Acupuncture point registration method based on feature fusion of median line and hip edge

Abstract

The invention discloses an acupoint registration method based on feature fusion of a midline and a hip edge, and relates to the field of graphic image processing, and the method comprises the steps: S1, constructing a standard coordinate system matched with the acupoints of a human body; s2, acquiring a depth image based on a structured light camera to generate three-dimensional point cloud data of the back of the human body; s3, recognizing a human body median line from the three-dimensional point cloud data based on symmetric analysis; s4, identifying a hip edge from the three-dimensional point cloud data based on the detection gradient extreme value; s5, establishing a local human body coordinate system based on the human body median line and the hip edge; and S6, registering the local human body coordinate system with the standard coordinate system by using a non-linear registration algorithm, and mapping the registered coordinates into the world coordinate system of the user to finish the acupuncture point registration of the crowds with different body types. The invention provides an acupoint registration method based on feature fusion of a median line and a hip edge, which is used for reducing errors caused by body type differences and improving the accuracy of acupoint recognition at the same time.

Description

Technical Field

[0001] The present invention relates to the field of graphics and image processing, and more specifically, to an acupoint registration method based on the fusion of midline and buttocks edge features. Background Art

[0002] Acupoint identification is an important part of traditional Chinese medicine treatment. Currently, it mainly relies on manual experience and manual positioning, which has problems such as large errors, low efficiency, and poor repeatability. With the development of computer vision and deep learning technology, the research and development of acupoint identification systems has become an important direction in the field of intelligent Chinese medicine. However, due to individual differences in the human body, changes in posture, and complex lighting conditions, it is difficult to achieve high-precision and automatic acupoint positioning with existing technologies.

[0003] At present, acupoint positioning technology is mainly divided into two categories: The first is a recognition method based on two-dimensional images, which uses a camera or ordinary RGB images to collect human surface information and combines image processing algorithms (such as edge detection and skin color segmentation) to locate acupoints. The problem is that this method relies on lighting conditions and cannot process three-dimensional surfaces. It is easily affected by posture changes and has large positioning errors. For example, Wang Hao et al. pointed out in "Computational Intelligence and Neuroscience" that the positioning accuracy of acupoint recognition is low in complex scenes (such as uneven lighting, occlusion, and skin color interference) (DOI: PMC10659787).

[0004] The second is the fixed template matching method, which establishes a standard human 3D point cloud model and matches the patient's body surface data through rigid registration. The problem is that it matches the human point cloud through a predefined 3D template, but cannot adapt to individual body shape differences (such as spinal curvature and uneven muscle distribution), resulting in poor clinical adaptability. For example, when respiratory movement causes the Zhongwan acupoint point cloud displacement to be greater than 4mm, the system needs to be recalibrated; In addition, there are also patents that propose to solve the problems of existing acupoint recognition algorithms by using an acupoint algorithm that measures the length of a human body's horizontal fingers. However, it can only be applied to acupoint recognition on the back of the human body. For example, the acupoint recognition algorithm that measures the length of a human body's horizontal fingers is proposed in patent number CN116864079A. It lacks the ability to extract hand data and is easily affected by lighting changes, finger occlusion or posture tilt, which can easily lead to measurement errors, and it is unable to handle three-dimensional deformation problems.

[0005] Therefore, it can be seen from the above that the defects of the prior art are mainly: 1. Insufficient recognition accuracy: The two-dimensional method ignores depth information, and the three-dimensional template method cannot dynamically correct individual differences.

[0006] 2. Low processing efficiency: Traditional symmetry analysis requires traversing the entire graph, which has high computational complexity (O(n2)O(n2)) and is difficult to process in real time.

[0007] 3. Poor robustness: Gradient detection is prone to failure under noise interference, resulting in reference point offset. Summary of the invention

[0008] An object of the present invention is to solve at least the above problems and / or disadvantages and to provide at least the advantages which will be described hereinafter.

[0009] In order to achieve these purposes and other advantages of the present invention, a method for acupoint registration based on the fusion of midline and buttock edge features is provided, comprising: S1. Based on the analysis of the TCM anatomical database and clinical measured data, a standard coordinate system adapted to human acupuncture points is constructed; S2, acquiring a depth image based on a structured light camera to generate three-dimensional point cloud data of the back of the human body; S3, identifying the human body midline from the three-dimensional point cloud data obtained in S2 based on symmetry analysis; S4, identifying the edge of the buttocks from the three-dimensional point cloud data obtained in S2 based on double threshold second-order gradient detection; S5, establishing a local human body coordinate system based on the human body midline and hip edge obtained in S3 and S4; S6. Use a nonlinear registration algorithm to register the local human body coordinate system obtained in S5 with the standard coordinate system obtained in S1, and map the registered coordinates to the user's world coordinate system to complete the acupoint registration for people of different body shapes.

[0010] Preferably, in S1, the standard coordinate system is obtained by principal component analysis PCA and kernel density estimation KDE, and is characterized by the following formula: In the above formula, P i Indicates i The standard coordinates of the acupuncture points, n Represents the total number of acupoint location samples in the data, K (⋅) is the kernel function, which is used to represent the point to be evaluated. P With sample points P j The Euclidean distance between h Indicates the bandwidth parameter.

[0011] Preferably, in S2, the three-dimensional point cloud data is characterized by the following formula: P w = K -1R [t] P c In the above formula, P w is the world coordinate, K is the camera intrinsic parameter matrix, R [t] is the viewpoint change matrix, P c is a 3D point in the camera coordinate system.

[0012] Preferably, in S3, the process of identifying the midline of the human body through symmetry analysis is: S30, based on the global highest point rough positioning, directly search for the point set with the smallest depth value in the three-dimensional point cloud data P global ,by P global The included points are used as reference points in turn to obtain the initial position of the candidate midline from the occipital region of the skull x c ; S31, based on a predetermined sliding step length s Set the width to W The depth image is divided into k overlapping windows, and s should be smaller than the width of each window Δ w ; S32, based on S30 x c The local symmetry error for each window E ( y , x c ) to perform calculations; S33. From every window k Find the local minimum error position , to obtain the global midline position by weighted average x center , complete the global error integration and the screening of the human body midline.

[0013] Preferably, in S30, the point set P global It is characterized by the following formula: In the above formula, I ( x , y ) represents the pixel point in the depth map ( x , y ) is the depth value of the skull, Ω is the predefined range of the occipital region of the skull; In S32, the local symmetry error E ( y , x c ) is characterized by the following formula: In the above formula, δ is the boundary buffer parameter in the sliding window calculation; In S33, the global midline position x center It is characterized by the following formula: In the above formula, the weight W k For Window k The inverse of the minimum error.

[0014] Preferably, in S4, the process of obtaining the buttocks edge by double-threshold second-order gradient detection is: S40, calculating the depth gradient along the depth map longitudinally to identify the hip edge position by a second-order difference operator G y ( y ); S41, based on the dual threshold mechanism G y ( y ) is determined to be the buttocks edge.

[0015] Preferably, in S40, the G y ( y ) is obtained by the following formula: In the above formula, I line ( y ) is the depth value of the hip edge detection line, I line ( y -1) is the depth value of the previous line of the hip edge detection line, I line ( y +1) is the depth value of the next line of the hip edge detection line; In S41, the dual threshold mechanism is determined by the following formula G y ( y ) is the buttock edge; In the above formula, is the gradient threshold, is the deep mutation threshold,y edge For the hip edge, is the depth value of the next line of the hip edge detection line, The depth value of the hip edge detection line.

[0016] Preferably, in S5, the local human body coordinate system The global midline position x center The connection line is x Shaft, hip edge y edge The connection line is y Axis is constructed by: In the above formula, x、 y is the coordinate of the pixel.

[0017] Preferably, in S6, the process of acupoint registration includes: S61, normalize the standard coordinate system by the following formula to obtain the standard acupuncture point position : In the above formula, W is the width of the depth image, H is the height of the depth image, is the abscissa of the local coordinate system, is the ordinate of the local coordinate system, is the normalized horizontal axis, is the normalized vertical coordinate; S62, coordinate transformation and normalization processing are performed on the patient's back depth image acquired by the camera to obtain the patient's acupuncture point position by the following formula: : In the above formula, x center is the global midline position, y edge For the hip edge, The original coordinates of the patient's back depth image collected by the camera; S63, using thin plate spline TPS model to define the deformation field T ( P ), the standard acupoint position is completed by the following coordinate system transformation To the patient's acupuncture point location The mapping: In the above formula, γis the smoothing term weight, T (.) is the thin plate spline TPS transformation function, n is the total number of samples of acupoint locations in the data, is the second-order gradient operator, P is a point in the coordinate system; S64, mapping the registered acupoint coordinates back to the patient's original coordinate system through the following formula: : .

[0018] The present invention has at least the following beneficial effects: Firstly, the present invention solves the problem of inaccurate positioning of acupuncture points on the back of the human body and inability to effectively cope with body shape differences in the prior art by integrating the dual characteristic coordinate system of the midline and the edge of the buttocks.

[0019] Secondly, the present invention solves the problem of low data acquisition and processing efficiency in the prior art, which affects clinical application, by using image processing technologies such as horizontal symmetry error analysis and second-order gradient detection.

[0020] Thirdly, the present invention solves the problem in the prior art that gradient detection is prone to failure under noise interference, resulting in reference point offset, through a gradient-depth dual threshold detection mechanism.

[0021] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the overall process when the method of the present invention is adopted; Figure 2 A schematic diagram of three-dimensional image acquisition and coordinate conversion when the method of the present invention is used; Figure 3 A schematic diagram of the connection line of the global midline position obtained after the depth image of the back of the human body is recognized by the method of the present invention; Figure 4 The figure is a schematic diagram of the buttock edge connection line obtained after the depth image of the back of a human body is recognized by the method of the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0024] The present invention relates to an acupoint positioning system based on structured light three-dimensional depth imaging and reference point detection, comprising: a three-dimensional image acquisition module, and a terminal connected to the three-dimensional image acquisition module for communication, on which acupoint registration software for acupoint identification is mounted, wherein the acupoint registration software is provided with a midline and buttocks edge recognition module and an acupoint registration and mapping module, so as to provide an acupoint registration method based on the fusion of midline and buttocks edge features to complete high-precision acupoint identification. Furthermore, the acupoint registration method is to achieve high-precision acupoint identification through symmetry analysis, gradient extreme value detection and nonlinear registration algorithm, and is suitable for intelligent auxiliary equipment for traditional Chinese medicine treatments such as acupuncture and moxibustion.

[0025] Furthermore, the present invention proposes an acupoint registration method based on the fusion of midline and buttocks edge features to achieve high-precision acupoint identification, providing technical support for the standardization and intelligence of traditional Chinese medicine treatment. Specifically, the acupoint registration method of the present invention is an acupoint identification method based on the image recognition technology of "highest point" selection. By establishing a standard acupoint database, symmetry analysis and gradient extreme value detection are performed to determine the midline of the back and the buttocks edge of the human body, and then acupoints are determined using a nonlinear registration algorithm, such as Figure 1 As shown, the processing flow includes the following steps: Step 1: Construct a human standard acupuncture point database and standard coordinate system: Based on the existing available knowledge of traditional Chinese medicine, construct a human acupuncture point depth coordinate system and normalize it as a standard coordinate system.

[0026] Step 2: Acquire the image of the back of the human body through the structured light camera or depth camera in the image acquisition module to generate three-dimensional point cloud data of the human body.

[0027] Step 3: Perform symmetry analysis and gradient extreme value detection through the midline and hip edge recognition module to determine the midline of the human back and the hip edge, and then establish a local human coordinate system.

[0028] Step 4: Use the nonlinear registration algorithm in the acupoint registration and mapping module to align the local human coordinate system with the standard coordinate system, and then map the aligned coordinates to the user's world coordinates.

[0029] Furthermore, in step 1, the construction of the standard acupoint template is based on the TCM anatomical database and clinical measured data. Indicates i Coordinates of acupuncture points. Standard acupuncture point template The acupoint distribution was modeled by principal component analysis (PCA) and kernel density estimation (KDE) method in the following formula: in, P i Indicatesi The standard coordinates of the acupoints, n represents the total number of acupoint location samples in the data, K(⋅) represents the kernel function, and the point to be evaluated P and the sample point P j The Euclidean distance between them, h represents the bandwidth parameter.

[0030] Furthermore, in step 2, if Figure 2 As shown, the 3D image acquisition module uses a structured light camera to obtain depth information Z(x, y) and generate 3D point cloud data of the back. The 3D coordinate calculation formula is: P w =K -1 R [t] P c in, P w is the world coordinate, K is the camera intrinsic parameter matrix, R [t] is the viewpoint change matrix, P c is a 3D point in the camera coordinate system.

[0031] Furthermore, in this step, statistical outlier filtering (SOR) is used to remove abnormal points, and Gaussian filtering is used to smooth the point cloud surface to reduce the impact of noise on subsequent symmetry analysis.

[0032] Further, in step three, the midline and hip edge recognition module may include, in terms of working mode: a midline recognition submodule and a hip edge recognition and feature extraction submodule; The workflow of the midline identification submodule includes: 1. Directly search for the point set with the smallest depth value in the 3D point cloud data by roughly locating the global highest point P global (Occipital Region of the Skull): In the above formula, I ( x , y ) represents the pixel point in the depth map ( x , y ) is the depth value, Ω is the predefined range of the occipital region of the skull (here it is 20), P global The included points are used as reference points in turn, and the columns where the reference points are located are x c As the initial position of the candidate midline.

[0033] 2. Calculate the horizontal symmetry error in small batches using the sliding window to determine the midline position ①Window division: Set the depth map width W Divide into k Overlapping windows, each with a width of Δ w , the sliding step length is s (s<Δ w , avoiding gaps and covering the entire image), then k The column ranges for each window are: ② Calculation of symmetry error within the window: For each window k , the candidate midline position is calculated by the following formula x c The local symmetry error of in, x c is the initial position of the candidate midline, I ( x , y ) represents the pixel point in the depth map ( x , y ), δ is the boundary buffer parameter in the sliding window calculation. When calculating the symmetry error within the sliding window, if x c Being too close to the window edge may cause some symmetrical points ( x c - x , y )or( x c + x , y ) is outside the current window.

[0034] ③ Global error integration and median line screening: For each window k , find the local minimum error position As the optimal solution for this window: The weighted average of the optimal solutions of all windows is used as the global midline position x center : Among them, the weight W k For Window k The minimum inverse error of , ensuring that high confidence windows dominate the results.

[0035] The workflow of the hip edge recognition and feature extraction submodule includes: (1) The depth gradient is calculated along the longitudinal direction of the depth map, and the hip edge position is identified by the second-order difference operator.

[0036] Define the longitudinal gradient at the hip edge G y ( y )for: (2) The buttocks edge is positioned as follows: In the above formula, is the gradient threshold, screening significant mutations (here =10mm), is the deep mutation threshold (here =15mm), excluding minor fluctuations.

[0037] After the midline and hip edge recognition module completes the hip edge and midline recognition, the human body local coordinate system is constructed. The human body local coordinate system is based on the global midline position. x center The connecting line is the x-axis, with the hip edge y edge The connecting line is constructed as the y-axis, that is: In this step, the second-order gradient is more sensitive to depth mutations to reduce noise interference, and the dual threshold mechanism is used to ensure the robustness of hip edge positioning. Specifically, the local human coordinate system constructed in this step can effectively integrate the midline and the hip edge to reduce the error caused by body shape differences and improve the accuracy of acupoint recognition. At the same time, in the application of the midline and hip edge recognition module, the gradient-depth dual threshold detection is used to enhance the robustness of hip edge recognition and resist noise interference.

[0038] Furthermore, in step 4, the workflow of the acupoint registration and mapping module includes: (1) Normalize the standard acupoint template in the coordinate system ,Right now: Similarly, the patient's actual original coordinates The above changes are also made to obtain the patient's acupuncture point location , then: (2) Using the thin plate spline TPS model to define the deformation field T ( P ),Right now: ①Linear term: , describing global affine transformations (translation, rotation, scaling).

[0039] ② Nonlinear term: , describing the local deformation, P j For the control point (take the standard acupoint P i location, also called a sample point), P is a point in the coordinate system, φ (.) is the radial basis function, W j is the weight coefficient, m is the number of control points, r For point P With control point P j The Euclidean distance between .

[0040] ③Radial basis function: , minimizing bending energy and ensuring smooth deformation.

[0041] Through coordinate system transformation, the standard acupoint positions are Mapping to patient acupuncture point locations , will soon T ( P ) is substituted into the following formula to optimize the objective: In the above formula, Indicates the location of the control point (take the key acupoints of the standard template), γ Smooth term weight (here γ = 0.1), suppressing excessive deformation. Taking into account the local deformation of the human body, the TPS model can simultaneously describe the global linear transformation and local nonlinear deformation, adapt to the spinal curvature and muscle distribution differences, and achieve accurate alignment of the standard acupoint template with the patient's body. Normalized coordinates eliminate size differences and improve the generalization of registration.

[0042] (3) Map the registered acupoint coordinates back to the patient's original coordinate system, that is: To restore the scaled coordinates to the absolute position in the patient's actual coordinate system.

[0043] The acupoint registration and mapping module in this scheme eliminates size differences and improves cross-individual generalization capabilities by normalizing TPS registration.

[0044] Example 1: Acupuncture point identification on the back of adults The depth image is used by the structured light camera to input the depth image of W=512, H=1024 into the algorithm processing software of the terminal; The midline and buttocks edge recognition module of the present invention is used to detect the midline and buttocks edge of the back of the human body in the input image, and the following is obtained: Figure 3-Figure 4 shown x center =256, y edge =320; The acupoint registration and mapping module of the present invention is first normalized to convert the standard acupoint of Shenshu acupoint into P i =(256,200) is converted to (0.5,0.25).

[0045] The acupoint registration and mapping module of the present invention is used to perform TPS registration and output the registration result: TPS output =(0.51,0.24), the actual coordinates after inverse transformation are (256.5,324), completing the acupoint registration operation of the coordinates.

[0046] The traditional method (such as the existing acupoint recognition method based on deep learning) is compared with the method of the present invention from different indicators (such as processing time, algorithm complexity, average error, and maximum error) to prove the usefulness of the method of the present invention. The comparison results are shown in Table 1. It can be seen that the present invention has significant improvements in all indicators compared with the existing traditional methods.

[0047] Table 1 The above solution is only an illustration of a preferred embodiment, but is not limited thereto. When implementing the present invention, appropriate replacement and / or modification can be performed according to user needs.

[0048] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily realized. Therefore, without departing from the general concept defined by the claims and equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. An acupoint registration method based on the fusion of midline and buttock edge features, characterized in that: include: S1. Based on the analysis of the TCM anatomical database and clinical measured data, a standard coordinate system adapted to human acupuncture points is constructed; S2, acquiring a depth image based on a structured light camera to generate three-dimensional point cloud data of the back of the human body; S3, identifying the human body midline from the three-dimensional point cloud data obtained in S2 based on symmetry analysis; S4, identifying the edge of the buttocks from the three-dimensional point cloud data obtained in S2 based on double threshold second-order gradient detection; S5, establishing a local human body coordinate system based on the human body midline and hip edge obtained in S3 and S4; S6. Use a nonlinear registration algorithm to register the local human body coordinate system obtained in S5 with the standard coordinate system obtained in S1, and map the registered coordinates to the user's world coordinate system to complete the acupoint registration for people of different body shapes.

2. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S1, the standard coordinate system is obtained by principal component analysis PCA and kernel density estimation KDE, and is characterized by the following formula: In the above formula, P i Indicates i The standard coordinates of the acupuncture points, n Represents the total number of acupoint location samples in the data, K (⋅) is the kernel function, which is used to represent the point to be evaluated. P With sample points P j The Euclidean distance between h Indicates the bandwidth parameter.

3. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S2, the three-dimensional point cloud data is represented by the following formula: P w = K -1 R [t] P c In the above formula, P w is the world coordinate, K is the camera intrinsic parameter matrix, R [t] is the viewpoint change matrix, P c is a 3D point in the camera coordinate system.

4. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S3, the process of identifying the human body midline through symmetry analysis is as follows: S30, based on the global highest point rough positioning, directly search for the point set with the smallest depth value in the three-dimensional point cloud data P global ,by P global The included points are used as reference points in turn to obtain the initial position of the candidate midline from the occipital region of the skull x c ; S31, based on a predetermined sliding step length s Set the width to W The depth image is divided into k overlapping windows, and s should be smaller than the width of each window Δ w ; S32, based on S30 x c The local symmetry error for each window E ( y , x c ) to perform calculations; S33. From every window k Find the local minimum error position , to obtain the global midline position by weighted average x center , complete the global error integration and the screening of the human body midline.

5. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 4, characterized in that: In S30, the point set P global Characterized by the following formula: In the above formula, I ( x , y ) represents pixel points ( x , y ), A predefined range for the occipital region of the skull; In S32, the local symmetry error E ( y , x c ) is characterized by the following formula: In the above formula, δ is the boundary buffer parameter in the sliding window calculation; In S33, the global midline position x center Characterized by the following formula: In the above formula, the weight W k For Window k The inverse of the minimum error.

6. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S4, the process of obtaining the buttocks edge through double-threshold second-order gradient detection is: S40, calculating the depth gradient along the depth map longitudinally to identify the hip edge position by a second-order difference operator G y ( y ); S41, based on the dual threshold mechanism G y ( y ) is the hip edge y edge Make a judgment.

7. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 6, characterized in that: In S40, the G y ( y ) is obtained by the following formula: In the above formula, I line ( y ) is the depth value of the hip edge detection line, I line ( y -1) is the depth value of the previous line of the hip edge detection line, I line ( y +1) is the depth value of the next line of the hip edge detection line; In S41, the dual threshold mechanism is determined by the following formula G y ( y ) is the buttock edge; In the above formula, is the gradient threshold, is the deep mutation threshold, y edge For the hip edge, is the depth value of the next line of the hip edge detection line, The depth value of the hip edge detection line.

8. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S5, the local human body coordinate system The global midline position x center The connection line is x Shaft, hip edge y edge The connection line is y The axis is constructed by: In the above formula, x , y is the coordinate of the pixel point.

9. The acupoint registration method based on the fusion of midline and buttock edge features as claimed in claim 1, characterized in that: In S6, the acupoint registration process includes: S61, normalize the standard coordinate system by the following formula to obtain the standard acupuncture point position : In the above formula, W is the width of the depth image, H is the height of the depth image, is the abscissa of the local coordinate system, is the ordinate of the local coordinate system, is the normalized horizontal axis, is the normalized vertical coordinate; S62, coordinate transformation and normalization processing are performed on the patient's back depth image acquired by the camera to obtain the patient's acupuncture point position by the following formula: : In the above formula, x center is the global midline position, y edge For the hip edge, The original coordinates of the patient's back depth image collected by the camera; S63, using thin plate spline TPS model to define the deformation field T ( P ), the standard acupoint position is completed by the following coordinate system transformation To the patient's acupuncture point location The mapping: In the above formula, γ is the smoothing term weight, T (.) is the thin plate spline TPS transformation function, n is the total number of samples of acupoint locations in the data, is the second-order gradient operator, P is a point in the coordinate system; S64, mapping the registered acupoint coordinates back to the patient's original coordinate system through the following formula: : 。

Images