Multi-scene oriented intelligent displacement machine system

Abstract

This invention discloses an intelligent transfer machine system for multiple scenarios, including a sensor system, a lifting device, a walking aid device, a transfer device, and a control system. The sensor system includes an armrest pressure sensor, a seat pressure sensor, and an ultrasonic sensor. The armrest pressure sensor is located inside the handle and is used to detect the pressure FS1 on the handle and the pressure FS2 on the hand gripping the armrest. The pressure sensor is located on the seat of the intelligent transfer machine and is used to measure the left and right hip pressures FT1 and FT2, respectively. The ultrasonic sensor is used to detect the distance C between the user and the front edge of the seat. When FS1≠0, FS2=0, FT1, FT2=0, and C≠0, the control system activates the standing function mode. When FS1≠0, FS2≠0, FT1, FT2=0, and C≠0, the control system activates the walking aid function mode. When FS1≠0, FS2=0, FT1, FT2≠0, and C=0, the control system activates the transfer function mode.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation assistive devices and the technology of getting up and transferring, and in particular to an intelligent transfer machine system for multiple scenarios. Background Technology

[0002] Self-care reflects human independence and self-respect, and is even more important for the daily life and mental health of disabled elderly people. Currently available assistive devices for daily living are limited by complex application environments and difficult operation, resulting in low intelligence, poor safety, and insufficient human-machine integration, which greatly restricts their widespread use.

[0003] This invention addresses the urgent needs of the elderly for a healthy life. Specifically, it addresses the prominent characteristics of typical "assisted living" scenarios, such as spatial environment transitions, close human-machine interaction, and diverse disability conditions. It designs a multi-scenario intelligent mobility aid system to assist the elderly with daily activities such as getting up, moving, toileting, and bathing. This multi-scenario intelligent mobility aid system is a safer and easier-to-use intelligent elderly care robot product that will significantly improve the elderly's ability to live independently. Summary of the Invention

[0004] To achieve the objectives of this invention, the following technical solution is adopted:

[0005] A multi-scenario intelligent transfer machine system includes a sensor system, a lifting device, a walking aid device, a transfer device, and a control system, wherein the sensor system includes a handrail pressure sensor, a seat pressure sensor, and an ultrasonic sensor.

[0006] The armrest pressure sensor is used to detect the pressure FS1 on the handrail handle and the pressure FS2 on the hand grip; the pressure sensor is used to measure the pressure FT1 and FT2 on the left and right hips respectively; the ultrasonic sensor is used to detect the value C of the distance C from the user to the front edge of the seat;

[0007] When FS1≠0, FS2=0, FT1、FT2=0, C≠0, the control system starts the standing function mode;

[0008] When FS1≠0, FS2≠0, FT1, FT2=0, and C≠0, the control system activates the walking assistance function mode.

[0009] When FS1≠0, FS2=0, FT1、FT2≠0, and C=0, the control system starts the shift function mode.

[0010] The aforementioned intelligent transfer machine system, in the stand-up function mode:

[0011] When FS = 1 / 2G, the lifting motor stops when it raises the seat to the highest point, the lifting motor is locked, and the shift drive module is locked. The user's weight is G, and FS = FS1 + FS2.

[0012] The aforementioned intelligent transfer machine system, in the assisted walking function mode:

[0013] When the ultrasonic sensor reading is C, the control system determines that the user is within a safe range. The walking aid then activates the drivers of the left and right moving motors, which then begin to move forward, providing power for the user's movement. If the user is too close to or too far from the intelligent transfer machine, the control system determines that the user's position is unsafe and there is a risk of falling. The walking aid then deactivates the drivers of the left and right moving motors, and the motors brake.

[0014] The aforementioned intelligent shifting machine system, wherein in shifting function mode:

[0015] The shifting device separates and closes the seat.

[0016] In the aforementioned intelligent transfer machine system, during the transfer function mode: when the seat opening function is activated, the intelligent transfer machine's driving state is prohibited, and only the armrest's raising and lowering function is retained to adjust the seat height.

[0017] The aforementioned intelligent shifting machine system, wherein in shifting function mode:

[0018] When the seat pressure FT1+FT2 reaches the user's weight threshold G, it is confirmed that the user is seated and the seat is ready to open. The seat opening function is activated, and the values ​​of the two seat pressure sensors FT1 and FT2 are monitored. When FT1 and FT2 gradually decrease, the seat opening function is confirmed to be safe and the opening process continues. If either FT1 or FT2 increases, the seat opening function is paused, and the user or operator adjusts their body position to a suitable position to activate the seat opening function.

[0019] The aforementioned intelligent transfer machine system, wherein: the walking aid device includes a motor drive system, and the control system controls the motor drive system through an algorithm based on sampled hip pressure values ​​FT1 and FT2. Let the power spectral density ratio be KP, then the calculation formula is:

[0020]

[0021] Where P(f) is the power spectrum range, δ f Let f0 be the frequency at which the power spectrum reaches its maximum value, and f0 be the integration range. The solution, if If the number of solutions is greater than 1, we take f0 that makes P(f) the maximum value, where P0 is the area of ​​the power spectrum in f0±σ, σ is a fixed increment, and P is the area of ​​the entire power spectrum.

[0022] Set the speed control current of the left hip track motor under this seat opening function to I.t1 Then there is,

[0023]

[0024] Where A is the amplification factor of the motor drive current sampling amplifier, FT1 is the pressure on the left hip, and G is the estimated weight.

[0025] The equivalent motor feedback reference current is I tc1 , then there is

[0026]

[0027] Where B is the motor drive current sampling feedback coefficient, N is the number of sampling points within this time period, and X i Let be the voltage signal amplitude at the i-th sampling point;

[0028] The speed control current of the right hip track motor under this function is set to I. t2 Then there is,

[0029]

[0030] Where A is the amplification factor of the motor drive current sampling amplifier, FT2 is the pressure on the right hip, and G is the estimated weight.

[0031] The equivalent motor feedback reference current is I tc2 , then there is

[0032]

[0033] Where B is the motor drive current sampling feedback coefficient, N is the number of sampling points within this time period, and X i Let be the voltage signal amplitude at the i-th sampling point.

[0034] In the aforementioned intelligent shifting machine system, the sampling current control strategy is to subtract the equivalent motor feedback reference current from the speed regulation current to obtain the current control gain value ΔI. t1 ΔI t2 This is input as a control variable to the main controller to control the speed adjustment of the corresponding motor.

[0035] ΔI t1 =I t1 -I tc1

[0036] ΔI t2 =I t2 -I tc2 .

[0037] The aforementioned intelligent transfer system includes a formula for estimating human body weight G:

[0038] First, we set up a weight model data preprocessing step. The distance Cx between the human body and the seat of the intelligent transfer machine is measured by an ultrasonic sensor. K is the model convolution operation coefficient, W is the convolution matrix coefficient, B is the bias operation subtraction factor, and σ is the pooling operation coefficient. Then we have:

[0039] The characteristic value M of the user's general weight data is:

[0040] M = σ[K(W,Cx,“volid”)+B], where void is the default convolution operation type;

[0041] Secondly, the generalized weight data feature values, the ADL scale used in data preprocessing, the user's self-measured weight, and the initial information from the above steps are used to perform a second matrix model operation to derive the generalized weight data model value Q:

[0042] Q = ε[M,Q] ADL Q sel [,Q0], where ε is the pooling operation coefficient, Q ADL Q is used to estimate the ability to perform daily living activities. sel Q0 represents the self-measured weight, and Q0 represents the initial weight.

[0043] Secondly, a weight estimation model algorithm based on multi-data fusion was established to derive a user-specific weight-based index:

[0044]

[0045] Wherein, P is the weight-specific index, LF is the equivalent value of the pressure on the left armrest, RF is the equivalent value of the pressure on the right armrest; Qcl is the characteristic value of the weight-specific data on the left, Qcr is the characteristic value of the weight-specific data on the right, Mcl is the model value of the weight-specific data on the left, Mcl is the model value of the weight-specific data on the left, and λ is the correction coefficient.

[0046] When the user's left hand is the primary power source, λ = e LC / RC

[0047] When the user's right hand is the primary power source, λ = e 1-LC / RC

[0048] Where LC = ∑i, j = 0, 1, 2, 3∑ lj＝li+1 M l(ij) , RC=∑ i，j＝0,1,2,3 ∑ rj＝ri+1 M r(ij)

[0049] M l(ij) Let M be the coordinates of the pressure feature node on the left handrail, which is a linear coordinate system between li and the pressure feature node lj on the left handrail; r(ij)Let ri be the coordinate of the pressure feature node on the right handrail, and rj be the linear coordinate system of the pressure feature node on the right handrail.

[0050] Let k i and k j If is the shortest path value between network nodes at coordinates (i, j), then:

[0051]

[0052]

[0053] Finally, the sensor fusion-based weight algorithm yields an estimated weight for the user:

[0054]

[0055] Where Qi is the generalized weight model value at coordinate i, Mi is the generalized weight feature value at coordinate i, Qj is the generalized weight model value at coordinate j, Mj is the generalized weight feature value at coordinate j, and df(i,j) is the integral of G based on the generalized weight feature model.

[0056] The intelligent transfer machine system further includes a structured scene object recognition system. This system comprises a data analysis module and a data preprocessing module. The data analysis module analyzes and processes the signals obtained from the data preprocessing module, including: extracting features from the preprocessed image information using a convolutional network model containing three convolutional layers and three max-pooling layers. The image information is acquired in real-time by multiple cameras mounted on the intelligent transfer machine.

[0057] Let X0 be the feature vector of the object extracted by the first convolutional layer.

[0058]

[0059] [Nw T] = Feedback(J0 B0 M) c M p Y)

[0060] Where conv3 represents the convolution operation, J0 represents the convolution kernel matrix parameters, valid represents the convolution operation type, μ represents the input image matrix, B0 represents the bias parameter, α represents the pooling operation; β represents the output function of the convolutional neural network, Nw represents the trained convolutional neural network, Nw0 represents the initial trained convolutional neural network layer; T represents the parameters of the trained convolutional neural network, T0 represents the parameters of the initial trained convolutional neural network layer; Y represents the input structured scene image data, Y0 represents the initial input structured scene image data; M c M represents the kernel size and the number of layers.p The maximum pooling kernel size and number of layers are given by `<maxpooling>`, and `Feedback()` is the feature function of the feedback network model.

[0061] Let X1 be the feature vector of the object extracted by the second convolutional layer.

[0062]

[0063] [Nw T] = Feedback(J1 B1 M) c M p Y)

[0064] Where conv3 is the convolution operation, J1 is the convolution kernel matrix parameter, valid is the convolution operation type, μ is the input image matrix, B1 is the bias parameter, α is the pooling operation; β is the output function of the convolutional neural network, Nw is the trained convolutional neural network, Nw1 is the initial trained convolutional neural network of the second layer; T is the parameters of the trained convolutional neural network, T1 is the parameters of the initial trained convolutional neural network of the second layer; Y is the input structured scene image data, Y1 is the initial input structured scene image data of the second layer; M c M represents the kernel size and the number of layers. p This defines the max-pooling kernel size and the number of layers. Feedback() represents the feature functions of the feedback network model.

[0065] Let X2 be the feature vector of the object extracted by the third convolutional layer.

[0066]

[0067] [Nw T] = Feedback(J2 B2 M) c M p Y)

[0068] Where conv3 is the convolution operation, J2 is the convolution kernel matrix parameter, valid is the convolution operation type, μ is the input image matrix, B2 is the bias parameter, α is the pooling operation; β is the output function of the convolutional neural network, Nw is the trained convolutional neural network, Nw2 is the convolutional neural network after initial training of the third layer; T is the parameters of the trained convolutional neural network, T2 is the parameters of the convolutional neural network after initial training of the third layer; Y is the input structured scene image data, Y2 is the structured scene image data of the initial input of the second layer; M c M represents the kernel size and the number of layers. p This defines the max-pooling kernel size and the number of layers. Feedback() represents the feature functions of the feedback network model.

[0069] The formula for calculating the weighted parameter Cs in the precipitation model is as follows:

[0070]

[0071] scre is the alignment function, and mod is the feature recognition pattern function;

[0072] The feature vector of the object to be identified extracted by the multiplication layer is fused with the image recognition feature, and the fused recognition feature value C is set. fig ,but

[0073] C fig =[C1 C2...CN], (N is an integer), perform full-chain fusion to obtain the final fused feature D. fig Represented by a matrix:

[0074]

[0075] Where i and j are the dimensions of the spatial dimension fusion feature and the image recognition feature, respectively, and N is the number of features.

[0076] An intelligent transfer machine includes the intelligent transfer machine system described above. Attached Figure Description

[0077] Figure 1 It is a design scheme for an intelligent transfer machine system for multiple scenarios;

[0078] Figure 2 It is a design method for the mechanical body of an intelligent transfer machine system for multiple scenarios;

[0079] Figure 3 This is a diagram showing the seat with the seat open.

[0080] Figure 4 This is a schematic diagram of the seat's hip support section. Detailed Implementation

[0081] The following is in conjunction with the appendix Figure 1-4 The specific embodiments of the present invention will be described in detail below. These embodiments are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Obviously, the embodiments described in this invention are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0082] The terms "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include the specific features, structures, or characteristics described in connection with that embodiment. Therefore, the terms "comprising," "including," "having," and variations thereof in this specification mean "including but not limited to," unless otherwise specifically emphasized.

[0083] like Figure 1 , 2 As shown, this invention constructs a human dynamics model for the body of an intelligent transfer machine oriented to multiple scenarios. For the motion mechanism, it uses finite element modeling simulation, motion analysis, and human experimental measurement to analyze body characteristics, force changes, and biomechanical characteristics of the human-machine interface during posture transition. Based on the analysis results, it guides the transition mode during posture transition. Using spatial kinematics theories such as spinor theory and feature orientation sets, it develops the design of compliant posture adjustment mechanism and obstacle crossing drive mechanism, performs mechanism synthesis, optimizes structural parameters, performs dynamic analysis, verifies load capacity, and completes modular and engineering design.

[0084] The intelligent patient transfer system, designed for multiple scenarios, includes a sensor system, a lifting device, a walking aid, a transfer device, and a control system. The intelligent patient transfer system is installed on an intelligent patient transfer machine, which includes a frame, a seat plate, and wheels. Two wheels are rotatably mounted on either side of the frame. The seat plate is mounted on the frame and includes a first seat plate on the left and a second seat plate on the right. The first and second seat plates are driven to rise and fall by lifting motors 1 and 2, respectively. The first and second wheels are driven by a left motor and a right motor, respectively.

[0085] The aforementioned intelligent transfer machine system for multiple scenarios includes a sensor system comprising an armrest pressure sensor, a seat pressure sensor, and an ultrasonic sensor.

[0086] The handrail pressure sensor is located inside the handle and is used to detect the pressure generated by the hand holding the handle. This invention defines the pressure value of holding the handrail (half-grip, palm pressure) as FS1; and the pressure value of gripping the handrail tightly (full grip, finger pressure) as FS2. There are two types of handrail pressure sensors: one is a sensor that measures the handrail pressure value, typically a switch-type pressure sensor, measuring the specific force value, and is installed at the connection between the handle connecting rod and the main body of the transfer machine; a three-dimensional force sensor that measures the handrail torque is also located here, used to measure the magnitude of the handrail pressure torque. The other type is a thin-film pressure sensor that measures the handrail gripping force, installed at the user's gripping point at the front of the handrail, used to measure whether there is a change in pressure over a certain area. A three-dimensional force sensor that measures motor torque is installed at the connection between the motor mounting shaft and the main body mechanism, used to measure the torque data of the left and right motor mounting shafts, which is used as basic motor data and saved in the main controller as reference data for future improvements.

[0087] The seat (hip) pressure sensor is located on the first and second seat plates of the intelligent transfer machine. When someone sits down, the left and right hip pressure sensors generate electrical signals, with values ​​set as FT1 and FT2 respectively. The ultrasonic sensor is located at the front end of the seat of the intelligent transfer machine system for multiple scenarios. It is used to detect the distance of the person in front of the intelligent transfer machine. Let C be the value of the distance from the person in front to the front edge of the seat detected by the ultrasonic sensor.

[0088] The intelligent transfer machine system for multiple scenarios, wherein: when FS1≠0, FS2=0, FT1, FT2=0, and C≠0, the control system starts the stand-up function mode;

[0089] The intelligent transfer machine system for multiple scenarios, wherein: when FS1≠0, FS2≠0, FT1, FT2=0, and C≠0, the control system activates the walking assistance function mode;

[0090] The aforementioned intelligent shifting machine system for multiple scenarios includes the following: when FS1≠0, FS2=0, FT1、FT2≠0, and C=0, the control system activates the shifting function mode.

[0091] 1. Stand up function mode

[0092] At this point, the user holds onto the handrail but does not fully grip it, and the user's weight is set to G. Lifting motors 1 and 2 lower to their lowest positions.

[0093] Let FS = FS1 + FS2. When FS = 1 / 2G, lifting motor 1 and lifting motor 2 start to rise, raising the first and second seat plates, and stop when they reach the highest point. Lifting motor 1 and lifting motor 2 are locked, and the shift drive module is locked (the driver controlling the left and right travel motors is prohibited from working to prevent the shift machine from moving and causing the user to lose balance).

[0094] 2. Walking assistance function mode

[0095] If the user grips the handrails with both hands and actively pushes the intelligent mobility aid to walk, activating the walking assistance mode, the intelligent mobility aid assists the user's walking according to the user's operating mode. An ultrasonic sensor is located at the front of the seat of the multi-scenario intelligent mobility aid system to detect the distance to the person in front of the machine. Let the ultrasonic sensor value be C. When C is within the range of 50cm-100cm, the intelligent mobility aid determines that the user is within a safe range, and the walking aid controls the drivers of the left and right driving motors to operate, causing the motors to move forward and provide power for the user's walking. When C is outside the 50cm-100cm range, meaning the user is too close or too far from the intelligent mobility aid, the intelligent mobility aid determines that the user's position is unsafe and there is a risk of falling. The walking aid then disables the drivers of the left and right driving motors, braking them to ensure the user's safety.

[0096] 3. Shift function mode

[0097] If the user chooses to sit in the seat of the intelligent transfer machine, FT1 and FT2 ≠ 0, C = 0, and the intelligent transfer machine activates its transfer function mode. At this time, the intelligent transfer machine operates according to the user's usual habits when using a wheelchair and transfer machine. That is, the user turns 180 degrees, sits in the seat of the intelligent transfer machine, and operates the controller on the armrest to control the movement of the intelligent transfer machine and the transfer function of the seat's hip board.

[0098] Special design of the seating area: Figure 3 This is a schematic diagram of the seat in its open state. The invention features a hip-opening and closing support structure and a physical support interface, providing "feel-free" contact support. The first and second seat plates, when joined together, form the entire seat for the user to sit on. The opening mechanism can separate and close the seat, dividing it into two seat plates (left and right) or merging them into a single seat plate. This allows the user's hips to be moved from a bed or chair to a toilet or shower chair in the bathroom. A pressure sensor and a gyroscope are installed on each of the left and right seat plates.

[0099] When the seat opening function is activated, the intelligent transfer machine's movement is disabled; only the armrest's raising and lowering function is retained to adjust the seat height (the armrest and the main body of the intelligent transfer machine are fixed in position, and the armrest changes with the seat height when the seat is adjusted). When the seat pressure FT1+FT2 reaches the user's weight threshold G (G is an estimated user weight), it is confirmed that the user is seated correctly, and the seat opening function is ready to be activated. The seat opening function is then activated, and the values ​​of the two seat pressure sensors, FT1 and FT2, are analyzed. If FT1 and FT2 gradually decrease, the seat opening function is confirmed to be safe and the opening process continues. If either FT1 or FT2 increases, it indicates that the user's posture is incorrect and there is a risk of tipping over; the seat opening function is then paused, and the user or operator adjusts their posture to a suitable position before reactivating the seat opening function.

[0100] The seat opening mechanism includes a drive motor for driving the seat to open and close.

[0101] The control system uses sampled hip pressure values ​​FT1 and FT2 to control the motor drive system via an algorithm. Let φ be the angle between the force directions of FT1 and FT2 measured by the gyroscope and the vertical line. Figure 4 As shown.

[0102] In the frequency domain, since the power spectrum distribution is relatively stable, the proportion of power spectrum energy near the maximum value in the overall signal is also relatively stable and is not affected by the specific location of the maximum value. Therefore, the power spectrum ratio method in frequency domain analysis is used to extract the feature values ​​for circuit sampling. Let the power spectrum ratio be KP, then the calculation formula is:

[0103]

[0104] Where P(f) is the power spectrum range, δ f Let f0 be the frequency at which the power spectrum reaches its maximum value, and f0 be the integration range. The solution, if If the number of solutions is greater than 1, then we take f0 that maximizes P(f). P0 is the area of ​​the power spectrum within f0 ± σ, where σ is a fixed increment. P is the area of ​​the entire power spectrum.

[0105] Set the speed control current of the left hip track motor under this seat opening function to I. t1 Then there is,

[0106]

[0107] Where A is the amplification factor of the motor drive current sampling amplifier, FT1 is the pressure on the left hip, and G is the estimated weight.

[0108] The equivalent motor feedback reference current is I tc1 , then there is

[0109]

[0110] Where B is the motor drive current sampling feedback coefficient, N is the number of sampling points within this time period, and X i Let be the voltage signal amplitude at the i-th sampling point.

[0111] The speed control current of the right hip track motor under this function is set to I. t2 Then there is,

[0112]

[0113] Where A is the amplification factor of the motor drive current sampling amplifier, FT2 is the pressure on the right hip, and G is the estimated weight.

[0114] The equivalent motor feedback reference current is I tc2 , then there is

[0115]

[0116] Where B is the motor drive current sampling feedback coefficient, N is the number of sampling points within this time period, and Xi Let be the voltage signal amplitude at the i-th sampling point. The sampling current control strategy is to subtract the equivalent motor feedback reference current from the speed control current to obtain the current control gain value ΔI. t1 ΔI t2 This is input as a control variable to the main controller to control the speed adjustment of the corresponding motor.

[0117] ΔI t1 =I t1 -I tc1

[0118] ΔI t2 =I t2 -I tc2

[0119] Formula for estimating human body weight G:

[0120] First, we set up a weight model data preprocessing step. The distance Cx between the human body and the seat of the intelligent transfer machine is measured by an ultrasonic sensor. K is the model convolution operation coefficient, W is the convolution matrix coefficient, B is the bias operation subtraction factor, and σ is the pooling operation coefficient. Then we have:

[0121] User general weight data feature value M is

[0122] M = σ[K(W,Cx,“volid”)+B], where void is the default convolution operation type.

[0123] Secondly, the generalized weight data feature values, the ADL (Activities of Daily Living) scale, user self-measured weight, and initial information from the above steps are used to perform a second matrix model operation to derive the generalized weight data model value Q:

[0124] Q = ε[M,Q] ADL Q sel [,Q0], where ε is the pooling operation coefficient, Q ADL Q is used to estimate the ability to perform daily living activities. sel Q0 represents the self-measured weight, and Q0 represents the initial information weight. Can be (User inputs a value).

[0125] Secondly, a weight estimation model algorithm based on multi-data fusion was established to derive a user-specific weight-based index:

[0126]

[0127] Where P is the weight-specific index, LF is the equivalent value of pressure on the left armrest, and RF is the equivalent value of pressure on the right armrest. Qcl is the characteristic value of the weight-specific data on the left side, Qcr is the characteristic value of the weight-specific data on the right side, Mcl is the model value of the weight-specific data on the left side, Mcr is the model value of the weight-specific data on the left side, and λ is the correction coefficient.

[0128] When the user's left hand is the primary power source, λ = e LC / RC

[0129] When the user's right hand is the primary power source, λ = e 1-LC / RC

[0130] Where LC = ∑i, j = 0, 1, 2, 3∑ lj＝li+1 M l(ij) , RC=∑ i，j＝0,1,2,3 ∑ rj＝ri+1 M r(ij)

[0131] M l(ij) Let M be the coordinates of the pressure feature node on the left handrail, which is a linear coordinate system between li and the pressure feature node lj on the left handrail; r(ij) Let ri be the coordinate of the pressure feature node on the right handrail, and let rj be the linear coordinate system of the pressure feature node on the right handrail.

[0132] Let k i and k j If is the shortest path value between network nodes at coordinates (i, j), then:

[0133]

[0134]

[0135] Finally, the sensor fusion-based weight algorithm yields an estimated weight for the user:

[0136]

[0137] Where Qi is the generalized weight model value at coordinate i, Mi is the generalized weight feature value at coordinate i, Qj is the generalized weight model value at coordinate j, Mj is the generalized weight feature value at coordinate j, and df(i,j) is the integral of G based on the generalized weight feature model.

[0138] This invention relates to an intelligent moving machine designed for multiple scenarios, capable of recognizing structured objects in indoor scenes such as bathroom toilets, washbasins, and shower rooms. Therefore, a structured scene object recognition system is included. This system comprises a data analysis module and a data preprocessing module. The data analysis module analyzes and processes the signals obtained from the data preprocessing module, including extracting features from the preprocessed image information using a convolutional network model containing three convolutional layers and three max-pooling layers. The image information is acquired in real-time by multiple cameras mounted on the intelligent moving machine.

[0139] Let X0 be the feature vector of the object extracted by the first convolutional layer.

[0140]

[0141] [Nw T] = Feedback(J0 B0 M) c M p Y)

[0142] Where conv3 represents the convolution operation, J0 represents the convolution kernel matrix parameters, valid represents the convolution operation type, μ represents the input image matrix, B0 represents the bias parameter, α represents the pooling operation; β represents the output function of the convolutional neural network, Nw represents the trained convolutional neural network, Nw0 represents the initial trained convolutional neural network layer; T represents the parameters of the trained convolutional neural network, T0 represents the parameters of the initial trained convolutional neural network layer; Y represents the input structured scene image data, Y0 represents the initial input structured scene image data; M c M represents the kernel size and the number of layers. p This represents the max-pooling kernel size and the number of layers. Feedback() is the feature function of the feedback network model.

[0143] Let X1 be the feature vector of the object extracted by the second convolutional layer.

[0144]

[0145] [Nw T] = Feedback(J1 B1 M) c M p Y)

[0146] Where conv3 is the convolution operation, J1 is the convolution kernel matrix parameter, valid is the convolution operation type, μ is the input image matrix, B1 is the bias parameter, α is the pooling operation; β is the output function of the convolutional neural network, Nw is the trained convolutional neural network, Nw1 is the initial trained convolutional neural network of the second layer; T is the parameters of the trained convolutional neural network, T1 is the parameters of the initial trained convolutional neural network of the second layer; Y is the input structured scene image data, Y1 is the initial input structured scene image data of the second layer; M c M represents the kernel size and the number of layers. p This represents the max-pooling kernel size and the number of layers. Feedback() is the feature function of the feedback network model.

[0147] Let X2 be the feature vector of the object extracted by the third convolutional layer.

[0148]

[0149] [Nw T] = Feedback(J2 B2 M) c M pY)

[0150] Where conv3 is the convolution operation, J2 is the convolution kernel matrix parameter, valid is the convolution operation type, μ is the input image matrix, B2 is the bias parameter, α is the pooling operation; β is the output function of the convolutional neural network, Nw is the trained convolutional neural network, Nw2 is the convolutional neural network after initial training of the third layer; T is the parameters of the trained convolutional neural network, T2 is the parameters of the convolutional neural network after initial training of the third layer; Y is the input structured scene image data, Y2 is the structured scene image data of the initial input of the second layer; M c M represents the kernel size and the number of layers. p This represents the max-pooling kernel size and the number of layers. Feedback() is the feature function of the feedback network model.

[0151] The formula for calculating the weighted parameter Cs in the precipitation model is as follows:

[0152]

[0153] scre is the alignment function, and mod is the feature recognition pattern function.

[0154] The feature vector of the object to be identified extracted by the multiplication layer is fused with the image recognition feature, and the fused recognition feature value C is set. fig ,but

[0155] C fig =[C1 C2...CN], (N is an integer), perform full-chain fusion to obtain the final fused feature D. fig Represented by a matrix:

[0156]

[0157] Where i and j are the dimensions of the spatial dimension fusion feature and the image recognition feature, respectively, and N is the number of features.

[0158] The network model of this invention is trained using an existing training set of indoor structured scenes. First, pre-data acquisition and input are performed. Identifiable objects with features in the indoor structured scene are collected and uploaded to the database, such as the entrance to the bathroom or bedroom, and the toilet, washbasin, and shower chair in the bathroom. Second, the intelligent transfer device undergoes adaptive learning. The user or caregiver operates the intelligent transfer device to complete a practice session of an indoor structured usage scenario, recording the image information of the actual scene and providing feedback on the recognition results. Third, the intelligent transfer device is used according to actual needs, and after each operation, the intelligent transfer device autonomously learns and iterates its information.

[0159] This invention also proposes three control strategies for elderly people using intelligent mobility aids for getting up, moving around, weight reduction support, and smooth transfers, which can achieve natural and smooth human-computer interaction and determine the usage status of the intelligent mobility aid.

[0160] As mentioned earlier, the intelligent transfer machine has three states: 1. Standing function mode; 2. Walking assistance function mode; 3. Transfer and transfer mode.

[0161] (1) When the intelligent shifter is in stand-up function mode, the system generates an auxiliary stand-up control strategy method, including the following steps:

[0162] The handrail pressure and drive motor torque information are labeled as Ff and Fq, respectively. The force and torque information output by the three-dimensional force sensor in its custom coordinate system (xyz) includes the force components Ffx, Ffy, Ffz, Fqx, Fqy, Fqz in the x, y, and z directions, and the torque components ε in the x, y, and z directions. fx ε fy ε fz , ε qx ε qy ε qz . That is, Ff = [Ffx, Ffy, Ffz, ε fx , ε fy , ε fz ], Fq=[Fqx, Fqy, Fqz, ε qx , ε qy , ε qz ];

[0163] The kinematic matrix from the three-dimensional force sensor to the DC drive motor is J. fq The pose transfer matrix from the user's standing position center point to the three-dimensional force sensor is T. fq The control gain adjustment coefficient matrix is ​​K fq Then the auxiliary walking control strategy controls the force ε stand The calculation formula is as follows:

[0164]

[0165] in, J is the transfer matrix from getting up to standing stably. fq The transformation matrix.

[0166] Gain control coefficient matrix K fq :

[0167]

[0168] |X m |、|Y m |、|Z m| represents the absolute value of the position of the center point of the handrail force sensor in the three-dimensional coordinate system xyz, within the corresponding system coordinate system xyz. |A t |、|B t |、|C t | represents the absolute value of the position of the center point of the forward force of the drive motor in the three-dimensional coordinate system xyz and the corresponding system coordinate system xyz.

[0169] Gain control coefficient matrix K fq This is to trigger the control model to output control gain in the diagonal matrix dimension when the user confirms their intention to stand up. The system coordinates the intelligent shifting machine to assist in standing, ensuring that the intelligent shifting machine stably provides the handrail lifting torque output, while ensuring that the drive motor brakes and locks, thus completing the control mechanism for assisting in standing up.

[0170] (2) When the intelligent shifter is in the assistive mode, the system generates a shift assistive control strategy method, including the following steps:

[0171] In the assisted walking mode, the handrail pressure output is stable, while the drive motor output varies with gait speed and road conditions. Let the drive motor output torque information be denoted as Fw, and the coordinate vector corresponding to the origin of the traversal coordinate system as the motor output torque coordinate system vector is Rw = (Rw... x Rw y Rw z )

[0172] The motor output torque damping coefficients are respectively

[0173]

[0174] Where ωx, ωy, and ωz are the angular velocity components corresponding to the x, y, and z axes, respectively.

[0175] Displacement-assisted walking control strategy control model ε move The calculation also needs to take into account the real-time active thrust F of the user. zd The components in the X, Y, and Z directions are denoted as [τ]. x , τ y , τ z To ensure effective and constant thrust control adjustment by the user, ε move The calculation is as follows:

[0176]

[0177] The user applies a boosting force—the combined force of the pressure from the handrail and the forward thrust—to assist in starting the intelligent mobility aid. This boosting force is greater when the user begins to operate the aid, decreasing to a steady level as the gait stabilizes. During stopping, the boosting force slightly increases and then gradually decreases to zero. Throughout this process, the motor provides its rated output current to the control system during the user's stable walking phase, achieving the stability of the control force and keeping the user's walking movement within a normal distance range. This range can be personalized based on the individual user's physical characteristics, thus realizing the assistive function of mobility aid.

[0178] (3) When the intelligent shift machine is in shift transfer mode, the system generates a transfer control strategy method, which includes the following steps:

[0179] ①The control force of the hip support, transfer and switching control strategy includes two parts: the first part is the static pressure of the seat board when the transfer is supported, and the second part is the dynamic shear force of the seat board when the switching is supported.

[0180] ②The vertical gravity of the user under static seating conditions is G0, and the static coefficient IF1 is set, 0≤IF1≤1.

[0181] Then transfer the static pressure of the support

[0182] ③ During the transfer process, the user's buttocks are affected by the dynamic movement of the intelligent transfer machine's opening mechanism. The force on the user's buttocks in the vertical direction changes dynamically. Set the transfer support dynamic coefficient IF2, IF2>0; set the transfer support dynamic correction coefficient IF3, 0≤IF3≤1. The dynamic force F of this part of the support is... shear The calculation formula is:

[0183]

[0184] Among them, F x F y F z These are the components of the force in the X, Y, and Z directions, respectively, d(V x ), d(V y ), d(V z (x) represents the acceleration in the x-direction.

[0185] ④ Control strategy for dynamic shear force during transfer support ε shear The calculation formula is:

[0186]

[0187] Where n = 1, 2, 3..., and t is a time constant.

[0188] By combining three control strategies, a control system for standing up, moving, and transferring can be implemented, satisfying the following characteristic control model:

[0189]

[0190] Among them, M damp For the adjustable motor damping coefficient, F Co This is the damping correction factor. ε represents the average operating speed of the hip track mechanism. general The comprehensive control model is the sum of the assisted walking control strategy model, the displacement assistance control strategy model, and the displacement assistance control strategy model.

[0191] The advantages of this invention are:

[0192] (1) Using a comprehensive sampling current and feedback current calculation method in the time and frequency domains helps the control system to effectively identify weak current changes, increases the spatial dimension of the control system to execute commands, and effectively improves the accuracy of the control strategy algorithm.

[0193] (2) Use a multi-source feature deep neural network fusion algorithm to perform feature fusion on structured object recognition data information, so as to give full play to the correlation of multi-source data and the complementarity of heterogeneous data.

[0194] (3) Based on the indoor structured information input, the input data has obvious feature values, which makes the prediction of the object recognition result of the intelligent transfer machine in the indoor environment more targeted and has a higher recognition rate.

[0195] (4) By using the method and system of the present invention, it is possible to collect indoor fixed structured data information, adopt different path planning schemes according to the feature value level of the information, and complete the adaptive object recognition scheme through machine learning, so as to achieve accurate and continuous docking of transfer devices and improve the age-friendly interaction level of the equipment.

[0196] (5) The present invention adopts the integration of assisted walking control strategy, transfer assisted walking control strategy and transfer assisted walking control strategy and its comprehensive integrated control strategy, which greatly facilitates users to complete daily self-care activities by relying on intelligent transfer machine, avoids excessive participation by human or user during transfer, and avoids discomfort in training patients caused by switching multiple control strategies; at the same time, the three control methods are a continuous control strategy model, which is conducive to the overall coordinated control of the system and improves the comfort of use.

[0197] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-scenario intelligent transfer machine system, comprising a sensor system, a lifting device, a walking aid device, a transfer device, and a control system, characterized in that: The sensor system includes armrest pressure sensors, seat pressure sensors, and ultrasonic sensors; the intelligent transfer machine system is installed on the intelligent transfer machine, which includes a first seat plate on the left and a second seat plate on the right, with the seat pressure sensors located on the first and second seat plates; The armrest pressure sensor is used to detect the pressure FS1 of the handrail handle and the pressure FS2 of the hand gripping the armrest. The pressure value of the hand gripping the armrest halfway and applying force with the palm is set as FS1, and the pressure value of the hand gripping the armrest fully and applying force with the fingers is set as FS2. The seat pressure sensor is used to measure the pressure FT1 and FT2 of the left and right buttocks respectively. The ultrasonic sensor is used to detect the distance C between the user and the front edge of the seat. When FS1≠0, FS2=0, FT1, FT2=0, and C≠0, the control system starts the standing function mode; When FS1≠0, FS2≠0, FT1, FT2=0, and C≠0, the control system activates the walking assistance function mode. When FS1≠0, FS2=0, FT1, FT2≠0, and C=0, the control system starts the shift function mode; In the shift function mode: When the seat pressure FT1+FT2 reaches the user's weight threshold G, it is confirmed that the user is seated and the seat is ready to open. The seat opening function is activated, and the values ​​of the two seat pressure sensors FT1 and FT2 are monitored. When FT1 and FT2 gradually decrease, the seat opening function is confirmed to be safe and the opening process continues. If either FT1 or FT2 increases, the seat opening function is paused, and the user or operator adjusts their body position to a suitable position to activate the seat opening function.

2. The intelligent transfer machine system according to claim 1, characterized in that... In the stand-up function mode: When FS=1 / 2G, the lifting motor stops when it raises the seat to the highest point, the lifting motor is locked, and the shift drive module is locked. The user's weight is G, and FS=FS1+FS2.

3. The intelligent transfer machine system according to claim 1, characterized in that... In walking assistance mode: An ultrasonic sensor detects the distance C between the user and the front edge of the seat. The control system determines that the user is within a safe range based on the value C. The walking aid then controls the drivers of the left and right motors to work, and the left and right motors begin to move forward, providing power for the user to walk.

Images