Intelligent walking aid method and device for sarcopenia rehabilitation

Abstract

The invention discloses an intelligent walking aid training method and device for sarcopenia rehabilitation. The intelligent walking aid training method comprises the steps that grip strength data of a rehabilitation person is collected in real time through a grip strength sensor; collecting pressure data applied to the armrest by the rehabilitation person through a pressure sensor; acquiring real-time image data of a rehabilitation person through a camera module; based on the collected multi-modal data, adaptive damping control, human body posture recognition and fall risk prediction are executed; training parameters are adjusted according to the analysis result, and real-time feedback is provided through a display screen. According to the method, data results recognized by a sensor and an algorithm can be fully utilized, the fall probability of the rehabilitation person can be predicted through an improved Transform structure according to the data, and if the fall probability is larger than a threshold value, the outside world can be reminded to intervene in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical rehabilitation equipment, in particular to an intelligent walking aid method and device for sarcopenia rehabilitation. BACKGROUND

[0002] With the intensification of social aging, it is estimated that the proportion of the elderly population will exceed 20% in 2035, and sarcopenia has gradually become an important problem affecting the quality of life of the elderly. Sarcopenia patients usually show insufficient muscle strength, difficulty walking, and decreased balance ability, increasing the risk of falling. Studies have shown that long-term regular rehabilitation training can effectively improve the muscle strength and walking balance ability of elderly sarcopenia patients. In view of this demand, it is necessary to develop an intelligent walking aid system suitable for sarcopenia rehabilitation. However, the current general rehabilitation walking frame solution has problems such as complex structure, inconvenient operation, etc. for elderly sarcopenia patients. The present application provides an intelligent walking aid system for sarcopenia rehabilitation for elderly sarcopenia patients, which can greatly reduce the influence of complex external structures on the rehabilitation of elderly sarcopenia patients, and the operation method is simple, which improves the efficiency of rehabilitation to a certain extent. SUMMARY

[0003] To overcome the shortcomings of the prior art, the present application proposes an intelligent walking aid method and device for sarcopenia rehabilitation, which provides real-time feedback for the rehabilitation process, helps the rehabilitation patient to understand the current exercise state, and provides safe and intelligent rehabilitation equipment by combining the basic support frame body and sensors and other intelligent devices.

[0004] To achieve the above-mentioned purpose, the present application proposes an intelligent walking aid training method for sarcopenia rehabilitation, comprising the following steps:

[0005] S1: Real-time acquisition of grip strength data of rehabilitation personnel by grip sensor;

[0006] S2: Acquisition of pressure data applied by the rehabilitation personnel on the handrail by the pressure sensor;

[0007] S3: Acquisition of real-time image data of the rehabilitation personnel by the camera module;

[0008] S4: Based on the collected multi-modal data, adaptive damping control, human body posture recognition and fall risk prediction are performed;

[0009] S5: Adjusting the training parameters according to the analysis results and providing real-time feedback through the display screen.

[0010] Further, the adaptive damping control method in step S4 comprises:

[0011] Y1: Real-time acquisition of the current resistance value of the wheel;

[0012] Y2: Synchronize the collection of grip sensor values, pressure sensor average values, and movement time, and perform normalization processing;

[0013] Y3: Every five seconds, based on the normalized grip, pressure, and movement time data, calculate the comprehensive ability evaluation value of the rehabilitation personnel using a weighted fusion method, with the pressure weight set to 0.4, and the grip and movement time weights each set to 0.3;

[0014] Y4: Calculate the deviation of the current ability evaluation value from the preset target value, and the change rate relative to the previous moment;

[0015] Y5: Based on the deviation and change rate, dynamically adjust the wheel resistance to achieve adaptive control of the rehabilitation training intensity;

[0016] The human posture recognition method in step S4 includes:

[0017] S4-1: Obtain real-time color video streams through a camera, and extract each frame of picture;

[0018] S4-2: Input the frame picture into the improved YOLOv8 human upper body key point recognition algorithm;

[0019] S4-3: Identify nine key points of the human upper body: left shoulder, right shoulder, shoulder midpoint, left hip, right hip, hip midpoint, left elbow, right elbow, and spine midpoint;

[0020] S4-4: Visualize the key point results on the display screen for the rehabilitation personnel to correct their posture.

[0021] Further, the improved YOLOv8 algorithm uses an SFB module to replace the original basic module, and the processing steps of the SFB module include:

[0022] Step A: The input features are transformed through a 1x1 convolution;

[0023] Step B: After depth separable convolution, including 3x3 depth convolution and 1x1 point convolution;

[0024] Step C: Obtain output features through a second 1x1 convolution;

[0025] Step D: Introduce residual connection to add the input and output, reducing information loss.

[0026] Further, a TAU module is added before the model output for feature enhancement, including:

[0027] Channel dimension reduction is performed on the feature map, with a reduction ratio of r;

[0028] Layer normalization processing is performed on the reduced features;

[0029] Two branches are processed: the first branch passes through a fully connected layer and a ReLU activation function, and the second branch passes through a fully connected layer and a Sigmoid activation function;

[0030] The outputs of the two branches are multiplied element by element, and the channel attention weight is obtained through a fully connected layer.

[0031] Further, the fall risk prediction method in step S4 comprises:

[0032] Step P1: reduce the human posture two-dimensional information matrix to a one-dimensional vector;

[0033] Step P2: splice the posture vector with grip data and pressure data to form a 20-dimensional multi-modal input vector;

[0034] Step P3: add position encoding and input into the Transformer encoder;

[0035] Step P4: process and output the fall probability value through the decoder;

[0036] Step P5: trigger an alarm when the probability value is greater than 65%.

[0037] Further, the fall risk prediction adopts an improved Transformer architecture, which processes multi-modal input data in parallel through 5 attention heads, wherein:

[0038] The human posture, grip and pressure data are fused as a unified input;

[0039] The encoder-decoder structure is used to extract temporal features and inter-modal correlations;

[0040] The output layer generates a fall risk probability value between 0 and 1, and triggers an early warning when the probability exceeds 0.65.

[0041] Further, the encoder and decoder each include a feedforward network module, which adopts a two-layer fully connected network structure, with the middle layer dimension expanded to 4 times the input dimension, and uses a ReLU activation function to realize nonlinear transformation.

[0042] Further, it further comprises an abnormal situation detection step:

[0043] When it is detected that the grip data has decreased by more than 30% in a short period of time, it is determined that the upper limb muscle is fatigued;

[0044] When it is detected that the rehabilitation personnel continuously maintain the midpoint of the spine deviating from the center line by more than 15 degrees for 30 seconds, it is determined that the posture is abnormal;

[0045] When an abnormality is detected, a video is automatically recorded and a reminder message is sent;

[0046] Further comprising the step of evaluating the lower limb muscle state through electromyography:

[0047] Collecting surface electromyography of the calf;

[0048] Calculating the root mean square value of the electromyography to evaluate the muscle activation level;

[0049] According to the root mean square value, it is judged whether the lower limb muscle force is in place, and when the root mean square value is lower than the preset threshold, it is prompted to strengthen the training intensity;

[0050] The evaluation results are stored for the reference of doctors.

[0051] An intelligent walking aid device for sarcopenia rehabilitation is suitable for the method, and the structure comprises a support frame body (1) made of lightweight aluminum alloy material and having a foldable structure. A grip strength sensor (2) is installed at the handle of the support frame body (1) to reflect the change of the hand strength of the rehabilitation personnel, and an SOS button (8) is also installed for emergency call.

[0052] A main control unit (12) is installed at the left handle of the support frame body (1) to receive data transmitted by a Bluetooth module.

[0053] An LED three-color indicator light (9), a buzzer (10) and a detachable display screen (3) are arranged on the cross beam of the support frame body (1), a camera module (13) is embedded beside the detachable display screen (3) to shoot the road conditions in front, and a connecting rod (6) is used to connect the structure at the lower part of the support frame body (1), and a detachable weight-bearing frame (11) is arranged in the middle part of the connecting rod (6) to add weights to realize weight-bearing training.

[0054] A pressure sensing assembly (7) is arranged at the upper and lower joints of the support frame body (1) to detect the posture and center of gravity.

[0055] A damping adjuster (23) is arranged at the bottom of the support frame body (1) to control the resistance of the wheels below.

[0056] Further, the pressure sensing assembly (7) comprises a striking ball (14), a spring (15) and a pressure sensing block (16), the striking ball (14) is installed in the upper half pipe body (16) of the support frame body (1), the two ends of the striking ball (14) are connected with the springs (15), the end of the spring (15) away from the striking ball is connected with the pressure sensing block (18), and the pressure sensing block (18) is fixed on the inner wall of the upper half pipe body (16).

[0057] When the rehabilitation personnel exerts force downward, the pressure sensing block (18) button (19) in the lower half pipe body (17) will be triggered, and when the downward pressure disappears, the spring (15) will pop up and return to the initial state.

[0058] The grip sensor (2) comprises an upper surface (20) and a lower surface (21), when the rehabilitation personnel hold the upper surface (20) and the lower surface (21) of the handle at the position where the grip sensor (2) is located at the same time, and apply pressure at the same time, the grip sensor (2) will record the detailed data of the grip of the rehabilitation personnel;

[0059] The damping adjuster (23) is connected with the wheel (5) and the support frame body (1) at both ends respectively, and the resistance is generated by adjusting the knob (22) of the damping adjuster clockwise, so that the training needs of the rehabilitation personnel at different stages are met.

[0060] Compared with the prior art, the beneficial effects of the present application are:

[0061] 1. The present application provides an intelligent walking aid method and device for sarcopenia rehabilitation, a pressure sensor is installed in the inside of the support frame body, and a grip sensor is installed at the handle, so that the posture and gravity detection of the rehabilitation personnel can be monitored, and the possibility of falling of the rehabilitation personnel during the rehabilitation process is reduced.

[0062] 2. The present application provides an intelligent walking aid method and device for sarcopenia rehabilitation, a damping roller is installed below the support frame body of the rehabilitation walking aid frame, and the rehabilitation personnel can customize the damping coefficient by the adjustable knob, so that the rehabilitation training with different resistance can be realized.

[0063] 3. The present application provides an intelligent walking aid method and device for sarcopenia rehabilitation, which can capture the coordinate information of the upper body limbs of the human body in real time through the camera, and display the visual results in real time through the display screen, which is helpful for the rehabilitation personnel to correct the bad posture in time.

[0064] 4. The present application provides an intelligent walking aid method and device for sarcopenia rehabilitation, a display screen interface is added, which can provide a visual interface for the standing time, grip condition and posture stability of the patient during the movement, and an SOS button is also provided, so that the patient can make an emergency call in time when an emergency occurs. DETAILED DESCRIPTION

[0065] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0066] Figure 1 It is a flowchart of the present application

[0067] Figure 2Figure for the intelligent walking aid device for sarcopenia rehabilitation of the application;

[0068] Figure 3 Figure for the installation position of the pressure sensor of the application;

[0069] Figure 4 Figure for the installation position of the grip sensor;

[0070] Figure 5 Figure for the installation position of the damping adjuster;

[0071] Figure 6 Improved yolov8 human upper body key point recognition algorithm;

[0072] Figure 7 Figure for the algorithm detection steps;

[0073] Figure 8 Figure for the fall probability prediction network structure

[0074] In the figure: support frame body 1, grip sensor 2, detachable display screen 3, height adjustment hole 4, wheel 5, connecting rod 6, pressure sensing assembly 7, SOS emergency button 8, LED three-color indicator light 9, buzzer 10, detachable weight support 11, main control unit 12, camera module 13, impact ball 14, spring 15, pressure sensing block 16, lower half pipe body 17, pressure sensing block 18, button 19, upper surface 20, lower surface 21, knob 22 and damping adjuster 23. DETAILED DESCRIPTION

[0075] The technical solutions of the application will be described more clearly and completely by combining the drawings and the description of the preferred embodiments of the application.

[0076] As Figure 1 shown, the intelligent walking aid training system for sarcopenia rehabilitation of the application realizes intelligent rehabilitation training through the cooperative work of multiple sensors. The system first acquires the real-time state information of the rehabilitation personnel through four data acquisition modules, including the grip data collected by the grip sensor, the handrail pressure data collected by the pressure sensor, the human image data captured by the camera module, and the lower limb electromyographic signal collected by the electromyographic signal detection module. These multi-source heterogeneous data are gathered to the main control unit for multi-modal data fusion processing.

[0077] The main control unit, as the core processing center of the system, sends the received data to three parallel algorithm processing modules. The adaptive damping control module calculates the comprehensive ability value of the rehabilitation personnel every five seconds based on the normalized grip strength, pressure, and movement time data. By analyzing the deviation and rate of change of the ability value from the target value, the resistance of the walker wheels is dynamically adjusted to achieve personalized adaptation of training intensity. The human posture recognition module uses an improved YOLOv8 algorithm to detect the 9 key points of the upper body in real time, including shoulders, hips, elbows, shoulder midpoint, hip midpoint, and spine midpoint, providing accurate skeletal positioning information for posture evaluation. The fall risk prediction module uses a Transformer architecture to process the fused multi-modal data through 5 attention heads in parallel. When the predicted fall probability exceeds the 65% threshold, the system will immediately trigger the warning mechanism.

[0078] The processing results of the three algorithm modules are sent to the abnormal situation detection module, which continuously monitors two types of key abnormalities: when the grip strength data decreases by more than 30% in a short period of time, it is determined to be upper limb muscle fatigue; when the posture of the rehabilitation personnel deviates from the center line by more than 15 degrees and lasts for more than 30 seconds, it is determined to be a posture anomaly. The system performs corresponding output actions based on different processing results, including real-time adjustment of wheel resistance through the damping adjuster, providing visual feedback information on the display screen, issuing a sound alarm through the buzzer, and automatically recording a video when an abnormal situation is detected. All training data and evaluation results are finally stored in the rehabilitation database, providing objective evidence for doctors' subsequent diagnosis and rehabilitation program adjustment. The entire system forms a complete closed loop from data acquisition, intelligent analysis, abnormality detection to execution feedback, achieving intelligent and personalized management of sarcopenia patient rehabilitation training.

[0079] As a specific embodiment, as shown in Figure 2 The hardware structure includes a support frame body 1 made of lightweight aluminum alloy material with a foldable structure. The support frame body 1 is installed with a grip strength sensor 2 at the handle for reflecting the hand strength changes of the rehabilitation personnel, and an SOS button 8 for emergency call. A main control unit 12 is installed at the left handle for receiving data transmitted by the Bluetooth module. The support frame body 1 is provided with an LED three-color indicator light 9, a buzzer 10, and a detachable display screen 3 on the cross beam. A camera module 13 is embedded beside the screen for shooting the road conditions in front. A connecting rod 6 is used to connect the structure at the lower part of the walker frame, and a detachable weight rack 11 is provided in the middle part of the connecting rod for adding weights when needed for weight training. A pressure sensing component 7 is assembled at the upper and lower interfaces of the support frame body 1 for posture and gravity center detection. A damping adjuster 23 is assembled at the bottom of the support frame body 1 to control the resistance of the wheels 5 below.

[0080] Referring to Figure 3 , in order to reduce the interference of the surface sensor to the rehabilitation personnel, the present application installs the impact ball 14 in the upper half pipe body 16 of the support frame body 1, which will trigger the pressure sensor 18 button 19 in the lower half pipe body 17 when the rehabilitation personnel forces downward, and the spring 15 will pop up and return to the initial state when the downward pressure disappears.

[0081] Referring to Figure 4 , the grip sensor is embedded in the handle, and when the rehabilitation personnel holds the upper surface 20 and the lower surface 21 of the handle at the same time, the grip sensor 2 will record the detailed data of the grip of the rehabilitation personnel.

[0082] Referring to Figure 5 , the damping adjuster 23 is connected with the wheel 5 and the support frame body, and by adjusting the knob 22 of the damping adjuster clockwise, the resistance is generated to meet the training needs of the rehabilitation personnel at different stages.

[0083] The main control unit has user data receiving and storage functions, can display the data obtained from the sensor on the display screen, and can continuously track the training effect of the patient and generate a rehabilitation report.

[0084] The system adds an adaptive damping control algorithm, which will be used when the rehabilitation personnel selects the automatic damping adjustment state on the display screen, and the algorithm process includes:

[0085] Y1: The main control unit determines and obtains the resistance P of the wheel at this time t ;

[0086] Y2: Obtain the grip sensor value, the average value of the pressure sensor, and the movement time, and normalize them to obtain the final parameters F w , F y and T respectively;

[0087] Y3: Calculate once every five seconds, and the rehabilitation personnel's ability value U t , take the difference between the ability value and the target ability value U p , and take the difference with the last measured ability value U t-1 ;

[0088] Y4: Set the reward coefficient and α, combine the difference obtained in the last step and the resistance P of the wheel t+1 , and calculate the new resistance, and the main control unit adjusts the resistance according to the result.

[0089] The calculation formula of the rehabilitation personnel's ability value in the process is shown in formula (1):

[0090] U t = w1·Fw +w2·F y +w3·T(1)

[0091] Further, the resistance size that needs to be updated is obtained as shown in formula (2):

[0092]

[0093] The system adds an improved yolov8 upper body key point recognition algorithm, can capture nine key point information of the upper body of the human body in real time through the camera module, including left shoulder, right shoulder, shoulder midpoint, left hip, right hip, hip midpoint, left elbow, right elbow and spine midpoint, and the relative position information is transmitted into the main control unit, and the main control unit prompts on the display screen interface. Figure 7 The detection process thereof comprises the following steps:

[0094] S1: acquiring real-time color video stream through a camera, and extracting pictures of each frame;

[0095] S2: taking the frame picture as the input of the improved yolov8 upper body key point recognition algorithm;

[0096] S3: obtaining a detection result, and displaying the human body key point information visualization result on the display screen, so that the rehabilitation personnel can better observe the current body state, and if the rehabilitation personnel observes that the posture is not good at this time, the current poor posture can be corrected in time.

[0097] Referring to Figure 5 In the application, the original basic module is replaced by an SFB module, which first obtains features through a Conv operation with a kernel size of 3x3 and a step size of 1, then inputs the features into a depth separable convolution (DW Conv), including a depth convolution operation with a kernel size of 5x5 and a step size of 1 and a point convolution operation with a kernel size of 1x1 and a step size of 1, and finally obtains the final output through a Conv operation with a kernel size of 3x3 and a step size of 1, while introducing a residual structure to reduce the probability of learning errors of the model. Compared with the traditional scheme, the method of the application can reduce the parameters and calculation amount of the model and improve the frame rate of real-time inference.

[0098] Before the model output, a TAU module is added. The module first reduces the feature map information (dimension size is (bs, c, h, w)) to a lower dimension, and the vector dimension becomes (bs, c, h x w) after reduction. Then, the set of vectors is subjected to a layer normalization operation (LayerNorm). The normalized information enters two branches for feature enhancement, respectively. The two branches are subjected to a full connection operation (Linear), respectively. The output of the first branch is subjected to a ReLu function to realize nonlinearization, and the output of the second branch is subjected to a Sigmoid function operation to obtain the importance degree of each channel. The output results of the two paths are multiplied, and finally a full connection layer is used to obtain the module output. Compared with the traditional visual Transformer module, the method can avoid introducing a large number of parameters, and can also analyze the importance degree of different channels of the features through the multi-branch structure.

[0099] Table 1

[0100]

[0101] Referring to Table 1, it can be seen that the method achieves good performance in the AP50 index, verifying the effectiveness of the method. In addition, the parameter amount of the method is the lowest among several comparison methods, which can also play a greater help in algorithm deployment.

[0102] Referring to Figure 8 , the system can construct multi-modal input data based on human skeleton key point data, pressure sensor data and grip sensor data, and realize fall probability prediction through an improved Transformer. For the two-dimensional human posture information obtained by the algorithm, the information matrix size is 7 x 2. First, it is reduced to one-dimensional data with a size of 1 x 14. Then, the one-dimensional data is spliced with the grip sensor data (data dimension is 1 x 2) and the pressure sensor data (data dimension is 1 x 4) to serve as the input of the model (data dimension is 1 x 20). The input information is first subjected to position information coding, and then passes through the encoder and decoder modules. The encoder and decoder structures are consistent, and both include a multi-head attention mechanism (five heads), a normalization operation and a feedforward network. The formula of the multi-head attention mechanism is shown in formula (3):

[0103]

[0104] In the formula, sco i is the output of the i-th head, X is the input vector, Attention is a single-head attention operation, softmax is an activation operation, W i Q , W i K , W i Vis the parameter matrix of query, key, value vector respectively, d is the dimension, W p is the coefficient matrix, concat is the concatenation operation, and out is the total output. Then, layer normalization and superposition operations are performed to obtain the output out half , and the specific steps are shown in formula (4):

[0105] out half = out + LN(out) (4)

[0106] In the formula, LN represents the layer normalization operation. The result will enter the feedforward network, which is implemented using a multilayer perceptron, and the formula is shown in formula (5):

[0107]

[0108] In the formula, z is the output of the feedforward network, ReLu is the activation operation to realize nonlinearization, w j is the jth weight matrix, and b is the bias parameter matrix. Then, the comprehensive output z enc of the encoder is obtained again by passing through a superposition and layer normalization operation, and the formula is shown in formula (6):

[0109] z enc = z + LN(z) (6)

[0110] The structure of the decoder is consistent with that of the encoder. The obtained encoder output will also be subjected to similar operations as in formulas (3)-(6), and the output feature dimension after passing through the decoder is BSx1x512. The last two dimensions are combined, and a fully connected operation is performed to reduce the dimension to 2. Then, the classification probability is obtained through the Softmax function. After passing through the network, the probability value of the rehabilitation personnel falling is obtained. If the value is greater than the threshold value of 65%, it is determined to be high risk, and an alarm information is given through the buzzer, so as to facilitate the intervention of doctors and family members.

[0111] The data set of the algorithm is collected in the First Affiliated Hospital of Suzhou University, and the sensor data and posture data of the patient using the invention are recorded. According to the doctor's experience, whether the patient has a risk of falling at this time is judged based on the body shaking state of the patient, and the label is set artificially (among them, the label of the situation with greater risk of falling is set to 1, and the smaller one is set to 0). A total of 1037 data were collected, and the data set was divided according to 8:2.

[0112] In addition to judging whether the rehabilitation personnel has a fall probability through the multimodal fall probability prediction model, the application also sets an intelligent processing function of abnormal situation. When detecting that the grip data of the rehabilitation personnel suddenly drops, it is judged that the upper limb muscle is tired; when detecting that the rehabilitation personnel maintains an unhealthy posture for a long time, it is judged that the upper body has an abnormality. When the above abnormal situation occurs, the main control unit will automatically upload the recorded video and remind the family members and doctors to intervene safely.

[0113] The system further comprises an electromyographic signal detection module installed on the lower leg of the rehabilitation personnel, which is used to collect surface electromyographic signals and transmit them to the main control unit, and the main control unit is used to evaluate whether the lower limb muscle force is in place according to the electromyographic signals, and store data information to provide a reference for subsequent evaluation by doctors.

[0114] The pressure sensor used in the system is PT124G-210, the grip sensor is Forsentek FFK, the damping adjuster is FRT-C2-201, the camera uses the model Intel RealsenseD435i, and the electromyographic signal detection module uses the model Myoware Muscle Sensor 2.0.

[0115] As a specific embodiment, when using the sarcopenia rehabilitation intelligent walking aid system, the rehabilitation personnel first needs to fully unfold the folded lightweight aluminum alloy support frame body 1 to ensure the stability of the structure. Then, according to the rehabilitation stage and physical condition of the rehabilitation personnel, the height is adjusted through the height adjusting hole 4 first, and then the rolling resistance of the wheel 5 is adjusted by rotating the knob 22 of the bottom damping adjuster 23, clockwise rotation increases the resistance, and counterclockwise rotation reduces the resistance. If weight training is needed, an appropriate amount of weight can be placed on the detachable weight rack 11 in the middle position of the connecting rod 6.

[0116] After preparation, the main control unit 12 located at the left handle is turned on, at this time the detachable display screen 3 on the cross beam will light up and display the initial parameters. The rehabilitation personnel holds the handles on both sides of the walking aid with both hands, the fingers and palms will contact the upper surface 20 and lower surface 21 of the grip sensor 2 at the same time, and the system starts to record the grip data.

[0117] When the rehabilitation personnel starts to push the walking aid to walk, the weight of the body will be transmitted downward through the support frame body 1. At this time, the pressure sensing assembly 7 located at the upper and lower junctions starts to work, the impact ball 14 in the upper half of the pipe body 16 is compressed by the downward pressure, which in turn triggers the button 19 on the pressure sensing block 18 in the lower half of the pipe body 17, and the system monitors the posture and center of gravity changes of the user in real time in this way. When the pressure disappears, the spring 15 will pop the impact ball 14 back to its original position.

[0118] During walking, the grip sensor 2 continuously collects the grip data of both hands, which are transmitted to the main control unit 12 in real time through the Bluetooth module for analysis and processing. At the same time, the camera module 13 embedded next to the detachable display screen 3 constantly takes pictures of the road ahead, helping the rehabilitation personnel to discover and avoid obstacles in time. In case of emergency, the SOS emergency button 8 can be pressed to trigger the buzzer 10 to seek help.

[0119] The system provides real-time feedback to the rehabilitation personnel in various ways: the LED three-color indicator light 9 indicates the current training status and physical condition with different colors, the buzzer 10 emits an alarm sound when an abnormal situation is detected, and the detachable display screen 3 displays detailed rehabilitation data such as walking speed, grip size, center of gravity offset, walking distance, etc.

[0120] If the rehabilitation personnel feels unwell or encounters an emergency during training, the SOS button 8 at the handle can be pressed immediately, and the system will automatically send a distress signal to inform medical personnel or family members.

[0121] After training, the main control unit 12 will save all the data of the entire training process for doctors and rehabilitation therapists to analyze the rehabilitation progress. The user needs to remove the weights on the weight rack 11, take off the detachable display screen 3 for charging, and finally fold up the support frame 1 for storage. The entire system provides scientific, safe and effective rehabilitation training support for patients with sarcopenia through intelligent monitoring and feedback.

[0122] The above specific embodiments only describe the preferred embodiments of the present application, and do not limit the protection scope of the present application. Any modifications, substitutions and improvements of the technical solutions of the present application made by those skilled in the art according to the description and drawings provided by the present application without departing from the design concept and spirit of the present application shall belong to the protection scope of the present application. The protection scope of the present application is determined by the claims.

Claims

1. A smart walker training method for sarcopenia rehabilitation, characterized in that, The method comprises the following steps: S1: Real-time acquisition of the grip strength data of the rehabilitation personnel through the grip sensor; S2: Acquisition of the pressure data applied by the rehabilitation personnel on the handrail through the pressure sensor; S3: Acquisition of the real-time image data of the rehabilitation personnel through the camera module; S4: Based on the collected multi-modal data, adaptive damping control, human posture recognition and fall risk prediction are performed; S5: The training parameters are adjusted according to the analysis results, and real-time feedback is provided through the display screen.

2. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 1, wherein, The adaptive damping control method in step S4 comprises: Y1: Real-time acquisition of the current resistance value of the wheel; Y2: Synchronous acquisition of the grip sensor value, pressure sensor average value and exercise time, and normalization processing; Y3: Based on the normalized grip, pressure and exercise time data, the comprehensive ability evaluation value of the rehabilitation personnel is calculated every five seconds using a weighted fusion method, wherein the pressure weight is set to 0.4, and the grip and exercise time weights are each set to 0.3; Y4: Calculation of the deviation of the current ability evaluation value from the preset target value, and the change rate relative to the previous moment; Y5: Based on the deviation and the change rate, the wheel resistance is dynamically adjusted to realize adaptive control of the rehabilitation training intensity; The human posture recognition method in step S4 comprises: S4-1: Real-time color video stream is acquired through the camera, and each frame of picture is extracted; S4-2: The frame picture is input into the improved YOLOv8 human upper body key point recognition algorithm; S4-3: Nine key points of the human upper body are recognized: left shoulder, right shoulder, shoulder midpoint, left hip, right hip, hip midpoint, left elbow, right elbow and spine midpoint; S4-4: The key point visualization result is displayed on the display screen for the rehabilitation personnel to correct the posture.

3. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 2, wherein, The improved YOLOv8 algorithm replaces the original basic module with an SFB module, and the processing steps of the SFB module comprise: Step A: The input features are subjected to channel transformation through 1×1 convolution; Step B: After depth separable convolution, including 3×3 depth convolution and 1×1 point convolution; Step C: The output features are obtained through a second 1×1 convolution; Step D: Residual connection is introduced to add the input and the output, reducing information loss.

4. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 2, wherein, A TAU module is added before the model output for feature enhancement, including: Channel dimension reduction is performed on the feature map, and the dimension reduction ratio is r; Layer normalization processing is performed on the reduced features; Two branches are processed: the first branch passes through a fully connected layer and a ReLU activation function, and the second branch passes through a fully connected layer and a Sigmoid activation function; The outputs of the two branches are multiplied element by element, and the channel attention weight is obtained through a fully connected layer.

5. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 1, wherein, The fall risk prediction method in step S4 comprises: Step P1: The two-dimensional information matrix of the human posture is reduced to a one-dimensional vector; Step P2: The posture vector is spliced with the grip data and the pressure data to form a 20-dimensional multi-modal input vector; Step P3: After adding position encoding, the input is input into the Transformer encoder; Step P4: The decoder is processed and the fall probability value is output; Step P5: When the probability value is greater than 65%, an alarm is triggered.

6. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 5, wherein, The fall risk prediction adopts an improved Transformer architecture, which processes multi-modal input data in parallel through 5 attention heads, wherein: Human posture, grip strength and pressure data are fused as a unified input; Temporal features and inter-modal correlations are extracted through an encoder-decoder structure; The output layer generates a fall risk probability value between 0 and 1, and triggers an early warning when the probability exceeds 0.

65.

7. The intelligent walking stick training method for sarcopenia rehabilitation according to claim 5, wherein, Both the encoder and the decoder include a feedforward network module, which adopts a two-layer fully connected network structure, with the middle layer dimension expanded to 4 times the input dimension, and uses a ReLU activation function to realize nonlinear transformation. 8.The smart walking training method for sarcopenia rehabilitation of claim 1, wherein, It also includes an abnormal situation detection step: When the grip strength data drops by more than 30% in a short period of time, it is determined that the upper limb muscles are fatigued; When the rehabilitation personnel keep the midpoint of the spine deviating from the center line by more than 15 degrees for 30 seconds, it is determined that the posture is abnormal; When an abnormality is detected, a video is automatically recorded and a reminder message is sent; It also includes a step of evaluating the lower limb muscle state through electromyographic signals: Collect the surface electromyographic signals of the lower leg; Calculate the root mean square value of the electromyographic signals to evaluate the muscle activation level; Determine whether the lower limb muscle force is in place according to the root mean square value, and prompt to increase the training intensity when the root mean square value is below the preset threshold; Store the evaluation results for doctors to refer to.

9. An intelligent walking aid device for sarcopenia rehabilitation, suitable for the intelligent walking training method for sarcopenia rehabilitation according to any one of claims 1-8, characterized in that, The structure includes a support frame body (1) made of lightweight aluminum alloy material, which has a foldable structure; The support frame body (1) is provided with a grip strength sensor (2) at the handle for reflecting the hand strength change of the rehabilitation personnel, and is also provided with an SOS button (8) for emergency call; The support frame body (1) is provided with a main control unit (12) at the left handle for receiving data transmitted by the Bluetooth module; The support frame body (1) is provided with LED three-color indicator lights (9), a buzzer (10) and a detachable display screen (3) on the cross beam, and a camera module (13) is embedded beside the detachable display screen (3) for shooting the road conditions in front; The support frame body (1) is connected by a connecting rod (6) at the lower part, and the connecting rod (6) is provided with a detachable weight-bearing frame (11) in the middle part to add weights for weight training; A pressure sensing assembly (7) is arranged at the upper and lower joints of the support frame body (1) for posture and gravity center detection; A damping adjuster (23) is arranged at the bottom of the support frame body (1) to control the resistance of the lower wheel (5).

10. The intelligent walking aid device for sarcopenia rehabilitation according to claim 9, wherein the pressure sensing assembly (7) comprises a striking ball (14), a spring (15) and a pressure sensing block (16), the striking ball (14) is installed in the upper half pipe body (16) of the support frame body (1), the striking ball (14) is connected with the spring (15) at both ends, the spring (15) is connected with the pressure sensing block (18) away from the striking ball, and the pressure sensing block (18) is fixed on the inner wall of the upper half pipe body (16). When the rehabilitation personnel force down, it will trigger the pressure sensing block (18) button (19) in the lower half of the tube (17), when the downward pressure disappears, the spring (15) will pop up and return to the initial state; The grip sensor (2) includes an upper surface (20) and a lower surface (21), when the rehabilitation personnel hold the upper surface (20) and the lower surface (21) of the grip at the same time, and apply pressure at the same time, the grip sensor (2) will record the detailed data of its grip; The damping adjuster (23) is connected with the wheel (5) and the support frame body (1) respectively at both ends, and the resistance is generated by adjusting the knob (22) of the damping adjuster clockwise to meet the training needs of the rehabilitation personnel at different stages.

Images