Respiratory rehabilitation evaluation exercise analysis system based on internet of things

Abstract

The application discloses a respiratory rehabilitation evaluation motion analysis system based on the Internet of Things, which comprises an image acquisition module, a respiratory rehabilitation motion region enhancement module, a respiratory micro-change decomposition module, a respiratory motion contour enhancement module, a structure preservation fusion target construction module, a respiratory motion artifact suppression module, a rehabilitation state parameter extraction module and a motion state classification module. The application belongs to the field of image processing, and the scheme is characterized in that the environmental brightness benchmark statistics are used for adaptive light compensation, the image difference is reduced, the respiratory related region response is enhanced by using the saliency weight, the slight deformation is made more obvious, the human body chest and abdomen space structure distortion is avoided by using the structure preservation constraint, the real respiratory motion change is enhanced by using the edge continuity constraint, the adaptive truncation optimization method based on the local feature distribution is adopted to suppress the abnormal response caused by body shaking, environmental change and image acquisition noise, and finally the respiratory rehabilitation evaluation motion analysis effect is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing, specifically to a respiratory rehabilitation assessment and motion analysis system based on the Internet of Things. Background Technology

[0002] Respiratory rehabilitation assessment and motion analysis systems are technical systems that acquire videos of patients' breathing training, extract image features using image processing algorithms, and then use classification models to analyze rehabilitation status. However, general respiratory rehabilitation assessment and motion analysis systems suffer from several problems: significant differences in brightness of motion images due to changes in the acquisition environment; instability of chest and abdominal region features; and the easy subtraction of subtle respiratory movement changes by background noise, all of which affect the accuracy of rehabilitation status analysis. Furthermore, these systems often exhibit issues such as unclear changes in respiratory movement boundaries, strong interference from changes in body posture, and abnormal movement responses that can influence the results. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an IoT-based respiratory rehabilitation assessment and motion analysis system. Addressing the problems of general respiratory rehabilitation assessment and motion analysis systems, such as significant differences in motion image brightness due to changes in the acquisition environment, unstable chest and abdominal region features, and the easy obscuring of subtle respiratory motion changes by background noise, thus affecting the accuracy of rehabilitation status analysis, this solution uses environmental brightness benchmark statistics for adaptive illumination compensation to reduce image differences caused by different training scenarios and acquisition devices. It also utilizes saliency weights to enhance the response of respiratory-related regions, making the subtle deformations caused by chest and abdominal expansion and contraction more apparent. Furthermore, it separates basic human structural information through multi-scale motion feature decomposition and edge enhancement. By integrating respiratory dynamic changes with information, the stability of respiratory motion feature extraction is improved, thereby enhancing the accuracy of respiratory rehabilitation assessment motion analysis. Addressing the issues of unclear respiratory motion boundary changes, strong interference from changes in body posture, and the tendency for abnormal motion responses to affect results in general respiratory rehabilitation assessment motion analysis systems, this solution integrates the basic structural features of the chest and abdomen region with enhanced respiratory motion edge features. Structure preservation constraints prevent distortion of the human chest and abdomen spatial structure, while edge continuity constraints enhance realistic respiratory motion changes. An adaptive truncation optimization method based on local feature distribution is employed to suppress abnormal responses caused by body swaying, environmental changes, and image acquisition noise. Ultimately, this improves the effectiveness of respiratory rehabilitation assessment motion analysis.

[0004] The technical solution adopted by the present invention is as follows: The respiratory rehabilitation assessment and motion analysis system based on the Internet of Things provided by the present invention includes an image acquisition module, a respiratory rehabilitation motion region enhancement module, a respiratory micro-change decomposition module, a respiratory motion contour enhancement module, a structure preservation and fusion target construction module, a respiratory motion artifact suppression module, a rehabilitation state parameter extraction module, and a motion state classification module.

[0005] The image acquisition module acquires a sequence of continuous motion images of the patient during respiratory rehabilitation training through an IoT vision acquisition terminal, and constructs a neutral brightness reference after preprocessing.

[0006] The respiratory rehabilitation motion area enhancement module obtains key points of the patient's body based on the human posture estimation algorithm, locates the chest and abdomen area, and performs dynamic compensation on the pixels of the chest and abdomen area based on a neutral brightness reference.

[0007] The respiratory micro-change decomposition module generates respiratory motion saliency weights based on key points in the chest and abdomen, performs spatial weighting on the enhanced chest and abdominal motion region, extracts low-frequency structures, obtains the basic structural features of the chest and abdominal region, and then reconstructs the image.

[0008] The breathing motion contour enhancement module constructs a multi-scale edge layer for the reconstructed image and finally extracts the breathing motion edge enhancement features through the difference between adjacent scales.

[0009] The structure-preserving fusion target construction module fuses the basic structural features of the thoracic and abdominal regions with the edge enhancement features of respiratory motion, and applies structure-preserving constraints and edge continuity constraints.

[0010] The breathing motion artifact suppression module introduces scene-aware truncation based on local grayscale distribution: by analyzing the grayscale change trend in local areas, the enhancement intensity is adaptively adjusted to optimize the fusion result and achieve breathing motion artifact suppression;

[0011] The rehabilitation status parameter extraction module extracts respiratory rhythm stability, postural stability, and motor coordination indicators based on the optimized respiratory movement characteristics, as rehabilitation status features.

[0012] The motion state classification module takes the normalized rehabilitation state features as input to the XGBoost classification model and achieves the classification of the patient's respiratory rehabilitation motion state through model training.

[0013] Furthermore, after preprocessing, the image acquisition module calculates the average brightness value H of the three color channels of the current patient motion image. R H G and H B , serving as the pixel brightness benchmark for the red, green, and blue channels; constructing a neutral brightness benchmark for the current training environment: ; It is the neutral brightness benchmark for the current rehabilitation training environment.

[0014] Furthermore, the respiratory rehabilitation motion area enhancement module obtains key points of the human body through a human posture estimation algorithm to determine the chest and abdominal motion areas; and calculates the current color offset of the chest and abdominal areas based on a neutral brightness reference. ; ; ; This refers to the compensation amount for each channel; dynamic compensation of the chest and abdomen motion region is performed for any pixel within the chest and abdomen region: ; is the pixel value of the patient's motion image before compensation of the three color channels; (x,y) is the pixel coordinate; These are the pixel values ​​of the patient's chest and abdominal motion region after compensation; the compensation result is then truncated. ; These are the pixel values ​​of the enhanced respiratory motion region; It is a pixel range limitation function.

[0015] Furthermore, the respiratory micro-change decomposition module specifically includes:

[0016] The saliency weights of chest and abdominal movements are generated by determining the center of chest and abdominal movement based on the results of human key point detection. ;Construct saliency weights for chest and abdominal movements based on the center position of the chest and abdomen: ; It is the significance weight of chest and abdominal movements; These are Gaussian parameters that control the spatial scope of interest; and spatial augmentation is performed, expressed as: ; These are the image pixel values ​​after spatial enhancement;

[0017] Respiratory micromotor decomposition was used to extract low-frequency structures from spatially enhanced images, obtaining basic structural features of the thoracic and abdominal regions. ;

[0018] The image is reconstructed, and wavelet transform is used to decompose the breathing region into different scales to obtain respiratory motion change coefficients. Motion details are then reconstructed based on these coefficients. ; ; is the respiratory motion variation coefficient; a is the scale parameter; b is the spatial displacement parameter; p is the image spatial sampling position; It is a two-dimensional Haar wavelet basis function; It is to reconstruct the image; It refers to the dynamic changes in respiration.

[0019] Furthermore, the respiratory motion contour enhancement module specifically includes:

[0020] A Gaussian scale space is established based on the reconstructed image: ; It is the nth Gaussian-scaled image of the reconstructed image; c and d are the convolution window offsets; It is the image at the next higher Gaussian scale; It is a Gaussian convolution kernel;

[0021] Multi-scale respiratory edge feature extraction: The respiratory motion edge layer is obtained by differencing adjacent scales. ; It is an upsampling operation; It is the nth respiratory movement edge layer;

[0022] Multi-scale edge fusion involves weighted fusion of edge layers at various scales to obtain enhanced respiratory motion edge features. .

[0023] Furthermore, the structure-preserving fusion target construction module, based on the basic structural features and the enhanced features of respiratory motion edges, fuses and generates respiratory motion features, and establishes a fusion optimization target through structure preservation constraints and edge continuity constraints: ; ; These are the respiratory movement characteristics after final fusion; It is a spatial gradient operator; The structural features maintain proportion; It is the edge continuity constraint strength; It refers to the fusion ratio.

[0024] Furthermore, the respiratory motion artifact suppression module specifically includes:

[0025] Local grayscale change feature calculation: For the final fused respiratory motion features, the grayscale range is calculated within a local window. ; ; and These are the maximum and minimum eigenvalues ​​of the local window; calculate the local variation magnitude. ;

[0026] Adaptive enhanced slope calculation, based on the local cumulative distribution function Constructing dynamic enhancement factors: ; It is the gradient of grayscale distribution; It is a dynamic enhancement factor;

[0027] Dynamic truncation threshold calculation, generating the truncation range based on the local window size: ; It is the cutoff threshold; It refers to the size of a local window; It is a window adjustment weight; It is the characteristic distribution coefficient;

[0028] Respiratory motion features were optimized by adjusting the fusion features based on a dynamic enhancement factor and a cutoff threshold. ; These are optimized respiratory movement characteristics; It is the average feature value of the local window; It is a truncation function.

[0029] Furthermore, the rehabilitation status parameter extraction module specifically includes:

[0030] Based on the edge features of respiratory motion, the differences in chest and abdominal motion changes between consecutive video frames are calculated, and then the stability of respiratory rhythm is calculated.

[0031] Based on human keypoints, the changes in the patient’s torso spatial posture during training are obtained, and stability is calculated based on the posture changes in consecutive frames.

[0032] Based on respiratory movement characteristics, the differences in motor response between the chest and abdominal regions were calculated separately as indicators of motor coordination; these were used to form rehabilitation status characteristics and then normalized.

[0033] Furthermore, the motion state classification module uses the normalized rehabilitation state features as input to the classification model. By learning the respiratory motion feature distribution of patients at different rehabilitation stages, it achieves the classification of patients' rehabilitation motion state, using the XGBoost classification model.

[0034] After the model is trained, the continuous motion image sequence of the patient during respiratory rehabilitation training is acquired in real time through the Internet of Things visual acquisition terminal. After obtaining standardized rehabilitation state features, the patient's rehabilitation motion state is classified using a classification model.

[0035] The beneficial effects achieved by the present invention using the above solution are as follows:

[0036] (1) In view of the problems of general respiratory rehabilitation assessment motion analysis system, such as the significant difference in brightness of motion images caused by changes in the acquisition environment, the instability of chest and abdominal region features, and the easy submersion of small respiratory motion changes by background noise, which affects the accuracy of rehabilitation status analysis, this solution uses environmental brightness benchmark statistics to perform adaptive illumination compensation to reduce the image differences caused by different training scenarios and different acquisition devices; and uses saliency weights to enhance the response of respiratory related regions, making the small deformations caused by chest and abdominal expansion and contraction more obvious; and separates the basic human body structure information and respiratory dynamic change information through multi-scale motion feature decomposition and edge enhancement, thereby improving the stability of respiratory motion feature extraction and thus improving the accuracy of respiratory rehabilitation assessment motion analysis.

[0037] (2) In view of the problems of unclear changes in respiratory movement boundaries, strong interference from changes in human posture, and the easy influence of abnormal movement response on the results of general respiratory rehabilitation assessment movement analysis systems, this scheme integrates the basic structural features of the chest and abdomen region with the edge enhancement features of respiratory movement. The structure preservation constraint avoids the distortion of the human chest and abdomen spatial structure, while the edge continuity constraint enhances the real respiratory movement changes. An adaptive truncation optimization method based on local feature distribution is adopted to suppress abnormal responses caused by body swaying, environmental changes, and image acquisition noise. Finally, the effect of respiratory rehabilitation assessment movement analysis is improved. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the IoT-based respiratory rehabilitation assessment and exercise analysis system provided by the present invention.

[0039] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0040] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0042] Example 1, see Figure 1 The present invention provides an IoT-based respiratory rehabilitation assessment and motion analysis system, which includes an image acquisition module, a respiratory rehabilitation motion region enhancement module, a respiratory micro-change decomposition module, a respiratory motion contour enhancement module, a structure-preserving fusion target construction module, a respiratory motion artifact suppression module, a rehabilitation state parameter extraction module, and a motion state classification module.

[0043] The image acquisition module acquires a sequence of continuous motion images of the patient during respiratory rehabilitation training through an IoT vision acquisition terminal, and constructs a neutral brightness reference after preprocessing.

[0044] The respiratory rehabilitation motion area enhancement module obtains key points of the patient's body based on the human posture estimation algorithm, locates the chest and abdomen area, and performs dynamic compensation on the pixels of the chest and abdomen area based on a neutral brightness reference.

[0045] The respiratory micro-change decomposition module generates respiratory motion saliency weights based on key points in the chest and abdomen, performs spatial weighting on the enhanced chest and abdominal motion region, extracts low-frequency structures, obtains the basic structural features of the chest and abdominal region, and then reconstructs the image.

[0046] The breathing motion contour enhancement module constructs a multi-scale edge layer for the reconstructed image and finally extracts the breathing motion edge enhancement features through the difference between adjacent scales.

[0047] The structure-preserving fusion target construction module fuses the basic structural features of the thoracic and abdominal regions with the edge enhancement features of respiratory motion, and applies structure-preserving constraints and edge continuity constraints.

[0048] The breathing motion artifact suppression module introduces scene-aware truncation based on local grayscale distribution: by analyzing the grayscale change trend in local areas, the enhancement intensity is adaptively adjusted to optimize the fusion result and achieve breathing motion artifact suppression;

[0049] The rehabilitation status parameter extraction module extracts respiratory rhythm stability, postural stability, and motor coordination indicators based on the optimized respiratory movement characteristics, as rehabilitation status features.

[0050] The motion state classification module takes the normalized rehabilitation state features as input to the XGBoost classification model and achieves the classification of the patient's respiratory rehabilitation motion state through model training.

[0051] Example 2, see Figure 1 This embodiment is based on the above embodiment. The image acquisition module acquires a continuous motion image sequence of the patient during respiratory rehabilitation training through an Internet of Things visual acquisition terminal deployed in the rehabilitation training area. The acquired images are preprocessed, including spatial unification and brightness standardization, to reduce the differences in image quality caused by different training environments, different equipment parameters and different acquisition times.

[0052] The continuous motion image sequence includes information on the expansion and contraction of the patient's chest and abdominal region, as well as changes in body posture.

[0053] After preprocessing, the average brightness value H of the three color channels of the current patient motion image is calculated. R H G and H B , serving as the pixel brightness benchmark for the red, green, and blue channels; constructing a neutral brightness benchmark for the current training environment: ; It is the neutral brightness benchmark for the current rehabilitation training environment.

[0054] Example 3, see Figure 1 This embodiment is based on the above embodiment. The respiratory rehabilitation motion region enhancement module locates the patient's body area, extracts the chest and abdominal motion regions related to breathing as regions of interest from continuous motion images, and dynamically compensates the pixels of the chest and abdominal regions based on the neutral brightness reference of the current rehabilitation training environment, so that the respiratory motion region maintains a stable brightness and color distribution under different acquisition environments.

[0055] Human pose estimation algorithms are used to obtain key points of the human body and determine the movement areas of the chest and abdomen.

[0056] Calculate the current color shift in the chest and abdomen region based on a neutral brightness benchmark: ; ; ; This refers to the compensation amount for each channel; dynamic compensation of the chest and abdomen motion region is performed for any pixel within the chest and abdomen region: ; These are the pixel values ​​of the patient's motion image before compensation for the three color channels; These are the pixel values ​​of the patient's chest and abdominal motion region after compensation; (x, y) are the pixel coordinates; to avoid the compensated pixels exceeding the effective range, the compensation result is truncated: ; These are the pixel values ​​of the enhanced respiratory motion region; It is a pixel range limitation function.

[0057] Example 4, see Figure 1 This embodiment is based on the above embodiment. After the respiratory rehabilitation exercise area is enhanced, the respiratory micro-change decomposition module determines the main respiratory exercise area based on the key points of human posture, constructs the salience weight of chest and abdominal movement, and spatially enhances the respiratory-related area, so that the micro deformations generated by chest and abdominal expansion and contraction can obtain a higher response.

[0058] Low-frequency structure extraction was performed on the weighted respiratory motion region image to obtain the basic structural features of the chest and abdomen region; the specific operation was as follows:

[0059] The saliency weights for chest and abdominal movements are generated by determining the center of chest and abdominal movement based on the results of human keypoint detection. ; It is the location of the respiratory movement center. Here, m represents the coordinates of the i-th keypoint; m is the number of keypoints in the chest and abdomen region; the salience weights of chest and abdominal movements are constructed based on the center position of the chest and abdomen. ; It is the significance weight of chest and abdominal movements; This is a Gaussian parameter controlling the spatial focus area, with a value ranging from 20 to 80; and spatial augmentation is performed, expressed as: ; These are the image pixel values ​​after spatial enhancement;

[0060] Respiratory micro-motion decomposition and low-frequency structure extraction of spatially enhanced images yield basic structural features of the thoracic and abdominal regions: ; These are basic structural features; It is a two-dimensional Gaussian low-pass filter kernel; It is a convolution operation;

[0061] To reconstruct the image, since respiratory motion is periodic and continuous, wavelet transform is used to decompose the respiratory region into different scales, obtaining respiratory motion variation coefficients, and then reconstructing motion details based on these coefficients. ; ; is the respiratory motion variation coefficient; a is the scale parameter; b is the spatial displacement parameter; p is the image spatial sampling position; It is a two-dimensional Haar wavelet basis function; It is to reconstruct the image; It is a characteristic of dynamic changes in respiration. .

[0062] Example 5, see Figure 1 This embodiment is based on the above embodiment. After obtaining the reconstructed image, the respiratory motion contour enhancement module needs to further extract the continuous deformation boundaries of the patient's chest and abdomen region during respiration: a multi-scale edge pyramid is constructed based on the reconstructed respiratory motion feature image. Unstructured noise is gradually reduced through Gaussian smoothing at different scales, and the edge change features of the chest and abdomen region are obtained using the difference between adjacent scales, thereby enhancing the real chest expansion and abdominal contraction boundaries. The large scale is used to describe the overall chest and abdomen contour changes, and the small scale is used to capture the local small deformations generated during respiration. The specific operation is as follows:

[0063] A Gaussian scale space is established based on the reconstructed image: ; It is the nth Gaussian-scaled image of the reconstructed image; c and d are the convolution window offsets; It is the image at the next higher Gaussian scale; It uses a Gaussian convolution kernel and a two-dimensional 5×5 template;

[0064] Multi-scale respiratory edge feature extraction: The respiratory motion edge layer is obtained by differencing adjacent scales. ; It is an upsampling operation; It is the nth respiratory movement edge layer;

[0065] Multi-scale edge fusion involves weighted fusion of edge layers at various scales. ; It represents the enhanced edge features of respiratory motion; N is the number of edge scales; It is the scale fusion weight, and the total weight is 1.

[0066] By performing the above operations, this solution addresses the problems of general respiratory rehabilitation assessment motion analysis systems, such as significant differences in motion image brightness due to changes in the acquisition environment, unstable chest and abdominal region features, and the easy submersion of subtle respiratory motion changes by background noise, which affect the accuracy of rehabilitation status analysis. This solution uses environmental brightness benchmark statistics for adaptive illumination compensation to reduce image differences caused by different training scenarios and acquisition devices. It also utilizes saliency weights to enhance the response of respiratory-related regions, making the subtle deformations caused by chest and abdominal expansion and contraction more apparent. Furthermore, through multi-scale motion feature decomposition and edge enhancement, it separates basic human structural information from dynamic respiratory changes, improving the stability of respiratory motion feature extraction and thus enhancing the accuracy of respiratory rehabilitation assessment motion analysis.

[0067] Example 6, see Figure 1 This embodiment is based on the above embodiment. The structure-preserving fusion target construction module fuses respiratory motion features based on the basic structural features and respiratory motion edge enhancement features. In order to maintain the stability of the thoracic and abdominal motion spatial structure, a fusion optimization target is established through structure preservation constraints and edge continuity constraints. ; ; These are the respiratory movement characteristics after final fusion; It is a spatial gradient operator; The structural features are maintained in proportion, with a value ranging from 0.4 to 0.7; It is the edge continuity constraint strength, with a value ranging from 0.1 to 0.3; This is the fusion ratio, ranging from 0.5 to 0.8;

[0068] The first constraint is structural preservation, which preserves the human thoracic and abdominal structure in the fusion result; the second constraint is edge continuity, which preserves respiratory edge changes in the fusion result; the third constraint prevents edge position drift.

[0069] Example 7, see Figure 1 This embodiment is based on the above embodiment. Since the fused respiratory features may exhibit local abnormal responses, the respiratory motion artifact suppression module introduces scene-aware truncation based on local grayscale distribution. By analyzing the grayscale change trend in local areas, the enhancement intensity is adaptively adjusted to suppress abnormally high-response areas while preserving the true respiratory motion edge information. Specifically, the operation is as follows:

[0070] Local grayscale change feature calculation: For the final fused respiratory motion features, the grayscale range is calculated within a local window. ; ; and These are the maximum and minimum eigenvalues ​​of the local window; calculate the local variation magnitude. ;

[0071] Adaptive enhanced slope calculation, based on the local cumulative distribution function Constructing dynamic enhancement factors: ; It is the gradient of grayscale distribution; It is a dynamic enhancement factor; when there is an abnormal concentrated response in a local area, the CDF change gradient increases, which inhibits the enhancement factor through the denominator; when the area contains changes at the edge of effective respiratory motion, the CDF change is relatively gentle, maintaining a high enhancement response;

[0072] Dynamic truncation threshold calculation, generating the truncation range based on the local window size: ; It is the cutoff threshold; It refers to the size of a local window; This is the window adjustment weight, with a value ranging from 0.6 to 0.9; It is the feature allocation coefficient, with a value of 128~200, which controls the feature truncation range;

[0073] Respiratory motion features were optimized by adjusting the fusion features based on a dynamic enhancement factor and a cutoff threshold. ; These are optimized respiratory movement characteristics; It is the average feature value of the local window; It is a truncation function. z and P are the eigenvalues ​​to be truncated and the upper limit of truncation of the truncation function.

[0074] By performing the above operations, this solution addresses the problems of unclear changes in respiratory movement boundaries, strong interference from changes in human posture, and the tendency of abnormal movement responses to affect the results in general respiratory rehabilitation assessment motion analysis systems. It integrates the basic structural features of the chest and abdomen region with the edge enhancement features of respiratory movements. By using structure preservation constraints to avoid distortion of the human chest and abdomen spatial structure, and edge continuity constraints to enhance realistic respiratory movement changes, it employs an adaptive truncation optimization method based on local feature distribution to suppress abnormal responses caused by body swaying, environmental changes, and image acquisition noise. Ultimately, this improves the effectiveness of respiratory rehabilitation assessment motion analysis.

[0075] Example 8, see Figure 1This embodiment, based on the above embodiment, extracts rehabilitation assessment parameters from three dimensions: respiratory cycle stability, changes in human posture, and coordination of chest and abdominal movements, constructing a multidimensional respiratory movement feature set; specifically including:

[0076] Based on respiratory motion edge features, the differences in chest and abdominal motion changes between consecutive video frames are calculated: Then, the stability of the respiratory rhythm is calculated: ;in, U is the change in respiratory movement at time t; U is the number of effective respiratory characteristics. and These are the breathing edge features of the i-th feature point in frame t and frame (t-1), respectively. It is an indicator of respiratory rhythm stability; It is a variance calculation; It is a smoothing term;

[0077] Based on human keypoints, the changes in the patient's torso spatial posture during training are obtained, and stability is calculated based on the posture changes in consecutive frames. ;in, It is an attitude stability index; and These are the attitude angles at frame t and frame (t-1), respectively; attitude angles Including the left and right rotation angles of the human body Angle of forward and backward tilt of the human body and body tilt angle , represented as ; It is the maximum allowable attitude change angle;

[0078] Based on respiratory movement characteristics, the differences in motor response between the chest and abdominal regions were calculated separately and used as an indicator of motor coordination. This was obtained through regional gradient changes. ;in, It is an indicator of motor coordination; It represents the gradient of respiratory motion characteristics; mean(·) is the region pooling; and These are the chest ROI and abdominal ROI regions, obtained through human pose key point localization. It is the correlation coefficient; it constitutes the characteristics of the recovery state. And then normalize it.

[0079] Example 9, see Figure 1This embodiment is based on the above embodiment. The motion state classification module takes the normalized rehabilitation state features as input to the classification model. By learning the respiratory motion feature distribution of patients at different rehabilitation stages, the classification of patients' rehabilitation motion state is realized. The XGBoost classification model is adopted. The model adopts multi-class cross-entropy loss, and the parameters are obtained by iterative optimization through historical training samples. The categories of patients' rehabilitation motion state include excellent rehabilitation state, normal training state, suspected abnormal state, and abnormal training state.

[0080] After the model is trained, the continuous motion image sequence of the patient during respiratory rehabilitation training is acquired in real time through the Internet of Things visual acquisition terminal. After obtaining standardized rehabilitation state features, the patient's rehabilitation motion state is classified using a classification model. If the output patient rehabilitation motion state is a suspected abnormal state, the image sampling frequency is increased; if it is an abnormal training state, an early warning is issued to the management personnel.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0082] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A respiratory rehabilitation assessment and motion analysis system based on the Internet of Things, characterized in that: The system includes an image acquisition module, a respiratory rehabilitation motion region enhancement module, a respiratory micro-change decomposition module, a respiratory motion contour enhancement module, a structure-preserving fusion target construction module, a respiratory motion artifact suppression module, a rehabilitation state parameter extraction module, and a motion state classification module. The image acquisition module acquires a sequence of continuous motion images of the patient during respiratory rehabilitation training through an IoT vision acquisition terminal, and constructs a neutral brightness reference after preprocessing. The respiratory rehabilitation motion area enhancement module obtains key points of the patient's body based on the human posture estimation algorithm, locates the chest and abdomen area, and performs dynamic compensation on the pixels of the chest and abdomen area based on a neutral brightness reference. The respiratory micro-change decomposition module generates respiratory motion saliency weights based on key points in the chest and abdomen, spatially weights the enhanced chest and abdominal motion region, extracts low-frequency structures, obtains the basic structural features of the chest and abdominal region, and then reconstructs the image. The breathing motion contour enhancement module constructs a multi-scale edge layer for the reconstructed image and finally extracts the breathing motion edge enhancement features through the difference between adjacent scales. The structure-preserving fusion target construction module fuses the basic structural features of the thoracic and abdominal regions with the edge enhancement features of respiratory motion, and applies structure-preserving constraints and edge continuity constraints. The breathing motion artifact suppression module introduces scene-aware truncation based on local grayscale distribution: by analyzing the grayscale change trend in local areas, the enhancement intensity is adaptively adjusted to optimize the fusion result and achieve breathing motion artifact suppression; The rehabilitation status parameter extraction module extracts respiratory rhythm stability, postural stability, and motor coordination indicators based on the optimized respiratory movement characteristics, as rehabilitation status features. The motion state classification module takes the normalized rehabilitation state features as input to the XGBoost classification model and achieves the classification of the patient's respiratory rehabilitation motion state through model training.

2. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 1, characterized in that: The respiratory rehabilitation motion area enhancement module obtains key points of the human body through a human posture estimation algorithm to determine the chest and abdominal motion areas; it calculates the current color offset of the chest and abdominal areas based on a neutral brightness benchmark. ; ; ; This refers to the compensation amount for each channel; It is the neutral brightness benchmark for the current rehabilitation training environment; H R H G and H B This is the average brightness value of the three color channels; dynamic compensation for chest and abdominal motion regions is performed on any pixel within the chest and abdomen region. ; These are the pixel values ​​of the patient's motion image before compensation for the three color channels; (x,y) are pixel coordinates; These are the pixel values ​​of the patient's chest and abdominal motion region after compensation; The compensation results are truncated: ; These are the pixel values ​​of the enhanced respiratory motion region; It is a pixel range limitation function.

3. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 2, characterized in that: The respiratory micro-change decomposition module specifically includes: The saliency weights of chest and abdominal movements are generated by determining the center of chest and abdominal movement based on the results of human key point detection. ;Construct saliency weights for chest and abdominal movements based on the center position of the chest and abdomen: ; It is the significance weight of chest and abdominal movements; These are Gaussian parameters that control the spatial scope of interest; and spatial augmentation is performed, expressed as: ; These are the image pixel values ​​after spatial enhancement; Respiratory micromotor decomposition was used to extract low-frequency structures from spatially enhanced images, obtaining basic structural features of the thoracic and abdominal regions. ; The image is reconstructed, and wavelet transform is used to decompose the breathing region into different scales to obtain respiratory motion change coefficients. Motion details are then reconstructed based on these coefficients. ; ; is the respiratory motion variation coefficient; a is the scale parameter; b is the spatial displacement parameter; p is the image spatial sampling position; It is a two-dimensional Haar wavelet basis function; It is to reconstruct the image; It refers to the dynamic changes in respiration.

4. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 3, characterized in that: The respiratory motion contour enhancement module specifically includes: A Gaussian scale space is established based on the reconstructed image: ; It is the nth Gaussian-scaled image of the reconstructed image; c and d are the convolution window offsets; It is the image at the next higher Gaussian scale; It is a Gaussian convolution kernel; Multi-scale respiratory edge feature extraction: The respiratory motion edge layer is obtained by differencing adjacent scales. ; It is an upsampling operation; It is the nth respiratory movement edge layer; Multi-scale edge fusion involves weighted fusion of edge layers at various scales to obtain enhanced respiratory motion edge features. .

5. The IoT-based respiratory rehabilitation assessment and motion analysis system according to claim 4, characterized in that: The structure-preserving fusion target construction module, based on the basic structural features and respiratory motion edge enhancement features, fuses and generates respiratory motion features, and establishes a fusion optimization target through structure preservation constraints and edge continuity constraints: ; ; These are the respiratory movement characteristics after final fusion; It is a spatial gradient operator; The structural features maintain proportion; It is the edge continuity constraint strength; It refers to the fusion ratio.

6. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 5, characterized in that: The respiratory motion artifact suppression module specifically includes: Local grayscale change feature calculation: For the final fused respiratory motion features, the grayscale range is calculated within a local window. ; ; and These are the maximum and minimum eigenvalues ​​of the local window; calculate the local variation magnitude. ; Adaptive enhanced slope calculation, based on the local cumulative distribution function Constructing dynamic enhancement factors: ; It is the gradient of grayscale distribution; It is a dynamic enhancement factor; Dynamic truncation threshold calculation, generating the truncation range based on the local window size: ; It is the cutoff threshold; It refers to the size of a local window; It is a window adjustment weight; It is the characteristic distribution coefficient; Respiratory motion features were optimized by adjusting the fusion features based on a dynamic enhancement factor and a cutoff threshold. ; These are optimized respiratory movement characteristics; It is the average feature value of the local window; It is a truncation function.

7. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 1, characterized in that: The rehabilitation status parameter extraction module specifically includes: Based on the edge features of respiratory motion, the differences in chest and abdominal motion changes between consecutive video frames are calculated, and then the stability of respiratory rhythm is calculated. Based on human keypoints, the changes in the patient’s torso spatial posture during training are obtained, and stability is calculated based on the posture changes in consecutive frames. Based on respiratory movement characteristics, the differences in motor response between the chest and abdominal regions were calculated separately as indicators of motor coordination; these were used to form rehabilitation status characteristics and then normalized.

8. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 1, characterized in that: The motor state classification module takes the normalized rehabilitation state features as input to the classification model. By learning the distribution of respiratory movement features of patients at different rehabilitation stages, it realizes the classification of patients' rehabilitation motor state and adopts the XGBoost classification model. After the model is trained, the continuous motion image sequence of the patient during respiratory rehabilitation training is acquired in real time through the Internet of Things visual acquisition terminal. After obtaining standardized rehabilitation state features, the patient's rehabilitation motion state is classified using a classification model.

9. The IoT-based respiratory rehabilitation assessment and movement analysis system according to claim 1, characterized in that: After preprocessing, the image acquisition module calculates the average brightness value H of the three color channels of the current patient motion image. R H G and H B , serving as the pixel brightness benchmark for the red, green, and blue channels; constructing a neutral brightness benchmark for the current training environment: ; It is the neutral brightness benchmark for the current rehabilitation training environment.

Images