Computer vision-based nine-turn method of prolonging life landing type abdominal massage robot

Abstract

The application discloses a computer vision-based Yanjian nine-turn method floor type abdomen kneading robot, and relates to the technical field of intelligent physiotherapy robots, comprising a floor type body, an independent vision cantilever, an image acquisition module, a massage execution mechanism comprising a six-axis mechanical arm, a wrist mechanism and an integrated spherical contact, and a control unit; the independent vision cantilever is physically separated from the massage execution mechanism, effectively eliminating the interference of mechanical vibration on visual positioning; the control unit generates a path based on image recognition of acupoints and automatically controls the massage mechanism according to the first eight forms of the Yanjian nine-turn method; the wrist mechanism can adjust the posture of the contact according to the massage type, switch to a vertical posture to execute a kneading method by using the end part, or switch to an inclined posture to execute a pushing method by using the side surface. The application can realize automatic switching of traditional Chinese medicine compound manipulation without manual replacement of massage heads, and realizes full automation and standardized reproduction of traditional massage manipulation.

Description

Technical Field

[0001] This invention relates to the field of intelligent robots and traditional Chinese medicine physiotherapy equipment technology, specifically to a computer vision-based longevity nine-turn abdominal massage robot. Background Technology

[0002] The Nine-Turn Longevity Method is a famous guiding and massage technique created by Fang Kai in the Qing Dynasty. It is recorded in "Yishenji" and is also known as "Mofu Qiyun Tukao" or "Xianren Rufu Shu". It was later included in classics such as "Weisheng Yaoshu" and "Neigong Tushuo". This method integrates guiding exercises and abdominal massage. The complete set of exercises includes eight kneading and pushing movements and one upper body shaking method. It focuses on regulating the body, breath, and mind to nourish essence, qi, and spirit, so as to make essence and qi abundant and harmonize the internal and external. It is a unique longevity-promoting exercise.

[0003] According to the clinical study published by Xie Fangfang et al. in *Shi Zhen Guo Yi Guo Yao* (Volume 31, Issue 12, 2020) (ethics approval number: 2018-043; US Clinical Trials Registry number: NCT03496961), the Longevity Nine-Turn Method has multiple clinical effects on patients with chronic fatigue syndrome (CFS): (1) Improvement of fatigue: After 12 weeks of intervention, the scores of all dimensions of the MFI-20 scale (overall fatigue, physical fatigue, reduced activity, reduced interest, and mental fatigue) in the exercise group were significantly improved (P<0.05), with a total effective rate of 62.22%.

[0004] (2) Improved sleep quality: The scores of the exercise group on the PSQI scale in the five dimensions of subjective sleep quality, sleep latency, sleep duration, sleep disorder and daytime dysfunction were significantly reduced (P<0.05).

[0005] (3) Improvement of anxiety and depression: The scores of anxiety and depression dimensions of the HADS scale in the exercise group were significantly reduced (P<0.05).

[0006] (4) Improved quality of life: The scores of the eight dimensions of the SF-36 scale (physiological function, physiological function, bodily pain, general health, energy, social function, emotional function, and mental health) of the exercise group were significantly improved (P<0.05), and the group was significantly better than the cognitive behavioral education group in terms of bodily pain and energy (P<0.05).

[0007] (5) Changes in brain function mechanism: fMRI studies show that the Longevity Nine Turns Method can change neuronal activity (ALFF) and enhance brain functional connectivity (FC) in CFS patients. Specifically, it manifests as enhanced neuronal activity in brain regions such as the anterior cingulate cortex, medial cingulate cortex, postcentral gyrus, and superior medial frontal gyrus, as well as enhanced functional connectivity between the medial cingulate cortex and the middle temporal gyrus, the angular gyrus and the fusiform gyrus, and the occipital lobe.

[0008] This study also elucidates the mechanism of action of the Nine-Turn Longevity Method from the perspective of traditional Chinese medicine theory: The first four movements begin at the solar plexus (upper abdomen, on the anterior midline, about two finger widths down from the depression at the lower end of the sternum; Juque acupoint (RN14)), and are kneaded and moved down to the pubic bone below the navel. Then, the kneading is performed from the pubic bone upwards, and the two hands meet at the solar plexus and push straight up the pubic bone again. The entire process involves kneading through the upper abdomen (RN13), middle abdomen (RN12), lower abdomen (RN10), and Shenque acupoint (RN14). 8) Guanyuan (RN4) connects the middle and lower jiao (digestive system), using abdominal massage to transform stillness into movement and movement into stillness, harmonizing the upper and lower parts of the body, nourishing the five internal organs, expelling external pathogens, and clearing internal ailments. The fifth and sixth movements involve circling the abdomen with both hands, horizontally connecting the kidney, stomach, spleen, and liver meridians, stimulating nerve conduction pathways and the neuro-emotional pathway, strengthening the connection between the digestive system and the brain. The seventh and eighth movements push directly from the left and right hypochondria to the groin, guiding qi downwards to nourish the limbs and bones. The ninth movement involves holding the hands firmly, sitting cross-legged, and rotating, protecting the qi, improving eyesight, prolonging life, relieving emotional distress, and regulating qi and blood. Each of the first eight movements is both independent and interconnected, correspondingly adjusting all five internal organs.

[0009] Modern research suggests that the Nine Turns Longevity Method utilizes the brain-gut axis for neural conduction connections, strengthening the link between the digestive system and brain nerves through visceral stimulation neural pathways and neuro-emotional pathways, thus regulating fatigue, sleep, and anxiety / depression in CFS patients. The brain-gut peptides produced by the abdominal manipulation can also rapidly relieve brain fatigue. Simultaneously, as a low-intensity aerobic exercise, the Nine Turns Longevity Method can enhance the activity of antioxidant enzymes, scavenge free radicals, stabilize the body's internal environment, and keep upper limb skeletal muscles in a relaxed state, effectively relieving both mental and physical fatigue.

[0010] According to the nationally compiled textbook "Tuina Exercises" for general higher education in Traditional Chinese Medicine, the specific operation method of the Longevity Nine-Turn Method is as follows: First movement: Press the heart with the middle three fingers of both hands and rub in a clockwise circular motion 21 times.

[0011] The second technique: Use the middle three fingers of both hands to rub downwards from the solar plexus, rubbing and moving, until you reach the high bone below the navel.

[0012] Third movement: Using the middle three fingers of both hands, knead upwards from the high bone towards both sides, kneading and moving, until reaching the solar plexus. The hands should then meet.

[0013] Fourth movement: Use the middle three fingers of both hands to push from the solar plexus down to the sternum, 21 times in total.

[0014] Fifth movement: Use your right hand to rub your navel and abdomen in a circular motion from left to right 21 times.

[0015] Sixth movement: Use your left hand to rub your navel and abdomen in a circular motion from right to left 21 times.

[0016] Seventh movement: Place your left hand on your left side, below your soft ribs and kidneys, with your thumb pointing forward and your four fingers supporting your back, and gently pinch it; use the middle three fingers of your right hand to push straight from below your left breast to your groin 21 times.

[0017] Eighth movement: Place your right hand on your right lower back, with your thumb pointing forward and your four fingers supporting your back, and gently pinch it; use the middle three fingers of your left hand to push straight down from below your right breast to your groin 21 times.

[0018] Ninth Movement: (Complete movements one through eight in sequence to form one cycle; perform seven cycles consecutively each time.) After completing the movement, sit cross-legged, clasping your hands firmly on your knees. Slightly bend your toes. Rotate your upper body clockwise 21 times. Then, rotate counter-clockwise 21 times.

[0019] Although various abdominal massage or acupoint positioning devices already exist, a search and analysis reveals that existing technologies still have the following five shortcomings: Category 1: Traditional abdominal massage devices without automatic acupoint recognition function.

[0020] These devices use purely mechanical structures to perform simple abdominal massage movements. A typical example is CN2022203800776 (publication number CN217448391U), an automatic abdominal massager using Bian stone that only performs a single rotational abdominal massage movement. This limited technique makes it impossible to apply differentiated pushing and kneading techniques to different acupoint areas. A common drawback of these devices is that the massage path does not distinguish between meridians and acupoints, making it impossible to perform traditional Chinese medicine acupoint selection along meridians, let alone traditional exercises like the Nine Turns Longevity Method, which have a clear sequence of acupoint pathways.

[0021] The second category: intelligent massage devices that use non-visual methods to locate acupoints.

[0022] CN2019202276438 (Publication No. CN210125000U) discloses an automatic acupoint massager based on a photoelectric recognition automatic acupoint positioning system (Qingdao University of Technology). This device uses a photoelectric recognition sensor for acupoint positioning and a three-axis moving platform to drive the massage device. Its shortcomings are: the photoelectric recognition sensor is essentially a photoelectric through-beam or through-beam detection, mainly used to detect surface markings or color differences. It cannot acquire complete three-dimensional contours and texture information of the abdomen like a depth camera, and its ability to identify acupoints without special markings (such as the depression of the Shenque acupoint, the bony landmark relationship between the Jiuwei acupoint and the xiphoid process, etc.) is limited; the massage head of the three-axis platform can only achieve point-like pressure and cannot continuously execute the large-scale, multi-track massage paths unique to the Longevity Nine Turns Method, such as meridian kneading, abdominal kneading around the navel, and straight chest and abdomen push.

[0023] The third category: Although computer vision positioning is used, the massage technique of the Nine Turns for Longevity is not integrated.

[0024] Taiwan Patent I367089 describes a massage system with video acupoint positioning (inventors James Tsing-Iuan Tsay and Pau-Choo Chung). It uses image capture and comparison to locate acupoints and then drives the massage device. However, it is a general acupoint positioning system, not specifically for the abdomen, does not generate meridian pathways, and does not provide massage path planning corresponding to the Longevity Nine Turns Method or any other traditional Chinese medicine exercise.

[0025] CN2021112959315 (Publication No. CN114022554A) A method for acupoint detection and localization of massage robots based on YOLO (Beihua Aerospace Industry College) The core drawback of this method is that it relies on acupoint markers. Users must attach multiple circular markers to their abdomen before the massage, which is cumbersome. Moreover, the placement of the markers depends on human experience, and it cannot achieve fully automatic acupoint recognition that users can remove and use immediately without marking.

[0026] CN2024105713066 (Publication No. CN118141681A) Abdominal Acupoint Determination Method, Device, Moxibustion Robot, and Storage Medium (Jianghan University, Application Date 2024) This is currently the patent most similar to the visual positioning technology route of this invention. The method determines abdominal meridians through key human feature points, and then uses the topological relationship between meridians and acupoints to determine the location of abdominal acupoints. However, it has the following key shortcomings: This patent is a method patent, targeting the application scenario of a moxibustion robot, which performs fixed-point moxibustion on acupoints through moxibustion components, without involving mechanical massage actions such as rubbing or pushing, and does not involve the massage path of the first eight postures of the Longevity Nine Turns Method; it does not mention the guidance of the traditional Chinese medicine theory of Qi circulation and the four phases of the spleen.

[0027] CN2023110007078 (Publication No. CN116721150B) describes a method for predicting human acupoints based on key points and a massage robot (Shenzhen Lichi Sensor Technology Co., Ltd., 2023). This method derives the coordinates of target acupoints through a few key acupoints, representing a general acupoint prediction method for massage robots. Its shortcomings include: it does not specifically target abdominal meridians, does not generate meridian pathways for the Ren, Kidney, and Stomach meridians, and is not suitable for exercises like the Nine Turns Longevity Method, which requires complex massage along multiple meridians.

[0028] CN2022110302337 (Publication No. CN115393437A) Automatic Acupoint Recognition and Massage Trajectory Planning Method and Home Massage Robot (Yangzhou University, 2022) uses 3D vision combined with acupoint maps to generate massage trajectories. The technical route is similar to the 3D visual positioning + meridian path generation of the present invention. However, its acupoint map is a general human standard template, which does not include the bone measurement method and body measurement method correction guided by the TCM meridian theory. Moreover, it is aimed at general whole body massage scenarios and does not target abdominal meridians and the Longevity Nine Turns Method path.

[0029] CN2022109383297 (Publication No. CN117582369A) describes a method, device, equipment, system, and medium for acupoint positioning in a massage robot (Shanghai Pudong New Area Gongli Hospital and Shanghai Jiao Tong University, 2022). This method uses a vision system to collect human images and feature markers, and utilizes these markers to assist in acupoint positioning. Its shortcomings are: acupoint positioning requires markers, which is fundamentally different from the markerless, fully automatic solution of this invention that directly identifies natural anatomical features of the body surface without any markers; furthermore, it is a general massage system and does not involve the "Nine Turns for Longevity" method.

[0030] CN2020104376071 (Publication No. CN111437185A) Automatic Meridian Location Method and Application for Robots (Nanjing Aiyi Intelligent Technology Co., Ltd., 2020) acquires human body orthographic projection images through visual recognition, obtains edge lines through color difference analysis, and generates meridian images based on a preset meridian location method. Its shortcomings are: edge detection based on color difference analysis of 2D orthographic projection images is highly susceptible to changes in lighting and skin tone; meridian location is a general method and does not match the operational standards of specific exercises such as the Nine Turns Longevity Method; it is geared towards general applications such as moxibustion and massage, and does not provide automatic switching for specific compound techniques such as kneading and pushing.

[0031] Category 4: Existing massage robots lack standardized pathway programs based on specific traditional Chinese medicine exercises.

[0032] While existing massage robots have made some progress in general massage or acupoint positioning, none of them have transformed specific traditional Chinese medicine techniques into standardized, executable robot programs.

[0033] CN2016103394894 (Publication No. CN105919796A) describes a Traditional Chinese Medicine massage robot system and its acupoint search method (Shandong Jianzhu University, 2016). This system includes a six-axis robotic arm and a humanoid massage hand, capable of performing various massage techniques such as finger pressing, finger kneading, palm pushing, palm pressing, palm kneading, pinching, and tapping. While its techniques are similar to the kneading and palm pushing techniques of this invention, it suffers from the following key shortcomings: the massage techniques performed are general TCM massage movements, without path planning and hand modeling based on the operational standards of any specific exercise such as the Nine Turns Longevity Method; acupoint positioning is based on BP neural networks and 2D images, without employing the multi-task CNNs dual-modal (RGB+depth) detection plus meridian path topology generation scheme of this invention; the massage path is formed by connecting general acupoints, without generating meridian paths such as the Ren Meridian, Kidney Meridian, Stomach Meridian, Spleen Meridian, and Liver Meridian, and it lacks the ability to perform a composite motion planning that combines massage and walking.

[0034] US Patent US20170266077A1 (Application No. 15465456), a robotic massager and its method of use (Christian Campbell Mackin, 2017), achieves adaptive adjustment of the massage path for users of different body types through deformation mapping of anatomical calibration points, which is similar to the idea of ​​this invention using body measurement for individual body type calibration. However, the essential difference is that the massage path of the patent is based on a pre-programmed general path, and the path deformation for the second body type user is only geometrically scaled, without involving the acupoint and meridian meanings of any specific traditional Chinese medicine exercise such as the Nine Turns of Longevity; while the path of this invention originates from the standard massage action sequence of the first eight movements of the Nine Turns of Longevity, and the body type adaptation is based on the overall coordinate calibration on the topology of the Ren Meridian, Kidney Meridian, Stomach Meridian, Spleen Meridian, and Liver Meridian, rather than simple geometric scaling of the path.

[0035] Category 5: Existing massage equipment lacks parameter design based on measured biomechanical data.

[0036] Yan Xiaohui et al. (2015) used a three-dimensional motion analysis system and a massage technique testing system to standardize the operational parameters of 12 massage techniques, including palm rubbing, and provided reference ranges for the force and frequency of the techniques based on expert measurements: vertical force of palm kneading: 8.728-105.431 N, frequency: 1.11-3.06 Hz; vertical force of palm rubbing: 5.964-66.460 N, frequency: 1.20-2.14 Hz; vertical force of palm rubbing: forward rubbing: 5.256-70.235 N, backward rubbing: 3.903-42.169 N, frequency: 1.27-1.81 Hz. However, no existing massage device has cited these research results to calibrate its own massage parameters. Summary of the Invention

[0037] The present invention provides a computer vision-based longevity nine-turn abdominal massage robot that can solve the above-mentioned problems.

[0038] To solve the above problems, the technical solution adopted by the present invention is as follows: A computer vision-based longevity massage robot for abdominal massage includes a floor-standing body, an independent vision cantilever, an image acquisition module, a massage actuator, and a control unit. The independent vision cantilever and the massage actuator are respectively mounted on the floor-standing body, with the independent vision cantilever physically separated from the massage actuator. The image acquisition module is mounted on the independent vision cantilever and is used to acquire image data of the user's abdomen and transmit it to the control unit. The massage actuator includes a six-axis robotic arm, a wrist mechanism, and an integrated spherical contact. The six-axis robotic arm is mounted on the floor-standing body, the wrist mechanism is mounted at the end of the six-axis robotic arm, and the integrated spherical contact is mounted on the wrist mechanism. The wrist mechanism is used to adjust the integrated spherical contact. The spherical contact head is positioned relative to the user's abdominal skin. The control unit is electrically connected to the image acquisition module and the massage actuator. The control unit is used to identify acupoints on the user's abdomen based on the image data and generate a massage path. It then controls the massage actuator to perform corresponding massage actions according to the preset massage path program of the first eight movements of the Longevity Nine Turns Method. When executing the massage path program of the first eight movements of the Longevity Nine Turns Method, the control unit controls the wrist mechanism to switch the integrated spherical contact head to a vertical position perpendicular to the abdominal skin to perform a kneading motion using the end of the integrated spherical contact head, or to switch the integrated spherical contact head to an inclined position tilted towards the abdominal skin to perform a pushing motion using the side of the integrated spherical contact head.

[0039] Furthermore, the image acquisition module includes a depth camera and a light source component, and the image data includes color images and depth images; the control unit has a built-in multi-task convolutional neural network model, which includes a feature extraction backbone and multiple task-specific output branches; when identifying acupoints on the user's abdomen, the multi-task convolutional neural network model uses modal fusion weights to perform weighted summation and fusion of the shallow feature maps extracted from the color image and the shallow feature maps extracted from the depth image, and the fused data is input to the feature extraction backbone to extract multi-scale features, and outputs in parallel through the multiple task-specific output branches the two-dimensional coordinates of the user's key abdominal acupoints in the image plane, the binary segmentation map of the abdominal boundary, the depth value of the user's key abdominal acupoints in the depth image, and the location confidence score of each of the user's key abdominal acupoints, including the Shenque acupoint; the modal fusion weights are dynamically adjusted according to the user's body mass index, increasing the modal fusion weights of the shallow feature maps corresponding to the color image and decreasing the modal fusion weights of the shallow feature maps corresponding to the depth image when the user's body mass index is high.

[0040] Furthermore, the control unit is equipped with a voice-guided assisted positioning function. When the location confidence score of the Shenque acupoint output by the multi-task convolutional neural network model is lower than a preset confidence threshold, the control unit outputs a voice prompt to guide the user to lightly touch the Shenque acupoint with their finger. The image acquisition module acquires an image containing the user's finger and performs fingertip detection to obtain the three-dimensional coordinates of the contact point between the finger and the abdomen as candidate coordinates of the Shenque acupoint. The control unit compares the candidate coordinates of the Shenque acupoint with the coordinates of the Shenque acupoint detected by the multi-task convolutional neural network model. When the deviation between the two is less than or equal to a preset deviation threshold, the coordinates of the Shenque acupoint detected by the multi-task convolutional neural network model are accepted. When the deviation between the two is greater than the preset deviation threshold, the candidate coordinates of the Shenque acupoint are used as the final coordinates of the Shenque acupoint.

[0041] Furthermore, the abdominal boundary represented by the binary segmentation map of the abdominal boundary detected by the multi-task convolutional neural network model includes the xiphoid process tip, the upper edge boundary of the pubic symphysis, the most prominent point of the left anterior superior iliac spine, and the most prominent point of the right anterior superior iliac spine. The control unit establishes a local coordinate system of the abdomen with the Shenque acupoint as the origin, and uses the horizontal distance between the detected most prominent points of the left and right anterior superior iliac spines as the reference for the transverse width of the abdomen. The absolute length of one inch is calculated as one-twelfth of the transverse width reference between the two anterior superior iliac spines. Based on the absolute length of one inch and the bone measurement method, the control unit calculates the engineering coordinates of other meridians and acupoints, and automatically generates the circulation trajectories of the Ren meridian, kidney meridian, stomach meridian, spleen meridian, and liver meridian in the abdominal region. The circulation trajectories are converted into spatial coordinates in the robotic arm base coordinate system through the hand-eye calibration matrix, forming a massage path for the massage execution mechanism to perform.

[0042] Furthermore, the massage actuator also includes a six-dimensional force sensor installed between the wrist mechanism and the integrated spherical contact head. The six-dimensional force sensor is used to collect the contact force between the integrated spherical contact head and the user's abdominal skin in real time. The base of the integrated spherical contact head integrates a heating component. The kneading method includes finger kneading and palm kneading, and the pushing method includes palm pushing. The control unit performs closed-loop force control based on an incremental proportional-integral-derivative control algorithm, and sets different multi-level target force values ​​for the finger kneading, palm kneading, and palm pushing methods respectively. When performing the kneading method, the integrated spherical contact head forms point contact with the user's abdominal skin in the vertical posture and performs circular or linear translational motion. When performing the pushing method, the integrated spherical contact head forms surface contact with the user's abdominal skin using the side of the sphere in the tilted posture and slides along the massage path. Furthermore, the set target force value is applied during the pushing stroke of the palm pushing method, and the contact force drops to a preset return force threshold during the return stroke.

[0043] Furthermore, the massage path program of the first eight movements of the Longevity Nine Turns Method specifically includes the first to eighth movements executed sequentially: First movement: The control unit controls the integrated spherical contact head to maintain the vertical posture, positioning it at the Juque acupoint, and performing a clockwise circular finger kneading technique; Second movement: The control unit controls the integrated spherical contact head to maintain the vertical posture, kneading downwards from the Juque acupoint along the Ren meridian and bilateral kidney meridian areas to the Qugu acupoint, performing a finger kneading technique in a kneading and walking manner; Third movement: The control unit controls the integrated spherical contact head to maintain the vertical posture, kneading upwards from the Qugu acupoint to both sides, performing a finger kneading technique in a kneading and walking manner along the bilateral kidney meridian and stomach meridian areas to the Juque acupoint; Fourth movement: The control unit controls the integrated spherical contact head to switch to the tilting position. The first movement involves a slanted posture, pushing directly from the Juque acupoint along the Ren meridian and the bilateral kidney meridian areas to the Qugu acupoint, performing a palm pushing technique; the second movement involves the control unit controlling the integrated spherical contact to switch to the vertical posture, performing a clockwise palm kneading technique around the navel and abdomen with the Shenque acupoint as the center; the third movement involves the control unit controlling the integrated spherical contact to maintain the vertical posture, performing a counterclockwise palm kneading technique around the navel and abdomen with the Shenque acupoint as the center; the fourth movement involves the control unit controlling the integrated spherical contact to switch to the inclined posture, pushing directly from below the right breast to the spleen and liver meridian areas of the right groin, performing a palm pushing technique; the fifth movement involves the control unit controlling the integrated spherical contact to maintain the inclined posture, pushing directly from below the left breast to the spleen and liver meridian areas of the left groin, performing a palm pushing technique.

[0044] Furthermore, in the kneading and walking mode of the second and third modes, the control unit superimposes the linear movement trajectory and the circular kneading trajectory, so that the position coordinates of the integrated spherical contact at the current moment are equal to: the starting acupoint coordinates of the current kneading segment, the product of the movement direction vector and the current moment, and the trigonometric function bias of the kneading circumference radius and the current moment, so as to generate a sinusoidal or spiral massage trajectory that moves along the meridian massage path.

[0045] Furthermore, the control unit also integrates a respiratory tracking adaptive compensation algorithm; the control unit uses the image acquisition module to acquire the depth mean fluctuation sequence of the region of interest near the Shenque acupoint in real time, and extracts the respiratory waveform phase and respiratory component through a bandpass filter; the control unit multiplies the respiratory component by a gain coefficient to obtain the feedforward compensation amount, and superimposes the feedforward compensation amount into the normal position control command at the end of the massage actuator, so that the integrated spherical contact head follows the abdominal breathing fluctuations in real time and maintains a stable normal contact force.

[0046] Furthermore, the control unit is also equipped with multiple safety protection mechanisms, including: Force over-limit protection: when the six-dimensional force sensor detects that the contact force in any direction exceeds the preset safety force threshold, the control unit controls the six-axis robotic arm to stop feeding and automatically retract in the opposite direction; Deformation monitoring protection: when the image acquisition module detects in real time that the deformation of the abdominal surface exceeds the set deformation threshold and the duration exceeds the set time, the control unit controls the six-axis robotic arm to pause the current massage operation and issue a voice prompt; Emergency stop protection: the computer vision-based longevity nine-turn method floor-mounted abdominal massage robot is equipped with a mechanical emergency stop button. After triggering the mechanical emergency stop button, the power supply of the six-axis robotic arm is cut off, the joints of the six-axis robotic arm are simultaneously de-energized and the brake is activated to allow the integrated spherical contact head to be manually pushed back.

[0047] Furthermore, the control unit also includes a voice prompt module and a treatment management module. After completing the first cycle of the first eight movements of the Longevity Nine Turns Method massage path, the voice prompt module prompts the user to maintain the correct posture, and the treatment management module automatically cycles through the first eight movements of the Longevity Nine Turns Method massage path a set number of times. After completing the first eight movements of the Longevity Nine Turns Method massage path a set number of times, the control unit confirms that the six-axis robotic arm has returned to a safe distance, reminds the user to sit up through the voice prompt module, and guides the user to complete the upper body swaying finishing movement according to the standard of the ninth movement of the Longevity Nine Turns Method.

[0048] Compared with the prior art, the beneficial effects of the present invention are: (1) The present invention adopts a split design that physically separates the independent visual cantilever and the massage execution mechanism (six-axis robotic arm), realizing global static positioning of the eye outside the hand, and completely eliminating the interference of the robotic arm movement and vibration on visual positioning; at the same time, the wrist mechanism automatically adjusts the posture of the integrated spherical contact head, flexibly switching between vertical posture (point contact) and tilted posture (surface contact), and realizes automatic seamless switching of various compound TCM techniques such as finger kneading, palm kneading and palm pushing without the need for manual replacement of the massage head throughout the process, and fully and standardly automatically reproduces the operation process of the first eight steps of the Longevity Nine Turns Method.

[0049] (2) RGB-D dual-modal fusion is performed using multi-task convolutional neural networks (CNNs), and the modal weights can be dynamically adjusted according to the user's body mass index (BMI). For cases with insufficient confidence, an auxiliary positioning mechanism of voice guidance for users to lightly touch the finger is introduced, which effectively overcomes the problem of acupoint positioning in extreme boundary cases such as obesity, abdominal scars, and body hair occlusion. At the same time, the traditional Chinese medicine bone measurement method is innovatively transformed into a precise engineering coordinate algorithm, which can adaptively generate meridian paths according to the user's abdominal contour.

[0050] (3) The six-dimensional force sensor is integrated. The force and frequency parameters of each technique are calibrated based on the actual measurement data of biomechanics of professional TCM massage, ensuring that the mechanical characteristics applied by the robot are highly consistent with the real expert techniques. Furthermore, a vision-based breathing tracking adaptive compensation algorithm is integrated to extract the breathing component as feedforward compensation, so that the robotic arm follows the abdominal undulation in real time, ensuring that the normal massage contact force remains stable and is not disturbed by the user's breathing movement.

[0051] (4) In a specific massage pattern, the control unit generates a sinusoidal or spiral dynamic massage trajectory by mathematically superimposing the linear movement interpolation trajectory with the circular friction trajectory, perfectly reproducing the delicate technique of kneading and moving in traditional massage.

[0052] (5) It has a built-in complete treatment status machine and voice reminder system. It can not only automatically cycle through the first eight massages seven times, but also guide the user to complete the ninth movement (upper body rotation) after the robotic arm is safely returned to its position. This ensures that the massage service provided by the device is completely consistent with the clinically verified seven-times-one-times-longevity-nine-turns method operation standard.

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, embodiments of the present invention are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a schematic diagram of a floor-standing abdominal massage robot application; Figure 2 This is a schematic diagram of a multi-task CNN network architecture; Figure 3 This is a flowchart of the automatic acupoint recognition and positioning module. Figure 4 This is a schematic diagram of the anchor point location for key acupoints in the abdomen; Figure 5 This is a schematic diagram of the topological generation of the Ren Meridian, Kidney Meridian, and Stomach Meridian pathways in the abdomen. Figure 6 This is a schematic diagram of the first massage path (circular finger kneading method for the Juque acupoint); Figure 7 This is a schematic diagram of the second massage path (downward finger kneading method in the area between the Ren meridian and the bilateral Stomach meridians); Figure 8This is a schematic diagram of the third massage technique (upward finger kneading method in the bilateral kidney meridian-stomach meridian area); Figure 9 This is a schematic diagram of the fourth massage technique (downward finger kneading method in the area between the two sides of the Stomach Meridian). Figure 10 This is a diagram of the fifth massage technique (clockwise palm kneading around the navel). Figure 11 This is a schematic diagram of the sixth massage technique (counterclockwise palm kneading around the navel). Figure 12 This is a schematic diagram of the seventh massage technique (palm pushing method in the right spleen meridian area). Figure 13 This is a schematic diagram of the massage path for the eighth movement (palm pushing method in the left spleen meridian area). Figure 14 This is a schematic diagram comparing the postures of an integrated bionic spherical contact (vertical-kneading method, inclined-pushing method). Figure 15 This is a flowchart of the three-level force closed-loop control. Figure 16 This is a flowchart of hand-eye calibration and tactile zero-position calibration. Figure 17 This is a schematic diagram of a respiratory tracking adaptive compensation system; Figure 18 This is a logic block diagram of a multi-layered security protection system; Figure 19 This is the overall block diagram of the control system; Figure 20 This is a sequence diagram of voice reminders for treatment management (first round of holding the position, seventh round of completing the ninth posture reminder).

[0056] In the diagram: 1- Counterweight base; 2- Column; 3- Independent visual cantilever; 4- RGB-D depth camera; 5- LED light source assembly; 6- Six-axis robotic arm; 7- Torque sensor; 8- Variable posture integrated spherical contact head; 9- Touch screen all-in-one machine; 10- Physiotherapy bed. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0058] I. System Hardware Architecture and Split Design This invention provides a computer vision-based, floor-mounted abdominal massage robot for the Longevity Nine Turns Method. During use, no human intervention is required. The user lies supine on the therapy bed, and the device automatically completes the following steps after startup: calibration and verification of image acquisition and acupoint location (with voice guidance if necessary); tactile zero-position calibration; generation of the massage path; execution of the first to eighth massage movements in sequence; and voice prompts for the ninth movement to conclude the exercise.

[0059] Combination Figure 1 and Figure 19 As shown, the computer vision-based Longevity Nine Turns Abdominal Massage Robot includes: a floor-standing body, an independent vision cantilever, an image acquisition module, a massage execution mechanism, and a control unit.

[0060] The floor-standing unit includes a counterweight base (weighing no less than 35kg to ensure overall stability) and a support column.

[0061] The independent visual cantilever and the massage actuator are respectively mounted on the uprights of the floor-standing unit, and are physically completely separated. The independent visual cantilever has vertical lifting and tilting adjustment capabilities, and its working height is set at no less than 1.5 meters above the treatment bed or no less than 1 meter above the user's body surface. This split design, where the eye is outside the hand, ensures that the entire range of motion of the massage actuator is outside the effective depth of field boundary of the camera, completely eliminating the interference of mechanical vibration on visual positioning.

[0062] An image acquisition module is installed at the end of the independent vision cantilever and includes a depth camera and a light source assembly (such as an RGB-D depth camera and an LED light source). It is used to acquire image data (including color and depth images) of the user's abdomen and transmit it to the control unit.

[0063] The massage actuator includes a six-axis robotic arm, a wrist mechanism, and an integrated spherical contact head. The six-axis robotic arm (with a working radius of at least 900 mm) is mounted on the side of the floor-standing unit; the wrist mechanism (such as a micro servo motor or stepper motor drive mechanism) is mounted at the end of the six-axis robotic arm; and the integrated spherical contact head (made of medical-grade silicone, approximately 35 mm in diameter) is mounted on the wrist mechanism. Under control unit commands, the wrist mechanism can adjust the pitch and tilt of the integrated spherical contact head relative to the user's abdominal skin within a range of 0 to 60 degrees.

[0064] The control unit (such as a touch-screen all-in-one machine with a built-in embedded AI computing platform) is electrically connected to both the image acquisition module and the massage actuator. The control unit processes image data, identifies acupoints, generates massage paths, and sequentially controls the massage actuator according to a preset program of the first eight massage steps of the Longevity Nine Turns Method. Figure 14 As shown, during program execution, the control unit controls the wrist mechanism to switch the integrated ball contact to a different type based on the current number. Figure 14The vertical posture shown in (a) is perpendicular to the abdominal skin (i.e., 0, point contact, used for performing finger or palm kneading techniques), or can be switched to... Figure 14 The tilted posture shown in (b) is inclined to the abdominal skin (e.g., at a 40-50 degree angle to the normal, using the side to form surface contact for performing the palm pushing method). This allows for automatic switching of TCM compound techniques without changing the contact point throughout the entire process.

[0065] II. Visual Positioning and Acupoint Recognition System To achieve fully automatic positioning that can be used immediately upon lying down, such as Figure 2 As shown, the control unit has a built-in multi-task convolutional neural network model (based on a ResNet-50 fine-tuned architecture), which includes a feature extraction backbone and four task-specific output branches (Head1 to Head4).

[0066] When identifying acupoints on a user's abdomen: 1. Dynamic Modal Fusion: The model utilizes modal fusion weights to perform element-wise weighted summation and fusion of shallow feature maps extracted from color images and shallow feature maps extracted from depth images. These modal fusion weights are dynamically adjusted based on the user's Body Mass Index (BMI): for users with a low BMI (e.g., BMI 24) who are underweight, the weight of depth image features is increased (due to their prominent navel depression); for users with a high BMI (e.g., BMI > 28) who are obese and whose navel is filled with fat, the modal fusion weight of the corresponding shallow feature map from the color image is increased (depending on pigmentation features), while the weight of the depth image is decreased, thus ensuring detection accuracy.

[0067] 2. Multi-task parallel output: The fused feature map is input to the feature extraction backbone to extract multi-scale features, and outputs through four branches respectively: (1) the two-dimensional coordinates of key acupoints (such as Shenque acupoint) on the user's abdomen in the image plane; (2) the binary segmentation map of the abdominal boundary; (3) the depth value of the acupoint in the depth map; (4) the location confidence score.

[0068] Voice-guided assisted localization function: For the rare cases where obesity, body hair obscuring the location, or postoperative scarring results in a localization reliability score lower than a preset confidence threshold (e.g., 0.85), combined with... Figure 3 As shown, the control unit triggers the voice-guided assisted positioning function: it outputs a voice prompt to guide the user to lightly touch their Shenque acupoint (navel) with their finger; it uses an image acquisition module to acquire an image containing the finger, and obtains the three-dimensional coordinates of the contact point as candidate coordinates for the Shenque acupoint through a fingertip detection algorithm. The control unit compares the candidate coordinates with the coordinates detected by the model: if the deviation is less than or equal to a preset deviation threshold (e.g., 15mm), the coordinates detected by the model are accepted; if the deviation is greater than 15mm, the candidate coordinates indicated by the user's finger are taken as the final coordinates of the Shenque acupoint.

[0069] Projective coordinate calculation and meridian path generation: as Figure 4 shown, when calculating engineering coordinates, based on the binary segmentation map of the abdominal boundary output by the model, the distal point of the xiphoid process, the boundary point of the upper margin of the pubic symphysis, the most convex point of the left anterior superior iliac spine and the most convex point of the right anterior superior iliac spine are extracted. The control unit establishes a local abdominal coordinate system with the Shenque acupoint as the coordinate origin, and takes the horizontal distance between the most convex points of the left and right anterior superior iliac spines as the abdominal transverse width reference. Using the traditional Chinese medicine bone measurement method, one-twelfth of this width reference is calculated as the absolute length of one cun.

[0070] as Figure 5 shown, the control unit calculates the coordinates of related acupoints such as Juque, Zhongwan and Tianshu based on the absolute length of one cun, and automatically generates the circulation trajectories of the Ren channel (abdominal midline), Kidney channel (0.5 cun lateral to the midline), Stomach channel (2 cun lateral to the midline), Spleen channel (4 cun lateral to the midline) and Liver channel. Then, as Figure 16 shown, the above trajectory points are uniformly converted into absolute spatial coordinates under the robotic arm base coordinate system via the calibrated hand-eye matrix combined with tactile zero calibration data, and a continuous three-dimensional massage path is generated by interpolation.

[0071] III. Closed-loop force control and manipulation execution parameters The massage actuator further comprises a six-dimensional force sensor installed between the wrist mechanism and the integrated spherical contact head. The sensor collects the normal and tangential contact forces between the contact head and the abdomen in real time. Meanwhile, the base of the integrated spherical contact head is integrated with a heating assembly (such as an annular PTC heating element, maintaining 37°C-45°C) to achieve the combined effect of warm moxibustion and abdominal kneading.

[0072] as Figure 15 shown, the control unit adopts an incremental PID control algorithm (sampling frequency 100Hz) for closed-loop force control. For different manipulation methods, the system queries the parameter table according to user settings, and presets three levels of target force values: light, medium and heavy: Finger kneading (formula 1 to 3): the contact head is in a vertical posture (point contact), and executes in circular or translational motion. Target force values: light gear 3.5N, medium gear 12.0N, heavy gear 25.0N.

[0073] Palm kneading (formula 5 and 6): the contact head is in a vertical posture, and performs circular motion around the Shenque acupoint. Target force values: light gear 8.0N, medium gear 22.0N, heavy gear 45.0N (the biomechanical range of self-kneading abdomen is equivalent to the biomechanical parameters of doctors' rubbing manipulation, so the biomechanical data of palm rubbing manipulation in literature is used as the basis for force grading of self-kneading abdomen).

[0074] Palm Pushing Method (Fourth, Seventh, and Eighth Techniques): Switch the contact head to an inclined posture (using side surface contact). Slide and push along the meridian. Apply the set target force value during the pushing stroke (same as palm kneading method), and reduce the contact force to the preset return force threshold during the return stroke (e.g., reduce it to less than 1 / 3 of the set force during the pushing stroke).

[0075] IV. The massage path procedure and the algorithm of "rubbing and walking" for the first eight movements of the Longevity Nine Turns Method The control unit executes the following procedures automatically in sequence: First Form (as shown) Figure 6 (As shown): Control the integrated spherical contact head to maintain a vertical posture, position it at the Juque acupoint, and perform a clockwise circular (diameter 4-6cm) finger kneading method for a total of 21 circles.

[0076] Second form (as shown) Figure 7 (As shown): Control the integrated spherical contact head to maintain a vertical posture, and rub down from the Juque acupoint along the Ren meridian and the bilateral kidney meridian areas to the Qugu acupoint. Use a rubbing and walking motion to perform the finger rubbing method, repeating 21 times.

[0077] The third form (such as) Figure 8 (As shown): Control the integrated spherical contact head to maintain a vertical posture, and knead from the Qugu acupoint to both sides upwards, using a kneading and walking motion along the bilateral kidney and stomach meridian areas to perform finger kneading to the Juque acupoint, alternating left and right sides 21 times each.

[0078] Fourth Form (such as) Figure 9 (As shown): Control the integrated spherical contact head to switch to an inclined posture, push it straight from the Juque acupoint along the Ren meridian and the bilateral kidney meridian areas to the Qugu acupoint, perform the palm pushing method, and repeat 21 times.

[0079] Fifth Form (such as) Figure 10 (As shown): Control the integrated spherical contact head to switch to a vertical posture, and perform a palm kneading method in a clockwise direction around the navel and abdomen with the Shenque acupoint as the center, for a total of 21 circles.

[0080] The sixth form (such as) Figure 11 (As shown): Control the integrated spherical contact head to maintain a vertical posture, and perform a counterclockwise palm kneading motion around the navel and abdomen with the Shenque acupoint as the center, for a total of 21 circles.

[0081] Seventh Form (as Figure 12 (As shown): Control the integrated spherical contact head to switch to an inclined posture, push it directly from under the right breast to the spleen and liver meridian areas of the right groin, perform the palm pushing method, and repeat 21 times.

[0082] Eighth Form (as shown) Figure 13 (As shown): Control the integrated spherical contact head to maintain an inclined posture, push it from below the left breast to the spleen and liver meridian areas of the left groin, perform the palm pushing method, and repeat 21 times.

[0083] In this embodiment, for the combined kneading and walking motion of the second and third movements, the control unit performs vector superposition of the linear movement trajectory and the circular kneading trajectory in kinematic planning. Specifically, the position coordinates of the integrated spherical contact head at the current moment are equal to the sum of three variables: the first variable is the coordinates of the starting acupoint of the current kneading segment; the second variable is the product of the movement direction vector and the current moment (reflecting meridian movement); and the third variable is the trigonometric function bias (sine and cosine, reflecting in-situ kneading) of the kneading circle radius and the current moment. Through this motion control algorithm, a sinusoidal or spiral massage trajectory moving along the meridian massage path is generated, highly replicating the essence of traditional Chinese massage.

[0084] V. Respiratory Tracking Adaptive Compensation Algorithm To prevent sudden changes in contact force caused by abdominal movement due to the user's breathing, combined with Figure 17 As shown, the control unit integrates a respiratory tracking adaptive compensation algorithm. The image acquisition module uses a high frame rate (e.g., 30fps) to acquire the depth mean fluctuation sequence of a region of interest (e.g., 10cm x 10cm) near the Shenque acupoint in real time. The respiratory waveform phase and respiratory component are extracted using a bandpass filter (e.g., a passband of 0.15Hz-0.5Hz, covering the resting respiratory frequency of adults). The control unit multiplies this respiratory component by the lung-abdominal motion gain coefficient to calculate the feedforward compensation. This feedforward compensation is superimposed on the normal position control command of the Z-axis at the end of the robotic arm, enabling the integrated spherical contact head to follow the abdominal respiratory fluctuations in real time, keeping the steady-state error of the normal contact force within a minimal range, unaffected by respiratory motion.

[0085] VI. Multiple security protection mechanisms To ensure absolute user safety, such as Figure 18 As shown, the control unit is equipped with three independent and parallel safety protection mechanisms: 1. Force over-limit protection: When the six-dimensional force sensor detects that the contact force in any direction exceeds the preset safety force threshold (e.g., 15N), the control unit controls the six-axis robotic arm to stop feeding within a very short time (e.g., 20ms) and automatically retract to a safe distance (e.g., 10mm).

[0086] 2. Deformation monitoring and protection: When the image acquisition module detects in real time that the deformation of the point cloud on the abdominal surface exceeds the set deformation threshold (e.g., 30mm) and the duration exceeds the set duration (e.g., 0.5 seconds) (indicating that the user's body is twisting abnormally or intends to leave), the control unit controls the robotic arm to immediately pause the current massage operation and issue a voice prompt.

[0087] 3. Emergency Stop Protection: The equipment control panel is equipped with a mechanical emergency stop button. Once triggered by the user, the power supply to the six-axis robotic arm will be immediately cut off. At this time, the joint motors of the six-axis robotic arm will lose power synchronously and activate the holding brake. Meanwhile, its Z-axis lead screw adopts a non-self-locking design, allowing the user to manually push the integrated ball contact back to the safe position in the power-off state.

[0088] VII. Treatment Course Management The control unit has a built-in voice prompt module and a treatment management module. For example... Figure 20 As shown, after completing the first eight movements of the Longevity Nine Turns Method, the voice prompt module reminds the user to maintain a supine position. Subsequently, the treatment management module automatically cycles through the remaining movements until the total number of movements set by the user is reached. After all the set number of cycles (e.g., the seventh cycle) is completed, the control unit, after confirming that the six-axis robotic arm has fully returned to a safe distance (more than 30cm from the user's body surface), reminds the user to sit up through the voice prompt module and guides the user to complete the ninth movement (sitting cross-legged, hands clenched, upper body rotated 21 times clockwise and 21 times counterclockwise) according to the standard movements to conclude the exercise, using a combination of voice and text.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A computer vision-based nine-turn method of prolonging life fall type abdominal massage robot, characterized in that, The device includes a floor-standing body, an independent visual cantilever, an image acquisition module, a massage actuator, and a control unit. The independent visual cantilever and the massage actuator are respectively mounted on the floor-standing body, with the independent visual cantilever physically separated from the massage actuator. The image acquisition module is mounted on the independent visual cantilever and is used to acquire image data of the user's abdomen and transmit it to the control unit. The massage actuator includes a six-axis robotic arm, a wrist mechanism, and an integrated spherical contact head. The six-axis robotic arm is mounted on the floor-standing body, the wrist mechanism is mounted at the end of the six-axis robotic arm, and the integrated spherical contact head is mounted on the wrist mechanism. The wrist mechanism is used to adjust the integrated spherical contact head relative to the user's abdominal skin. The control unit is electrically connected to the image acquisition module and the massage execution mechanism respectively. The control unit is used to identify the user's abdominal acupoints based on the image data and generate a massage path. According to the preset massage path program of the first eight movements of the Longevity Nine Turns Method, the control unit controls the wrist mechanism to switch the integrated spherical contact head to a vertical posture perpendicular to the abdominal skin to perform a kneading motion using the end of the integrated spherical contact head, or to switch the integrated spherical contact head to an inclined posture tilted to the abdominal skin to perform a pushing motion using the side of the integrated spherical contact head.

2. The computer vision-based longevity nine-turn abdominal massage robot according to claim 1, characterized in that, The image acquisition module includes a depth camera and a light source component. The image data includes color images and depth images. The control unit has a built-in multi-task convolutional neural network model, which includes a feature extraction backbone and multiple task-specific output branches. When identifying acupoints on the user's abdomen, the multi-task convolutional neural network model uses modal fusion weights to perform weighted summation and fusion of the shallow feature maps extracted from the color image and the shallow feature maps extracted from the depth image. The fused feature maps are then input to the feature extraction backbone to extract multi-scale features. The multiple task-specific output branches output in parallel the two-dimensional coordinates of the user's key abdominal acupoints on the image plane, the binary segmentation map of the abdominal boundary, the depth value of the user's key abdominal acupoints in the depth image, and the location confidence scores of each of the user's key abdominal acupoints, including the Shenque acupoint. The modal fusion weights are dynamically adjusted according to the user's body mass index (BMI). When the user's BMI is high, the modal fusion weights of the shallow feature maps corresponding to the color image are increased, while the modal fusion weights of the shallow feature maps corresponding to the depth image are decreased.

3. The computer vision-based longevity nine-turn abdominal massage robot according to claim 2, characterized in that, The control unit is equipped with a voice-guided assisted positioning function. When the location confidence score of the Shenque acupoint output by the multi-task convolutional neural network model is lower than a preset confidence threshold, the control unit outputs a voice prompt to guide the user to lightly touch the Shenque acupoint with their finger. The image acquisition module acquires an image containing the user's finger and performs fingertip detection to obtain the three-dimensional coordinates of the contact point between the finger and the abdomen as candidate coordinates of the Shenque acupoint. The control unit compares the candidate coordinates of the Shenque acupoint with the coordinates of the Shenque acupoint detected by the multi-task convolutional neural network model. When the deviation between the two is less than or equal to a preset deviation threshold, the coordinates of the Shenque acupoint detected by the multi-task convolutional neural network model are accepted. When the deviation between the two is greater than the preset deviation threshold, the candidate coordinates of the Shenque acupoint are used as the final coordinates of the Shenque acupoint.

4. The computer vision-based longevity nine-turn abdominal massage robot according to claim 2, characterized in that, The abdominal boundary represented by the binary segmentation map of the abdominal boundary detected by the multi-task convolutional neural network model includes the xiphoid process tip, the upper edge boundary of the pubic symphysis, the most prominent point of the left anterior superior iliac spine, and the most prominent point of the right anterior superior iliac spine. The control unit establishes a local coordinate system of the abdomen with the Shenque acupoint as the origin, and uses the horizontal distance between the detected most prominent points of the left and right anterior superior iliac spines as the reference for the transverse width of the abdomen. The absolute length of one inch is calculated as one-twelfth of the reference for the transverse width of the abdomen. Based on the absolute length of one inch and the bone measurement method, the control unit calculates the engineering coordinates of other meridians and acupoints, and automatically generates the circulation trajectories of the Ren meridian, kidney meridian, stomach meridian, spleen meridian, and liver meridian in the abdominal region. The circulation trajectories are converted into spatial coordinates in the robotic arm base coordinate system through the hand-eye calibration matrix, forming a massage path for the massage execution mechanism to perform.

5. The computer vision-based longevity nine-turn abdominal massage robot according to claim 1, characterized in that, The massage actuator also includes a six-dimensional force sensor installed between the wrist mechanism and the integrated spherical contact head, the six-dimensional force sensor being used to collect the contact force between the integrated spherical contact head and the user's abdominal skin in real time; The base of the integrated spherical contact head integrates a heating component; the kneading method includes finger kneading and palm kneading, and the pushing method includes palm pushing; the control unit performs closed-loop force control based on an incremental proportional-integral-derivative control algorithm, and sets different multi-level target force values ​​for the finger kneading, palm kneading, and palm pushing methods respectively; when performing the kneading method, the integrated spherical contact head forms point contact with the user's abdominal skin in the vertical posture, performing circular or linear translational motion; when performing the pushing method, the integrated spherical contact head forms surface contact with the user's abdominal skin using the side of the sphere in the tilted posture, and slides along the massage path; and the set target force value is applied during the pushing stroke of the palm pushing method, and the contact force drops to a preset return force threshold during the return stroke.

6. The computer vision-based longevity nine-turn abdominal massage robot according to claim 5, characterized in that, The massage path procedure for the first eight movements of the Longevity Nine Turns Method specifically includes the first to eighth movements performed sequentially: First technique: The control unit controls the integrated spherical contact to maintain the vertical posture, position it at the Juque acupoint, and perform a clockwise circular finger kneading method; Second technique: The control unit controls the integrated spherical contact to maintain the vertical posture, and rubs down from the Juque acupoint along the Ren meridian and the bilateral kidney meridian areas to the Qugu acupoint, using a rubbing and walking method to perform the finger rubbing technique; Third step: The control unit controls the integrated spherical contact to maintain the vertical posture, and kneads upwards from the Qugu acupoint to both sides, using a kneading and walking method along the bilateral kidney and stomach meridian areas to the Juque acupoint; Fourth move: The control unit controls the integrated spherical contact to switch to the tilted posture, and pushes it from the Juque acupoint along the Ren meridian and the bilateral kidney meridian areas to the Qugu acupoint, performing the palm pushing method; Fifth technique: The control unit controls the integrated spherical contact to switch to the vertical posture, and performs a palm kneading method of rubbing the navel and abdomen in a clockwise direction with the Shenque acupoint as the center; Sixth technique: The control unit controls the integrated spherical contact to maintain the vertical posture and performs a counterclockwise palm kneading technique around the navel and abdomen with the Shenque acupoint as the center; Seventh move: The control unit controls the integrated spherical contact to switch to the tilted posture, pushing directly from below the right breast to the spleen and liver meridian areas of the right groin, performing the palm pushing method; Eighth move: The control unit controls the integrated spherical contact to maintain the tilted posture and pushes it from below the left breast to the spleen and liver meridian areas of the left groin, performing the palm pushing method.

7. The computer vision-based longevity nine-turn abdominal massage robot according to claim 6, characterized in that, In the kneading and walking mode of the second and third modes, the control unit superimposes the linear movement trajectory and the circular kneading trajectory, so that the position coordinates of the integrated spherical contact head at the current moment are equal to: the starting acupoint coordinates of the current kneading segment, the product of the movement direction vector and the current moment, and the trigonometric function bias of the kneading circumference radius and the current moment, so as to generate a sinusoidal or spiral massage trajectory that moves along the meridian massage path.

8. The computer vision-based longevity nine-turn abdominal massage robot according to claim 1, characterized in that, The control unit also integrates a respiratory tracking adaptive compensation algorithm; the control unit uses the image acquisition module to acquire the depth mean fluctuation sequence of the region of interest near the Shenque acupoint in real time, and extracts the respiratory waveform phase and respiratory component through a bandpass filter; the control unit multiplies the respiratory component by a gain coefficient to obtain the feedforward compensation amount, and adds the feedforward compensation amount to the normal position control command at the end of the massage actuator, so that the integrated spherical contact head follows the abdominal breathing fluctuations in real time and maintains a stable normal contact force.

9. The computer vision-based longevity nine-turn abdominal massage robot according to claim 5, characterized in that, The control unit is also equipped with multiple safety protection mechanisms, including: ① Force over-limit protection: When the six-dimensional force sensor detects that the contact force in any direction exceeds the preset safety force threshold, the control unit controls the six-axis robotic arm to stop feeding and automatically retract in the opposite direction; ② Deformation monitoring protection: When the image acquisition module detects in real time that the deformation of the abdominal surface exceeds the set deformation threshold and the duration exceeds the set time, the control unit controls the six-axis robotic arm to pause the current massage operation and issue a voice prompt; ③ Emergency stop protection: The computer vision-based longevity nine-turn abdominal massage robot is equipped with a mechanical emergency stop button. After triggering the mechanical emergency stop button, the power supply of the six-axis robotic arm is cut off, the joints of the six-axis robotic arm are simultaneously de-energized and the brake is activated to allow the integrated spherical contact head to be manually pushed back.

10. The computer vision-based longevity nine-turn abdominal massage robot according to claim 1, characterized in that, The control unit also has a built-in voice prompt module and a treatment management module. After completing the first round of the first eight movements of the Longevity Nine Turns Method massage path, the voice prompt module prompts the user to maintain the body position, and the treatment management module automatically cycles through the first eight movements of the Longevity Nine Turns Method massage path for a set number of times. After completing the first eight movements of the Longevity Nine Turns Method massage path for a set number of times, the control unit confirms that the six-axis robotic arm has returned to a safe distance, reminds the user to sit up through the voice prompt module, and guides the user to complete the upper body swaying finishing movement according to the standard of the ninth movement of the Longevity Nine Turns Method.

Images