CN-121582550-B - Intelligent acupuncture point positioning method and system based on image recognition

CN121582550BCN 121582550 BCN121582550 BCN 121582550BCN-121582550-B

Abstract

The application provides an intelligent acupuncture point positioning method and system based on image recognition, and relates to the technical field of image processing. The method comprises the steps of collecting image data, depth point cloud data and posture data of a target part, aligning the image data, extracting anatomical feature points, constructing a feature map, correspondingly storing coordinates and feature vectors of the anatomical feature points, determining the sinking degree of each pixel point in the image data based on the depth point cloud data, screening candidate nodes to construct a candidate map, taking the feature vector of the anatomical feature point closest to the candidate node as the candidate feature vector, determining a matching template and a matching score of the candidate map based on the candidate feature vector and topological connection, determining stability weight of the candidate nodes based on contour differences of adjacent image data frames, calculating node confidence and determining a high-confidence candidate map based on the node confidence, and determining the point coordinates based on topological consistency of the high-confidence candidate map and a standard point topological map. The application can improve the accuracy of the acupoint positioning.

Inventors

WU JIANG
ZHANG XIN
GAO HANG

Assignees

辽宁得康药业集团有限公司

Dates

Publication Date: 20260512
Application Date: 20251128

Claims (9)

1. An intelligent acupuncture point positioning method based on image recognition is characterized by comprising the following steps: Collecting image data, depth point cloud data and attitude data of a target part, and performing space-time alignment to the same coordinate system; Extracting anatomical feature points from the image data and constructing a feature map, wherein the feature map correspondingly stores coordinates and feature vectors of each anatomical feature point, and the feature vectors comprise texture contrast, local dishing degree and detection confidence of the anatomical feature points; Determining the sinking degree of each pixel point in the image data based on the depth point cloud data to generate a sinking saliency map, screening initial candidate areas from the sinking saliency map, taking the pixel point with the largest sinking degree in each initial candidate area as a candidate node, and constructing a candidate map containing candidate feature vectors and topological connection, wherein the candidate feature vectors are the feature vectors of the anatomical feature points closest to the candidate node; Determining a matching template and a matching score of the candidate graph in a pre-built acupoint template library based on the candidate feature vector and the topological connection, determining stability weight of the candidate node based on contour difference of adjacent image data frames, determining node confidence degree based on the concave degree, the stability weight and the matching score, and determining a high-confidence candidate graph based on the node confidence degree; determining initial acupoint coordinates based on the topological consistency of the high-confidence candidate graph and a standard acupoint topological graph, and dynamically compensating the initial acupoint coordinates based on the gesture data to obtain final acupoint coordinates; The determining the stability weight of the candidate node based on the contour difference of the adjacent image data frames comprises the following steps: Extracting edge contours of different pixel areas in the adjacent image data frames by adopting an edge detection algorithm, and determining corresponding points between the edge contours of the pixel areas to form a corresponding point set; determining distances between the corresponding points in the corresponding point set, determining a displacement degree of the pixel region based on the maximum distance, and determining the stability weight of each candidate node in the pixel region based on a region type and the displacement degree.
2. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein the steps of collecting image data, depth point cloud data and posture data of a target part and performing space-time alignment include: Triggering an RGB camera to acquire the image data through a unified clock, acquiring the depth point cloud data through a depth sensor and acquiring the gesture data through an inertial measurement unit, and aligning the image data, the depth point cloud data and the gesture data to a unified sampling frequency based on a linear interpolation processing method; Calibrating internal parameters of the RGB camera and the depth sensor, calculating the relative pose between the RGB camera and the depth sensor, and converting the depth point cloud data into an image coordinate system of the image data based on the relative pose by taking the internal parameters as a reference; and establishing a right-hand coordinate system by taking a principal point of the image data as an origin, and mapping the image data, the depth point cloud data and the gesture data to the right-hand coordinate system.
3. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein the extracting anatomical feature points in the image data and constructing a feature map comprises: An edge detection algorithm is adopted to initially extract the edge contour of the target part in the image data, and morphological closing operation is adopted to optimize the edge contour; extracting the anatomic feature points from the optimized edge profile, and outputting two-dimensional coordinates of the anatomic feature points under an image coordinate system where the image data are located; determining a target peripheral window of the anatomical feature point, converting an image area corresponding to the target peripheral window into a local gray level image, and determining the texture contrast based on gray level values of all pixel points in the local gray level image; Extracting depth values from the corresponding depth point cloud data based on the two-dimensional coordinates of the anatomical feature points, and determining the local dishing degree based on the depth values corresponding to the anatomical feature points; Obtaining the detection confidence coefficient by carrying out edge continuity analysis on the optimized edge profile; and marking a vector formed by the texture contrast, the local dishing degree and the confidence as the feature vector of the anatomical feature point, and storing the anatomical feature point and the feature vector in a one-to-one correspondence manner to obtain the feature map.
4. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein determining the degree of concavity of each pixel point in the image data based on the depth point cloud data to generate a concavity saliency map comprises: extracting a depth value from the corresponding depth point cloud data based on two-dimensional coordinates of the pixel point in an image coordinate system for each pixel point in the image data; Obtaining target pixel points in the neighborhood of the current pixel point, calculating and averaging the difference value of the depth values of the current pixel point and each target pixel point, and recording the difference value as the recession degree of the current pixel point; And normalizing the concave degree of each pixel point in the image data to generate a corresponding concave saliency map.
5. The method for locating acupoints on intelligent acupuncture based on image recognition according to claim 4, wherein the step of screening initial candidate regions from the concave saliency map, using the pixel point with the largest concave degree in each initial candidate region as a candidate node, and constructing a candidate map comprising candidate feature vectors and topological connection comprises the steps of: acquiring the pixel points with the concave degree larger than a first preset threshold value to obtain an initial candidate pixel set; carrying out connected domain analysis on the initial candidate pixel set, and removing a fragment area to obtain the initial candidate area; and taking the pixel points in the initial candidate region as candidate nodes, constructing the candidate graph based on Euclidean distance between the candidate nodes by adopting k-nearest neighbor connection, and associating the feature vector of the anatomical feature point closest to the candidate nodes as the candidate feature vector.
6. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein the determining a matching template and a matching score of the candidate map in a pre-established acupuncture point template library based on the candidate feature vector and the topological connection comprises: determining a first candidate matching score of the degree of concavity of each candidate node in the candidate graph and the degree of concavity of a corresponding point in the posture standard point template and a second candidate matching score of the topological connection of the candidate node and the corresponding point for each posture standard point template in the pre-built point template library; Determining the candidate matching score of the candidate graph and the posture standard acupoint templates based on the first candidate matching score and the second candidate matching score of each candidate node, and determining the posture standard acupoint templates with the candidate matching score larger than a second preset threshold as candidate templates to obtain a candidate template set; if the highest candidate matching score is smaller than the second preset threshold, calculating the product of the local depression degree and the detection confidence degree of each candidate node in the candidate graph, and recording the product as a node optimization factor; removing the candidate node with the minimum node optimization factor from the candidate graph, and redefining the candidate matching score of the candidate graph with the candidate templates after the candidate node is removed; repeating the step of determining the candidate matching score of the candidate graph and each candidate template after removing the candidate node again if the highest candidate matching score is smaller than the second preset threshold value until the candidate matching score is not smaller than the second preset threshold value or the iteration number reaches the preset number; The candidate template with the highest candidate matching score is determined as the matching template, and the candidate matching score of the matching template is determined as the matching score.
7. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein the determining a node confidence based on the dishing degree, the stability weight, and the matching score, and determining a high confidence candidate map based on the node confidence, comprises: Calculating the product of the concave degree of the candidate node and the normalized value of the matching score based on a preset weight proportion to obtain a first confidence factor of the candidate node; Marking the stability weight of the candidate node as a second confidence factor, and calculating the product of the first confidence factor and the second confidence factor to obtain the node confidence; and determining the candidate nodes with the node confidence coefficient larger than a third preset threshold value as high-confidence nodes, and constructing the high-confidence candidate graphs corresponding to the high-confidence nodes.
8. The intelligent acupuncture point positioning method based on image recognition according to claim 1, wherein the determining initial acupuncture point coordinates based on the topological consistency of the high confidence candidate map and a standard acupuncture point topological map comprises: Obtaining the standard acupoint topological graph of the target part, and constructing a candidate anchor point topological graph of the high-confidence candidate graph; Acquiring actual distances of all connecting edges in the candidate anchor point topological graph, and determining a first topological factor based on the actual distances and standard distances of corresponding edges in the standard acupuncture point topological graph; Acquiring an actual included angle between connecting edges in the candidate anchor point topological graph, and determining a second topological factor based on the actual included angle and a corresponding standard included angle in the standard acupuncture point topological graph; determining the topological consistency of the candidate anchor point topological graph and the standard acupuncture point topological graph based on the first topological factor and the second topological factor; If the topological consistency is not smaller than a fourth preset threshold, determining the candidate node corresponding to the highest node confidence in the high-confidence candidate graph as the initial acupoint coordinate; If the topological consistency is smaller than the fourth preset threshold, correcting coordinates of other candidate nodes in the high-confidence candidate graph based on the relative distance and direction between the point corresponding to the standard point topological graph and other point points by taking the candidate node corresponding to the highest node confidence in the high-confidence candidate graph as a reference, and recalculating the topological consistency until the topological consistency is not smaller than the fourth preset threshold; and determining comprehensive confidence coefficient based on the topological consistency and the node confidence coefficient of each candidate node after coordinate correction, and determining the candidate node corresponding to the highest confidence coefficient as the initial acupoint coordinate.
9. An intelligent acupuncture point positioning system based on image recognition, which is characterized by comprising: the data acquisition module is used for acquiring image data, depth point cloud data and attitude data of the target part and performing space-time alignment to the same coordinate system; The image processing module is used for extracting anatomical feature points from the image data and constructing a feature map, wherein the feature map correspondingly stores coordinates of the anatomical feature points and feature vectors, and the feature vectors comprise texture contrast, local dishing degree and detection confidence of the anatomical feature points; The image processing module is further configured to determine a dishing degree of each pixel point in the image data based on the depth point cloud data, so as to generate a dishing saliency map, screen an initial candidate region from the dishing saliency map, and construct a candidate map including candidate feature vectors and topological connection by using the pixel point with the largest dishing degree in each initial candidate region as a candidate node, where the candidate feature vectors are feature vectors of the anatomical feature points closest to the candidate node; The image processing module is further configured to determine a matching template and a matching score of the candidate graph in a pre-built acupoint template library based on the candidate feature vector and the topological connection, determine a stability weight of the candidate node based on a contour difference of adjacent image data frames, determine a node confidence level based on the sag degree, the stability weight and the matching score, and determine a high confidence candidate graph based on the node confidence level; The acupoint positioning module is used for determining initial acupoint coordinates based on the topological consistency of the high-confidence candidate graph and the standard acupoint topological graph, and dynamically compensating the initial acupoint coordinates based on the gesture data to obtain final acupoint coordinates; The determining the stability weight of the candidate node based on the contour difference of the adjacent image data frames comprises the following steps: Extracting edge contours of different pixel areas in the adjacent image data frames by adopting an edge detection algorithm, and determining corresponding points between the edge contours of the pixel areas to form a corresponding point set; determining distances between the corresponding points in the corresponding point set, determining a displacement degree of the pixel region based on the maximum distance, and determining the stability weight of each candidate node in the pixel region based on a region type and the displacement degree.

Description

Intelligent acupuncture point positioning method and system based on image recognition Technical Field The application relates to the technical field of image processing, in particular to an intelligent acupuncture point positioning method and system based on image recognition. Background The acupuncture point positioning is a core technical link in the traditional Chinese medicine acupuncture diagnosis and treatment system, and the core aim is to position the acupuncture points for clinical treatment according to the classical positioning principles of the traditional Chinese medicine meridian theory and the human body surface anatomical mark, combining the bone degree division method, the finger-inch positioning method and the like. The positioning accuracy not only determines the air effect, stimulation targeting and clinical treatment efficiency of the needling operation, but also has key effects on avoiding operation risks and preventing and controlling puncture complications, and is a basic premise for promoting the development of acupuncture treatment from experience to standardization, individuation and intelligence. In the related art, the acupuncture point positioning auxiliary technology mostly uses static identification of a single visible light image as a core implementation path. The method is specifically realized by collecting two-dimensional images of local areas of a human body, extracting visual features such as skin textures, contour edges and the like in the images, and then matching with pre-stored acupoint image templates to further infer the approximate positions of acupoints. However, the method has the following problems that static identification of a single image is easily interfered by external environment and human physiological characteristics, so that the accuracy of visual characteristic extraction is reduced, positioning deviation is caused, the image identification range is limited to a local area of a human body, the cooperative positioning of multiple points of the whole body cannot be realized, verification links of physiological anatomical characteristics of the points are lacking, the clinical requirements of high accuracy and high reliability are difficult to meet, the positioning under a static scene can be realized only, the dynamic adaptability is not realized, and when a patient has slight limb movement or body position adjustment in the diagnosis and treatment process, the positioning result is easy to fail, and the real-time diagnosis and treatment requirements cannot be met. Disclosure of Invention In order to solve the problems of the related technology that the application requirements of clinical acupuncture point positioning are difficult to be met due to the dependence on single-mode data, lack of physiological verification and dynamic adaptability, insufficient fusion of individual differences and dynamic body position factors, the application provides an intelligent acupuncture point positioning method based on image recognition, which adopts the following technical scheme: Collecting image data, depth point cloud data and attitude data of a target part, and performing space-time alignment to the same coordinate system; Extracting anatomical feature points from the image data and constructing a feature map, wherein the feature map correspondingly stores coordinates and feature vectors of each anatomical feature point, and the feature vectors comprise texture contrast, local dishing degree and detection confidence of the anatomical feature points; Determining the sinking degree of each pixel point in the image data based on the depth point cloud data to generate a sinking saliency map, screening initial candidate areas from the sinking saliency map, taking the pixel point with the largest sinking degree in each initial candidate area as a candidate node, and constructing a candidate map containing candidate feature vectors and topological connection, wherein the candidate feature vectors are the feature vectors of the anatomical feature points closest to the candidate node; Determining a matching template and a matching score of the candidate graph in a pre-built acupoint template library based on the candidate feature vector and the topological connection, determining stability weight of the candidate node based on contour difference of adjacent image data frames, determining node confidence degree based on the concave degree, the stability weight and the matching score, and determining a high-confidence candidate graph based on the node confidence degree; And determining initial acupoint coordinates based on the topological consistency of the high-confidence candidate graph and the standard acupoint topological graph, and dynamically compensating the initial acupoint coordinates based on the gesture data to obtain final acupoint coordinates. The method comprises the steps of acquiring image data, depth point cloud data and gesture data of a target part