Automatic acupoint projection method, system and device and medium

Abstract

The invention relates to an automatic acupoint projection method, system and device and a medium, and belongs to the technical field of medical image processing. The projection method comprises the steps that a first front supine image of a current physiotherapy user is collected and preprocessed; inputting the preprocessed first front supine image into a key point detection model, and outputting an acupoint recognition coordinate set; generating a projection instruction according to the acupoint recognition coordinate set, and controlling a projector to generate initial projection light spots for marking the acupoints on the body of the current physiotherapy user to obtain a corresponding projection coordinate set; performing spatial registration on the projection coordinate set and the acupoint recognition coordinate set, calculating translation and rotation parameters of a plurality of matching point pairs, and performing primary correction on the projection coordinate set; collecting a second front supine image again; and detecting the actual coordinate of the projection light spot after primary correction in the second front supine image, and performing error compensation correction in combination with the acupoint identification coordinate set. According to the invention, the accuracy and clinical operability of automatic acupoint positioning can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to an automatic acupoint projection method, system, device and medium. BACKGROUND

[0002] With the increasing number of patients with chronic diseases and old-age diseases year by year, the demand for non-drug intervention therapy is increasing, and traditional Chinese medicine acupuncture is widely valued in the fields of chronic disease rehabilitation and sub-health conditioning due to its natural, green and effective advantages. However, in the traditional acupuncture operation, the accurate positioning of acupoints mainly depends on the experience and skills of the operator, and there are problems of positioning deviation, poor repeatability and strong human subjectivity, especially for beginners or operators with poor skills, it is difficult to stably realize the accurate identification and repeated needle insertion of standard acupoint positions. In addition, due to physiological factors such as posture change and body size difference, the surface markers of acupoints are not obvious, which further increases the difficulty of manual acupoint identification.

[0003] At the same time, with the continuous development of image recognition and deep learning algorithms, medical image automatic recognition has achieved remarkable results in many clinical scenarios such as skin disease recognition, tumor auxiliary diagnosis and bone positioning. Human posture estimation and key point detection models are becoming mature, which can efficiently extract main anatomical feature points of the body, providing the possibility for image-based acupoint recognition. However, as a functional point without clear anatomical structure markers, the recognition task of acupoints not only faces the general challenges of small target positioning and fuzzy representation features, but also has the special problem that the recognition results are difficult to directly convert into physical therapy operation execution parameters (such as projection position, acupoint selection path, etc.).

[0004] At present, in the existing research or products, most of the acupoint recognition technologies still stay at the image level output, that is, the acupoint coordinates are marked in the two-dimensional image, which cannot realize the mapping between the image and the real three-dimensional space, and also lacks a control mechanism to directly convert the recognition results into physical therapy projection or operation instructions. This makes there is a "discontinuity" between image recognition and actual application, which limits the clinical operability of the recognition results and the practicability of the system. At the same time, in the actual projection auxiliary equipment, the problem of how to ensure the spatial accuracy between the projection light point and the acupoint recognition point has not been effectively solved. SUMMARY

[0005] In order to improve the accuracy and clinical operability of automatic acupoint positioning, the present application provides an automatic acupoint projection method, system, device and medium.

[0006] In a first aspect, the present application provides an automatic acupoint projection method, which adopts the following technical scheme: An automatic acupoint projection method, the projection method comprising: acquire a first front supine image of a current physiotherapy user and pre-process the first front supine image to obtain a pre-processed first front supine image; input the pre-processed first front supine image into a pre-trained key point detection model to output a set of acupoint recognition coordinates; generate a projection instruction based on the set of acupoint recognition coordinates, control a projector to generate an initial projection light point highlighting an acupoint on the current physiotherapy user, and obtain a corresponding set of projection coordinates based on a projector coordinate system; perform spatial registration on the set of projection coordinates and the set of acupoint recognition coordinates, calculate translation and rotation parameters of a plurality of matching point pairs, and perform a first correction on the set of projection coordinates to output a first corrected projection light point; acquire a second front supine image of the current physiotherapy user, the second front supine image containing the first corrected projection light point; detect actual coordinates of the first corrected projection light point in the second front supine image, and perform error compensation correction in combination with the set of acupoint recognition coordinates.

[0007] By adopting the technical solution, a closed-loop control system is constructed based on deep learning to realize acupoint image recognition, mapping and registration between an image and a physical coordinate space, control and real-time feedback correction of optical projection, and to have high recognition accuracy and high physical projection precision. The technical solution can greatly reduce human intervention, improve acupoint positioning efficiency and consistency, effectively reduce acupuncture errors, and lay a solid foundation for the digitalization and intelligentization of traditional Chinese medicine physiotherapy.

[0008] Optionally, the step of acquiring a first front supine image of a current physiotherapy user and pre-processing the first front supine image includes: acquire a first front supine image of a current physiotherapy user using an industrial camera, the industrial camera being arranged above a physiotherapy bed; and perform pixel value normalization processing on the first front supine image; perform histogram equalization processing on the first front supine image after pixel value normalization to obtain a pre-processed first front supine image.

[0009] By adopting the technical solution, a high-performance industrial camera is used to accurately acquire a user's body surface image, and two classical and efficient image processing techniques, image normalization and histogram equalization, are combined to greatly improve the standardization degree and effective information density of the input image, providing a high-quality input basis for the subsequent acupoint recognition model, effectively suppressing the influence of external environmental factors or individual differences of the human body, and thus improving the accuracy and stability of acupoint recognition as a whole.

[0010] Optionally, before the step of inputting the pre-processed first front supine image into the pre-trained key point detection model, the method further includes: Based on a pre-built human target detection model, the current physiotherapy user located on the physiotherapy bed is subjected to human posture analysis; it is determined whether the current physiotherapy user's trunk axis is offset from the central axis of the physiotherapy bed by an angle less than a preset threshold and whether the limb endpoints are located within a preset area; If so, then confirm that the user is in the correct position and continue with the step of inputting the preprocessed first frontal supine image into the pre-trained key point detection model; If not, it is determined that the user is incorrectly positioned, and a positioning guidance instruction is output to the voice prompt module. After detecting that the user is correctly positioned, the first frontal supine image of the current physiotherapy user is re-acquired.

[0011] By adopting the above technical solutions, a posture detection and positioning judgment mechanism is introduced into the human-computer interaction process, and a pre-position evaluation system oriented towards the usage process is established, realizing a complete logical chain of posture standardization, image input standardization, and model recognition accuracy improvement.

[0012] Optionally, the steps of spatially registering the projection coordinate set with the acupoint identification coordinate set, calculating the translation and rotation parameters of multiple matching point pairs, performing initial correction on the projection coordinate set, and outputting the initially corrected projection light point include: Spatial registration is performed between the projected coordinate set and the acupoint identification coordinate set to obtain a mapping relationship set containing multiple matching point pairs; wherein each matching point pair includes an acupoint identification coordinate (x...). D ,y D ,θ D ) and a corresponding projection coordinate (x) d ,y d ,θ d ); Calculate translation and rotation parameters based on the matching point pairs; Based on the translation and rotation parameters, calculate the i-th unregistered projection coordinate (x) i ,y i ,θ i Correction amount According to the correction amount For the i-th projection coordinate (x) i ,y i ,θ i Perform initial calibration and output the projected light spot after initial calibration.

[0013] By adopting the technical solutions, the point-to-point mapping relationship between the recognition result and the projection instruction is constructed, the unified conversion of the coordinate space is realized in combination with the rigid body transformation model, and the error feedback mechanism is further introduced into the projection control, so that the preliminary but accurate light point correction is realized. The linkage mechanism from recognition to correction constitutes the "vision-space-control" closed loop which is crucial in the application, and not only guarantees the high precision and high consistency of the acupoint positioning, but also lays a solid foundation for subsequent more complex dynamic compensation and feedback control.

[0014] Optionally, the calculation formula of the translation and rotation parameters is: (Δx,Δy,Δθ) T =(x d ,y d ,θ d ) T -(x D ,y D ,θ D ) T ; In the above formula, T represents vector transposition, Δx and Δy represent the translation deviation between two coordinates, and Δθ represents the rotation angle deviation between two coordinates.

[0015] Optionally, the calculation formula of the correction amount is: In the above formula, represents a translation compensation term, represents a rotation matrix, and represents a deviation vector.

[0016] Optionally, the step of detecting the actual coordinates of the primary corrected projection light point in the second front supine image and performing error compensation correction in combination with the acupoint recognition coordinate set comprises: detecting the actual coordinates of the primary corrected projection light point in the second front supine image based on a pre-constructed light point target detection network; performing error value calculation on the actual coordinates of the primary corrected projection light point based on the acupoint recognition coordinate set; when the error value is greater than a preset error threshold, performing again compensation correction on the primary corrected projection light point based on the error value.

[0017] By adopting the technical solutions, the target detection network of deep learning is combined with the geometric error analysis technology to establish a closed-loop feedback control system with "perception-comparison-adjustment" as the core, and a multi-stage light point correction mechanism from the first rough registration to the second fine compensation is realized. By detecting the error between the projection result and the recognition result and making compensatory adjustment according to the quantitative index, the coincidence accuracy of the projected light points and the acupoints is significantly improved, and the whole system has the ability of adaptive correction, effectively overcoming the error accumulation caused by complex variables such as individual differences of human body, posture deviation or environmental light changes. This dynamic self-correcting mechanism is a key technical guarantee for realizing clinical-level accurate acupoint positioning and projection, and is an important embodiment of the application from "static projection" to "intelligent tracking correction".

[0018] In a second aspect, the application provides an automatic acupoint projection system, which adopts the following technical solutions: An automatic acupoint projection system, the projection system comprising a host computer, an image acquisition device and a projector; The image acquisition device is configured to acquire a front supine image of a current physiotherapy user; The projector is configured to receive a projection instruction and generate a projection light point for highlighting an acupoint on the current physiotherapy user; The host computer is configured to: send a first image acquisition instruction to the image acquisition device, receive a first front supine image of the current physiotherapy user sent by the image acquisition device and pre-process the first front supine image to obtain a pre-processed first front supine image; input the pre-processed first front supine image into a pre-trained key point detection model to output a set of acupoint recognition coordinates; generate a projection instruction based on the set of acupoint recognition coordinates and send the projection instruction to the projector to control the projector to generate an initial projection light point for highlighting an acupoint on the current physiotherapy user, and obtain a corresponding set of projection coordinates based on a projector coordinate system; perform spatial registration on the set of projection coordinates and the set of acupoint recognition coordinates, calculate translation and rotation parameters of a plurality of matching point pairs, and perform a primary correction on the set of projection coordinates to output a primary corrected projection light point; send a second image acquisition instruction to the image acquisition device, receive a second front supine image of the current physiotherapy user sent by the image acquisition device, and the second front supine image contains the primary corrected projection light point; detect the actual coordinates of the primary corrected projection light point in the second front supine image, and perform error compensation correction in combination with the set of acupoint recognition coordinates.

[0019] In a third aspect, the application provides a computer device, which adopts the following technical solutions: A computer device comprises a memory, a processor and a computer program stored on the memory, the processor executing the computer program to implement the steps of the method of the first aspect.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium stores a computer program capable of being loaded and executed by a processor to implement any one of the methods of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a first flowchart of an automatic acupoint projection method according to an embodiment of the present application.

[0022] Figure 2 is a second flowchart of an automatic acupoint projection method according to an embodiment of the present application.

[0023] Figure 3 is a third flowchart of an automatic acupoint projection method according to an embodiment of the present application.

[0024] Figure 4 is a fourth flowchart of an automatic acupoint projection method according to an embodiment of the present application.

[0025] Figure 5 is a fifth flowchart of an automatic acupoint projection method according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application. Figures 1-5 In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0027] The embodiments of the present application disclose an automatic acupoint projection method.

[0028] With reference to Figure 1 An automatic acupoint projection method, the projection method comprising: Step S101: acquiring a first front supine image of a current physiotherapy user and pre-processing to obtain a pre-processed first front supine image; The front supine image of the user is acquired by an image acquisition device (such as an industrial camera), and the camera device is erected above the physiotherapy bed at a fixed pitch angle to ensure that the field of view covers the trunk region of the human body.

[0029] First, a front image of the user in a supine position is acquired, which serves as the input basis for subsequent acupoint recognition. Due to the complex lighting conditions on the surface of the human body skin, weak texture features, and large differences in skin color, posture, and shooting environment of different users, the original image often contains a large amount of redundant information or even noise. Therefore, a series of preprocessing operations need to be performed on the image.

[0030] Specifically, the preprocessing operations can include image denoising (such as using Gaussian filtering or median filtering to eliminate salt and pepper noise), image enhancement (such as contrast stretching and edge sharpening), standardization (such as gray scale normalization), and posture correction, etc. The core logic of this process is to eliminate as much as possible the interference of non-acupoint features, improve the recognition efficiency and accuracy of the subsequent key point detection model, and especially in the background of blurred human skin features, enhance the model's perception ability to the target area.

[0031] Step S102, input the preprocessed first front supine image into the pre-trained key point detection model, and output an acupoint recognition coordinate set; The key point detection model is usually based on a convolutional neural network (CNN) architecture, such as OpenPose, HRNet, or a more lightweight MobileNet-Heatmap variant. The model has been pre-trained through a large number of labeled data sets to achieve automatic recognition of acupoint regions on the human body surface. The task of this key point detection model is to extract specific spatial points from the input image, which correspond to the acupoint positions in this technical solution. Since acupoints mostly lack obvious texture and shape features and are extremely small, the model uses a heatmap regression mechanism to generate a response map for each acupoint, and determines its two-dimensional image coordinates by finding the maximum response point. The acupoint recognition coordinate set is a set of points in the image coordinate system that represent the centers of acupoints, containing precise positioning information in the form of (x, y).

[0032] It can be understood that the advantage of this model lies in its "end-to-end" characteristics, which do not require human-designed feature extraction rules, but rather learn to recognize and distinguish acupoints and surrounding non-target areas independently through the network, with strong robustness and generalization ability.

[0033] Step S103, generate a projection instruction based on the acupoint recognition coordinate set to control the projector to generate an initial projection light point that highlights the acupoint on the current physiotherapy user, and obtain a corresponding projection coordinate set based on the projector coordinate system; Specifically, the acupoint recognition points in the image coordinate system are converted into projection instructions in the physical space. First, a mapping relationship between the image space and the projector control space needs to be established, and geometric modeling is usually performed using internal calibration parameters (such as the focal length, principal point, and distortion coefficient of the projector) and external parameters (the positional relationship between the projector and the camera). The generation process of the projection instructions is generally based on the perspective projection transformation matrix (Homography) or the camera-projector joint calibration (Camera-Projector Calibration) method, which converts the image space coordinates into projector control coordinates through inverse transformation, and then guides the projector to generate corresponding visible light spots on the surface of the user's body. The projection coordinate set output by this process is the physical projection point coordinates corresponding to each acupoint position in the current device coordinate system, which is used for subsequent spatial registration and dynamic correction.

[0034] In step S104, the projection coordinate set is spatially registered with the acupoint recognition coordinate set, the translation and rotation parameters of a plurality of matching point pairs are calculated, and the projection coordinate set is initially corrected, and the initially corrected projection light points are output. In this step, the geometric transformation relationship between the recognition coordinate set (image space) and the projection coordinate set (projection space) needs to be calculated to correspond the recognized acupoint positions to the actual projection positions one by one.

[0035] Specifically, the selection of the matching point pairs is usually based on the preset acupoint identification points or the local feature response maximum points, and the least squares method, rigid body transformation model, or affine transformation is used for registration solution. By solving the translation vector (Translation Vector) and rotation angle (Rotation Matrix) between the two-dimensional point sets, the projection coordinates are adjusted to ensure that the light points projected by the projector are closer to the true acupoint positions.

[0036] It can be understood that the initially corrected light points are the first correction result based on rough registration, and this result can achieve a high degree of fit, but due to individual posture, body size, or device error, etc., error compensation is still needed to improve the accuracy.

[0037] In step S105, a second front supine image of the current physiotherapy user is collected again, and the second front supine image contains the initially corrected projection light points. In this step, the system performs image verification on the initial correction result, that is, by taking the user's body image again to observe the actual light point position. The projected light spots should be clearly visible in the second image, and their image coordinates can be accurately extracted through binary segmentation, color recognition, or brightness detection of the light points.

[0038] It can be understood that this step not only verifies the initial correction result, but also provides a data basis for the error modeling of the next step. Due to the angle difference between the projected light points and the image acquisition camera, the acquisition device needs to complete joint calibration with the projector in advance to ensure the coordinate reversible projection.

[0039] Step S106, detecting the actual coordinates of the projected light points after initial correction in the second front supine image, and combining the acupoint recognition coordinate set for error compensation correction.

[0040] Among them, after detecting the actual coordinates of the projected light points, the original acupoint recognition coordinates are compared to form a set of residual vectors. By statistically analyzing the directionality and distribution characteristics of the residual values, an error model (such as a rigid body error model or a nonlinear fitting model) can be established to achieve overall or local projection adjustment. Specifically, the compensation method can be static correction (such as applying the same correction to all light points) or dynamic adjustment (such as independent correction for each acupoint). The final output projection instruction will make the light point position more closely coincide with the actual acupoint, achieving sub-millimeter acupoint projection accuracy, thereby ensuring the accuracy of acupuncture and moxibustion operation.

[0041] In the above embodiments, based on deep learning, acupoint image recognition, mapping and registration between image and physical coordinate space, control and real-time feedback correction of optical projection are realized, and a closed-loop control system is constructed, which has high recognition accuracy and high physical projection precision. The technical scheme of the present application can greatly reduce human participation, improve acupoint positioning efficiency and consistency, and effectively reduce acupuncture errors, laying a solid foundation for the digitalization and intelligentization of traditional Chinese medicine physiotherapy.

[0042] Reference Figure 2 As an embodiment of step S101, the step of acquiring and preprocessing the first front supine image of the current physiotherapy user includes: Step S201, using an industrial camera to acquire the first front supine image of the current physiotherapy user, and the industrial camera is erected above the physiotherapy bed; In the embodiments of the present application, an industrial camera is used for image acquisition, which is erected above the physiotherapy bed and vertically aligned with the user's supine body. Compared with consumer-level camera equipment, the industrial camera has significant advantages in imaging resolution, image acquisition frequency, light interference resistance, dynamic range, etc. It uses a high dynamic range image sensor (HDR Sensor), which can stably capture skin surface features under different lighting conditions, which is particularly important for identifying human acupoints with unclear contours and weak feature differences. By being fixedly erected above the bed, it ensures that each shot is taken from the same angle and distance, greatly reducing the interference of shooting posture on image features, which helps to improve model generalization and registration accuracy.

[0043] Step S202, pixel value normalization processing is performed on the first front supine image; In the step S202, the pixel value normalization processing is performed on the first front supine image. Pixel normalization is a basic preprocessing method in image processing, and the purpose is to map the pixel values of the original image (usually integers between 0-255) to the interval of floating-point values between 0 and 1, so as to eliminate the pixel range difference caused by different image sensor parameters, exposure time or light intensity.

[0044] Step S203, histogram equalization processing is performed on the first front supine image after pixel value normalization, to obtain a preprocessed first front supine image.

[0045] In the step S203, the histogram equalization is an image enhancement technique, aiming to improve the overall contrast of the image. The principle is to redistribute the gray levels of the image, so that the gray value distribution of the image is more uniform, thereby enhancing the detail performance of the low contrast area. Especially in the human acupoint recognition scene, due to the uniform skin color and weak texture in part of the acupoint area, the original dynamic range of the image is narrow, and it is difficult to be recognized by the neural network. Through the histogram equalization, the edges around the acupoint or the color difference is significantly strengthened, which enhances the perception ability of the model to the acupoint area and improves the accuracy and robustness of the subsequent key point detection.

[0046] In the present application, the normalization processing and the histogram equalization form a synergistic optimization structure before and after, the former standardizes the input data scale, and the latter enhances the feature saliency. When used together, it can significantly reduce the interference factors caused by environmental light, skin color difference and shooting resolution, so that the image reaches a unified and enhanced effect in both data scale and image structure. The processing flow is a typical preprocessing pipeline design in the deep learning scene, which has good universality and migratability.

[0047] In the above embodiment, the user's body surface image is accurately collected by using high-performance industrial camera equipment, and two classical and efficient image processing techniques of image normalization and histogram equalization are combined, which greatly improves the standardization degree and effective information density of the input image, provides a high-quality input basis for the subsequent acupoint recognition model, effectively suppresses the influence caused by external environmental factors or individual differences of human body, and thus improves the accuracy and stability of acupoint recognition in the whole system.

[0048] Reference Figure 3 As a further embodiment of the automatic acupoint projection method, before the step of inputting the preprocessed first front supine image into the pre-trained key point detection model, it further comprises: Step S301, performing human posture analysis on the current physiotherapy user located on the physiotherapy bed based on a pre-constructed human target detection model; The target detection model usually adopts a deep convolutional neural network structure such as YOLO (You Only Look Once) or Faster R-CNN, which is used to detect the overall body bounding box in the physiotherapy bed area and identify key parts such as the head, torso, hands and feet. Then, combined with the skeleton modeling technology, the system further infers the body axis information of the user in the supine position.

[0049] In step S302, it is determined whether the torso axis of the current physiotherapy user is offset from the central axis of the physiotherapy bed by an angle less than a preset threshold and whether the endpoints of the limbs are located within a preset area. If yes, it jumps to step S303, and if no, it jumps to step S304. Specifically, the torso axis refers to the line segment formed by the center point of the shoulder to the center point of the pelvis in the supine position, reflecting the alignment of the body axis and the bed body. The "central axis of the physiotherapy bed" is the center line of the bed body geometry. By comparing the angle between the two axes, the system can determine whether the user is tilted, and the tilt angle is generally achieved by calculating the angle vector between the two axes, which is usually completed by using the vector dot product formula or the Hough transform auxiliary line detection mechanism. In addition, the system also detects whether the endpoints of the limbs (such as wrist joints and ankle joints) fall within a preset area, which is modeled according to the actual size of the physiotherapy bed and the standard human anatomy posture.

[0050] In step S303, it is determined that the user is correctly positioned, and the preprocessed first supine image is input into the pre-trained key point detection model. When the system determines that the torso axis of the current physiotherapy user is offset from the central axis of the physiotherapy bed by an angle less than a preset threshold and that the limbs are located within a preset area, it is considered that the user's posture is standard and aligned, and the subsequent key point detection model input process can be continued. The "preset threshold" is set according to clinical experience, which is generally between 5° and 10°, aiming to give a certain tolerance to the human posture to adapt to patients of different body types while ensuring treatment effect.

[0051] In step S304, it is determined that the user is incorrectly positioned, and a positioning guide instruction is output to the voice prompt module, and the first supine image of the current physiotherapy user is reacquired after the user is correctly positioned.

[0052] If the user is not positioned, i.e., does not meet the above geometric relationship requirements, the system outputs a guide instruction through the voice prompt module, which usually integrates a TTS (Text to Speech) speech synthesis engine to broadcast the positioning prompt in natural language form, realizing non-contact interactive guidance for the patient. For example, the instructions "please adjust your body 3 cm to the left" and "please place your hands flat on the side of your body" are automatically generated by the system according to the detected offset direction, amplitude and angle and are broadcast in real time.

[0053] Finally, after the system detects that the user readjusts and meets the in-place condition, the first front supine image of the user is re-acquired, and the image preprocessing and recognition process is re-entered. This feedback mechanism constitutes a closed-loop process of feedforward detection-interactive guidance-feedback acquisition, ensuring that the image data received by the key point detection model has uniform posture standardization, greatly reducing the recognition bias, coordinate error and subsequent physical projection error caused by posture differences.

[0054] In the above embodiments, a posture detection and in-place judgment mechanism is introduced in the human-computer interaction link, a preposed posture evaluation system is established, and a complete logical chain of posture standardization-image input standardization-model recognition accuracy improvement is realized.

[0055] Reference Figure 4 As an embodiment of step S104, the projection coordinate set and the acupoint recognition coordinate set are spatially registered, the translation and rotation parameters of a plurality of matching point pairs are calculated, and the projection coordinate set is initially corrected. The step of outputting the initially corrected projection light points comprises: Step S401, spatially register the projection coordinate set and the acupoint recognition coordinate set to obtain a mapping relationship set containing a plurality of matching point pairs; wherein each matching point pair includes an acupoint recognition coordinate (x D ,y D ,θ D ) and a corresponding projection coordinate (x d ,y d ,θ d ); Specifically, the point sets defined in the two spaces (i.e., the acupoint recognition coordinate set in the image coordinate system and the projection coordinate set in the device projection coordinate system) are one-to-one corresponding, and a function relationship between them is established. In this scheme, since the image recognition result is derived from the camera angle of view, and the projection coordinate is based on the position of the projector to control the light beam, the coordinate systems of the two are inconsistent, so a set of transformation mechanism must be constructed to align them.

[0056] In the embodiments of the present application, the construction of the mapping relationship needs to rely on a group of corresponding point pairs with actual semantics (usually significant acupoint), which are respectively derived from the two sets and have stable spatial correspondence in the preset anatomical region. The initial matching point set can be generated by the method based on the minimum Euclidean distance of the adjacent region, the shape consistency constraint, or the automatic matching based on the known human body anatomy template library.

[0057] Step S402, calculate the translation and rotation parameters based on the matching point pairs; Wherein, the calculation formula of the translation and rotation parameters is: (Δx,Δy,Δθ) T =(x d ,y d ,θ d ) T -(x D ,y D ,θ D ) T ; In the above formula, T represents the vector transpose, Δx and Δy represent the translational deviation between the two coordinates, and Δθ represents the rotational angle deviation between the two coordinates.

[0058] Step S403: Calculate the i-th unregistered projection coordinate (x, y) based on the translation and rotation parameters. i ,y i ,θ i Correction amount Among them, the correction amount The calculation formula is: In the above formula, Indicates the translation compensation item. Represents the rotation matrix. This represents the deviation vector.

[0059] Specifically, the correction amount refers to the spatial offset vector between the original projection point and its position under ideal recognition coordinates. Essentially, it is a two-dimensional vector representing the direction and magnitude of adjustment required for each point. This vector can be understood as a local directional guide in the error field, reflecting deviations introduced by inaccurate initial recognition, projection distortion, or slight movements of the user's body.

[0060] Step S404, based on the correction amount For the i-th projection coordinate (x) i ,y i ,θ i Perform initial calibration and output the projected light spot after initial calibration.

[0061] In the above embodiments, a point-to-point mapping relationship is constructed between the recognition results and the projection commands. A unified transformation of the coordinate space is achieved by combining a rigid body transformation model. Furthermore, an error feedback mechanism is introduced into the projection control, realizing preliminary but accurate light spot correction. This linkage mechanism from recognition to correction constitutes the crucial "vision-space-control" closed loop in this application, which not only ensures high precision and high consistency in acupoint positioning but also lays a solid foundation for subsequent more complex dynamic compensation and feedback control.

[0062] Reference Figure 5As an embodiment of step S106, detecting the actual coordinates of the projection light points after the initial correction in the second front supine image and performing error compensation correction in combination with the acupoint recognition coordinate set comprises: Step S501, detecting the actual coordinates of the projection light points after the initial correction in the second front supine image based on a pre-constructed light point target detection network; Wherein, the system reacquires a second image of the user in a supine position, which contains the initial correction light points that have been projected on the user's body surface. The main purpose of this image is to obtain feedback information of the actual projection effect and provide the true projection position for subsequent error calculation. The task of the light point target detection network in this stage is to identify the accurate position coordinates of these light points in the image. Due to the characteristics of high brightness, strong contrast, and regular shape of the light points, the network can be trained using a lightweight convolutional neural network structure (such as YOLOv5-tiny, EfficientDet, etc.) to quickly and stably extract the light spot center.

[0063] It should be noted that in order to cope with different lighting environments, skin color changes and body surface curvature interference, the target detection network needs to introduce a special brightness enhancement feature channel and combine image space attention mechanism (such as SE-Block or CBAM module) to improve the model's perception ability of small targets (i.e. light points). At the same time, the light point coordinates output by the network are usually pixel positions in the image space, and need to be converted back to the same spatial reference system as the recognition coordinates through inverse mapping to ensure the homogeneity of error calculation.

[0064] Step S502, calculating the error value of the actual coordinates of the projection light points after the initial correction based on the acupoint recognition coordinate set; Wherein, the error value of the actual coordinates of the projection light points after the initial correction is calculated based on the acupoint recognition coordinate set. The core of this link is to quantify the spatial error. The difference between each light point actual coordinate and its should-be position (i.e. acupoint recognition coordinate) can be represented by the Euclidean distance formula.

[0065] Step S503, when the error value is greater than a preset error threshold, performing again compensation correction on the projection light points after the initial correction based on the error value.

[0066] Specifically, the system introduces a preset error threshold as a quantitative basis for judging whether the current correction is effective. The threshold is set according to the system accuracy target, generally less than the average diameter of human acupoints (about 5-10mm), such as within 2mm, to ensure that the light point positioning accuracy is higher than the error level of traditional manual marking.

[0067] When the system detects that the error value of any light spot exceeds the threshold value, the first corrected projection light spot is compensated and corrected again based on the error value. The compensation method usually adopts a local fine-tuning strategy, that is, based on the known error vector, the output control parameters of the projector are updated in real time, so that the next time the projection light spot makes a "reverse deviation" in this direction, thereby approaching the true acupoint position.

[0068] It can be understood that this correction not only includes translation correction, but also can introduce a micro-angle adjustment according to the system setting to deal with the visual distortion caused by the change of body curvature. In addition, in order to avoid over-correction or system oscillation, the compensation amplitude can be set to a gradual convergence model, that is, the compensation proportion is dynamically adjusted according to the error value (for example, using a PID control strategy or an exponential weighted average method), to ensure that the compensation is both rapid and stable.

[0069] In the above embodiment, the target detection network of deep learning is combined with the geometric error analysis technology to establish a closed-loop feedback control system with "perception-comparison-adjustment" as the core, and a multi-stage light point correction mechanism from the first coarse registration to the second fine compensation is realized. By detecting the error between the projection result and the recognition result, and making a decision compensation according to the quantitative index, not only the coincidence accuracy of the projection light spot and the acupoint is significantly improved, but also the whole system has the ability of adaptive correction, which effectively overcomes the error accumulation caused by complex variables such as individual differences of human body, posture deviation or environmental light change. This dynamic self-correction mechanism is a key technical guarantee for realizing clinical-level accurate acupoint positioning and projection, and is also an important embodiment of the application from "static projection" to "intelligent tracking correction".

[0070] The application further discloses an automatic acupoint projection system.

[0071] The automatic acupoint projection system comprises a host computer, an image acquisition device and a projector. The image acquisition device is used for acquiring a front supine image of a current physiotherapy user. The projector is used for receiving a projection instruction and generating a projection light spot for highlighting an acupoint on the current physiotherapy user. The host computer is configured to: send a first image acquisition instruction to the image acquisition device, receive a first front supine image of the current physiotherapy user sent by the image acquisition device and perform preprocessing to obtain a preprocessed first front supine image; input the preprocessed first front supine image into a pre-trained key point detection model to output an acupoint recognition coordinate set; generate a projection instruction based on the acupoint recognition coordinate set and send the projection instruction to the projector to control the projector to generate an initial projection light spot for highlighting an acupoint on the current physiotherapy user, and obtain a corresponding projection coordinate set based on a projector coordinate system; The set of projection coordinates is spatially registered with the set of acupoint recognition coordinates, translation and rotation parameters of a plurality of matching point pairs are calculated, the set of projection coordinates is initially corrected, and the initially corrected projection light points are output; The second image acquisition instruction is sent to the image acquisition device, and a second front supine image of the current physiotherapy user is received, which contains the initially corrected projection light points; The actual coordinates of the initially corrected projection light points in the second front supine image are detected, and error compensation correction is performed in combination with the set of acupoint recognition coordinates.

[0072] The automatic acupoint projection system can implement any of the automatic acupoint projection methods described above, and the specific working processes of the modules in the automatic acupoint projection system can refer to the corresponding processes in the method embodiments described above.

[0073] In several embodiments provided in the present application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are only schematic; for example, the division of a certain module is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0074] The embodiments of the present application also disclose a computer device.

[0075] The computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the automatic acupoint projection method as described above when executing the computer program.

[0076] The embodiments of the present application also disclose a computer readable storage medium.

[0077] The computer readable storage medium stores a computer program capable of being loaded and executed by a processor to implement any of the automatic acupoint projection methods as described above.

[0078] The computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus; the program code contained in the computer readable medium can be transmitted by any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any appropriate combination of the above.

[0079] It should be noted that the computer device and the storage medium of the embodiments of the present application are electronic devices and storage media to which the automatic acupoint projection method is applied, and all the embodiments of the automatic acupoint projection method are applicable to the computer device and the storage medium, and can achieve the same or similar beneficial effects. For the computer device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the description of the method embodiments.

[0080] In the present application, the terms "first", "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0081] Although the present application is described herein in conjunction with various embodiments, other variations of the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from the description, the drawings and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit can implement several of the functions recited in the claims. Measures recited in mutually different dependent claims do not exclude each other from being combined in one embodiment.

[0082] The above are the preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, any feature disclosed in the specification (including the abstract and the drawings) can be replaced by other equivalent or similar features, unless specifically stated. That is, each feature is only an example of a series of equivalent or similar features, unless specifically stated.

Claims

1. An automatic acupoint projection method, characterized by, The projection method comprises: Collecting a first front supine image of a current physiotherapy user and pre-processing to obtain a pre-processed first front supine image; inputting the pre-processed first front supine image into a pre-trained key point detection model to output a set of acupoint recognition coordinates; Generating a projection instruction according to the set of acupoint recognition coordinates to control a projector to generate an initial projection light point highlighting an acupoint on the current physiotherapy user, and obtaining a corresponding set of projection coordinates based on the projector coordinate system; Spatially registering the set of projection coordinates and the set of acupoint recognition coordinates, calculating translation and rotation parameters of a plurality of matching point pairs, and performing a primary correction on the set of projection coordinates to output a primary corrected projection light point; Collecting a second front supine image of the current physiotherapy user again, the second front supine image containing the primary corrected projection light point; Detecting the actual coordinates of the primary corrected projection light point in the second front supine image and performing error compensation correction in combination with the set of acupoint recognition coordinates.

2. The automatic acupoint projection method according to claim 1, wherein, The step of collecting a first front supine image of a current physiotherapy user and pre-processing comprises: Collecting a first front supine image of a current physiotherapy user using an industrial camera, the industrial camera being mounted above a physiotherapy bed; performing pixel value normalization processing on the first front supine image; Performing histogram equalization processing on the first front supine image after pixel value normalization to obtain a pre-processed first front supine image.

3. The automatic acupoint projection method according to claim 2, wherein, Before the step of inputting the pre-processed first front supine image into a pre-trained key point detection model, it further comprises: Performing human posture analysis on the current physiotherapy user located on the physiotherapy bed based on a pre-constructed human target detection model; determining whether the trunk axis of the current physiotherapy user is offset from the central axis of the physiotherapy bed by an angle less than a preset threshold and whether the endpoints of the limbs are located within a preset area; If yes, it is determined that the user is correctly positioned, and the step of inputting the pre-processed first front supine image into a pre-trained key point detection model is continued to be executed; If no, it is determined that the user is incorrectly positioned, a positioning guide instruction is output to a voice prompt module, and a first front supine image of the current physiotherapy user is re-collected after detecting that the user is correctly positioned.

4. The automatic acupoint projection method according to claim 1, wherein, The step of spatially registering the set of projection coordinates and the set of acupoint recognition coordinates, calculating translation and rotation parameters of a plurality of matching point pairs, and performing a primary correction on the set of projection coordinates to output a primary corrected projection light point comprises: The projection coordinate set and the acupoint recognition coordinate set are spatially registered to obtain a mapping relationship set containing a plurality of matching point pairs; wherein each matching point pair includes an acupoint recognition coordinate (x D ,y D ,θ D ) and a corresponding projection coordinate (x d ,y d ,θ d ); Calculating translation and rotation parameters based on the matching point pairs; According to the translation-rotation parameters, a correction amount (Δx i ,Δy i ,Δθ i ) of the unregistered i-th projection coordinate (x i ,y i ,θ i ) is calculated According to the correction amount The i-th projection coordinate (x i ,y i ,θ i ) is initially corrected, and the initially corrected projection light point is output.

5. The automatic acupoint projection method according to claim 4, wherein, The calculation formula of the translation and rotation parameters is: (x, y, θ) T = (x d ,y d ,θ d ) T - (x D ,y D ,θ D ) T ; In the above formula, T represents vector transposition, Δx and Δy represent translation deviations between two coordinates, and Δθ represents a rotation angle deviation between two coordinates.

6. The automatic acupoint projection method according to claim 4, wherein, the correction amount The calculation formula is: In the above formulae, denotes a translation compensation term, denotes a rotation matrix, denotes a deviation vector.

7. The automatic acupoint projection method according to any one of claims 1 to 6, characterized in that, The step of detecting the actual coordinates of the primary corrected projection light point in the second front supine image and performing error compensation correction in combination with the set of acupoint recognition coordinates comprises: Detecting the actual coordinates of the primary corrected projection light point in the second front supine image based on a pre-constructed light point target detection network; Calculating error values for the actual coordinates of the primary corrected projection light point based on the set of acupoint recognition coordinates; When the error value is greater than a preset error threshold, the primary corrected projection light point is re-compensated and corrected based on the error value.

8. An automatic acupoint projection system, characterized in that, The projection system comprises a host computer, an image acquisition device and a projector. The image acquisition device is configured to acquire a front supine image of a current physiotherapy user. The projector is configured to receive a projection instruction and generate a projection light point for highlighting an acupoint on the current physiotherapy user. The host computer is configured to: send a first image acquisition instruction to the image acquisition device, receive a first front supine image of the current physiotherapy user sent by the image acquisition device, and pre-process the first front supine image to obtain a pre-processed first front supine image; input the pre-processed first front supine image into a pre-trained key point detection model to output an acupoint recognition coordinate set; generate a projection instruction based on the acupoint recognition coordinate set and send the projection instruction to the projector, control the projector to generate an initial projection light point for highlighting an acupoint on the current physiotherapy user, and obtain a corresponding projection coordinate set based on a projector coordinate system; perform spatial registration on the projection coordinate set and the acupoint recognition coordinate set, calculate translation and rotation parameters of a plurality of matching point pairs, and perform primary correction on the projection coordinate set to output a primary corrected projection light point; send a second image acquisition instruction to the image acquisition device, receive a second front supine image of the current physiotherapy user sent by the image acquisition device, and the second front supine image comprises the primary corrected projection light point; detect an actual coordinate of the primary corrected projection light point in the second front supine image, and perform error compensation correction in combination with the acupoint recognition coordinate set.

9. A computer device, characterized by: A computer program product comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program product comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method of any one of claims 1 to 7.

Images