecutive bitmap images at a frequency of 25 frames per second. For each frame, a pre-trained YOLOv8 deep learning object detection model is used to locate the bounding box of the nursing staff's body. Within the bounding box, specific action features are identified, such as "infusion operation," "turning over for care," and "vital sign measurement." The area in which the action occurs is determined by combining the coordinates of the patient's bed position. Simultaneously, the timestamp in the metadata of each frame is read as a time tag. Based on the QR code recognition results of the bedside card, the patient ID is extracted as the behavioral target number, for example, the target number. Generate from the order in which the actions occur. Starting with an incrementing integer as a marker for the order of actions, the extracted "time tag: “Behavioral Target Number:” "Action type: Infusion preparation", "Sequence marker: Fields such as "" are stored in a structured database, and all identified action items are indexed and reorganized according to the ascending order of time tags to form a linked list structure with temporal logic, generating a list of nursing action node sequences.

[0064] S112: Based on the behavior target number and time tag in the nursing action node sequence list, retrieve the corresponding nursing task item number and standard operation sequence number from the patient nursing task list, construct the standard action sequence and match it one-to-one with the sequence mark in the nursing action node sequence list, calculate the difference between the two numbers and accumulate the offset value as the offset judgment index, and obtain the sequence number offset comparison result set.

[0065] Read the target number in the list And the time tag is Action nodes within the scope, such as detected action sequences containing "patient verification (sequence marker 1)", "intravenous puncture (sequence marker 2)", and "hand hygiene (sequence marker 3)", are then accessed in the electronic nursing task sheet of the hospital information system (HIS) to retrieve data within that time period. Nursing task item number corresponding to patient number [number] (Intravenous infusion task) Obtain the standard operating procedure (SOP) sequence numbers defined for this task, namely "Hand hygiene (standard number 1)", "Patient verification (standard number 2)", and "Venucleation (standard number 3)". The system matches the detected action nodes with the standard action items by name to establish a mapping relationship. Then, it calculates the absolute difference between the two sequence numbers. The specific calculation process is as follows: the offset value of the hand hygiene item is... Check the patient item offset value The offset value for the intravenous puncture item is The above individual offset values ​​are summed to obtain the cumulative offset value. This value quantifies the degree of discrepancy between the actual operation process and the standard specifications. The system stores the calculation results of each comparison task into an array to obtain a set of sequential number offset comparison results.

[0066] S113: Based on the cumulative offset value of each item in the sequential number offset comparison result set, determine whether it exceeds the preset sequential offset threshold, filter the segment index corresponding to the number whose cumulative offset value exceeds the sequential offset threshold, and organize the action node segments in the corresponding time range into a set to establish a nursing behavior misalignment segment set.

[0067] Call the preset order offset threshold Make a judgment, this threshold The settings are based on the average deviation level of historical compliant nursing operation data. The setting method selects 1000 nursing operation records that have been rated as qualified in the most recent month and calculates the arithmetic mean of their cumulative deviation values. with standard deviation ,set up This represents the allowable range of normal operational fluctuations. In this embodiment, the cumulative offset value calculated above is... Obviously greater than The operation was determined to have a significant sequence anomaly. The paragraph index corresponding to the cumulative offset value was then selected, and the corresponding time range was identified. to All action nodes within the time period are extracted as independent units to be analyzed and labeled with "sequence misalignment" to establish a set of nursing behavior misalignment segments.

[0068] Please see Figure 3 The specific steps for obtaining the outline completion structure of the occluded area are as follows:

[0069] S211: Call the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extract the image pixel matrix in the current frame, perform Canny edge detection on the edge region in the image, identify the gray-scale change region of the boundary between the nursing staff and the bed, and convert the region into a binary edge map structure to generate the edge pixel structure map of the monitoring image.

[0070] Frame-by-frame read time range to The image data within is used to convert RGB color images into single-channel grayscale image matrices. Apply a Gaussian filter (kernel size) to the matrix , To smooth out noise, the Canny edge detection algorithm is then executed, calculating the gradient magnitude and direction of image pixels, and setting the lower threshold in the dual-threshold detection to [value missing]. The high threshold is Strong edge pixels are selected, and gray-scale abrupt change areas are identified at the junction of the nurse's limbs and the bed rails and sheets. Morphological closing operations are used to connect the broken edge lines, and the pixel values ​​of the edge areas are set to 255 (white) and the background is set to 0 (black) to form a binary edge distribution map. This map clearly outlines the contact contour between the nurse and the environment, generating an edge pixel structure map of the monitoring image.

[0071] S212: Based on the location coordinates of the bed edge region and the pixel distribution of neighboring regions in the edge pixel structure diagram of the monitoring image, extract the Sobel gradient vector set inside and outside the occluded block, and obtain the gradient values ​​in the x and y directions corresponding to each pixel point, using the formula:

[0072] ;

[0073] The gradient direction difference value is obtained by calculation, and the point pairs with the smallest gradient direction difference value are matched to generate occlusion edge direction comparison information;

[0074] in, This represents the normalized gradient vector of the pixel within the occluded region of the i-th pair of edge points. The data source is the gradient value of that pixel after magnitude normalization. This represents the normalized gradient vector of the pixel outside the occluded region in the i-th pair of edge points. The data source is the local window gradient normalization sequence of neighboring boundary pixels. Let represent the magnitude of the normalized gradient vector within the occlusion region at the i-th pair of edge points. This represents the magnitude of the normalized gradient vector outside the occlusion in the i-th pair of edge points. Let represent the normalized Euclidean distance between the coordinates of the i-th pair of edge points, obtained by normalizing the coordinate differences between the two points after division by the resolution scale. This represents the spatial distance suppression constant, used to adjust the exponential decay relationship between the gradient direction difference and the spatial location difference. This represents the difference in gradient direction between the i-th pair of edge points;

[0075] Define the inner point of the occlusion at the detected boundary (e.g., where the nurse's arm covers the bed rail). and the corresponding outer occlusion point ,use The Sobel operator calculates these two points respectively. and gradient components of direction The gradient direction difference value is calculated according to the formula, and the point pairs with the smallest gradient direction difference value are matched. The formula is as follows:

[0076] ;

[0077] This formula aims to find the optimal matching point by quantifying the "directional consistency" and "spatial proximity" of the inner and outer edges within the occluded area, thereby restoring the occluded contour relationship. In the formula, and Representing the first For candidate points, the normalized gradient vectors of pixels inside and outside the occluded region are calculated. These two steps are performed by computing the Sobel gradient vector. Divide it by its modulus to obtain the direction of edge extension. The cosine similarity is calculated by dividing the dot product of two vectors by their magnitudes. Its value ranges from [value missing]. When the inner and outer edges are completely aligned (collinear), this term approaches 1. 3. It is the normalized Euclidean distance between the coordinates of two points. If point and Then the Euclidean distance Normalization is performed by dividing it by the length of the image diagonal. (like ),Right now 4. It is the spatial distance suppression constant, used to adjust the influence of spatial distance on matching weights. The term indicates that the value decreases exponentially with increasing distance. The value is set based on the expectation that the effect of this factor will significantly decrease when the distance exceeds 1% of the image size. For example, setting... hour If the value drops to 0.5, then Solving for This embodiment takes 5. It is the gradient direction difference value. The smaller the value, the more consistent the direction between the inner and outer points and the closer they are, and the more likely they are the two ends of the same occluded edge.

[0078] The example illustrates this: select a pair of pixels near the occlusion edge, the inner point... gradient vector , module length After normalization Outer point gradient vector , module length After normalization .point coordinate ,point coordinate Pixel distance Normalized distance Calculate the cosine similarity term: Calculate the distance weight term: Calculate the difference value: The result A value of "extremely small" indicates that points A and B have a very high probability of matching, and the system determines them as connection points on the same edge.

[0079] S213: Based on the coordinate position sequence of the matching point pairs in the occlusion edge direction comparison information, construct the pixel displacement path between adjacent frame images, construct the contour connection trajectory according to the contour change trend of continuous frames, identify the spatial changes of connected segments in the point pair path and record the closed edge position index, and generate the contour completion structure of the occlusion area.

[0080] As in the previous example, the point With point The system tracks the displacement of these two points in five consecutive adjacent frames. If in the first frame... Frame to During the frame, point Move to ,point Move to This constructs a smooth pixel displacement path, connects broken contour segments according to the Bézier curve algorithm, fills the visual gap caused by body occlusion, identifies the spatial change rate of connected segments in the path, records the vertex index of closed edges, and thus reconstructs the complete nursing action area in each frame, generating the contour completion structure of the occluded area.

[0081] Please see Figure 4 The specific steps for obtaining the nursing behavior stability control data group are as follows:

[0082] S311: Based on the time sequence label of each segment in the contour completion structure of the occluded area, extract each nursing action node corresponding to the segment, obtain the frame index length between the start frame and the end frame of the node, calculate the corresponding action duration value in combination with the frame frequency, and extract the action number as the behavior label. After summarizing, establish an action duration and label number table.

[0083] Extract the completed action segment "venipuncture" and obtain the starting frame index of this action. With end frame index Calculate the frame index length as Frame, combined with video frame rate Calculate the duration of the action. Seconds, extract the number corresponding to the action. (Piercing action), to " "and" "Binding, and repeating this process for all identified actions, as shown in Table 1 (only partial data is shown), and then summarizing to create a table of action duration and tag number."

[0084] Table 1. Examples of Nursing Action Duration and Label Numbering

[0085] Serial Number Action tag number (ID) Action Name Start time End time Duration (s) 1 01 Verify patient 10：05：01 10：05：06 5.0 2 03 Venous puncture 10：05：15 10：05：24 9.0 3 02 Hand hygiene 10：05：30 10：05：40 10.0

[0086] S312: Call the action duration and tag number sequence in the tag number table, count the cumulative number of repeated number items between adjacent nodes, calculate the standard deviation of the action duration time interval between adjacent nodes, construct the interval stability parameter set based on the repetition frequency and time fluctuation degree, and generate the behavior sequence stability evaluation matrix.

[0087] Count the cumulative number of duplicate numbered items between adjacent nodes, for example, during a long period of monitoring. The phrase "(venipuncture)" appeared 3 times (possibly due to repeated attempts after failed attempts), with the corresponding durations being as follows: , , Calculate the standard deviation of the duration of these three operations. The calculation process is as follows: mean ,variance Standard deviation Based on the repetition frequency (3 times) and the degree of time fluctuation (1.56), an interval stability parameter is constructed. The lower the value of this parameter, the more unstable the action is, and the system generates a behavior sequence stability evaluation matrix accordingly.

[0088] S313: Based on the time window of each segment in the behavioral sequence stability assessment matrix, extract the respiratory rate value and heart rate interval sequence synchronously collected in the nursing monitoring data, count the number of fluctuations of physiological signals within the segment, and bind them with the segment corresponding to the stability parameter. After fusion, generate a control structure data table and establish a nursing behavior stability control data group.

[0089] For example, regarding the time period during which the second "venipuncture" occurred... The system retrieves the patient's physiological signals synchronously recorded by the bedside monitor, extracts the respiratory rate and heart rate interval sequences within this 9-second segment, and statistically analyzes them. It finds that the patient's heart rate surges from 75 bpm to 110 bpm within this segment. A heart rate fluctuation exceeding 20% ​​is defined as a severe fluctuation, and two such fluctuations were recorded. The respiratory rate showed no significant abnormalities and zero fluctuations. The system then correlates "heart rate fluctuation: 2 times" and "respiratory fluctuation: 0 times" with stability parameters. Corresponding binding was performed to show that the instability of nursing operations is related to the patient's physiological stress. After fusion, a control structure data table was generated to establish a control data group for the stability of nursing behavior.

[0090] Please see Figure 5 The specific steps for obtaining the occlusion region behavior offset label mapping set are as follows:

[0091] S411: Based on the mapping relationship between the time index and node number in the set of misaligned nursing behavior segments and the occluded area contour completion structure, locate the corresponding image frame number for each mapped node, retrieve the set of pixel coordinates of the closed contour area in the image frame, identify the spatial position of the action behavior annotation point in the area, extract the coordinate point set of the action node in the area, and generate the coordinate group of the action node in the contour.

[0092] Regarding the number The "venipuncture" node was located in the image frame. Retrieve the set of pixel coordinates of the completed closed contour region in the frame. Within this area, the system identifies predefined key points of motion (such as the tip of a needle or the nurse's wrist) and extracts the coordinate sets of these key points, for example, extracting the wrist coordinates. Generate the coordinate set of the action node within the contour.

[0093] S412: Call the image frame position coordinates of each node in the action node coordinate group within the contour, calculate the Euclidean distance with the reference center coordinates of the task target area corresponding to the action label, determine whether the Euclidean distance exceeds the allowable offset threshold for task execution, filter the nodes that exceed the threshold and record the offset position and label number, and generate a node spatial offset annotation sequence.

[0094] The task execution allowable offset threshold is set by calculating the sum of the average value of the coordinate distribution range and the standard deviation of the coordinate distribution range based on the coordinate distribution range of the action node position coordinate set in the image frame, which is used as the task execution allowable offset threshold.

[0095] For example, wrist coordinate points Get the action tag Corresponding SOP standard task target area reference center coordinates (i.e., the standard puncture site center), calculate the Euclidean distance. Pixels. Determine if this distance exceeds the task execution allowable offset threshold. Threshold Setting process: Select the set of coordinates of all action nodes in the past 100 qualified execution records of this action, and calculate the average value of the distribution range of these coordinates (set the average distribution radius after normalization to be 1). (pixels) and standard deviation ( (pixels). Settings Pixels. In this example, the calculated offset distance. Therefore, the system determines that the node's spatial location is abnormal, identifies the node, and records its offset position. and label number Generate a sequence of node spatial offset labels.

[0096] S413: Based on the coordinate point information, label number and frame number index in the node spatial offset labeling sequence, construct a mapping structure to record the coordinates of the abnormal location points corresponding to the behavior labels, organize the index path of the offset nodes by frame, and cross-identify them with the corresponding number of the task target area to establish the occlusion area behavior offset label mapping set.

[0097] Record tags Corresponding abnormal location Occurred in frame And organize the index paths of offset nodes by frame, for example from arrive Offset coordinates from Move to By combining the corresponding number of the task target area (left hand dorsal vein), and cross-identifying it as "left hand dorsal area - puncture deviation", a set of behavior offset label mapping for occluded areas is established.

[0098] Please see Figure 6 The specific steps for obtaining information from the ward nursing monitoring behavioral abnormality focus group are as follows:

[0099] S511: Call the action number, action duration value and spatial offset distance value corresponding to each data segment in the nursing behavior stability control data group and the occluded area behavior offset label mapping set, construct a two-dimensional mapping matrix for each behavior node with time index as the horizontal axis and stability parameter and spatial offset as the vertical axis, calculate the behavior risk score for each node, and generate a behavior node score matrix.

[0100] For action Time Index The vertical axis contains stability parameters. (The normalized value is too low, denoted as a risk factor of 0.8) and spatial offset (Exceeding the standard, risk factor 0.9). Calculate the comprehensive risk score based on the preset risk classification criteria (as shown in Table 2). The scoring formula is set as follows: ,set up . The node received a score of 9.6, generating a behavior node rating matrix.

[0101] Table 2 Comparison Table of Behavioral Risk Scoring Standards

[0102] Risk level Rating range describe Response measures Low risk [0，3) Stable operation and accurate positioning Normal archiving Medium risk [3，7) Slight deviation or duration fluctuation Note High risk [7，10] Severe misalignment or physiological abnormality Trigger alarm

[0103] S512: Based on the behavior node scoring matrix, filter the node numbers that exceed the preset anomaly judgment threshold, mark the nodes in time index order, identify the connection relationship between continuously distributed nodes, construct the trajectory connection path structure between nodes, and record the time series index and spatial coordinate sequence of the abnormal nodes to obtain the trajectory connection set of abnormal nodes.

[0104] Node numbers exceeding the preset anomaly threshold (set to 7 points) are filtered out. In this example, the node... A score of 9.6 exceeds the threshold, so it is marked as an abnormal keyframe. The connection relationships between consecutively distributed nodes are then identified in chronological order. If subsequent frames... , If all scores are high-risk, an "abnormal trajectory" is constructed, recording the time series index of the occurrence of the anomaly. With spatial coordinate sequence This yields the connection set of abnormal node trajectories.

[0105] S513: Based on the trajectory sequence in the trajectory connection set of abnormal nodes, structurally map and bind the abnormal path with the corresponding action number, patient unique identifier and nursing unit number, construct the monitoring graph structure index according to the attribution relationship of each trajectory segment, extract high-frequency nodes in the trajectory path for focused annotation, and establish a ward nursing monitoring behavior abnormal focused information group.

[0106] Extract the abnormal path and action number (Venus puncture), Patient unique identifier Nursing Unit Number Perform structure mapping and binding, build a graph structure index in the UI layer of the monitoring view, and extract the coordinates points that appear most frequently in the trajectory path (such as...). The system will highlight the abnormal behavior of patients by marking it with a red box and display "Abnormal intravenous puncture action in bed P03 of ICU-01 unit" in a pop-up window on the system interface, thus establishing a ward nursing monitoring behavior abnormality focus information group.

[0107] A big data-based ward nursing monitoring data processing system is used to implement the aforementioned big data-based ward nursing monitoring data processing method. The system includes:

[0108] The behavior misalignment recognition module acquires continuous frame images from ward nursing monitoring videos, extracts the corresponding standard action sequence numbers from the patient nursing task list based on nursing big data samples, determines whether the cumulative difference of the offset between adjacent numbers exceeds the set sequence offset threshold, filters out the position segment sequences with offsets, and generates a set of nursing behavior misalignment segments.

[0109] The occlusion region analysis module calls the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extracts the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculates the direction deviation of the gradient values ​​of the Sobel operator inside and outside the occlusion region edge, compares the trend of occlusion boundary path change in continuous frames, records the information of the reconstructed contour closed line segment, and generates the contour completion structure of the occlusion region.

[0110] The behavioral parameter analysis module completes the structure based on the outline of the occluded area, calculates the standard deviation of the number of repeated numbers between adjacent nodes and the time interval, and extracts the respiratory rate and heart rate variability consistent with the time segment from the key physiological indicators monitored during the nursing period. It combines and generates time continuity parameters and physiological fluctuation data for each action sequence to generate a nursing behavior stability control data set.

[0111] The behavior offset segmentation module extracts action nodes within the contour-closed area of ​​the monitoring image based on the contour completion structure of the occluded area and the set of misaligned nursing behavior segments. It then compares the Euclidean distance between the marked position of each action node in the image frame and the position of the task target area of ​​the corresponding action label, filters out the coordinate points of nodes that exceed the offset range, and generates a set of behavior offset label mappings for the occluded area.

[0112] The abnormal information processing module calls the nursing behavior stability control data set and the occluded area behavior offset label mapping set to construct the risk score of each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, establish a monitoring label structure corresponding to patient identifiers and nursing unit identifiers, and generate a ward nursing monitoring behavior abnormality focus information set.

[0113] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for processing ward nursing monitoring data based on big data, characterized in that, Includes the following steps: S1: Obtain continuous frame images from ward nursing monitoring videos, extract the corresponding standard action sequence numbers from the patient nursing task list based on nursing big data samples, determine whether the cumulative difference between adjacent numbers exceeds the set sequence offset threshold, filter out the position segment sequences with offsets, and generate a set of nursing behavior misalignment segments. S2: Call the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extract the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculate the direction deviation of the gradient values ​​of the Sobel operator inside and outside the edge of the occluded area, compare the trend of the occluded boundary path change in continuous frames, record the information of the reconstructed contour closed line segment, and generate the contour completion structure of the occluded area. S3: Based on the outline completion structure of the occluded area, the standard deviation of the number of repeated numbers between adjacent nodes and the time interval is calculated, and the respiratory rate and heart rate variability consistent with the time segment of the key physiological indicators monitored during the nursing period are extracted. The time continuity parameters and physiological fluctuations of each action sequence are combined to generate a control data set of nursing behavior stability. S4: Based on the occluded area contour completion structure and the set of nursing behavior misalignment segments, extract the action nodes within the contour closed area in the monitoring image, and compare the Euclidean distance between the marked position of each action node in the image frame and the task target area position of the corresponding action label, filter the node coordinates that exceed the offset range, and generate an occluded area behavior offset label mapping set. S5: Call the nursing behavior stability control data set and the occluded area behavior offset label mapping set to construct the risk score of each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, establish a monitoring label structure corresponding to patient identifiers and nursing unit identifiers, and generate a ward nursing monitoring behavior abnormality focus information group.

2. The method for processing ward nursing monitoring data based on big data according to claim 1, characterized in that, The set of misaligned nursing behavior segments includes execution node index number, action path time segment, and sequence misalignment offset value. The occluded area contour completion structure includes edge closure connection segment pixel coordinates, occlusion boundary extension path vector, and frame sequence matching sliding window length. The nursing behavior stability control data set includes behavior label stability coefficient, physiological index fluctuation number, and action execution cycle standard deviation. The occluded area behavior offset label mapping set includes label position offset difference, coordinate deviation node index, and task target area offset marker. The ward nursing monitoring behavior abnormality focusing information set includes scoring node number sequence, abnormal path segment trajectory coordinates, and trajectory risk level identifier.

3. The method for processing ward nursing monitoring data based on big data according to claim 2, characterized in that, The specific steps for obtaining the set of misaligned nursing behavior segments are as follows: S111: Acquire continuous frame images from ward nursing monitoring video, identify the action behavior area of ​​nursing staff in the frame, extract the time label, behavior target number and behavior sequence mark from the image frame, sort all extracted data according to the time of action occurrence, establish an action time sequence list, and generate a nursing action node sequence list. S112: Based on the behavior target number and time tag in the nursing action node sequence list, retrieve the corresponding nursing task item number and standard operation sequence number from the patient nursing task list, construct the standard action sequence and match it one-to-one with the sequence mark in the nursing action node sequence list, calculate the difference between the two numbers and accumulate the offset value as the offset judgment index, and obtain the sequence number offset comparison result set. S113: Based on the cumulative offset value of each item in the sequence number offset comparison result set, determine whether it exceeds the preset sequence offset threshold, filter the segment index corresponding to the number whose cumulative offset value exceeds the sequence offset threshold, and organize the action node segments in the corresponding time range into a set to establish a nursing behavior misalignment segment set.

4. The ward nursing monitoring data processing method based on big data according to claim 3, characterized in that, The specific steps for obtaining the contour completion structure of the occluded region are as follows: S211: Call the image frame corresponding to each action node in the set of misaligned nursing behavior segments, extract the image pixel matrix in the current frame, perform Canny edge detection on the edge region in the image, identify the gray-scale change region between the nursing staff and the bed, and convert the region into a binary edge map structure to generate the edge pixel structure map of the monitoring image. S212: Based on the location coordinates of the edge region of the hospital bed and the pixel distribution of the neighboring region in the edge pixel structure diagram of the monitoring image, extract the Sobel gradient vector group inside and outside the occluded block, obtain the gradient values ​​of the x and y directions corresponding to each pixel, calculate the gradient direction difference value, and match the point pair with the smallest gradient direction difference value to generate occluded edge direction comparison information. S213: Based on the coordinate position sequence of the matching point pairs in the occlusion edge direction comparison information, construct the pixel displacement path between adjacent frame images, construct the contour connection trajectory according to the contour change trend of continuous frames, identify the spatial changes of connected segments in the point pair path and record the closed edge position index, and generate the contour completion structure of the occlusion area.

5. The ward nursing monitoring data processing method based on big data according to claim 4, characterized in that, The specific formula for obtaining the gradient direction difference value is as follows: ； in, This represents the normalized gradient vector of the pixels within the occluded region of the i-th pair of edge points. This represents the normalized gradient vector of the pixel outside the occluded region in the i-th pair of edge points. Let represent the magnitude of the normalized gradient vector within the occlusion region at the i-th pair of edge points. This represents the magnitude of the normalized gradient vector outside the occlusion in the i-th pair of edge points. This represents the normalized Euclidean distance between the coordinates of the i-th pair of edge points. Represents the spatial distance suppression constant. This represents the difference in gradient direction between the i-th pair of edge points.

6. The ward nursing monitoring data processing method based on big data according to claim 5, characterized in that, The specific steps for obtaining the nursing behavior stability control data group are as follows: S311: Based on the time sequence label of each segment in the contour completion structure of the occluded area, extract each nursing action node corresponding to the segment, obtain the frame index length between the start frame and the end frame of the node, calculate the corresponding action duration value in combination with the frame frequency, and extract the action number as the behavior label. After summarizing, establish an action duration and label number table. S312: Call the action duration and the tag number sequence in the tag number table, count the cumulative number of repeated number items between adjacent nodes, calculate the standard deviation of the action duration time interval between adjacent nodes, construct an interval stability parameter set based on the repetition frequency and the degree of time fluctuation, and generate a behavior sequence stability evaluation matrix. S313: Based on the time window of each segment in the behavioral sequence stability assessment matrix, extract the respiratory rate value and heart rate interval sequence synchronously collected in the nursing monitoring data, count the number of fluctuations of physiological signals within the segment, and bind them with the segment corresponding to the stability parameter. After fusion, generate a control structure data table and establish a nursing behavior stability control data group.

7. The ward nursing monitoring data processing method based on big data according to claim 6, characterized in that, The specific steps for obtaining the occlusion region behavior offset label mapping set are as follows: S411: Based on the mapping relationship between the time index and node number in the set of misaligned nursing behavior segments of the occluded area contour completion structure, locate the corresponding image frame number for each mapped node, retrieve the set of pixel coordinates of the closed contour area in the image frame, identify the spatial position of the action behavior annotation point in the area, and extract the coordinate point set of the action node in the area to generate the coordinate group of the action node in the contour. S412: Call the image frame position coordinates of each node in the action node coordinate group within the contour, calculate the Euclidean distance with the reference center coordinates of the task target area corresponding to the action label, determine whether the Euclidean distance exceeds the allowable offset threshold for task execution, filter the nodes that exceed the threshold and record the offset position and label number, and generate a node spatial offset label sequence. S413: Based on the coordinate point information, label number and frame number index in the node spatial offset label sequence, construct a mapping structure to record the coordinates of the abnormal location points corresponding to the behavior labels, organize the index paths of the offset nodes by frame, and perform cross-identification by the corresponding number of the task target area to establish a behavior offset label mapping set for the occluded area.

8. The ward nursing monitoring data processing method based on big data according to claim 7, characterized in that, The specific steps for obtaining the abnormal behavior focus information group in the ward nursing monitoring are as follows: S511: Call the action number, action duration value and spatial offset distance value corresponding to each data segment in the nursing behavior stability control data group and the occluded area behavior offset label mapping set, construct a two-dimensional mapping matrix for each behavior node with time index as the horizontal axis and stability parameter and spatial offset as the vertical axis, calculate the behavior risk score for each node, and generate a behavior node score matrix. S512: Based on the behavior node scoring matrix, the node numbers that exceed the preset abnormal judgment threshold are filtered, and the nodes are marked in time index order. The connection relationship between continuously distributed nodes is identified, the trajectory connection path structure between nodes is constructed, and the time series index and spatial coordinate sequence of abnormal nodes are recorded to obtain the abnormal node trajectory connection set. S513: Based on the trajectory sequence in the trajectory connection set of the abnormal nodes, the abnormal path is structurally mapped and bound to the corresponding action number, patient unique identifier and nursing unit number. A monitoring graph structure index is constructed according to the attribution relationship of each trajectory segment. High-frequency nodes in the trajectory path are extracted for focused annotation, and a ward nursing monitoring behavior abnormal focused information group is established.

9. A ward nursing monitoring data processing system based on big data, characterized in that, The system is used to implement the ward nursing monitoring data processing method based on big data as described in any one of claims 1-8, and the system includes: The behavior misalignment recognition module acquires continuous frame images from ward nursing monitoring videos, extracts the corresponding standard action sequence numbers from the patient nursing task list based on nursing big data samples, determines whether the cumulative difference of the offset between adjacent numbers exceeds the set sequence offset threshold, filters out the position segment sequences with offsets, and generates a set of nursing behavior misalignment segments. The occlusion area analysis module calls the image frame corresponding to each action node in the set of nursing behavior misalignment segments, extracts the Canny edge detection results and the gradient direction of the bed edge pixels in the image, calculates the direction deviation of the gradient values ​​of the Sobel operator inside and outside the occlusion area edge, compares the trend of occlusion boundary path change in continuous frames, records the information of the reconstructed contour closed line segment, and generates the contour completion structure of the occlusion area. The behavioral parameter analysis module, based on the outline completion structure of the occluded area, calculates the standard deviation of the number of repeated numbers between adjacent nodes and the time interval, and extracts the respiratory rate and heart rate variability consistent with the time segment from the key physiological indicators monitored during the nursing period. It combines and generates time continuity parameters and physiological fluctuation quantity comparison data under each action sequence to generate a nursing behavior stability comparison data group. The behavior offset segmentation module extracts action nodes within the contour closed area in the monitoring image based on the occluded area contour completion structure and the set of nursing behavior misalignment segments. It then compares the Euclidean distance between the marked position of each action node in the image frame and the task target area position of the corresponding action label, filters out node coordinates that exceed the offset range, and generates an occluded area behavior offset label mapping set. The abnormal information processing module calls the nursing behavior stability control data set and the occluded area behavior offset label mapping set to construct a risk score for each behavior node on the time axis, identify and mark abnormal nodes, connect abnormal nodes to construct continuous trajectory segments, establish a monitoring label structure corresponding to patient identifiers and nursing unit identifiers, and generate a ward nursing monitoring behavior abnormality focus information set.

Images