KR100994408B1 - Method and device for deducting pinch force, method and device for discriminating muscle to deduct pinch force - Google Patents

Abstract

The present invention relates to a finger force estimating method and estimating apparatus, a muscle discriminating method for finger force estimating, and a muscle discriminating apparatus.

Finger force estimation method according to the present invention, the EMG signal acquisition step of acquiring the EMG signals of the seven muscles that govern the muscles of the thumb and index finger from a plurality of EMG sensors mounted to be in close contact with the skin of the underarm of the subject, EMG signal selection step of selecting EMG signals of four muscles having information for measuring the force among the EMG signals of the seven muscles by a predetermined discriminant, and converts the four selected EMG signals into EMG signals in the time window, respectively In order to calculate the bearance for each of the EMG signals in the time window converted by the predetermined variance algorithm, a force signal estimation for estimating the calculated bearance as a force signal by the artificial neural network algorithm And displaying the estimated force signal.

 EMG signal, discriminant, variance algorithm, time window

Description

METHOD AND DEVICE FOR DEDUCTING PINCH FORCE, METHOD AND DEVICE FOR DISCRIMINATING MUSCLE TO DEDUCT PINCH FORCE}

The present invention relates to a finger force estimating method and estimating apparatus, a muscle discriminating method for finger force estimating, and a muscle discriminating apparatus.

Man's versatile hand technology is controlled by the central nervous system of the brain and spinal cord, which govern the voluntary movements of the body. Excessive force when catching an object will destroy it and attempting to hold it with a weak force will not lift the object. When a person catches an object, he or she generates an optimal force that is neither excessive nor weak, and the magnitude of the force is adjusted differently according to the shape, weight, and texture of the object. That is, it is very important to properly control the force in handling the object by hand, and the magnitude of this force varies according to the degree of contraction of the muscle, which can be observed using electromyogram.

EMG is an electrophysiological signal that can measure muscle contraction. Human-machine interaction (HMI) such as robotic prosthesis, exoskeleton, and telemanipulation It is one of the representative bio signals widely used to extract human movement intention in the field. EMG can be acquired in a non-invasive way, so it is relatively simple to use compared to other nerve signals (electroneurogram, cortical signals), and the signal precedes before actual body movement occurs. The advantage is that the intention can be interpreted before the figure is predicted. Recently, Tenore et al. And Nagata et al. The team succeeded in analyzing more than ten movement intentions, such as bending and squeezing of five fingers from human EMG signals, and the results of the study indicate that users individually control the fingers of the robot to control the hand robot using EMG signals. It can be shown that it can be done. However, this study only extracted the intentions of "ON" and "OFF" states, such as finger bending and shaping, and did not extract force information on how much force to cause finger bending and shaping. Recently, there have been attempts to estimate muscle force from EMG signals, but mainly extracting torques of wrist and elbow joints, and the study of finger force does not exist according to our research. Recalling that the finger force adjustment function is very important for the versatility of the object, it is very important for the user to extract the control force for the purpose of controlling the hand robot.

Currently developed hand robots, such as the Utah / MIT dexterous hand, the shadow dexterous hand, and the Cyberhand, have five fingers like the human hand and provide more than 16 degrees of freedom. Japan's Okada hand demonstrated the ability to spin a rod between fingers, demonstrating that it is technically possible for hand robots to perform complex movements as well. Although robots that can mimic human's delicate and precise hand movements have been actively developed, researches on extracting user's power intention to deal with delicate and delicate objects have been relatively insufficient.

The following difficulties exist in estimating finger force from EMG signals. Many of the 39 muscles contribute to finger force, most extrinsic muscles are located deep in the arm, making it impossible to acquire signals with surface electrodes, and most intrinsic muscles are too small to be observed. The muscles that govern finger movement are less bulky than the lower extremity and shoulder muscles, and crosstalk is a serious problem. The human nervous system does not control each muscle individually in generating finger force, but rather generates a force by controlling several muscles into several groups. Is not yet known. In rehabilitation and motor physiology, there are studies that estimate the force from EMG, but these are results of off-line analysis or invasive methods using needle electrodes, which are not suitable for HMI applications.

Surface EMG, on the other hand, is known to have limitations in viewing individual muscle contractions. Because, even if you attach the electrode to the skin of the muscle area that you want to observe because the contraction of the muscles in the side or inside also affect the electrode. In addition, different muscles have different anatomical structures, and the thickness of the skin also has a big impact. In addition, there is a problem that the signal-to-noise ratio of the EMG signal drops significantly if the muscle to be viewed is located too deep in the skin.

Therefore, in order to solve the above problems, the present invention uses pinch force of the finger by using a predetermined discriminant that can select only the muscles that are most likely to predict the force when observed with the surface electrode. An object of the present invention is to provide a finger force estimating method and estimating apparatus capable of effectively determining the position of a surface electrode, an muscle discriminating method for finger force estimating, and a muscle discriminating apparatus.

In addition, the present invention, by using an artificial neural network algorithm using the number of various hidden nodes to determine the neural network structure, finger force estimation that can estimate the pinch force despite the very short learning time An object of the present invention is to provide a method and estimating apparatus, a muscle discriminating method for finger force estimation, and a muscle discriminating apparatus.

In addition, the present invention can be used to handle the object elaborately in the use of the prosthesis by people whose thumb and index finger are cut, and the force to perform directly by the person to control dangerous operations when performing the remote control robot An object of the present invention is to provide a finger force estimating method and estimating apparatus, a muscle discriminating method for finger force estimating, and a muscle discriminating apparatus.

The finger force estimating method according to claim 1 is an EMG signal acquiring step of acquiring an EMG signal of seven muscles that govern the muscles of the thumb and forefinger from a plurality of EMG sensors mounted to be in close contact with the skin of the underarm of the person to be measured. EMG signal selection step of selecting the EMG signals of the four muscles having information for measuring the force of the EMG signals of the seven muscles by a predetermined discriminant equation, each of the four selected EMG signals as EMG signals in the time window A transform calculation step of calculating a bearance for each of the EMG signals in the time window converted by the predetermined dispersion algorithm, and a force signal for estimating the calculated bearance as a force signal by an artificial neural network algorithm Estimating, displaying the estimated force signal.

The finger force estimation method of the invention according to claim 2 is the finger force estimation method according to the invention according to claim 1, wherein, in the EMG signal acquisition step, the EMG signal is a Gaussian process having an average of zero.

The finger force estimation method of the invention according to claim 3 is the finger force estimation method according to the invention according to claim 1, in the force signal estimation step, the artificial neural network algorithm presets a force signal corresponding to the bearance of the EMG signal in advance, The set force signal is estimated as a force signal corresponding to a bearance for each of the EMG signals in the transformed time window.

The finger force estimating apparatus of the invention according to claim 4 is a plurality of EMG sensor units which are mounted to be in close contact with the skin of the underarm of the person to be measured and acquire EMG signals of seven muscles that govern the muscles of the thumb and the index finger, and predetermined determination. EMG signal selection unit for selecting the EMG signal of the four muscles having information for measuring the force of the EMG signals of the seven muscles obtained by the equation, by converting the selected four EMG signals into EMG signals in the time window, respectively A bearance calculator for calculating a bearance for each of the EMG signals in the time window converted by a predetermined variance algorithm, and a force signal estimator for estimating the calculated bearance as a force signal by an artificial neural network algorithm And a display for displaying the estimated force signal.

The finger force estimating apparatus of the invention according to claim 5 is the finger force estimating apparatus according to the invention according to claim 4, wherein the EMG signal is a Gaussian process having an average of zero.

The finger force estimating apparatus according to claim 6 is the finger force estimating apparatus according to claim 4, wherein the artificial neural network algorithm presets a force signal corresponding to the bearance of the EMG signal, and converts the preset force signal. Estimate the force signal corresponding to the bearance for each of the EMG signals in the time window.

The muscle discrimination method for finger force estimation according to the invention according to claim 7 is an EMG signal which is mounted to be in close contact with the skin of the underarm of the subject under test and acquires EMG signals for seven muscles that govern the thumb and the index finger in a plurality of times. Acquisition step, the reference function result calculation step of calculating each reference function result value by inputting the EMG signal acquired in a plurality of times to the reference function, four muscles are identified in order of the calculated result of each reference function result Muscle discrimination step.

The muscle discrimination method for finger force estimation according to the invention of claim 8 is the muscle discrimination method for finger force estimation according to the invention according to claim 7, wherein the EMG signal is a Gaussian process having an average of zero.

The muscle discrimination apparatus for finger force estimation according to the present invention is an EMG signal, which is mounted to be in close contact with the skin of the underarm of the subject under test and obtains EMG signals for seven muscles that govern the thumb and the index finger in a plurality of times. Acquisition unit, the reference function result calculation unit for inputting the EMG signal obtained in a plurality of times to the reference function to calculate the respective reference function result value, the four muscles in order to determine the four muscles in order It includes a muscle discriminating unit.

The muscle discrimination apparatus for finger force estimation of the invention according to claim 10 is the muscle discrimination apparatus for finger force estimation according to the invention of claim 9, wherein the EMG signal is a Gaussian process having an average of zero.

As described above, according to the finger force estimating method and estimating apparatus, the muscle discriminating method for estimating finger force, and the muscle discriminating apparatus according to the present invention, the pinch force of the finger is increased by using a predetermined discrimination equation. The position of the surface electrode can be effectively determined for estimation.

In addition, according to the present invention, by using an artificial neural network algorithm using the number of various hidden nodes to determine the neural network structure, the pinch force can be estimated even though the learning time is very short.

In addition, according to the present invention, a person whose thumb and index finger are cut can be used to precisely handle an object in using the prosthesis, and the force to be directly performed by a person is controlled when performing a dangerous operation using a remote control robot. It can be used as a way to.

Specific matters other than the problem to be solved, the problem solving means, and the effects of the present invention as described above are included in the following embodiments and the drawings. Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings. Like reference numerals refer to like elements throughout.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the accompanying drawings are only described in order to more easily disclose the contents of the present invention, but the scope of the present invention is not limited to the scope of the accompanying drawings that will be readily available to those of ordinary skill in the art. You will know.

1 is a view for explaining the peripheral structure of the finger force estimation apparatus according to an embodiment of the present invention, Figure 2 is a view for explaining the structure of the finger force estimation apparatus according to an embodiment of the present invention, 3 is a diagram illustrating an arm of a person to be measured equipped with an EMG sensor in a finger force estimating apparatus according to an embodiment of the present invention.

On the other hand, in one embodiment of the present invention, the most delicate and widely used form of the form of holding a person by hand is selected, which is called a pinch (pinch) as a form of holding the object with the fingertips using the thumb and index finger. . That is, in the embodiment of the present invention, the finger force will be referred to as pinch force.

1 is a view for explaining the peripheral structure of the finger force estimation apparatus according to an embodiment of the present invention.

As shown in FIG. 1, the environment used in the finger force estimating apparatus includes the arm and the wrist 40 of the subject to be arm armed (not shown) and the wrist height in order to implement an isometric experiment environment. Secure using a wrist brace (not shown). Since the fingers are difficult to fix using the fasteners, use a motion camera (Micron Tracker S60, Claron Technology Inc., Canada) (10) to observe finger movements. The test subject pinches the thumb and index finger with a force sensor (NANO 17, ATI Industrial Automation, USA) attached to an aluminum rod. EMG signals are measured using a surface electrode (DE-2.1, Delsys, USA) 20 and connected to a signal collector (PCI 6034e, National InstrumentsTM, USA) (not shown) to signal the computer 30 at 1 kHz. To be transmitted.

2 is a view for explaining the structure of the finger force estimation apparatus according to an embodiment of the present invention.

As shown in FIG. 2, the finger force estimating apparatus according to an embodiment of the present invention includes an EMG sensor unit 100, an EMG signal selecting unit 200, a bearance calculator 300, and a force signal estimating unit. 300, the display unit 400 is included.

The EMG sensor unit 100 is configured so as to be in close contact with the skin of the lower arm of the person to be measured and acquires EMG signals of seven muscles that govern the muscles of the thumb and the index finger. The EMG signal is considered a Gaussian process with an average of zero. Here, a total of 15 muscles govern the movement of the thumb and index finger. Only seven muscles are located close to the skin surface using ADAM Interactive Anatomy (A.D.A.M. Inc., USA) software. The seven selected muscles were Extensor Digitorium (ED), Abductor Pollicis Longus (APL), Flexor Digitorum Superficialis (FDS), Dorsal Interosseous (DI), Abductor Pollicis Brevis (APB), Flexor Pollicis Brevis (FPB), Adductor Pollicis (AP These muscles can be confirmed muscle contraction through the surface electrode. Therefore, seven surface electrodes are attached to the underarm of the test subject to obtain electromyography of the selected muscle as shown in FIG. 3. Here, the wires between the surface electrodes are not shown for accurate description of the positional relationship between the surface electrodes.

The EMG signal selecting unit 200 selects EMG signals of four muscles having information for measuring force among the EMG signals of the seven muscles obtained by a predetermined discriminant.

The bearance calculation unit 300 converts the four EMG signals selected by the EMG signal selecting unit 200 into EMG signals in the time window, respectively, to convert the EMG signals in the time window converted by a predetermined dispersion algorithm. Calculate the bearance for each. In other words, the signal variance was obtained using a moving window to extract the characteristics of the signal. The window length was set to 200 msec and the window increment was set to 50 msec.

The force signal estimator 400 estimates the force calculated by the bearer calculator 300 as a force signal by an artificial neural network algorithm. The neural network algorithm presets a force signal corresponding to the bearance of the EMG signal, and estimates the preset force signal as a force signal corresponding to the bearance for each of the EMG signals in the converted time window.

The display unit 500 displays the force signal estimated by the force signal estimator 400. The display unit is, for example, a display.

A method of estimating a finger force of a subject using a finger force estimating apparatus according to an embodiment of the present invention configured as described above will be described with reference to FIGS. 4 to 8.

4 is a view for explaining the order of the finger force estimation method according to another embodiment of the present invention, Figure 5 is a view for explaining a dispersion algorithm of the finger force estimation method according to another embodiment of the present invention, 6a to 6c are diagrams for explaining an artificial neural network algorithm of the finger force estimation method according to another embodiment of the present invention, Figure 7 is a pinch force measured from the force sensor in the finger force estimation method according to another embodiment of the present invention And the pinch force estimated through the neural network.

As shown in Figure 4, the finger force estimation method according to another embodiment of the present invention, the EMG signal acquisition step (S100), EMG signal selection step (S200), bearance calculation step (S300), force signal estimation A step S400 and a display step S500 are included.

EMG signal acquisition step (S100), and obtains the EMG signal of the seven muscles that govern the muscles of the thumb and index finger from a plurality of EMG sensors mounted to be in close contact with the skin of the lower arm of the subject to be measured.

In the EMG signal selection step (S200), the EMG signals of the four muscles having information for measuring the force among the EMG signals of the seven muscles are selected by a predetermined discriminant.

In the bearance calculation step S300, the four EMG signals selected in the EMG signal selection step S200 are converted into EMG signals in the time window, and each of the EMG signals in the time window converted by the predetermined dispersion algorithm. Calculate the bearance for. Here, the dispersion algorithm, as shown in Figure 5, can see the EMG signal obtained in the first channel. Due to the nature of the EMG signal, the signal is too noisy, so the time window is used to soften the signal. Since the EMG signal is known as a Gaussian process, the variance of the EMG signal in the time window is calculated and used as a signal feature. At this time, a noisy raw EMG signal (upper graph in FIG. 5) is obtained as a smooth signal (lower graph in FIG. 5). In other words, this method is a kind of low-pass filter. Using this, a method of obtaining the bearance is as shown in Equation 1 below.

는 표준편차이고,

는 분산이며, N은 시간이고, M_i는 시간i에서의 근전도 신호의 크기이며,

는 타임윈도우 내에서의 근전도 신호들의 평균값이다.here,

Is the standard deviation,

Is the variance, N is the time, M _i is the magnitude of the EMG signal at time i,

Is the average value of the EMG signals in the time window.

In the distributed algorithm, N = 200 (the length of the time window). The EMG signal is acquired at 1 kHz, and the time window is moved every 20 msec. In other words, the process of estimating the force is called every 20 msec.

In the force signal estimating step S400, the bearance calculated in the bearance calculation step S300 is estimated as a force signal by an artificial neural network algorithm. In this case, the artificial neural network is very useful when mapping a certain input and output when it is not clearly known and simply treating it as a black box. Used.

Finally, the display step S500 displays the force signal estimated in the force signal estimation step S400 through display means such as a monitor of a computer.

As shown in FIG. 6A, the number of neurons of the hidden layer is set to 10 for the clear structure of the artificial neural network according to another embodiment of the present invention. In order to create this black box model, learning is required. This learning is to obtain EMG and force signals in advance, and to model the desired force automatically when inputting EMG signals. When the learning is properly completed, the EMG signal is acquired in real time and the neural network algorithm outputs the force value estimated by the neural network.

On the other hand, the performance of the multi-layer neural network (ie, artificial neural network) is determined by various factors, the most important is the complexity of the structure of the neural network, which can be determined depending on the complexity of the input signal and the output signal relationship. As the structure of the neural network grows, it is possible to estimate signals even in complex nonlinear relations, but there is a disadvantage in that an overfitting problem may occur and it takes a long time to learn. If the structure of the neural network is small, it takes a short time to learn, but if the relationship between the input and output signals is very complicated, they cannot be estimated well, which leads to the problem of underfitting. However, since it is difficult to know the quantitative nonlinearity of EMG and muscle force, as shown in FIG. 6B, the simulation is performed using various numbers of hidden nodes from 1 to 20 as shown in FIG. 6B. Optimized results are shown in the count. Here, two criteria, namely, NRMSE (Normalized Root Mean Squared Error) and CORR (CORRelation) are performed.

For example, we describe the results of an experiment conducted on five males with an average age of 26.4 ± 2.3 years. Each set is divided into two parts (training part, test part), and one set is performed in real time over 10 times. The training part is to perform a kind of learning process to properly estimate the pinch force when the EMG signal is input to the multilayer neural network. The training part outputs a graph as shown in FIG. 2, 4, 6, 8 N) and dynamic sine wave. The pinch force and EMG signals collected during the training part train the multilayer neural network using a backpropagation algorithm. At the end of the training session, the subject applies the pinch force as he wishes for 1 minute and compares the pinch force estimated by the EMG signal with the pinch force measured by the force sensor. . Hereinafter, the experimental results of the pinch force measured by the force sensor and the pinch force estimated by the multilayer recognition neural network as the input of the comparative analysis EMG signal will be described with reference to FIG.

As shown in FIG. 7, in the experimental results, the gray line is the actual pinch force measured from the force sensor, and the black line is the pinch force estimated through the multilayer neural network. In order to quantitatively analyze the results, the following [Equation 2] and [Equation 3] were used.

[Table 1] below shows the test results of five test subjects, and it can be seen that the pinch force can be estimated on the average of NRMSE = 0.112 ± 0.082 and CORR = 0.932 ± 0.058.

Item NRMSE CORR S1 0.130 ± 0.064 0.924 ± 0.042 S2 0.102 ± 0.024 0.940 ± 0.014 S3 0.124 ± 0.030 0.915 ± 0.014 S4 0.104 ± 0.019 0.935 ± 0.029 S5 0.101 ± 0.027 0.945 ± 0.018 Average 0.112 ± 0.082 0.932 ± 0.018

Prior to experimenting in real time, the time taken to collect EMG signals to learn multilayered neural networks is 85 seconds. The number of iterations to train multilayer recognition neural networks was limited to 100 times, and the computation time required for the experiment (Pentium 4, 2.93 GHz processor) was 64 seconds. Therefore, the time taken for EMG signal collection and learning is 2 minutes 29 seconds, and it can be seen that pinch force can be extracted from the signal with EMG of a specific user in a very short time. The reason for the shortening of the learning time is because the use of Fisher discriminant analysis not only selects channels that can efficiently perform the learning but also reduces the number of input nodes.

8 is a view for explaining the structure of the muscle determination device for finger force estimation according to another embodiment of the present invention.

As shown in FIG. 8, the muscle discrimination apparatus for finger force estimation according to another embodiment of the present invention includes an EMG signal acquisition unit 210, a reference function result calculation unit 220, and a muscle determination unit 230. ).

The EMG signal acquisition unit 210 is mounted to be in close contact with the skin of the underarm of the subject to be acquired multiple times of EMG signals for the seven muscles that govern the thumb and the index finger.

The reference function result calculator 220 inputs the EMG signal obtained a plurality of times to the reference function to calculate the respective reference function result values.

The muscle discriminating unit 230 discriminates four muscles in order of the calculated respective reference function results.

Hereinafter, a muscle discrimination method for finger force estimation according to another embodiment of the present invention using a muscle discrimination apparatus for finger force estimation according to another embodiment of the present invention configured as described above will be described with reference to FIGS. This will be described with reference to 10b.

9 is a flowchart illustrating a procedure of a muscle discrimination method for finger force estimation according to another embodiment of the present invention, and FIGS. 10A and 10B illustrate a muscle discrimination method for finger force estimation according to another embodiment of the present invention. A diagram for explaining a distributed algorithm.

As shown in FIG. 9, in the muscle determination method for finger force estimation according to another embodiment of the present invention, an EMG signal acquisition step S210, a reference function result calculation step S220, and a muscle determination step S230 are performed. ).

EMG signal acquisition step (S210) is mounted so as to be in close contact with the skin of the underarm of the subject to be acquired multiple times the EMG signal for the seven muscles that govern the thumb and index finger.

In the reference function result calculation step (S220), the EMG signal obtained a plurality of times is input to the reference function to calculate the respective reference function result values. Here, as described above, the EMG signal is regarded as a Gaussian process with an average of zero. In the embodiment of the present invention, in order to extract the characteristics of the signal, the dispersion of the signal is obtained by using a moving window, the window length of the window is 200 msec, and the window increment is 50 msec. Set to.

A total of 15 muscles govern the movement of the thumb and index finger. Only seven muscles are located close to the skin surface using ADAM Interactive Anatomy (A.D.A.M. Inc., USA) software. The seven muscles selected were Extenensor Digitorium (ED), Abductor Pollicis Longus (APL), Flexor Digitorum Superficialis (FDS), Dorsal Interosseous (DI), Short Abdominor Pollicis Brevis (APB), Short Flexor Pollicis Brevis (FPB), and Adductor Pollicis (AP). These muscles can be identified through surface electrodes. As shown in FIG. 2, seven surface electrodes were attached to the underarm of one test subject, and as shown in FIG. 10A, pinch force was generated and EMG signals were acquired five times in total. The electrode of channel 6 was used to acquire FPB electromyography, but in FIG. 10A, considering that the muscle force decreases when pinch force is generated and that it is similar to the signal of channel 5, the signal measured in channel 6 is It can be seen that the signal is measured in the APB muscle, not the FPB muscle. It is presumed that anatomical FPB muscles were largely obscured by APB muscles, so no EMG signal was obtained. Although the signals acquired in channels 5 and 6 are measured from one muscle of APB, they are activated by a large number of motor units of one muscle, each of which is the rate, force of muscle contraction. Because of the different characteristics, the measured signal does not exactly match. Fisher discriminant analysis was used to select the channels with the most effective information among the seven measured signals to estimate the pinch force. That is, the Fisher discrimination equation is a reference function of [Equation 4], and is evaluated using it.

와

는 각각 i번째 클래스 특징신호의 평균과 분산을 의미한다. 기준함수에서 분자는 클래스간의 특징신호 차이가 클수록 커지게 되며, 분모는 각 클래스간의 특징신호 편차가 작을수록 작아져서 전반적으로 클래스간의 분리도가 크게 되면 기준함수의 결과가 크게 된다. 따라서, k번째 기준함수 결과가 가장 크게 나왔다면 k번째 채널에서 측정된 신호는 핀치 힘을 추정하는데 가장 효과적인 정보를 제공한다고 생각할 수 있다.This function is useful for solving two pattern recognition problems. The k on the left side of the equation and the numbers 1 and 2 on the right side represent the measured channel number and pattern class, respectively.

Wow

Denote the mean and variance of the i th class feature signal, respectively. In the reference function, the numerator becomes larger as the feature signal difference between classes becomes larger, and the denominator becomes smaller as the feature signal deviation between classes becomes smaller. Therefore, if the k th reference function result is the largest, it can be considered that the signal measured in the k th channel provides the most effective information for estimating the pinch force.

Using this method, the signal measured in FIG. 10A was applied to the reference function of Equation [4], class 1 was classified as when the pinch force was 0 N, and class 2 was classified as when the pinch force was 8 N. do. EMG signals are proportional to muscle force, so channels that can be well classified at low and high pinch forces are also effective in estimating the continuous force between 0 and 8 N. FIG. 10B shows a result of Fisher discriminant analysis using signals acquired five times, and it can be seen that signals of channels 4 to 7 provide the most effective information for estimating pinch force. Therefore, in the exemplary embodiment of the present invention, pinch force is measured in real time from a plurality of subjects using only electrodes of channels 4 to 7.

Then, the muscle determination step (S230), the four muscles are discriminated in order of the calculated respective reference function results.

Therefore, in the embodiment of the present invention configured as described above, the finger force is estimated from the EMG in real time, and the muscles governing the thumb and the index motion can be selected and observed by the surface electrode in order to estimate the pinch force from the EMG signal. Signals are obtained from seven muscles, four of which are most efficient for estimating forces through Fisher linear discriminant analysis, and finger force estimates using multilayer perceptrons. can do.

For this reason, the finger force estimation method and the estimation device according to the present invention, the muscle determination method and the muscle determination device for finger force estimation determine the electrode position effective for estimating the pinch force using Fisher discriminant analysis. By estimating the force using the number of various hidden nodes to determine the neural network structure, the pinch force can be estimated even though the learning time is very short.

The results of this study can be used to manipulate objects in the use of prostheses by those whose thumb and index finger are cut, and the power of human beings to directly control dangerous tasks when performing remote control robots. It may also be used in a manner.

As such, the technical configuration of the present invention described above can be understood by those skilled in the art that the present invention can be implemented in other specific forms without changing the technical spirit or essential features of the present invention.

Therefore, the exemplary embodiments described above are to be understood as illustrative and not restrictive in all respects, and the scope of the present invention is indicated by the following claims rather than the detailed description, and the meaning and scope of the claims and their All changes or modifications derived from equivalent concepts should be construed as being included in the scope of the present invention.

1 is a view for explaining the peripheral structure of the finger force estimation apparatus according to an embodiment of the present invention.

2 is a view for explaining the structure of the finger force estimation apparatus according to an embodiment of the present invention.

3 is a diagram illustrating an arm of a person to be measured equipped with an EMG sensor in a finger force estimating apparatus according to an embodiment of the present invention.

4 is a view for explaining the procedure of the finger force estimation method according to another embodiment of the present invention.

5 is a view for explaining a dispersion algorithm of the finger force estimation method according to another embodiment of the present invention.

6A to 6C are diagrams for explaining an artificial neural network algorithm of a finger force estimation method according to another embodiment of the present invention.

7 is a view showing the pinch force measured from the force sensor and the pinch force estimated through the neural network in the finger force estimation method according to another embodiment of the present invention.

8 is a view for explaining the structure of the muscle discriminating device for finger force estimation according to another embodiment of the present invention.

9 is a flowchart illustrating a procedure of a muscle discrimination method for finger force estimation according to another embodiment of the present invention.

10A and 10B illustrate a dispersion algorithm of a muscle discrimination method for finger force estimation according to another embodiment of the present invention.

Claims (10)

An EMG signal acquiring step of acquiring EMG signals of seven muscles that govern the muscles of the thumb and the index finger from a plurality of EMG sensors mounted to closely contact the skin of the underarm of the person to be measured; An EMG signal selecting step of selecting an EMG signal of four muscles having information for measuring force among the EMG signals of the seven muscles according to a predetermined discriminant; A luminance calculation step of converting the selected four EMG signals into EMG signals in a time window, and calculating a bearance for each of the EMG signals in the converted time window by a predetermined dispersion algorithm; A force signal estimating step of estimating the calculated bearance as a force signal by an artificial neural network algorithm; And Displaying the estimated force signal; Including, finger force estimation method. The method of claim 1, In the EMG signal acquiring step, the EMG signal is a Gaussian process having an average of 0, Finger force estimation method. The method of claim 1, In the force signal estimating step, the artificial neural network algorithm presets a force signal corresponding to the bearance of the EMG signal, and sets the preset force signal for each of the EMG signals in the converted time window. Finger force estimation method. A plurality of EMG sensor units mounted in close contact with the skin of the underarm of the person to be measured and acquiring EMG signals of seven muscles that govern the muscles of the thumb and the index finger; An EMG signal selecting unit for selecting EMG signals of four muscles having information for measuring force among the acquired EMG signals of the seven muscles by a predetermined discriminant; A bias calculator for converting the selected four EMG signals into EMG signals in a time window, and calculating a bearance for each of the EMG signals in the converted time window by a predetermined dispersion algorithm; A force signal estimator for estimating the calculated bearance as a force signal by an artificial neural network algorithm; And A display unit which displays the estimated force signal; Including, finger force estimation device. The method of claim 4, wherein The EMG signal is a Gaussian process with an average of zero, Finger force estimation device. The method of claim 4, wherein The neural network algorithm presets a force signal corresponding to the bearance of the EMG signal, and converts the preset force signal into a force signal corresponding to the bearance for each of the EMG signals in the converted time window. Estimate, finger force estimation device. An EMG signal acquiring step of acquiring an EMG signal for seven muscles mounted to be in close contact with the skin of the underarm of the subject under control of the thumb and the index finger; A reference function result calculating step of inputting an EMG signal obtained a plurality of times to a reference function to calculate each reference function result; Muscle discrimination method for finger force estimation comprising a muscle discrimination step of discriminating the four muscles in the order that the calculated respective reference function results are larger. The method of claim 7, wherein The EMG signal is a Gaussian process with an average of zero, Muscle discrimination method for finger force estimation. An EMG signal acquiring unit, which is mounted to be in close contact with the skin of the underarm of the person to be measured and acquires EMG signals for the seven muscles that govern the thumb and the index finger a plurality of times; A reference function result calculator for inputting an EMG signal obtained a plurality of times to a reference function to calculate respective reference function result values; Muscle discriminating apparatus for finger force estimation comprising a muscle discriminating unit for discriminating the four muscles in the order that the calculated respective reference function results are larger. 10. The method of claim 9, The EMG signal is a Gaussian process with an average of zero, Muscle Discrimination Device for Finger Force Estimation.

Images