A remote synchronized video recognition wound care guidance system

Abstract

The application belongs to the technical field of wound care, and discloses a wound care guidance system for remote synchronous video recognition, which comprises a video acquisition module, an AR measurement module, a double-channel calculation module, a data fusion module and a designated suggestion generation module; the video acquisition module is used for video data acquisition work; the AR measurement module can dynamically superimpose a preset virtual scale template on a corrected video image frame; the double-channel calculation module is used for calculating a wound necrosis recession rate and a granulation growth slope through a dynamic index, quantifying a wound change trend, a depth of tunneling and a sinus node through a parallel processing mechanism. The AR measurement module of the application superimposes a virtual scale on augmented reality, accurately labels the length, width, area, volume, tunneling, sinus and sinus parameters of a wound in a real-time video stream, and calculates the wound necrosis recession rate and the granulation growth slope through the double-channel calculation module, quantifies the wound change trend, the depth of tunneling and the sinus node, and is convenient for monitoring the rate evolution trajectory of a wound of a patient.

Description

Technical Field

[0001] This application belongs to the field of video recognition technology, specifically relating to a remote synchronous video recognition wound care guidance system. Background Technology

[0002] With the continuous advancement of medical technology and the increasing trend of population aging, wound care has become an indispensable part of clinical medical care and home care. However, uneven distribution of medical resources, inconvenience for patients to access medical care, and inconsistent quality of care have led to numerous challenges for traditional wound care models. In particular, for chronic wounds (such as lower extremity venous ulcers and diabetic foot), which have long treatment cycles and many complications, continuous professional guidance and monitoring are required. Therefore, a wound care guidance system is needed.

[0003] However, most wound care guidance systems currently available only have basic video consultation functions, which can easily lead to distortion in wound size measurement during remote identification, resulting in inaccurate dosage of nursing medication and affecting the patient's recovery. At the same time, they lack dynamic analysis and identification of various aspects of the wound, making it difficult to dynamically adjust the amount of nursing dressings. Summary of the Invention

[0004] This application provides a remote synchronous video recognition wound care guidance system, which aims to solve the problems of existing technologies that only have basic video consultation functions, which can easily lead to distortion of wound size measurement during remote recognition, resulting in inaccurate dosage of nursing medication and affecting the patient's recovery effect; at the same time, it lacks dynamic analysis and recognition of multiple aspects of the wound.

[0005] A remote synchronous video recognition wound care guidance system includes a video acquisition module, an AR measurement module, a dual-channel calculation module, a data fusion module, and a specified suggestion generation module;

[0006] The video acquisition module interfaces with the patient's mobile terminal to collect video data and uses a built-in calibration unit to synchronize different video data over time. The video acquisition module also includes an environmental perception unit and a physiological parameter acquisition unit.

[0007] The AR measurement module can dynamically overlay a preset virtual ruler template onto the corrected video image frame. After the patient selects anatomical landmarks, key parameters are automatically annotated. Key parameters include wound length, width, area, and sinus tract structure.

[0008] The dual-channel computing module uses a parallel processing mechanism to calculate the wound necrosis regression rate and granulation tissue growth slope through dynamic indicators, and to quantify the wound change trend, infiltration depth and atrophic vascular nodes.

[0009] The wound necrosis regression rate Rnecrosis The calculation formula is as follows:

[0010]

[0011] Where A necrosis (t) represents the necrotic area at the current moment, A necrosis (t-7) represents the area of ​​necrosis one week prior;

[0012] The granulation tissue growth slope K granulation The calculation formula is as follows:

[0013]

[0014] Where, x i The percentage of red on day i. The average percentage of red, t i Let i be the time of the i-th measurement. The average measurement time is n, and the number of detections is n.

[0015] The data fusion module can project the feature vectors of wound necrosis regression rate and granulation growth slope to a unified semantic space through a deep learning model to obtain the wound risk value.

[0016] The specified suggestion generation module can quickly search and match the corresponding level of nursing plan framework in the preset nursing strategy library according to the size of the wound risk value.

[0017] Furthermore, the scene perception unit is used to continuously monitor the scene illumination distribution and independently process the highlight overflow area and shadow detail area through the partition histogram equalization technology to distinguish skin color and wounds;

[0018] The physiological parameter acquisition unit is connected to the ABI detector and blood glucose meter via an interface to collect and measure the patient's lower limb hemodynamic parameters and blood glucose concentration information, respectively.

[0019] Furthermore, the AR measurement module includes a deformation elimination unit, a superposition unit, and a labeling unit.

[0020] Furthermore, the distortion elimination unit is used to eliminate lens distortion through a matrix compensation algorithm, establish the relationship between the actual imaging point and the pixel coordinate system using standard markers, calculate the correction coordinates, and perform topological modeling of the sinus tract structure.

[0021] The virtual ruler template of the overlay unit identifies planar feature points of video image frames based on computer vision algorithms and constructs a homography transformation matrix from the coordinate system to the screen pixel coordinate system.

[0022] The annotation unit can automatically annotate the wound length, width, and area information based on the marker points selected by the user.

[0023] Furthermore, the calibration unit utilizes preset spatial markers to achieve synchronization in spatial dimensions.

[0024] Furthermore, the correction coordinates are calculated based on the Harris corner detection algorithm, and the calculation formula is as follows:

[0025] x corr =x raw (1+k1r 2 +k2r 4 )

[0026] y corr =y raw (1+k1r 2 +k2r 4 )

[0027] Where k1 and k2 are both distortion coefficients, r is the defect value of the principal point coordinates, (x raw y raw () represents the vertex coordinates.

[0028] Furthermore, the deep learning model is built on an encoder-decoder architecture, where the encoder receives two feature vectors as input: the wound necrosis regression rate and the granulation tissue growth slope; the encoder is composed of multiple layers of neural networks.

[0029] Furthermore, the nursing care framework is presented in a structured form, which includes phased operation guidelines, consumable consumption prediction, and early warning threshold settings.

[0030] Furthermore, the specific functions of the dual-channel calculation module include calculating the wound necrosis regression rate, calculating the granulation tissue growth slope, calculating the daily change rate of wound volume, analyzing the depth of penetration, and evaluating vascular hemodynamics.

[0031] Furthermore, the specific content of the infiltration depth analysis is as follows: The dual-channel calculation module introduces a gradient vector field algorithm to track the direction of tissue infiltration. By analyzing the pixel displacement field between consecutive frames, a path cost map from the epidermis to deep tissues is constructed, and the lowest resistance diffusion channel is identified as the main infiltration channel. Combined with the spatial calibration data of the AR scale, the coordinates of the infiltration front end and the relative body surface depth are automatically marked, and a three-dimensional heat map is generated to show the spread range of the hidden infection.

[0032] The specific content of the fistula hemodynamic assessment is as follows: The dual-channel calculation module also integrates the pulse waveform data of the ABI detector with video blood flow imaging technology, uses the principle of optical coherence tomography to extract the microcirculation perfusion density distribution map, combines the frequency domain analysis method to calculate the peak blood flow and pulsatility index at the fistula inlet, and identifies abnormal shunt patterns through convolutional neural networks to automatically mark high-flow fistula nodes.

[0033] Compared with the prior art, this application has at least the following beneficial effects:

[0034] Based on further analysis and research of existing technical problems, this application uses an AR measurement module to augment reality and overlay a virtual ruler, enabling the system to accurately mark the length, width, area, volume, and sinus tract structure of wounds in real-time video streams. At the same time, it uses a perspective correction algorithm to eliminate shooting angle errors, reduce the deviation between remote synchronous video data and actual values, facilitate subsequent wound assessment by the system, and make it easier to accurately match nursing strategies.

[0035] Simultaneously, the dual-channel calculation module calculates the wound necrosis regression rate and granulation tissue growth slope, quantifies the wound change trend, infiltration depth, and atrophic nodes, which facilitates monitoring the rate evolution trajectory of the patient's wound, allows for dynamic adjustment of the amount of nursing dressings, and improves the effectiveness of nursing guidance. Attached Figure Description

[0036] Figure 1 This is a block diagram of a remote synchronous video recognition wound care guidance system provided in one embodiment of this application. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0038] like Figure 1 As shown, the wound care guidance system for remote synchronous video recognition provided in this application includes a video acquisition module, an AR measurement module, a dual-channel calculation module, a data fusion module, and a specified suggestion generation module;

[0039] The video acquisition module interfaces with the patient's mobile terminal to collect video data. Through a built-in calibration unit, it synchronizes different video data over time. This allows for accurate identification of the visual state of the patient's wound and their physiological parameters at a specific moment during subsequent analysis, ensuring consistency and accuracy of the time-series data and avoiding data analysis errors caused by time asynchrony.

[0040] The hardware triggering mechanism ensures that each piece of acquired data (including image frames and physiological signals) is precisely timestamped at the start of video acquisition, recording the specific moment the data was generated. When the video acquisition module acquires image frames and physiological signals at different times, it can perform millisecond-level synchronization mapping between the image frames and the corresponding physiological signals by comparing their timestamps.

[0041] Simultaneously, the calibration unit utilizes preset spatial markers to achieve spatial synchronization. Before data acquisition, a 1cm² rectangular standard marker is placed in a suitable position as a spatial reference standard. This ensures the image includes the complete marker.

[0042] The video acquisition module also includes an environmental perception unit and a physiological parameter acquisition unit. The environmental perception unit continuously monitors the scene's illumination distribution and independently processes highlight overflow areas and shadow detail areas using partitioned histogram equalization technology. When the average brightness of the skin area is detected to be below a preset threshold, a multi-level gain enhancement circuit is automatically activated, while simultaneously limiting the noise amplification factor to within the quantum error tolerance. For mixed light source scenes, a spectral separation algorithm is used to identify the color temperature of the dominant light source, driving the automatic white balance module to perform color correction, ensuring accurate reproduction of tissue color under different skin types.

[0043] The physiological parameter acquisition unit connects to an ABI analyzer and a blood glucose meter via an interface. The ABI analyzer measures the patient's lower limb hemodynamic parameters. The blood glucose meter collects the patient's blood glucose concentration information. The physiological parameter acquisition unit integrates data from different devices, providing multi-dimensional physiological information support for subsequent comprehensive assessment of the patient's condition and wound evolution.

[0044] The AR measurement module dynamically overlays a pre-set virtual ruler template onto corrected video image frames, automatically annotating key parameters after the patient selects anatomical landmarks. Key parameters include wound length, width, area, and sinus tract structure. The AR measurement module includes distortion correction units, overlay units, and annotation units.

[0045] The distortion correction unit is used to eliminate lens distortion through a matrix compensation algorithm. It establishes the relationship between the actual imaging point and the pixel coordinate system using standard markers and calculates the correction coordinates. The corrected image eliminates distortion errors, making the measurement results closer to the actual values. Details are as follows:

[0046] 1) Using an improved Harris corner detection algorithm, the vertex coordinates on the calibration board are accurately located. Based on the principal point coordinates, the corrected coordinates of the actual imaging point are calculated. The calculation formula is as follows:

[0047] x corr =x raw (1+k1r 2 +k2r 4 )

[0048] y corr =y raw (1+k1r 2 +k2r 4 )

[0049] Where k1 and k2 are both distortion coefficients, r is the defect value of the principal point coordinates, (x raw y raw Let be the coordinates of the vertex. The formula for calculating r is as follows:

[0050] r 2 =(xc x ) 2 +(yc y ) 2

[0051] Among them, (c x c y (x, y) are the coordinates of the principal point, and (x, y) are the coordinates of the actual imaging point.

[0052] 2) Sinus tract topology modeling

[0053] When a tubular cavity is detected within the wound, the system initiates a multi-section scanning mode. Video sequences from different angles are acquired by rotating the mobile terminal, and a shape memory algorithm is used to stitch them together to form a panoramic endoscopic view. The boundaries of the inner and outer walls of the sinus tract are segmented based on the level set method, a central skeleton model is constructed, and the radius of curvature distribution is calculated. A map of cross-sectional area changes along the tract is output simultaneously, accurately locating the narrowed segment and the dilated area, providing biomechanical parameter support for drainage tube selection.

[0054] The virtual ruler template of the overlay unit identifies planar feature points in video image frames using computer vision algorithms, constructing a homography transformation matrix from the coordinate system to the screen pixel coordinate system. When the user selects anatomical landmarks (such as the corners of wound edges) via multi-touch, the AR measurement module uses optical flow tracking technology to lock onto the target area, ensuring that the virtual ruler template remains dynamically bound to the actual tissue morphology. This process achieves precise alignment between digital tools and physical space, laying the foundation for subsequent quantitative analysis.

[0055] The annotation unit can automatically annotate the length, width, and area of ​​a wound based on user-selected markers. Length and width describe the wound's planar dimensions, while the area reflects the wound's size range. Through automatic annotation, medical staff can quickly obtain detailed wound data, avoiding the tediousness and potential errors of manual measurement, thus improving the accuracy and efficiency of data acquisition.

[0056] To further assess the three-dimensional morphology of the wound, the AR measurement module integrates 3D reconstruction capabilities based on monocular or multi-view geometry. Using video image frames of the wound captured from different angles, the module can generate a 3D mesh model of the wound. On this model, by calculating the spatial volume between the mesh and a fitted reference plane of surrounding healthy skin, the wound volume (cm³) can be accurately calculated, providing a more sensitive quantitative indicator for monitoring wound healing than two-dimensional area.

[0057] The dual-channel computing module, through a parallel processing mechanism, is used to calculate the wound necrosis regression rate and granulation tissue growth slope using dynamic indicators, quantifying wound change trends, infiltration depth, and atrophic vascular nodes. Specific details are as follows:

[0058] 1) Calculate the wound necrosis regression rate

[0059] Detailed wound area data is received from the AR measurement module, and a sliding window mechanism is used in the time dimension to calculate the regression rate R of necrotic tissue under the sliding window mechanism. necrosis R necrosis The value represents area reduction. Ensure each analysis period includes complete observation data for seven consecutive days to provide a stable time-series basis for subsequent calculations. The calculation formula is as follows:

[0060]

[0061] Where A necrosis (t) represents the necrotic area at the current moment, A necrosis (t-7) represents the area of ​​necrosis one week prior.

[0062] 2) Calculate the granulation tissue growth slope

[0063] Image frames were converted to the HSV color model. The HSV space describes color in three dimensions: hue, saturation, and value. In granulation tissue analysis, granulation tissue typically appears red. By setting a hue range, the red areas in the image were accurately extracted. After constructing a dynamic sequence of red proportions containing data from the last 7 days, the granulation tissue growth slope K was calculated. granulation The resulting slope directly reflects the proliferative activity of granulation tissue. This process weakens outlier interference through an adaptive weighting strategy, ensuring the robustness of trend judgment. The calculation formula is as follows:

[0064]

[0065] Where, x i The percentage of red on day i. The average percentage of red, t i Let i be the time of the i-th measurement. denoted as the average measurement time, and n represents the number of measurements.

[0066] Granulation tissue growth slope reflects the growth trend of granulation tissue over a period of time. For example, when the granulation tissue growth slope is >0, it indicates healthy granulation tissue growth. If the granulation tissue growth slope is >3%, it indicates that the granulation tissue is growing rapidly, and the system will recommend using hydrogel dressings to promote epithelialization.

[0067] 3) Calculate the daily rate of change of wound volume.

[0068] The dual-channel computing module initiates a volume change rate calculation pipeline by integrating real-time 3D mesh model data generated by the AR measurement module. This process first aligns 3D models of the same wound from two consecutive days along the timeline and employs an Iterative Closest Point (ICP) algorithm for high-precision registration to eliminate errors caused by minute differences in daily shooting angles.

[0069] After registration, the module uses the reference plane fitted to the healthy skin around the wound as a unified reference, and calculates the spatial volume of the two models relative to this reference plane to obtain yesterday's volume V. t-1 Compared to today's volume V t The volume change rate is calculated, quantifying the daily relative change in wound volume. A positive value indicates an increase in volume due to granulation tissue filling, while a negative value indicates a decrease in volume due to the regression of necrotic tissue or wound contraction. The specific calculation formula is as follows:

[0070]

[0071] After obtaining the volume change rate, the dual-channel calculation module jointly models it with the area change rate, necrosis regression rate, and granulation tissue growth slope.

[0072] Among them, the area change rate is derived from the edge detection algorithm of the two-dimensional projection image, representing the closing speed of the wound surface; the necrosis regression rate is obtained by comparing the pixel proportion decay curve of the necrotic area in consecutive frames; and the granulation tissue growth slope is based on the temporal analysis of the red channel in the HSV color space. A weighted fusion strategy is used to map these four indicators to a unified quantification space to construct a dynamic evaluation matrix, which comprehensively quantifies the healing or deterioration trend of the wound.

[0073] For example, when volume reduction is accompanied by area reduction and granulation slope increase, it is considered benign healing; if the volume increases abnormally but the necrosis rate is slowed, an infection warning is triggered. This multi-dimensional cross-validation mechanism effectively avoids the one-sidedness of a single indicator and provides clinical decision support with both spatiotemporal resolution and pathophysiological significance.

[0074] 4) Stealth Depth Analysis

[0075] The dual-channel computation module introduces a gradient vector field algorithm to track the direction of tissue infiltration. By analyzing the pixel displacement field between consecutive frames, a path cost map from the epidermis to deep tissues is constructed, identifying the lowest resistance diffusion channel as the main infiltrating pathway. Combined with spatial calibration data from an AR scale, the coordinates of the infiltrating front and its relative depth to the body surface are automatically labeled, generating a 3D heat map to display the spread range of occult infection, overcoming the limitations of traditional two-dimensional assessment.

[0076] 5) Vascular hemodynamic assessment

[0077] The dual-channel computing module also integrates pulse waveform data from the ABI detector with video blood flow imaging technology. It uses optical coherence tomography (OCT) to extract microcirculation perfusion density distribution maps, and combines this with frequency domain analysis to calculate the peak blood flow and pulsatility index at the fistula inlet. A convolutional neural network identifies abnormal shunt patterns and automatically marks high-flow fistula nodes.

[0078] The data fusion module can project the feature vectors of wound necrosis regression rate and granulation growth slope to a unified semantic space through a deep learning model to obtain wound risk values, thus laying a common foundation for subsequent comprehensive analysis.

[0079] The details are as follows:

[0080] The deep learning model is built on an encoder-decoder architecture. The encoder receives two feature vectors as input: the wound necrosis regression rate and the granulation tissue growth slope. The encoder consists of multiple layers of neural networks, each applying a non-linear activation function.

[0081] Once the feature vectors are input, the nonlinear transformation will change the representation of the data layer by layer, gradually peeling away the superficial information and digging deeper to uncover the profound semantic features hidden inside the feature vectors.

[0082] For example, for the feature vector of wound necrosis regression rate, the encoder can capture the tissue repair capacity information reflected behind its changing trend; for the feature vector of granulation tissue growth slope, deep semantics related to cell proliferation activity can be extracted.

[0083] After the encoder completes its processing, the resulting semantic features are fed into the decoder. The decoder, also built on a multi-layered neural network, has the task of accurately mapping the semantic features output by the encoder into a unified semantic space. In this mapping process, the decoder adjusts network weights and biases to map features from two different sources—the wound necrosis regression rate and the granulation tissue growth slope—into a unified semantic space, providing a standardized data foundation for subsequent wound care guidance.

[0084] The specified suggestion generation module can quickly search and match the corresponding level of nursing plan framework in the preset nursing strategy library based on the wound risk value.

[0085] For example, high-risk wounds are automatically associated with negative pressure wound therapy (NPWT) preparation procedures, while medium-risk wounds are matched with silver ion dressing usage guidelines. This stage achieves an initial mapping from objective indicators to clinical pathways, ensuring that basic protective measures are matched with the degree of injury.

[0086] The nursing strategy library contains specific guidance programs for complex wound structures.

[0087] For example, when burrowing features are identified, the protocol recommends the use of packing materials such as alginate strips or negative pressure wound therapy (NPWT); for detected sinus tracts or fistulas, drainage dressings (such as Utop SSD) will be matched and the importance of thorough exploration and drainage will be emphasized. All recommendations are combined with wound volume, area, and dynamic trend indicators to generate highly personalized operational guidelines.

[0088] The nursing care protocol framework is presented in a structured form, including phased operational guidelines (such as the order of debridement tool selection and the timing of growth factor application), consumable consumption prediction, and early warning threshold settings. It also generates a three-dimensional anatomical annotation map, using AR overlay technology to guide bedside procedures in real time. The module incorporates a therapeutic feedback loop, continuously collecting tissue response data during actual treatment and dynamically adjusting subsequent treatment parameters to form a closed-loop quality control system.

[0089] In the aforementioned remote synchronous video recognition wound care guidance system, an AR measurement module is used to overlay a virtual ruler, enabling the system to accurately mark the length, width, area, volume, undermining, sinus tract, and fistula parameters of the wound in the real-time video stream. At the same time, a fluoroscopic correction algorithm is used to eliminate shooting angle errors, reduce the deviation between remote synchronous video data and actual values, facilitate subsequent wound judgment by the system, and make it easier to accurately match care strategies.

[0090] Simultaneously, the dual-channel calculation module calculates the wound necrosis regression rate and granulation tissue growth slope, quantifies the wound change trend, infiltration depth, and atrophic nodes, which facilitates monitoring the rate evolution trajectory of the patient's wound, allows for dynamic adjustment of the amount of nursing dressings, and improves the effectiveness of nursing guidance.

[0091] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A remote synchronous video recognition wound care guidance system, characterized in that, It includes a video acquisition module, an AR measurement module, a dual-channel calculation module, a data fusion module, and a specified suggestion generation module; The video acquisition module interfaces with the patient's mobile terminal to collect video data and uses a built-in calibration unit to synchronize different video data over time. The video acquisition module also includes an environmental perception unit and a physiological parameter acquisition unit. The AR measurement module can dynamically overlay a preset virtual ruler template onto the corrected video image frame, and automatically annotate key parameters after the patient selects anatomical landmarks. Key parameters include wound length, width, area, and sinus tract structure; The dual-channel computing module uses a parallel processing mechanism to calculate the wound necrosis regression rate and granulation tissue growth slope through dynamic indicators, and quantifies the wound change trend, infiltration depth and atrophic vascular nodes. The wound necrosis regression rate R necrosis The calculation formula is as follows: Where A necrosis (t) represents the necrotic area at the current moment, A necrosis (t-7) represents the area of ​​necrosis one week prior; The granulation tissue growth slope K granulation The calculation formula is as follows: Where, x i The percentage of red on day i. The average percentage of red, t i Let i be the time of the i-th measurement. The average measurement time is n, and the number of detections is n. The data fusion module can project the feature vectors of wound necrosis regression rate and granulation growth slope to a unified semantic space through a deep learning model to obtain the wound risk value. The specified suggestion generation module can quickly search and match the corresponding level of nursing plan framework in the preset nursing strategy library according to the size of the wound risk value.

2. The wound care guidance system for remote synchronous video recognition according to claim 1, characterized in that, The environmental perception unit is used to continuously monitor the scene illumination distribution and independently process the highlight overflow area and shadow detail area through the partition histogram equalization technology to distinguish skin color and wounds. The physiological parameter acquisition unit is connected to the ABI detector and blood glucose meter via an interface to collect and measure the patient's lower limb hemodynamic parameters and blood glucose concentration information, respectively.

3. The wound care guidance system for remote synchronous video recognition according to claim 1, characterized in that, The AR measurement module includes a deformation elimination unit, a superposition unit, and a labeling unit.

4. The wound care guidance system for remote synchronous video recognition according to claim 3, characterized in that, The distortion elimination unit is used to eliminate lens distortion through a matrix compensation algorithm, establish the relationship between the actual imaging point and the pixel coordinate system using standard markers, calculate the correction coordinates, and perform topological modeling of the sinus tract structure. The virtual ruler template of the overlay unit identifies planar feature points of video image frames based on computer vision algorithms and constructs a homography transformation matrix from the coordinate system to the screen pixel coordinate system. The annotation unit can automatically annotate the wound length, width, and area information based on the marker points selected by the user.

5. A remote synchronous video recognition wound care guidance system according to claim 1, characterized in that, The calibration unit uses preset spatial markers to achieve synchronization in spatial dimensions.

6. A remote synchronous video recognition wound care guidance system according to claim 4, characterized in that, The correction coordinates are calculated based on the Harris corner detection algorithm, and the calculation formula is as follows: x corr =x raw (1+k1r 2 +k2r 4 and corr / and raw (1+k1r 2 +k2r 4 ) Where k1 and k2 are both distortion coefficients, r is the defect value of the principal point coordinates, (x raw y raw () represents the coordinates of the vertex.

7. A remote synchronous video recognition wound care guidance system according to claim 1, characterized in that, The deep learning model is built on an encoder-decoder architecture. The encoder receives two feature vectors as input: the wound necrosis regression rate and the granulation tissue growth slope. The encoder is composed of multiple layers of neural networks.

8. A remote synchronous video recognition wound care guidance system according to claim 1, characterized in that, The nursing care plan framework is presented in a structured form, which includes phased operation guidelines, consumable consumption prediction, and early warning threshold settings.

9. A remote synchronous video recognition wound care guidance system according to claim 1, characterized in that, The specific functions of the dual-channel calculation module include calculating the wound necrosis regression rate, calculating the granulation tissue growth slope, calculating the daily change rate of wound volume, analyzing the depth of penetration, and evaluating vascular hemodynamics.

10. A remote synchronous video recognition wound care guidance system according to claim 9, characterized in that, The specific content of the infiltration depth analysis is as follows: The dual-channel calculation module introduces the gradient vector field algorithm to track the direction of tissue infiltration. By analyzing the pixel displacement field between consecutive frames, a path cost map from the epidermis to the deep tissue is constructed, and the lowest resistance diffusion channel is identified as the main infiltration channel. Combined with the spatial calibration data of the AR scale, the coordinates of the infiltration front end and the relative body surface depth are automatically marked, and a three-dimensional heat map is generated to show the spread range of the hidden infection. The specific content of the fistula hemodynamic assessment is as follows: The dual-channel calculation module also integrates the pulse waveform data of the ABI detector with video blood flow imaging technology, uses the principle of optical coherence tomography to extract the microcirculation perfusion density distribution map, combines the frequency domain analysis method to calculate the peak blood flow and pulsatility index at the fistula inlet, and identifies abnormal shunt patterns through convolutional neural networks to automatically mark high-flow fistula nodes.

Images