Real-time ai visual analysis and compliance detection method and system for in-home care services

Abstract

The application relates to the technical field of artificial intelligence, and provides a real-time AI visual analysis and compliance detection method and system for home nursing service, which can detect the presence of a nursing staff to avoid service by unauthorized personnel; in the process of nursing by the nursing staff, real-time action detection is performed on the nursing staff and a nursing object based on a light-weight skeleton point extraction network to provide an accurate data basis; action compliance of the nursing staff is detected in real time according to core skeleton point coordinates of the nursing staff, without subjective bias; the state of the nursing object is detected in real time according to core skeleton point coordinates of the nursing object, so that real-time response can be performed according to the state of the nursing object, and serious injury to the nursing object is avoided; the nursing staff is subjected to multi-dimensional fusion action verification according to the core skeleton point coordinates of the nursing staff, full-dimensional quantification of action compliance is realized, and the verification accuracy is improved; and response measures are generated according to the verification result, so that timely intervention and injury prevention in advance are realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for real-time AI visual analysis and compliance detection of in-home nursing services. Background Technology

[0002] With the accelerating aging of the population, the demand for home-based elderly care services continues to grow, and in-home nursing care is gradually becoming an important component of the elderly care service system. In-home nursing care typically involves caregivers visiting the homes of elderly people or those with limited mobility to provide services such as daily living assistance, rehabilitation care, and health monitoring. The quality of these services directly affects the health and safety of the individuals being cared for.

[0003] However, in the current management of in-home nursing services, the authenticity and standardization of nursing practices are difficult to manage effectively. Most institutions mainly rely on caregivers' manual reporting, sign-in records, or simple photo uploads for service recording, lacking objective and real-time verification methods. Some institutions send inspectors to conduct random checks to strengthen management, but this method not only has limited coverage of management dimensions, but also requires a large investment of manpower and time, resulting in low overall management efficiency and making it difficult to form a continuous and comprehensive supervision mechanism.

[0004] Therefore, how to conduct intelligent and automated detection and real feedback of the entire process of home care services has important application value. Summary of the Invention

[0005] In view of the above, it is necessary to provide a method and system for real-time AI visual analysis and compliance detection of in-home nursing services, which aims to solve the problem of the inability to conduct intelligent and automated detection and real feedback on the entire process of in-home nursing services.

[0006] A method for real-time AI visual analysis and compliance detection of in-home nursing services, comprising: In response to the instruction to inspect in-home nursing services provided to the recipients, on-site inspections are conducted on the caregivers. During the care process, the caregiver performs real-time motion detection on the caregiver and the patient based on a lightweight skeletal point extraction network to obtain the core skeletal point coordinates of the caregiver and the patient. The compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, and the action compliance detection result is obtained. The status of the object being cared for is detected in real time based on the coordinates of the core skeletal points of the object being cared for, and the status detection result is obtained. When the action compliance detection result and the status detection result reach the action verification trigger condition, the caregiver is subjected to multi-dimensional fusion action verification based on the coordinates of the caregiver's core skeletal points to obtain the verification result. Response measures are generated based on the verification results.

[0007] A real-time AI visual analysis and compliance detection system for in-home nursing services is disclosed. The system includes multi-source sensors deployed at various service locations, an edge computing device for executing real-time AI visual analysis and compliance detection methods for in-home nursing services, and a business platform. The multi-source sensors, the edge computing device, and the business platform communicate with each other.

[0008] As can be seen from the above technical solutions, this invention can detect the presence of caregivers to prevent unauthorized personnel from providing services; during the caregiving process, a lightweight skeletal point extraction network is used to perform real-time motion detection on both the caregiver and the patient, providing an accurate data foundation; the compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, eliminating subjective bias; the patient's status is detected in real time based on the coordinates of the patient's core skeletal points, enabling real-time responses to avoid serious harm to the patient; multi-dimensional fusion motion verification of the caregiver is performed based on the coordinates of the caregiver's core skeletal points, achieving full-dimensional quantification of motion compliance and improving verification accuracy; response measures are generated based on the verification results for timely intervention and early prevention of injury. Attached Figure Description

[0009] Figure 1 This is a flowchart of a preferred embodiment of the method for real-time AI visual analysis and compliance detection of in-home nursing services according to the present invention.

[0010] Figure 2 This is a functional block diagram of a preferred embodiment of the in-home nursing service real-time AI visual analysis and compliance detection system of the present invention.

[0011] Figure 3 This is a schematic diagram of the computer device that is a preferred embodiment of the method for real-time AI visual analysis and compliance detection of in-home nursing services according to the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] like Figure 1The diagram shown is a flowchart of a preferred embodiment of the real-time AI visual analysis and compliance detection method for in-home nursing services according to the present invention. The order of the steps in this flowchart can be changed, and some steps can be omitted, depending on different needs.

[0014] The real-time AI visual analysis and compliance detection method for in-home nursing services is applied to one or more computer devices. The computer device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0015] The computer device can be any electronic product that can interact with the user, such as a personal computer, tablet computer, smartphone, personal digital assistant (PDA), game console, interactive network television (IPTV), smart wearable device, etc.

[0016] The computer equipment may also include network equipment and / or user equipment. The network equipment includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0017] The server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0018] Artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0019] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0020] The network in which the computer device is located includes, but is not limited to, the Internet, wide area network, metropolitan area network, local area network, and virtual private network (VPN).

[0021] S10, in response to the instruction to inspect in-home nursing services for the person being cared for, conducts on-site inspection of the caregiver.

[0022] In this embodiment, the person being cared for can be an elderly person, a person with limited mobility, or someone who requires in-home care.

[0023] In this embodiment, the in-home nursing service detection instruction can be triggered after it is determined that services need to be provided to the person being cared for, and the in-home nursing service detection instruction can be triggered by the platform to which the caregiver belongs.

[0024] In this embodiment, the on-site detection of caregivers includes: The home care service detection instruction is analyzed to obtain the service start time, target service location, and caregiver characteristic information; At a preset time before the service start time arrives, the multi-source sensor deployed at the target service location is activated, and data is collected in real time using the multi-source sensor. When a user is detected entering the target service location, data collected by the multi-source sensors is acquired as data to be processed. Based on the data to be processed, feature extraction is performed to obtain the target features; The target features are compared with the caregiver feature information; When the target characteristic is the same as the caregiver characteristic information, it is determined that the caregiver has arrived on time.

[0025] The target service location can be the residence of the person being cared for, etc. Multi-source sensors are deployed within the residence.

[0026] The multi-source sensors may include, but are not limited to, video acquisition devices, audio acquisition devices, positioning devices, timing devices, etc.

[0027] Of course, in other embodiments, other multimodal technologies can also be used for in-home detection, such as address detection via satellite positioning and other time detection methods. Different detection methods can be adapted according to different types of locations. For example, when it is inconvenient to directly install positioning equipment in a residential location, address detection can be performed via satellite positioning.

[0028] The caregiver's characteristics may include facial features, voiceprint features, body features, and other features that can assist in identity authentication.

[0029] The preset duration can be configured to 5 minutes or 3 minutes. That is, starting the multi-source sensors 5 minutes before the service start time can meet the on-site detection needs while avoiding waste of detection costs.

[0030] The above embodiments enable rapid and accurate on-site testing of caregivers, avoiding services provided by unauthorized personnel.

[0031] S11, during the nursing process, the caregiver performs real-time motion detection on the caregiver and the patient based on a lightweight skeletal point extraction network to obtain the core skeletal point coordinates of the caregiver and the patient.

[0032] In this embodiment, before performing real-time motion detection on the caregiver and the person being cared for based on a lightweight skeletal point extraction network, the method further includes: Construct an initial network; wherein the initial network includes a YOLOv8 detection branch, a Transformer enhancement branch, and a feature fusion layer. The Transformer enhancement branch takes the output of the YOLOv8 detection branch as input. The feature fusion layer is used to fuse the output of the YOLOv8 detection branch and the output of the Transformer enhancement branch, and outputs the fused features to the detection head of the YOLOv8 detection branch for precise bone point localization and bone point coordinate output. A total loss function is constructed, comprising skeletal point localization loss, spatial attention consistency loss, and fusion feature regression loss. The skeletal point localization loss constrains the YOLOv8 detection branch, and is based on Object Keypoint Similarity (OKS) loss to constrain the consistency between predicted and actual values ​​of each skeletal point. The spatial attention consistency loss constrains the attention mechanism of the Transformer enhancement branch. The fusion feature regression loss constrains the fusion effect between the YOLOv8 detection branch and the Transformer enhancement branch. The baseline model of the YOLOv8 detection branch is pre-trained; Freeze the human basic morphological feature extraction layer of the YOLOv8 detection branch after pre-training, and train the Transformer enhancement branch and the feature fusion layer. Unfreeze the human basic morphological feature extraction layer, and train the unfrozen initial network based on the total loss function to obtain the initial bone point extraction network; Each layer of the initial skeleton point extraction network is scored based on the γ parameter and L2 norm of the batch normalized layer. The initial skeletal point extraction network is structured and pruned according to a preset ratio and the score of each layer to obtain a pruned network. The pruning network is compressed using 8-bit integer quantization to obtain the lightweight skeleton point extraction network.

[0033] The YOLOv8 detection branch can coarsely locate 21 core skeletal points, including the head, shoulders, elbows, wrists, hips, knees, and ankles. However, some skeletal points may have coordinate deviations due to occlusion, insufficient lighting, or other factors.

[0034] Furthermore, the Transformer enhancement branch receives the coarsely localized bone point coordinates output by the YOLOv8 detection branch, and models the global spatial dependencies between bone points (such as the association between the shoulder and elbow, and the association between the hip and knee) by calculating attention weights to correct the coordinates of the biased bone points.

[0035] Furthermore, the feature fusion layer fuses the coarsely located bone point coordinates output by the YOLOv8 detection branch with the corrected coordinates output by the Transformer enhancement branch, and inputs the fused features into the detection head of the YOLOv8 detection branch, thereby obtaining accurate bone point coordinates.

[0036] The core of the skeletal point localization loss is to combine skeletal point visibility and scale normalization to solve the localization deviation problem caused by different human body sizes and occlusion.

[0037] Specifically, for each skeletal point, the Euclidean distance between the predicted value and the true value is calculated; the distance is converted into a similarity of 0 to 1 (the smaller the distance, the closer the similarity is to 1); a weighted average is calculated only for visible skeletal points to exclude the interference of occluded skeletal points; the average similarity is subtracted from 1 to obtain the final loss value (the smaller the loss, the more accurate the localization).

[0038] The skeletal point localization loss can directly optimize the regression accuracy of skeletal point coordinates, ensuring pixel-level localization accuracy of core skeletal points (shoulders, hips, knees, etc.), and can adapt to different body types and different occlusion scenarios (such as clothing covering the elderly), avoiding localization deviations caused by differences in human body scale.

[0039] The skeletal point localization loss can be configured with a high weight, which is the basis for ensuring compliance verification and status evaluation of subsequent actions.

[0040] The spatial attention consistency loss ensures that its modeling of spatial relationships between skeletal points conforms to the physiological structure of the human body, thus avoiding chaotic attention weight distribution.

[0041] Specifically, the attention weight matrix output by the Transformer multi-head attention layer is extracted; a standard attention matrix based on human physiological structure is loaded (which can be obtained through expert annotation or statistical analysis of a large number of compliant samples); the element-wise L1 distance between the two matrices is calculated, and the average value is taken as the loss value (the smaller the loss, the more the attention distribution conforms to human structure).

[0042] The spatial attention consistency loss constrains the attention focus direction of the Transformer enhancement branch, ensuring it prioritizes physiological connections between skeleton points (such as shoulder-elbow-wrist linkage) rather than irrelevant background. It also corrects local biases in YOLOv8 coarse localization, strengthens global spatial connection modeling of skeleton points, and improves localization stability in occluded or low-light scenes. Furthermore, the spatial attention consistency loss prevents the Transformer enhancement branch from overfitting noise (such as wrinkles in elderly clothing or background clutter), ensuring the effectiveness of attention correction.

[0043] The fusion feature regression loss ensures that the two types of features are complementary rather than conflicting, thus improving the robustness of the final skeleton point regression.

[0044] Specifically, the coarse features output by YOLOv8 Neck can be concatenated with the features corrected by Transformer, and a fused feature can be obtained through 1×1 convolution; the true coordinates of the skeleton points can be converted into a Gaussian heatmap (each skeleton point corresponds to a Gaussian kernel to simulate the feature response); linear regression can be performed on the fused feature, and the mean square error between it and the true heatmap can be calculated as the loss value (the smaller the loss, the closer the fused feature is to the feature distribution of the true skeleton points).

[0045] The fusion feature regression loss serves as a bridge loss between the YOLOv8 and Transformer branches. It can optimize the parameters of the feature fusion layer, ensuring the complementary advantages of the two types of features (local features of YOLOv8 and global features of Transformer). At the same time, it can reduce the dimensional redundancy of the fused features, avoid overfitting after feature concatenation, and further improve the accuracy of skeleton point localization, making up for the feature defects of a single branch (such as the lack of global correlation in YOLOv8 and the lack of local details in Transformer).

[0046] The total loss function can be a weighted sum of the skeleton point localization loss, spatial attention consistency loss, and fusion feature regression loss.

[0047] The human body basic morphological feature extraction layer can be the first three shallow feature extraction modules of the Backbone in the YOLOv8 detection branch, which is mainly used to extract basic visual information such as general texture and edge features.

[0048] Specifically, when scoring each layer of the initial skeleton point extraction network based on the γ parameter and L2 norm of the batch normalization layer, for convolutional layers with batch normalization layers (such as YOLOv8 Backbone / Neck), the γ parameter score can be used as the primary method, with the L2 norm score as a secondary method; for convolutional layers without batch normalization layers (such as Transformer attention layers and 1×1 convolutional layers), the L2 norm score can be used. Finally, a unified channel importance score is generated through weighted fusion, and pruning is performed according to the score (e.g., removing the lowest 30% of channels).

[0049] In the above embodiments, structured pruning (channel-level pruning) can significantly compress the volume without destroying the model structure (it can protect the Transformer attention mechanism and skeleton point detection structure during the pruning process), and it is the only compression method that does not introduce additional computation.

[0050] In the above embodiments, 8-bit integer quantization compression enables hardware-level acceleration at the edge, which is the only fixed-point operation supported by low-power embedded devices, and does not damage the channel structure, and can be perfectly superimposed with structured pruning.

[0051] S12, the compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, and the action compliance detection result is obtained.

[0052] In this embodiment, the step of detecting the compliance of the caregiver's actions in real time based on the coordinates of the caregiver's core skeletal points, and obtaining the action compliance detection result, includes: Obtain the basic threshold library for skeletal points; The basic threshold library of skeletal points is adjusted according to the disability level of the person being cared for to obtain the target action threshold; The deviation rate between the coordinates of the caregiver's core skeletal points and the target action threshold is calculated in real time to obtain the action deviation rate; When the action deviation rate exceeds a first threshold, the action compliance detection result is determined to indicate that the caregiver has a risk of non-compliance in their actions, and a warning message is issued; or When the action deviation rate is greater than or equal to the second threshold and less than or equal to the first threshold, the action compliance detection result is determined to be that the caregiver has a slight action deviation, and a correction prompt is issued; or When the action deviation rate is less than the second threshold, the action compliance detection result is determined to be that the caregiver's action is compliant.

[0053] The skeletal point basic threshold library can preset skeletal point angles, movement ranges, and trajectory standards for services such as massage, bathing assistance, and turning over.

[0054] Specifically, when adjusting the basic threshold library of skeletal points according to the disability level of the person being cared for, the higher the disability level, the greater the relaxation of the action threshold, thereby achieving adaptive adjustment of the threshold and avoiding misjudgment of compliance due to a one-size-fits-all threshold.

[0055] The action deviation rate can be expressed as: Dev = (|Act real - Th dynamic | / Th dynamic ) × 100%.

[0056] Where Dev represents the action deviation rate; Act real This indicates the actual action corresponding to the coordinates of the caregiver's core skeletal points; Th dynamic This represents the target action threshold.

[0057] For example: when the action deviation rate is greater than 20%, the action compliance detection result is determined to indicate that the caregiver has a risk of non-compliance in their actions, and a warning message is issued; or when the action deviation rate is greater than or equal to 10% and less than or equal to 20%, the action compliance detection result is determined to indicate that the caregiver has a slight action deviation, and a correction prompt is issued; or when the action deviation rate is less than 10%, the action compliance detection result is determined to indicate that the caregiver's actions are compliant.

[0058] Through the above embodiments, the action threshold can be dynamically adjusted according to the disability level, and the compliance of the action can be quantitatively determined by the action deviation rate. This eliminates the industry pain point of only judging whether there is an action without judging whether it is standardized, thus eliminating subjective bias. Subsequently, the skeletal point deviation data can be traced as evidence.

[0059] S13, the state of the object being cared for is detected in real time based on the coordinates of the core skeletal points of the object being cared for, and the state detection result is obtained.

[0060] In this embodiment, the step of detecting the state of the object being cared for in real time based on the coordinates of the core skeletal points to obtain the state detection result includes: Quantifiable multimodal features are selected based on physical and mental state, and a baseline model of the state of the person being cared for is constructed based on the quantifiable multimodal features. The quantifiable multimodal features are used as the data acquisition dimension to collect real-time features of the cared-for object, thereby obtaining real-time features; The real-time status score of the cared-over object is calculated based on the real-time features and the coordinates of the core skeletal points of the cared-over object. The deviation rate of the real-time state score relative to the state baseline model is calculated in real time. When the deviation rate exceeds a preset threshold, the state detection result is determined to be abnormal, and an abnormality warning is issued; or When the deviation rate is less than or equal to the preset threshold, the state detection result is determined to be normal.

[0061] Among these, features related to physical condition can include visual features such as limb movement, facial expressions, and body posture.

[0062] Among them, features related to mental state can include audio features such as tone of voice, initiative of response, and facial expressions, as well as visual features.

[0063] The quantifiable multimodal features can be normalized to the [0,1] interval to unify the magnitude of features of different dimensions. For example, facial expression of pain corresponds to a pain level with a feature value of 0-1; limb range of motion can correspond to the range of motion angles of the shoulder-elbow-wrist joints and be normalized to the [0,1] interval; voice tone can correspond to the average volume of speech and be normalized to the [0,1] interval.

[0064] Furthermore, by weighting and summing the feature values ​​of all quantifiable multimodal features according to their respective weights, a single-dimensional state score can be obtained, including the physical state dimension score and the mental state dimension score. Then, by weighting and summing the physical state dimension scores and the mental state dimension scores according to their weights (e.g., physical state weight 0.6, mental state weight 0.4), the total score for a single frame can be calculated.

[0065] Furthermore, the total number of frames collected within 1-2 minutes prior to the care can be collected, and the average of the total single-frame scores for all frames can be calculated to construct the state baseline model.

[0066] Specifically, the absolute difference between the real-time state score and the state baseline model can be calculated, and the percentage of the quotient of the absolute difference and the state baseline model can be calculated to determine the degree of deviation of the real-time state score from the state baseline model.

[0067] For example, when the deviation rate is greater than 30%, the state detection result is determined to be an abnormal state; when the deviation rate is less than or equal to 30%, the state detection result is determined to be a normal state.

[0068] Through the above embodiments, the state of the person being cared for (such as pain, discomfort, abnormal reactions, etc.) can be detected in real time during the nursing process, so as to make targeted responses based on the real-time state of the person being cared for (such as early warning of risks and timely prevention of harm) and ensure service safety.

[0069] In this embodiment, when a real-time anomaly is detected (such as a caregiver's non-compliant actions or an abnormal state of the person being cared for), the video can be truncated and preserved as evidence. Privacy protection can also be implemented simultaneously, such as redacting key information.

[0070] S14, when the action compliance detection result and the status detection result reach the action verification trigger condition, the caregiver is subjected to multi-dimensional fusion action verification based on the coordinates of the caregiver's core skeletal points to obtain the verification result.

[0071] In this embodiment, before performing multidimensional fusion motion verification on the caregiver based on the coordinates of the caregiver's core skeletal points, the method further includes: When it is determined, based on the action compliance detection results, that a preset number of consecutive video frames all have the risk of non-compliant actions, and / or, based on the state detection results, it is determined that an abnormal state exists, the action verification trigger condition is met.

[0072] For example, when 3-5 consecutive frames are marked as having a risk of non-compliance in actions, or when an abnormal state of the person being cared for is detected, multi-dimensional fusion action verification can be triggered in a timely manner to detect whether there is indeed a non-compliance issue in actions.

[0073] In this embodiment, the step of performing multi-dimensional fusion motion verification on the caregiver based on the coordinates of the caregiver's core skeletal points, and obtaining the verification result, includes: Based on the coordinates of the caregiver's core skeletal points, motion path analysis was performed to obtain trajectory verification results; Based on the coordinates of the caregiver's core skeletal points, the contact points of the limbs are checked to determine whether they are within the compliance range, and the contact verification results are obtained. Based on the Dynamic Time Warping (DTW) algorithm, the execution order of actions is checked for compliance according to the coordinates of the core skeletal points of the caregiver, and the timing verification result is obtained. The trajectory verification result, the contact verification result, and the timing verification result are weighted according to a preset weighting coefficient system to obtain the target verification score. When the target verification score is greater than the third threshold, the verification result is determined to be that the caregiver's actions are compliant; or When the target verification score is greater than or equal to the fourth threshold and less than or equal to the third threshold, the verification result is determined to be that the caregiver has a slight movement deviation; or When the target verification score is less than the fourth threshold, the verification result is determined to indicate that the caregiver has a risk of non-compliant actions.

[0074] Trajectory verification is a temporal-dimensional path integrity verification. It analyzes the changes in skeletal point coordinates over time to determine whether the motion path is consistent with the standard template. The motion path analysis may include motion amplitude deviation analysis (i.e., motion deviation rate analysis), trajectory direction deviation analysis, trajectory continuity deviation analysis, and motion start and end point deviation analysis.

[0075] Contact verification is used to identify the contact area between the caregiver's limbs and the elderly person's body, ensuring that operations are only performed on compliant areas (such as the back and limbs), and prohibiting contact with private or dangerous areas.

[0076] For example, small bounding boxes can be calculated based on skeletal points (e.g., generating small bounding boxes for the caregiver's limbs and the elderly person's body respectively), and the intersection-union ratio (IU) of every two small bounding boxes can be calculated and compared with a threshold to determine the contact status. Then, an image semantic segmentation model can be used to accurately locate the contact pixel region and map it to the elderly person's body part labels (e.g., head, torso, limbs, private parts, etc.). compliant contact items are pre-set (e.g., turning over allows contact with the back and shoulders, massage allows contact with the limbs, shoulders, and neck). If the contact area is within a compliant contact item, the contact is considered compliant; if the contact area is not within a compliant contact item, it is considered non-compliant, and an alert can be triggered immediately. Simultaneously, the contact duration can be calculated. If the contact duration is too short (e.g., less than 2 seconds), it is considered a false touch; if it is too long (e.g., greater than 10 seconds), it needs to be determined whether it is a standard pressure application based on the action type. Contact pressure can also be calculated based on skeletal point displacement; if the displacement abruptly exceeds a threshold, it is considered excessive force. Combining all the above detections yields the contact verification result.

[0077] Among them, the timing verification is used to compare the timing similarity between the actual action sequence and the standard action template sequence to determine whether the execution order and rhythm of the actions are compliant.

[0078] The preset weighting coefficient system may include weights corresponding to the trajectory verification result, contact verification result, and timing verification result, respectively. The trajectory verification result, contact verification result, and timing verification result may be quantified values ​​of the corresponding detection results, so as to calculate the target verification score according to the preset weighting coefficient system.

[0079] For example: when the target verification score is greater than 0.9, the verification result is determined to be that the caregiver's actions are compliant; when the target verification score is greater than or equal to 0.7 and less than or equal to 0.9, the verification result is determined to be that the caregiver has a slight deviation in actions; when the target verification score is less than 0.7, the verification result is determined to be that the caregiver has a risk of non-compliant actions.

[0080] The above embodiments enable the full-dimensional quantification of action compliance, covering multiple dimensions such as process, scope, location, and sequence, so as to completely eliminate non-standard operations.

[0081] S15, Generate response measures based on the verification results.

[0082] In this embodiment, the step of generating a response based on the verification result includes: When the verification result indicates that the caregiver has a slight movement deviation, a movement correction prompt is sent to the caregiver via the caregiver's wearable device; or When the verification result indicates that the caregiver has a risk of non-compliance in their actions, an error warning is issued to the caregiver through their wearable device, and the type of non-compliance is recorded.

[0083] The wearable device may include a wristband or similar device.

[0084] The above embodiments can provide prompts for correction when minor deviations occur, assisting caregivers in promptly correcting erroneous operations. When there is a risk of non-compliant actions, wearable devices such as wristbands can emit vibrations or sounds to alert caregivers to errors and record the type of non-compliance (such as trajectory, contact, or timing) for subsequent traceability.

[0085] In this embodiment, the method further includes: When a home care service termination instruction is received, the current time is used as the service termination time. Calculate the time difference between the service end time and the service start time to obtain the service duration; The nursing care has been completed based on the coordinates of the core skeletal points of the caregiver. Obtain a list of nursing behaviors corresponding to the described patient; The completed nursing behaviors are matched against the list of nursing behaviors to obtain matching results; The nursing behavior completion rate is calculated based on the matching results; The quantifiable multimodal features are used as the data acquisition dimension to collect the features of the cared-over object after the service is completed as the post-service features, and the core skeletal point coordinates of the cared-over object are obtained by motion detection based on the lightweight skeletal point extraction network. The post-service status score of the cared-for object is calculated based on the post-service characteristics and the coordinates of the post-service core skeletal points. Calculate the rate of change of the post-service state score relative to the state baseline model; Obtain the state threshold, and generate a nursing effect level based on the state change rate and the state threshold; A service quality report for the caregiver is generated based on the service duration, the completion rate of the nursing behavior, and the nursing effect level, and the service quality report is sent to the target business platform to which the caregiver belongs.

[0086] By calculating the service duration, it is possible to detect whether caregivers have insufficient service time.

[0087] By calculating the completion rate of the nursing actions, it is possible to detect whether any nursing actions were not performed or whether any tasks were missed.

[0088] Wherein, the state change rate = (the post-service state score - the state baseline model) / the state baseline model × 100%.

[0089] When the status change rate is greater than 10%, the nursing effect level can be determined as excellent, and it is used as a bonus for the caregiver; when the status change rate is greater than or equal to -10% and less than or equal to 10%, the nursing effect level can be determined as good, and the caregiver can be given a normal score; when the status change rate is less than 10%, the nursing effect level can be determined as poor, points are deducted from the caregiver, and the information is pushed to the target business platform to which the caregiver belongs.

[0090] The service quality report can be in the form of a visual report or similar document, and can be encrypted and uploaded to the target business platform. The service quality report may include information such as nursing data statistics, behavioral analysis, and anomaly alerts.

[0091] In this embodiment, full service data can be acquired periodically. For example, feature indicators and judgment thresholds can be automatically updated every 10,000 data points to achieve continuous optimization, effectively improve detection accuracy, and form a closed-loop self-evolving system of recognition, learning, and verification.

[0092] As can be seen from the above technical solutions, this invention can detect the presence of caregivers to prevent unauthorized personnel from providing services; during the caregiving process, a lightweight skeletal point extraction network is used to perform real-time motion detection on both the caregiver and the patient, providing an accurate data foundation; the compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, eliminating subjective bias; the patient's status is detected in real time based on the coordinates of the patient's core skeletal points, enabling real-time responses to avoid serious harm to the patient; multi-dimensional fusion motion verification of the caregiver is performed based on the coordinates of the caregiver's core skeletal points, achieving full-dimensional quantification of motion compliance and improving verification accuracy; response measures are generated based on the verification results for timely intervention and early prevention of injury.

[0093] like Figure 2 The diagram shown is a functional block diagram of a preferred embodiment of the in-home nursing service real-time AI visual analysis and compliance detection system of the present invention.

[0094] The in-home nursing service real-time AI visual analysis and compliance detection system 2 includes multi-source sensors 21 deployed at various service locations, used to perform functions such as... Figure 1 The aforementioned in-home nursing service real-time AI visual analysis and compliance detection method includes an edge computing device 22 and a business platform 23; wherein the multi-source sensor 21, the edge computing device 22, and the business platform 23 communicate with each other.

[0095] The multi-source sensor 21 may include video acquisition devices (such as cameras), audio acquisition devices (such as microphones), positioning devices, timing devices, etc.

[0096] The edge computing device 22 may also be equipped with a monitoring and management box, which is used to push the results of AI-based detection to the business platform 23.

[0097] The business platform 23 can be connected to the monitoring and management system and monitoring and management mini-program of an authoritative institution to achieve effective monitoring and management of in-home nursing services.

[0098] The edge computing device 22 may include a detection unit 220, a verification unit 221, and a generation unit 222. The module / unit referred to in this invention is a series of computer program segments that can be executed by a processor and perform a fixed function, and which are stored in memory. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0099] The detection unit 220 is used to detect the presence of caregivers in response to an in-home care service detection command for the care recipient. The detection unit 220 is also used to perform real-time motion detection on the caregiver and the person being cared for based on a lightweight skeletal point extraction network during the care process, so as to obtain the core skeletal point coordinates of the caregiver and the core skeletal point coordinates of the person being cared for. The detection unit 220 is also used to detect the compliance of the caregiver's actions in real time based on the coordinates of the caregiver's core skeletal points, and obtain the action compliance detection result; The detection unit 220 is also used to detect the state of the object being cared for in real time based on the coordinates of the core skeletal points of the object being cared for, and to obtain the state detection result; The verification unit 221 is used to perform multi-dimensional fusion action verification on the caregiver based on the coordinates of the caregiver's core skeletal points when the action compliance detection result and the status detection result reach the action verification triggering condition, and obtain the verification result. The generation unit 222 is used to generate response measures based on the verification result.

[0100] As can be seen from the above technical solutions, this invention can detect the presence of caregivers to prevent unauthorized personnel from providing services; during the caregiving process, a lightweight skeletal point extraction network is used to perform real-time motion detection on both the caregiver and the patient, providing an accurate data foundation; the compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, eliminating subjective bias; the patient's status is detected in real time based on the coordinates of the patient's core skeletal points, enabling real-time responses to avoid serious harm to the patient; multi-dimensional fusion motion verification of the caregiver is performed based on the coordinates of the caregiver's core skeletal points, achieving full-dimensional quantification of motion compliance and improving verification accuracy; response measures are generated based on the verification results for timely intervention and early prevention of injury.

[0101] like Figure 3 The diagram shown is a schematic representation of the computer device used in a preferred embodiment of the method for real-time AI visual analysis and compliance detection of in-home nursing services according to the present invention.

[0102] The computer device 1 may include a memory 12, a processor 13, and a bus (the arrow in the figure represents the bus), and may also include a computer program stored in the memory 12 and capable of running on the processor 13, such as a real-time AI visual analysis and compliance detection program for in-home care services.

[0103] Those skilled in the art will understand that the schematic diagram is merely an example of computer device 1 and does not constitute a limitation on computer device 1. Computer device 1 can be either a bus topology or a star topology. Computer device 1 may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, computer device 1 may also include input / output devices, network access devices, etc.

[0104] It should be noted that the computer device 1 described is merely an example. Other existing or future electronic products that are adaptable to this invention should also be included within the scope of protection of this invention and are incorporated herein by reference.

[0105] The memory 12 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the computer device 1, such as a portable hard drive of the computer device 1. In other embodiments, the memory 12 can be an external storage device of the computer device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the computer device 1. Furthermore, the memory 12 can include both internal and external storage units of the computer device 1. The memory 12 can be used not only to store application software and various types of data installed on the computer device 1, such as the code of a real-time AI visual analysis and compliance detection program for in-home care services, but also to temporarily store data that has been output or will be output.

[0106] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control unit of the computer device 1, connecting various components of the computer device 1 via various interfaces and lines. It executes programs or modules stored in the memory 12 (e.g., executing real-time AI visual analysis and compliance detection programs for in-home care services) and calls data stored in the memory 12 to perform various functions of the computer device 1 and process data.

[0107] The processor 13 executes the operating system of the computer device 1 and various installed applications. The processor 13 executes these applications to implement the steps in the above embodiments of the real-time AI visual analysis and compliance detection method for in-home nursing services, for example... Figure 1 The steps are shown.

[0108] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device 1. For example, the computer program may be divided into a detection unit 220, a verification unit 221, and a generation unit 222.

[0109] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute portions of the real-time AI visual analysis and compliance detection method for in-home nursing services described in various embodiments of the present invention.

[0110] If the modules / units integrated in the computer device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware devices. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.

[0111] The computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory, etc.

[0112] Furthermore, the computer-readable storage medium may primarily include a stored program area and a stored data area, wherein the stored program area may store the operating system, an application program required for at least one function, etc.; and the stored data area may store data created based on the use of blockchain nodes, etc.

[0113] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0114] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, in... Figure 3 The bus is represented by only one straight line, but this does not mean that there is only one bus or one type of bus. The bus is configured to enable communication between the memory 12 and at least one processor 13, etc.

[0115] Although not shown, the computer device 1 may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 13 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The computer device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0116] Furthermore, the computer device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the computer device 1 and other computer devices.

[0117] Optionally, the computer device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the computer device 1 and to display a visual user interface.

[0118] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0119] It will be understood by those skilled in the art that Figure 3 The structure shown does not constitute a limitation on the computer device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0120] Combination Figure 1 The memory 12 in the computer device 1 stores multiple instructions to implement a real-time AI visual analysis and compliance detection method for in-home nursing services, and the processor 13 can execute the multiple instructions to achieve the following: In response to the instruction to inspect in-home nursing services provided to the recipients, on-site inspections are conducted on the caregivers. During the care process, the caregiver performs real-time motion detection on the caregiver and the patient based on a lightweight skeletal point extraction network to obtain the core skeletal point coordinates of the caregiver and the patient. The compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, and the action compliance detection result is obtained. The status of the object being cared for is detected in real time based on the coordinates of the core skeletal points of the object being cared for, and the status detection result is obtained. When the action compliance detection result and the status detection result reach the action verification trigger condition, the caregiver is subjected to multi-dimensional fusion action verification based on the coordinates of the caregiver's core skeletal points to obtain the verification result. Response measures are generated based on the verification results.

[0121] Specifically, the processor 13's implementation method for the above instructions can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0122] It should be noted that all the data involved in this case was legally obtained.

[0123] If any AI models, software tools, or components not belonging to this company appear in the embodiments of this invention, they are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this invention has been obtained by an entity authorized (with the knowledge and consent) or fully authorized by all parties through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0124] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0125] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0126] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0127] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0128] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0129] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0130] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in this invention can also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for real-time AI visual analysis and compliance detection of in-home nursing services, characterized in that, The real-time AI visual analysis and compliance detection method for in-home nursing services includes: In response to the instruction to inspect in-home nursing services provided to the recipients, on-site inspections are conducted on the caregivers. During the care process, the caregiver performs real-time motion detection on the caregiver and the patient based on a lightweight skeletal point extraction network to obtain the core skeletal point coordinates of the caregiver and the patient. The compliance of the caregiver's actions is detected in real time based on the coordinates of the caregiver's core skeletal points, and the action compliance detection result is obtained. The status of the object being cared for is detected in real time based on the coordinates of the core skeletal points of the object being cared for, and the status detection result is obtained. When the action compliance detection result and the status detection result reach the action verification trigger condition, the caregiver is subjected to multi-dimensional fusion action verification based on the coordinates of the caregiver's core skeletal points to obtain the verification result. Response measures are generated based on the verification results.

2. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 1, characterized in that, The on-site testing of caregivers includes: The home care service detection instruction is analyzed to obtain the service start time, target service location, and caregiver characteristic information; At a preset time before the service start time arrives, the multi-source sensor deployed at the target service location is activated, and data is collected in real time using the multi-source sensor. When a user is detected entering the target service location, data collected by the multi-source sensors is acquired as data to be processed. Based on the data to be processed, feature extraction is performed to obtain the target features; The target features are compared with the caregiver feature information; When the target characteristic is the same as the caregiver characteristic information, it is determined that the caregiver has arrived on time.

3. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 1, characterized in that, Before performing real-time motion detection on the caregiver and the patient based on a lightweight skeletal point extraction network, the method further includes: Construct an initial network; wherein the initial network includes a YOLOv8 detection branch, a Transformer enhancement branch, and a feature fusion layer. The Transformer enhancement branch takes the output of the YOLOv8 detection branch as input. The feature fusion layer is used to fuse the output of the YOLOv8 detection branch and the output of the Transformer enhancement branch, and outputs the fused features to the detection head of the YOLOv8 detection branch for precise bone point localization and bone point coordinate output. A total loss function is constructed, comprising skeletal point localization loss, spatial attention consistency loss, and fusion feature regression loss. The skeletal point localization loss constrains the YOLOv8 detection branch, and constrains the consistency between predicted and actual values ​​of each skeletal point based on target keypoint similarity loss. The spatial attention consistency loss constrains the attention mechanism of the Transformer enhancement branch. The fusion feature regression loss constrains the fusion effect of the YOLOv8 detection branch and the Transformer enhancement branch. The baseline model of the YOLOv8 detection branch is pre-trained; Freeze the human basic morphological feature extraction layer of the YOLOv8 detection branch after pre-training, and train the Transformer enhancement branch and the feature fusion layer. Unfreeze the human basic morphological feature extraction layer, and train the unfrozen initial network based on the total loss function to obtain the initial bone point extraction network; Each layer of the initial skeleton point extraction network is scored based on the γ parameter and L2 norm of the batch normalized layer. The initial skeletal point extraction network is structured and pruned according to a preset ratio and the score of each layer to obtain a pruned network. The pruning network is compressed using 8-bit integer quantization to obtain the lightweight skeleton point extraction network.

4. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 1, characterized in that, The step of detecting the compliance of the caregiver's actions in real time based on the coordinates of the caregiver's core skeletal points, and obtaining the action compliance detection results, includes: Obtain the basic threshold library for skeletal points; The basic threshold library of skeletal points is adjusted according to the disability level of the person being cared for to obtain the target action threshold; The deviation rate between the coordinates of the caregiver's core skeletal points and the target action threshold is calculated in real time to obtain the action deviation rate; When the action deviation rate exceeds a first threshold, the action compliance detection result is determined to indicate that the caregiver has a risk of non-compliance in their actions, and a warning message is issued; or When the action deviation rate is greater than or equal to the second threshold and less than or equal to the first threshold, the action compliance detection result is determined to be that the caregiver has a slight action deviation, and a correction prompt is issued; or When the action deviation rate is less than the second threshold, the action compliance detection result is determined to be that the caregiver's action is compliant.

5. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 2, characterized in that, The step of detecting the state of the patient in real time based on the coordinates of the core skeletal points of the patient and obtaining the state detection result includes: Quantifiable multimodal features are selected based on physical and mental state, and a baseline model of the state of the person being cared for is constructed based on the quantifiable multimodal features. The quantifiable multimodal features are used as the data acquisition dimension to collect real-time features of the cared-for object, thereby obtaining real-time features; The real-time status score of the cared-over object is calculated based on the real-time features and the coordinates of the core skeletal points of the cared-over object. The deviation rate of the real-time state score relative to the state baseline model is calculated in real time. When the deviation rate exceeds a preset threshold, the state detection result is determined to be abnormal, and an abnormality warning is issued; or When the deviation rate is less than or equal to the preset threshold, the state detection result is determined to be normal.

6. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 1, characterized in that, Before performing multidimensional fusion motion verification on the caregiver based on the coordinates of the caregiver's core skeletal points, the method further includes: When it is determined, based on the action compliance detection results, that a preset number of consecutive video frames all have the risk of non-compliant actions, and / or, based on the state detection results, it is determined that an abnormal state exists, the action verification trigger condition is met.

7. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 1, characterized in that, The multi-dimensional fusion motion verification of the caregiver based on the coordinates of the caregiver's core skeletal points yields the following verification results: Based on the coordinates of the caregiver's core skeletal points, motion path analysis was performed to obtain trajectory verification results; Based on the coordinates of the caregiver's core skeletal points, the contact points of the limbs are checked to determine whether they are within the compliance range, and the contact verification results are obtained. Based on the dynamic time warping algorithm, the execution order of actions is checked for compliance according to the coordinates of the core skeletal points of the caregiver, and the time sequence verification result is obtained. The trajectory verification result, the contact verification result, and the timing verification result are weighted according to a preset weighting coefficient system to obtain the target verification score. When the target verification score is greater than the third threshold, the verification result is determined to be that the caregiver's actions are compliant; or When the target verification score is greater than or equal to the fourth threshold and less than or equal to the third threshold, the verification result is determined to be that the caregiver has a slight movement deviation; or When the target verification score is less than the fourth threshold, the verification result is determined to indicate that the caregiver has a risk of non-compliant actions.

8. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 7, characterized in that, The measures for generating a response based on the verification result include: When the verification result indicates that the caregiver has a slight movement deviation, a movement correction prompt is sent to the caregiver via the caregiver's wearable device; or When the verification result indicates that the caregiver has a risk of non-compliance in their actions, an error warning is issued to the caregiver through their wearable device, and the type of non-compliance is recorded.

9. The method for real-time AI visual analysis and compliance detection of in-home nursing services as described in claim 5, characterized in that, The method further includes: When a home care service termination instruction is received, the current time is used as the service termination time. Calculate the time difference between the service end time and the service start time to obtain the service duration; The nursing care has been completed based on the coordinates of the core skeletal points of the caregiver. Obtain a list of nursing behaviors corresponding to the described patient; The completed nursing behaviors are matched against the list of nursing behaviors to obtain matching results; The nursing behavior completion rate is calculated based on the matching results; The quantifiable multimodal features are used as the data acquisition dimension to collect the features of the cared-over object after the service is completed as the post-service features, and the core skeletal point coordinates of the cared-over object are obtained by motion detection based on the lightweight skeletal point extraction network. The post-service status score of the cared-for object is calculated based on the post-service characteristics and the coordinates of the post-service core skeletal points. Calculate the rate of change of the post-service state score relative to the state baseline model; Obtain the state threshold, and generate a nursing effect level based on the state change rate and the state threshold; A service quality report for the caregiver is generated based on the service duration, the completion rate of the nursing behavior, and the nursing effect level, and the service quality report is sent to the target business platform to which the caregiver belongs.

10. A real-time AI visual analysis and compliance detection system for in-home nursing services, characterized in that, The real-time AI visual analysis and compliance detection system for in-home nursing services includes multi-source sensors deployed at various service locations, an edge computing device for executing the real-time AI visual analysis and compliance detection method for in-home nursing services as described in any one of claims 1-9, and a business platform; wherein the multi-source sensors, the edge computing device, and the business platform communicate with each other.

Images