A computer vision-based plantar acupoint recognition method

Abstract

The application discloses a kind of based on computer vision's plantar acupoint identification method, it is related to computer vision technical field, including through high-density pressure distribution acquisition flat plate and optical integrated high-definition vision equipment;The application carries out data acquisition and space mapping by plantar image and pressure distribution data, provides the depth feature of biomechanics level based on the pressure distribution percentage, pressure center trajectory feature set extracted based on biomechanics theory, the mapping relationship between image coordinate system and pressure distribution flat plate coordinate system is established, the acupoint coordinates of visual positioning are accurately mapped to specific pressure sensing unit area, the pressure feature extraction of association is from the same physical area, the consistency of image and pressure data fusion result is realized, the image coordinates of acupoint and its corresponding area biomechanics feature are associated and fused, form the image coordinates of each acupoint and the static and dynamic pressure feature acupoint identification set of plantar region.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method for recognizing acupoints on the soles of the feet based on computer vision. Background Technology

[0002] Foot massage, as a traditional and popular health practice, works by stimulating specific reflex zones or acupoints on the soles of the feet to relieve fatigue, promote blood circulation, and regulate bodily functions. With increasing health awareness, the market demand for customized foot pads, insoles, and smart soles that provide personalized massage support is growing. Traditional Chinese medicine's foot reflexology theory holds that specific acupoints on the soles of the feet are related to the functions of various organs in the body. Stimulating the corresponding acupoints can play a regulatory and health-preserving role. Stimulating reflex zones on the feet related to the eyes to help relieve eye fatigue and intervene in the development of myopia is a popular auxiliary therapy. Based on this theory, various massage foot pads or insoles with fixed raised dot layouts have appeared on the market.

[0003] Existing methods have significant limitations:

[0004] Existing technologies typically capture foot images using cameras and then use image processing or machine learning algorithms to match pre-set acupoint maps to output acupoint coordinates. However, these technologies cannot reflect the human body's weight-bearing state when standing or walking. There is a lack of effective technological bridges to spatially and semantically map the pressure characteristics of a region to the precise acupoint coordinates corresponding to that region.

[0005] While visual scanning acquires the coordinate system of the foot surface and accurately maps and correlates it with the dynamic pressure distribution data collected by the foot pressure plate or sensor array, it cannot match the physiological hot spots that generate the maximum pressure and require fatigue relief when the user is actually walking or standing, thus reducing the accuracy of acupoint identification.

[0006] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0007] The purpose of this invention is to: obtain foot coordinate information and plantar biomechanical pressure data, map acupoint coordinates to biomechanical features, map to the plantar pressure distribution data space, and extract the static and dynamic biomechanical features of the corresponding area of ​​each acupoint, thereby generating a set of acupoint recognition results containing location information and mechanical state features.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a method for identifying acupoints on the sole of the foot based on computer vision, comprising the following steps:

[0009] Step 1: Using a high-density pressure distribution acquisition tablet and an optically integrated high-definition vision device, the user's plantar pressure distribution data and plantar images are acquired, and a set of target acupoint markers is received.

[0010] Step 2: Perform task-aware analysis on the plantar images, extract clarity features, foot posture features, and anatomical location complexity of target acupoints, and fuse them to generate an image comprehensive perception coefficient. Analyze the plantar pressure distribution data based on biomechanical theory to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. Evaluate the data quality to generate pressure data confidence. Based on the image comprehensive perception coefficient and the pressure data confidence, generate dynamic path decision factors.

[0011] Step 3: Based on the dynamic path decision factors, call the corresponding acupoint positioning model library to process the foot image and output the coordinate set of the overall target acupoints in the image coordinate system;

[0012] Step 4: Based on the spatial relationship between the high-density pressure distribution acquisition tablet and the high-definition vision device, spatially register the plantar images and plantar pressure distribution data, establish a coordinate system mapping relationship, and according to the coordinate system mapping relationship, map the coordinates of each acupoint in the overall target acupoint coordinate set to the corresponding area in the plantar pressure distribution data to generate a mapped acupoint corresponding area table.

[0013] Step 5: Based on the mapped acupoint corresponding region table, extract static and dynamic pressure feature values, and associate and fuse the image coordinates of each acupoint with the static and dynamic pressure feature values ​​of the corresponding region to generate an acupoint recognition set with acupoint coordinates and corresponding region biomechanical features.

[0014] Furthermore, the system collects plantar pressure distribution data and plantar images of the user, and receives a set of target acupoint identifiers. The specific process is as follows:

[0015] When a user stands on a high-density pressure distribution acquisition plate, the pressure values ​​of the user's soles and the contact surface between the plate are continuously collected at a set sampling frequency, and the image of the soles is captured to generate a set of sole pressure distribution data containing spatial location and time series.

[0016] The plantar pressure distribution dataset includes static pressure distribution data when the user is standing still, and dynamic pressure time series data continuously collected when the user is walking.

[0017] By using the established symptom-acupoint mapping knowledge base, the system maps and generates a corresponding set of target acupoint identifiers based on the symptom identifiers selected by the user.

[0018] Furthermore, image sharpness features, foot pose features, and anatomical location complexity of the target acupoints are extracted and fused to generate a comprehensive perceptual coefficient for the current recognition task. The specific process is as follows:

[0019] Image data of the foot area is acquired through a visual device and a prior anatomical knowledge base is received, where each entry represents a specific acupoint on the sole of the foot;

[0020] Parallel feature extraction is based on a multi-branch lightweight convolutional network, including image sharpness feature extraction branch, foot pose feature extraction branch and acupoint region feature extraction branch, to obtain sharpness feature vector, pose feature vector and region feature vector representing acupoint region;

[0021] The sharpness feature vector and pose feature vector are standardized, and the regional feature vector corresponding to the target acupoint is averaged by pooling to obtain a global average acupoint regional feature vector. The three are then concatenated into a comprehensive feature vector, which is mapped to the interval as the final comprehensive perception coefficient.

[0022] Furthermore, based on biomechanical theory, the plantar pressure distribution data is analyzed to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. The specific generation process is as follows:

[0023] Acquire plantar pressure distribution data to form a static plantar pressure distribution map and a dynamic plantar pressure sequence, and establish a coordinate system to divide the static plantar pressure distribution map into different functional areas, including the forefoot area and the hindfoot area;

[0024] Calculate the percentage of total pressure borne by the forefoot and hindfoot regions relative to the total pressure of a single foot, and calculate the average pressure per unit area within the pre-defined candidate key acupoints in the plantar image.

[0025] For each frame of pressure data in the dynamic sequence, calculate the plantar pressure center coordinates of the current frame, connect the pressure center coordinates of consecutive frames to obtain the pressure center trajectory;

[0026] The extracted percentage of forefoot pressure distribution, percentage of hindfoot pressure distribution, average pressure value of each region, and pressure center trajectory are structured to form a biomechanical feature set.

[0027] Furthermore, based on the image-integrated perception coefficient and the confidence level of the stress data, dynamic path decision factors are generated. The specific process is as follows:

[0028] Based on the biomechanical feature set, confidence analysis of the data is performed. The data completeness, pressure data variance, and physiological rationality of extracting biomechanical features are comprehensively analyzed and mapped into pressure data confidence.

[0029] The image comprehensive perception coefficient and pressure data confidence score are obtained, normalized, and then fused into a dynamic path decision factor.

[0030] Furthermore, based on the dynamic path decision factors, the corresponding acupoint location model library is invoked. The specific process is as follows:

[0031] Obtain the dynamic path decision factors for identification The classification mapping is determined, including the following three types of judgments:

[0032] like If so, the current task is determined to be a low-risk, high-determinism task, and the corresponding deterministic model category is the Fast Localization Network Library;

[0033] like If so, the current task is determined to be a medium-risk, medium-deterministic task, and the corresponding deterministic model category is heatmap generation network library;

[0034] If so, the current task is determined to be a high-risk, low-determinism task, and the corresponding deterministic model category is the Collaborative Reasoning Logic Network Library;

[0035] Based on the above three judgments, a structured preliminary deterministic call result is generated, and the set hierarchical threshold range is [ , ].

[0036] Furthermore, the plantar image is processed to output the coordinate set of the entire target acupoint in the image coordinate system. The specific process is as follows:

[0037] Obtain preliminary deterministic call results, identify the execution path, obtain the category identifier path of the current result, and execute the corresponding processing flow;

[0038] The fast localization network library path processing is coordinate regression. The preprocessed local foot image block is input into the loaded fast localization network library, and the coordinate vector is directly output as the two-dimensional coordinates of each acupoint in the target acupoint identifier set in a preset order. The coordinate vector is then parsed into the coordinate set of the entire target acupoint.

[0039] Input a local foot image patch into the loaded heatmap generation network library. For each acupoint in the target acupoint identifier set, output a high-resolution probability heatmap. The value of each pixel in the probability heatmap represents the probability that the pixel location is the center of the corresponding acupoint.

[0040] For each acupoint heatmap, the collaborative reasoning logic network library locates its global peak point. If there are multiple significant local peaks in the heatmap, the peak point with the highest heatmap response value and the strongest neighborhood spatial continuity is selected as the final location coordinate of the acupoint. The decision coordinates of all acupoints are summarized to form the coordinate set of the overall target acupoint.

[0041] Furthermore, after outputting the coordinate set of the overall target acupoints in the image coordinate system, a result verification step is also included:

[0042] For each identified acupoint, three indicators—peak intensity, regional concentration, and cross-scale stability—are extracted from its probability heatmap and weighted and fused to generate a single acupoint location reliability score.

[0043] Based on prior knowledge of foot anatomy, the mean deviation between the measured distance and the theoretical distance between all adjacent identified acupoints is calculated, and a topological consistency score of the acupoint group is generated by mapping through a negative exponential function.

[0044] The confidence score of a single acupoint location is weighted and fused with the topological homology score of acupoint group to obtain the comprehensive confidence score, and the recognition results are output and used for decision-making.

[0045] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0046] This computer vision-based foot acupoint recognition method acquires and spatially maps foot images and pressure distribution data. Based on a set of target acupoint identifiers, it dynamically identifies and adjusts acupoints. It provides biomechanical depth features by extracting pressure distribution percentage and pressure center trajectory features based on biomechanical theory. By establishing a mapping relationship between the image coordinate system and the pressure distribution plate coordinate system, it accurately maps the visually located acupoint coordinates to specific pressure sensing unit areas. The coordinate mapping generates the same acupoints identified by the same user under different times and conditions. The associated pressure features are extracted from the same physical area, achieving consistency in the fusion results of image and pressure data. Finally, it associates and fuses the image coordinates of acupoints with the biomechanical features of their corresponding areas, forming a set of acupoint recognition based on the image coordinates of each acupoint and the static and dynamic pressure features of the foot region. Attached Figure Description

[0047] Figure 1 A schematic diagram of the method flow structure of the present invention is shown. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1:

[0050] like Figure 1 As shown, a method for recognizing acupoints on the soles of the feet based on computer vision includes the following steps:

[0051] Step 1: Using a high-density pressure distribution acquisition tablet and an optically integrated high-definition vision device, the user's plantar pressure distribution data and plantar images are acquired, and a set of target acupoint markers is received.

[0052] Step 2: Perform task-aware analysis on the plantar images, extract clarity features, foot posture features, and anatomical location complexity of target acupoints, and fuse them to generate an image comprehensive perception coefficient. Analyze the plantar pressure distribution data based on biomechanical theory to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. Evaluate the data quality to generate pressure data confidence. Based on the image comprehensive perception coefficient and the pressure data confidence, generate dynamic path decision factors.

[0053] Step 3: Based on the dynamic path decision factors, call the corresponding acupoint positioning model library to process the foot image and output the coordinate set of the overall target acupoints in the image coordinate system;

[0054] Step 4: Based on the spatial relationship between the high-density pressure distribution acquisition tablet and the high-definition vision device, spatially register the plantar images and plantar pressure distribution data, establish a coordinate system mapping relationship, and according to the coordinate system mapping relationship, map the coordinates of each acupoint in the overall target acupoint coordinate set to the corresponding area in the plantar pressure distribution data to generate a mapped acupoint corresponding area table.

[0055] Step 5: Based on the mapped acupoint corresponding region table, extract static and dynamic pressure feature values, and associate and fuse the image coordinates of each acupoint with the static and dynamic pressure feature values ​​of the corresponding region to generate an acupoint recognition set with acupoint coordinates and corresponding region biomechanical features.

[0056] The process involves collecting data on plantar pressure distribution and images of the user's feet, and receiving a set of target acupoint identifiers. The specific steps are as follows:

[0057] When a user stands on a high-density pressure distribution acquisition plate, the pressure values ​​of the user's soles and the contact surface between the plate are continuously collected at a set sampling frequency, and the image of the soles is captured to generate a set of sole pressure distribution data containing spatial location and time series.

[0058] The plantar pressure distribution dataset includes static pressure distribution data when the user is standing still, and dynamic pressure time series data continuously collected when the user is walking.

[0059] By using the established symptom-acupoint mapping knowledge base, the system maps and generates a corresponding set of target acupoint identifiers based on the symptom identifiers selected by the user.

[0060] In this solution, after the user stands still, a high-definition camera captures a frontal view of the sole of the foot. Simultaneously, a pressure plate collects and calculates a stable static sole pressure distribution map, displaying the average pressure in each area. The user walks several steps on the plate at a normal pace, and the system continuously collects dynamic pressure sequences at a frequency, while simultaneously recording video images of the foot during the rolling process. Based on the user's selected adolescent myopia prevention mode, the system retrieves the corresponding target acupoint identifier set from the internal knowledge base. The static sole image, dynamic video frames, static pressure distribution map, dynamic sequence, and target acupoint identifier set are time-stamped and logically correlated to form a complete input data packet.

[0061] The image sharpness features, foot pose features, and anatomical location complexity of the target acupoint are extracted and fused to generate a comprehensive perceptual coefficient for the current recognition task. The specific process is as follows:

[0062] Image data of the foot area is acquired through a visual device and a prior anatomical knowledge base is received, where each entry represents a specific acupoint on the sole of the foot;

[0063] Parallel feature extraction is based on a multi-branch lightweight convolutional network, including image sharpness feature extraction branch, foot pose feature extraction branch and acupoint region feature extraction branch, to obtain sharpness feature vector, pose feature vector and region feature vector representing acupoint region;

[0064] The sharpness feature vector and pose feature vector are standardized, and the regional feature vector corresponding to the target acupoint is averaged by pooling to obtain a global average acupoint regional feature vector. The three are then concatenated into a comprehensive feature vector, which is mapped to the interval as the final comprehensive perception coefficient.

[0065] In this scheme, after acquiring the captured image of the sole of the foot, the sharpness is evaluated by the image gradient information, and a sharpness feature value is generated. The orientation standard is evaluated by analyzing the posture parameters of the foot contour, and a posture compliance score is generated. The inherent texture and boundary features of the anatomical region corresponding to the target acupoint are obtained by querying the prior anatomical knowledge base, and an anatomical complexity score is generated. The sharpness feature value and the score are fused to obtain the image comprehensive perception coefficient.

[0066] Based on biomechanical theory, plantar pressure distribution data was analyzed to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. The specific generation process is as follows:

[0067] Acquire plantar pressure distribution data to form a static plantar pressure distribution map and a dynamic plantar pressure sequence, and establish a coordinate system to divide the static plantar pressure distribution map into different functional areas, including the forefoot area and the hindfoot area;

[0068] Calculate the percentage of total pressure borne by the forefoot and hindfoot regions relative to the total pressure of a single foot, and calculate the average pressure per unit area within the pre-defined candidate key acupoints in the plantar image.

[0069] For each frame of pressure data in the dynamic sequence, calculate the plantar pressure center coordinates of the current frame, connect the pressure center coordinates of consecutive frames to obtain the pressure center trajectory;

[0070] The extracted percentage of forefoot pressure distribution, percentage of hindfoot pressure distribution, average pressure value of each region, and pressure center trajectory are structured to form a biomechanical feature set.

[0071] Based on the image-based integrated perception coefficient and the confidence level of the stress data, dynamic path decision factors are generated. The specific process is as follows:

[0072] Based on the biomechanical feature set, confidence analysis of the data is performed. The data completeness, pressure data variance, and physiological rationality of extracting biomechanical features are comprehensively analyzed and mapped into pressure data confidence.

[0073] The image comprehensive perception coefficient and pressure data confidence score are obtained, normalized, and then fused into a dynamic path decision factor.

[0074] In this scheme, after obtaining the static pressure distribution map, the foot is divided into zones, with the forefoot and hindfoot separated by the boundary line shown in the map. The calculated load percentages are: left forefoot 42%, hindfoot 58%; right forefoot 48%, hindfoot 52%. Within the preset acupoint candidate area, the average pressure value is calculated to be 18 N / cm². This data constitutes the static part of the biomechanical feature set. The relevant calculation process is as follows:

[0075] Let Q be the sum of all effective sensor pressure values ​​in the forefoot region, and Z be the sum of all effective sensor pressure values ​​on the sole of the foot. Then, the forefoot pressure percentage Y is: ;

[0076] Let S be the sum of the pressure values ​​from all valid sensors within the target area, and N be the number of valid sensors within the area. Then, the average pressure value of the area is P: ;

[0077] For dynamic pressure sequences, the system connects the center points of each frame to form a smooth trajectory that rolls from the outside of the heel to the inside during the user's walking process, thus forming the pressure center trajectory.

[0078] In the quality assessment, the integrity of the pressure map, the small variance of the static data, and the continuity of the dynamic trajectory are detected, and the ratio of the forefoot to the hindfoot is reasonable. Therefore, the confidence level of the pressure data is determined to be 0.95. The high score will be used together with the subsequent image perception coefficient to ensure that the most accurate acupoint coordinates are obtained under high-quality data.

[0079] The image comprehensive perception coefficient and the pressure data confidence level are obtained, and then weighted and fused to obtain the dynamic path decision factor.

[0080] In this solution, the corresponding acupoint location model library is called based on the dynamic path decision factors. The specific process is as follows:

[0081] Obtain the dynamic path decision factors for identification The classification mapping is determined, including the following three types of judgments:

[0082] like If so, the current task is determined to be a low-risk, high-determinism task, and the corresponding deterministic model category is the Fast Localization Network Library;

[0083] like If so, the current task is determined to be a medium-risk, medium-deterministic task, and the corresponding deterministic model category is heatmap generation network library;

[0084] If so, the current task is determined to be a high-risk, low-determinism task, and the corresponding deterministic model category is the Collaborative Reasoning Logic Network Library;

[0085] Based on the above three judgments, a structured preliminary deterministic call result is generated, and the set hierarchical threshold range is [ , ].

[0086] The foot image is processed to output the coordinate set of the entire target acupoint in the image coordinate system. The specific process is as follows:

[0087] Obtain preliminary deterministic call results, identify the execution path, obtain the category identifier path of the current result, and execute the corresponding processing flow;

[0088] The fast localization network library path processing is coordinate regression. The preprocessed local foot image block is input into the loaded fast localization network library, and the coordinate vector is directly output as the two-dimensional coordinates of each acupoint in the target acupoint identifier set in a preset order. The coordinate vector is then parsed into the coordinate set of the entire target acupoint.

[0089] Input a local foot image patch into the loaded heatmap generation network library. For each acupoint in the target acupoint identifier set, output a high-resolution probability heatmap. The value of each pixel in the probability heatmap represents the probability that the pixel location is the center of the corresponding acupoint.

[0090] For each acupoint heatmap, the collaborative reasoning logic network library locates its global peak point. If there are multiple significant local peaks in the heatmap, the peak point with the highest heatmap response value and the strongest neighborhood spatial continuity is selected as the final location coordinate of the acupoint. The decision coordinates of all acupoints are summarized to form the coordinate set of the overall target acupoint.

[0091] After outputting the coordinate set of the entire target acupoints in the image coordinate system, a result verification step is also included:

[0092] For each identified acupoint, three indicators—peak intensity, regional concentration, and cross-scale stability—are extracted from the probability heatmap and weighted and fused to generate a single acupoint location reliability score.

[0093] Based on prior knowledge of foot anatomy, the mean deviation between the measured distance and the theoretical distance between all adjacent identified acupoints is calculated, and a topological consistency score of the acupoint group is generated by mapping through a negative exponential function.

[0094] The confidence score of a single acupoint location is weighted and fused with the topological homology score of acupoint group to obtain the comprehensive confidence score, and the recognition results are output and used for decision-making.

[0095] In this scheme, the heatmap generation network library typically adopts an encoder-decoder structure. For each target acupoint, the network outputs a high-resolution probabilistic heatmap corresponding to the input image space. The value of each pixel on the heatmap represents the probability that the location is the center of the acupoint. The final acupoint coordinates are obtained by post-processing the heatmap to locate the global maximum point in the heatmap. The presence of multiple significant local peaks in the heatmap is detected by finding all points that satisfy the condition that the value is greater than the threshold and is the maximum value in the 3x3 neighborhood. If there are multiple candidate peaks, the candidate peak point with the highest heatmap response value is selected.

[0096] If the difference between the highest response values ​​is less than the preset tolerance, the continuity of the neighborhood space of each candidate peak point is further analyzed, and the average variance of the pixel values ​​in the neighborhood is calculated. The lower the value, the smoother and more continuous the peak area is. The peak point with the strongest continuity is selected, and the pixel position of the selected peak point is used as the final positioning coordinate of the acupoint. The decision coordinates of all acupoints are summarized to form the coordinate set of the overall target acupoint.

[0097] Specifically, after outputting the set of coordinates of the entire target acupoints in the image coordinate system, a result verification step is also included:

[0098] For each identified acupoint, three indicators—peak intensity, regional concentration, and cross-scale stability—are extracted from its probability heatmap and weighted and fused to generate a single acupoint location reliability score.

[0099] Based on prior knowledge of foot anatomy, the mean deviation between the measured distance and the theoretical distance between all adjacent identified acupoints is calculated, and a topological consistency score of the acupoint group is generated by mapping through a negative exponential function.

[0100] The confidence score of a single acupoint location is weighted and fused with the topological homology score of acupoint group to obtain the comprehensive confidence score, and the recognition results are output and used for decision-making.

[0101] In this solution, step five is implemented by using the coordinates of each target acupoint output in step four. and its corresponding high-resolution probability heatmap If generated by a collaborative reasoning logic network library and a heatmap generation network library, then the following evaluation is performed:

[0102] Peak response intensity : ;

[0103] Peak region spatial concentration: based on Taking a circular region Ω with radius r as the center, calculate the standard deviation of the heat map values ​​within the region. And calculate its ratio to the region mean:

[0104] ;

[0105] A value close to 1 indicates that the peak region is concentrated and clearly distributed;

[0106] In generation During the process, intermediate heatmaps generated at different upsampling stages are recorded. The peak coordinates of each intermediate heatmap are located, and the average Euclidean distance between the full-scale coordinates and the intermediate-scale peak coordinates is calculated. And mapped to a stability score:

[0107] ;

[0108] in, The attenuation coefficient;

[0109] Single acupoint location reliability score Generated by weighted fusion of the above three factors:

[0110] ;

[0111] Based on a standard foot anatomy model, any two acupoints in a predefined target acupoint set A are used. and Theoretical relative distance between ;

[0112] Based on all acupoint coordinates output in step four, calculate the measured Euclidean distance between any two acupoint coordinates. ;

[0113] Using the measured distance from the center of the heel to the tip of the second toe as the baseline length L of the individual foot, the relative distance deviation is calculated:

[0114] ;

[0115] Calculate the average deviation Δ for all acupoints and map it to a score. The scoring function is designed in a negative exponential form to be sensitive to larger deviations.

[0116] Overall confidence generation: Integrate single-site location confidence with topological consistency score, and then calculate the average single-site location confidence.

[0117] ;

[0118] Generate overall confidence score: ;

[0119] in, The single-point reliability score for the i-th acupoint;

[0120] The system does not use a fixed threshold, but rather a dynamic routing decision factor generated in step two. Dynamically adjust the qualified threshold ;

[0121] The rules are: ,in Based on the threshold, For adjustment coefficients, The larger the threshold, the higher the expected risk; the threshold T should be appropriately relaxed or tightened accordingly, and the strategy can be adjusted accordingly.

[0122] Recognition Result Output and Decision Making:

[0123] like ≥T: Determine that the recognition result is reliable. The system outputs the final set of acupoint coordinates and attaches as a credibility reference;

[0124] If <T: Determine that the result is unreliable. The system does not output coordinates, but triggers a high-confidence alarm and provides specific guiding instructions on the user interface, such as "If the foot positioning is blurred, please flatten the sole of the foot and re-acquire."

[0125] This method has successfully achieved high-precision adaptive recognition of plantar acupoints in complex shooting environments in practical applications. It can actively sense image clarity, foot posture, and anatomical complexity, and dynamically correct the recognition strategy based on local contour and texture features, effectively overcoming common problems such as hand shaking, uneven light, angular tilt, and local occlusion in user selfies, and significantly improving the recognition robustness under non-professional conditions;

[0126] In terms of safety guarantee, a dual mechanism of single acupoint position reliability evaluation and acupoint group topological relationship verification is integrated to ensure that each output coordinate is not only reliable in its own response but also conforms to the anatomical spatial layout of the human foot. By introducing a dynamic confidence threshold, the judgment standard can be adaptively adjusted according to the current task risk, fundamentally eliminating the output of incorrect coordinates and providing a reliable spatial guidance for subsequent massage and electrostimulation physical interventions;

[0127] The finally output result set can be directly used to guide the intelligent design of personalized insole and sole products. For the specific health care needs of improving myopia, it can ensure that the designed massage stimulation points are not only accurately positioned but also can be personalized according to the actual load characteristics of the user's sole, thereby improving the safety and comfort of the product in principle.

[0128] The setting of the interval and the size of the threshold are for the convenience of comparison. Regarding the size of the threshold, it depends on the amount of sample data and the number of base numbers set by those skilled in the art for each set of sample data; as long as it does not affect the proportional relationship between the parameters and the quantified values.

[0129] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation;

[0130] In the two embodiments provided in this application, it should be understood that the disclosed methods can be implemented in other ways; for example, the device embodiments described above are merely illustrative, and the division of modules is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed; another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or modules may be electrical, mechanical or other forms.

[0131] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for recognizing acupoints on the sole of the foot based on computer vision, characterized in that, Includes the following steps: Step 1: Using a high-density pressure distribution acquisition tablet and an optically integrated high-definition vision device, the user's plantar pressure distribution data and plantar images are acquired, and a set of target acupoint markers is received. Step 2: Perform task-aware analysis on the plantar images, extract clarity features, foot posture features, and anatomical location complexity of target acupoints, and fuse them to generate an image comprehensive perception coefficient. Analyze the plantar pressure distribution data based on biomechanical theory to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. Evaluate the data quality to generate pressure data confidence. Based on the image comprehensive perception coefficient and the pressure data confidence, generate dynamic path decision factors. Step 3: Based on the dynamic path decision factors, call the corresponding acupoint positioning model library to process the foot image and output the coordinate set of the overall target acupoints in the image coordinate system; Step 4: Based on the spatial relationship between the high-density pressure distribution acquisition tablet and the high-definition vision device, spatially register the plantar images and plantar pressure distribution data, establish a coordinate system mapping relationship, and according to the coordinate system mapping relationship, map the coordinates of each acupoint in the overall target acupoint coordinate set to the corresponding area in the plantar pressure distribution data to generate a mapped acupoint corresponding area table. Step 5: Based on the mapped acupoint corresponding region table, extract static and dynamic pressure feature values, and associate and fuse the image coordinates of each acupoint with the static and dynamic pressure feature values ​​of the corresponding region to generate an acupoint recognition set with acupoint coordinates and corresponding region biomechanical features.

2. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, The process involves collecting data on plantar pressure distribution and images of the user's feet, and receiving a set of target acupoint identifiers. The specific steps are as follows: When a user stands on a high-density pressure distribution acquisition plate, the pressure values ​​of the user's soles and the contact surface between the plate are continuously collected at a set sampling frequency, and the image of the soles is captured to generate a set of sole pressure distribution data containing spatial location and time series. The plantar pressure distribution dataset includes static pressure distribution data when the user is standing still, and dynamic pressure time series data continuously collected when the user is walking. By using the established symptom-acupoint mapping knowledge base, the system maps and generates a corresponding set of target acupoint identifiers based on the symptom identifiers selected by the user.

3. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, The image sharpness features, foot pose features, and anatomical location complexity of the target acupoint are extracted and fused to generate a comprehensive perceptual coefficient for the current recognition task. The specific process is as follows: Image data of the foot area is acquired through a visual device and a prior anatomical knowledge base is received, where each entry represents a specific acupoint on the sole of the foot; Parallel feature extraction is based on a multi-branch lightweight convolutional network, including image sharpness feature extraction branch, foot pose feature extraction branch and acupoint region feature extraction branch, to obtain sharpness feature vector, pose feature vector and region feature vector representing acupoint region; The sharpness feature vector and pose feature vector are standardized, and the regional feature vector corresponding to the target acupoint is averaged to obtain a global average acupoint regional feature vector. The three are then concatenated into a comprehensive feature vector, which is mapped to the interval as the final comprehensive perception coefficient.

4. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, Based on biomechanical theory, plantar pressure distribution data was analyzed to generate a biomechanical feature set including the percentage of pressure distribution in the forefoot and hindfoot, the average pressure value of the region, and the trajectory of the pressure center. The specific generation process is as follows: Acquire plantar pressure distribution data to form a static plantar pressure distribution map and a dynamic plantar pressure sequence, and establish a coordinate system to divide the static plantar pressure distribution map into different functional areas, including the forefoot area and the hindfoot area; Calculate the percentage of total pressure borne by the forefoot and hindfoot regions relative to the total pressure of a single foot, and calculate the average pressure per unit area within the pre-defined candidate key acupoints in the plantar image. For each frame of pressure data in the dynamic sequence, calculate the plantar pressure center coordinates of the current frame, connect the pressure center coordinates of consecutive frames to obtain the pressure center trajectory; The extracted percentage of forefoot pressure distribution, percentage of hindfoot pressure distribution, average pressure value of each region, and pressure center trajectory are structured to form a biomechanical feature set.

5. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, Based on the image-based integrated perception coefficient and the confidence level of the stress data, dynamic path decision factors are generated. The specific process is as follows: Based on the biomechanical feature set, confidence analysis of the data is performed. The data completeness, pressure data variance, and physiological rationality of extracting biomechanical features are comprehensively analyzed and mapped into pressure data confidence. The image comprehensive perception coefficient and pressure data confidence score are obtained, normalized, and then fused into a dynamic path decision factor.

6. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, Based on the dynamic path decision factors, the corresponding acupoint location model library is invoked. The specific process is as follows: Obtain the dynamic path decision factors for identification The classification mapping is determined, including the following three types of judgments: like If so, the current task is determined to be a low-risk, high-determinism task, and the corresponding deterministic model category is the Fast Localization Network Library; like If so, the current task is determined to be a medium-risk, medium-deterministic task, and the corresponding deterministic model category is heatmap generation network library; If so, the current task is determined to be a high-risk, low-determinism task, and the corresponding deterministic model category is the Collaborative Reasoning Logic Network Library; Based on the above three judgments, a structured preliminary deterministic call result is generated, and the set hierarchical threshold range is [ , ].

7. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, The foot image is processed to output the coordinate set of the entire target acupoint in the image coordinate system. The specific process is as follows: Obtain preliminary deterministic call results, identify the execution path, obtain the category identifier path of the current result, and execute the corresponding processing flow; The fast localization network library path processing is coordinate regression. The preprocessed local foot image block is input into the loaded fast localization network library, and the coordinate vector is directly output as the two-dimensional coordinates of each acupoint in the target acupoint identifier set in a preset order. The coordinate vector is then parsed into the coordinate set of the entire target acupoint. Input a local foot image patch into the loaded heatmap generation network library. For each acupoint in the target acupoint identifier set, output a high-resolution probability heatmap. The value of each pixel in the probability heatmap represents the probability that the pixel location is the center of the corresponding acupoint. For each acupoint heatmap, the collaborative reasoning logic network library locates its global peak point. If there are multiple significant local peaks in the heatmap, the peak point with the highest heatmap response value and the strongest neighborhood spatial continuity is selected as the final location coordinate of the acupoint. The decision coordinates of all acupoints are summarized to form the coordinate set of the overall target acupoint.

8. The method for identifying acupoints on the sole of the foot based on computer vision according to claim 1, characterized in that, After outputting the coordinate set of the entire target acupoints in the image coordinate system, a result verification step is also included: For each identified acupoint, three indicators—peak intensity, regional concentration, and cross-scale stability—are extracted from its probability heatmap and weighted and fused to generate a single acupoint location reliability score. Based on prior knowledge of foot anatomy, the mean deviation between the measured distance and the theoretical distance between all adjacent identified acupoints is calculated, and a topological consistency score of the acupoint group is generated by mapping through a negative exponential function. The confidence score of a single acupoint location is weighted and fused with the topological homology score of acupoint group to obtain the comprehensive confidence score, and the recognition results are output and used for decision-making.

Images