KR20160023976A - A Multiple Gait phase Recognition Method using Boosted Classifiers and Classification Matrix based on Surface EMG Signals - Google Patents

Abstract

(Heel strike, Mid stance, Toe off) using two classifiers (Heel LDA, Toe LDA) and only three surface EMG signals (sEMG, surface electromyography) without physical sensors The present invention relates to a multi-gait step recognition method using a surface EMG signal-based boost classifier and a classification matrix for classifying a swing of an EMG signal. The method includes: (a) receiving a sample signal of an EMG signal of a plurality of channels, Generating a classifier to classify a classifier; (b) recognizing the walking operation with the classifier.
By the above method, it is possible to classify the walking step using only the surface electromyogram (sEMG) signal without using a physical sensor.

Description

TECHNICAL FIELD [0001] The present invention relates to a multi-gait phase recognition method using a Boost classifier based on surface EMG signals and a classification matrix,

(Heel strike, Mid stance, Toe off) using two classifiers (Heel LDA, Toe LDA) and only three surface EMG signals (sEMG, surface electromyography) without physical sensors The present invention relates to a multi-gait step recognition method using a classifier matrix and a Boost classifier based on a surface electromyogram signal, which classifies swing stages.

Recently, the importance of convergence technology has expanded, and studies combining IT and BT have been actively conducted. Typically, EMG (electromyography) is used to control power prostheses used by limb amputees.

These studies are helping people with disabilities live their lives. General femoral prostheses are controlled using physical sensors such as angles and pressures [Non-Patent Document 1]. However, there is a disadvantage in that the method using the physical sensor must always walk at the same speed because it reproduces only the pre-trained walking motion.

In order to solve these problems, studies are underway to analyze the walking phase by applying EMG, which is a biological signal. EMG is a method of measuring the electrical activity that occurs when a muscle contracts. Therefore, it is possible to measure detailed motions such as subject's motion intention, walking speed, and walking stride [Non-Patent Document 2].

The gait phase classification performed for power athletic control is largely divided into stance and swing. It refers to the condition that the foot of the leg determined by the standard of the stance is touched to the ground. It is divided into five stages (Heel strike, Foot flat, Mid stance, Heel off, Toe off). It is classified into three stages (Acceleration, Mid swing, Deceleration) in detail.

[Non-Patent Document 1] E.C. Martinez-Villalpando and H. Herr, "Agonist-antaagonist active knee prosthesis: A preliminary study in level-ground walking," Journal of Rehabilitation Research & Development, vol. 46, pp. 361-373, 2009. [Non-Patent Document 2] C. Jensen, O. Vasseljen, R. H. Westgaard, "The Influence of Electrode Position on Bipolar Surface Electromyogram Recordings of the Upper Trapezius Muscle," J. Appl Physiol., Vol. 67, pp. 266-273, 1993. [Non-Patent Document 3] Zygote media groups, Inc., Zygote Body 3D data, 2012. [Non-Patent Document 4] A.N. Donald, N. David, "Kinesiology of the Musculoskeletal System," Mosby, 2009. [Non-Patent Document 5] JH Yu, DH Lee, D.H. Kim, "Surface EMG Signal Based Hand Motion Recognition Method Using Multi-class SVM with Electromyogram Signals", Proceedings of JCCIS 2012, The Third Conference on Computer and Information Science, Vol. 06, No. 2, pp. 51-53, 2012. [Non-Patent Document 6] H. Huang, TA Kuiken and RD Lipschutz, "A Strategy for Identifying Locomotion Modes Using Surface Electromyography," Biomedical Engineering, IEEE Transactions on, vol. 56, no. 1, pp. 65-73, 2009. [Non-patent Document 7] D. H. Lee, S. L. Lee and D. H. Kim, "Implementation of a prototype system for surface EMG analysis based on Gait Phases," submitted to Yanbian University of Science and Technology 2012 International Symposium, June, 2012.

SUMMARY OF THE INVENTION The object of the present invention is to solve the above-mentioned problems, and it is an object of the present invention to solve the above problems by using only a surface electromyography (sEMG) signal without using a physical sensor and using two classifiers (Heel LDA, Toe LDA) The present invention provides a multi-gait step recognition method using a classifier matrix and a Boost classifier based on the surface EMG signal, which classifies three stages (Heel strike, Mid stance, Toe off) and swing step.

According to an aspect of the present invention, there is provided a multi-gait step recognition method using a superficial class EMG signal-based boost classifier and a classification matrix, the method comprising the steps of: (a) receiving sample signals of EMG signals of a plurality of channels, Generating a classifier; (b) recognizing the walking operation with the classifier.

The present invention also provides a multi-gait step recognition method using a surface EMG signal-based boost classifier and a classification matrix, wherein (a) comprises: (a1) walking movement data and N Acquiring training data of the EMG signal of the channel; (a2) extracting characteristic values by M feature extraction algorithms for each channel (M is a natural number equal to or greater than 2) for each training data to obtain N × M training characteristic vectors; And (a3) generating a classifier using the training feature vector as an input value and the walking motion as an output value using the training feature vectors.

The present invention further provides a multi-gait step recognition method using a surface EMG signal based boost classifier and a classification matrix, wherein the step (b) comprises: (b1) receiving an EMG signal of the actual walking of the N channels in real time, Extracting EMG signal data of a predetermined window size; (b2) extracting characteristic values of the actual walking data by the M feature extraction algorithms for each channel with respect to the EMG signal data of the actual walking data to obtain characteristic vectors of the N * M dimensional actual walking data; And (b3) inputting the feature vector of the actual gait, and identifying the gait of the actual gait through the classifier.

Further, the present invention is a multi-gait step recognition method using a surface EMG signal-based boost classifier and a classification matrix, wherein the gait operation consists of on / off heel operation and on / off heel operation, And a heel sorter for sorting the heel action.

Further, the present invention is a multi-gait step recognition method using a classifier matrix and a superficial classifier based on a surface electromyogram signal. In the step (b3), the on / off operation of the forefoot and the on / Wherein a classification matrix indicating an operation is constructed so that the results of the forefoot classifier and the heel classifier are identified through the classification matrix.

Further, the present invention is a multi-gait step recognition method using a surface EMG signal-based boost classifier and a classification matrix, wherein in the step (a1), the walking operation data is obtained through a pressure sensor, The window size is extracted by averaging the time between operations for each of the forefoot operation and the heel operation, and the window sizes of the forefoot classifier and the heel classifier are obtained independently of each other.

Further, the present invention is characterized in that in the step (a1), the window size is independently determined for each of the four walking motions in the multi-gait step recognition method using the surface electromyogram signal-based boost classifier and the classification matrix.

Further, the present invention provides a multi-gait step recognition method using a superficial class EMF signal-based boost classifier and a classification matrix, wherein in step (b1), a window size of a corresponding gait operation is used in accordance with a gait step .

According to another aspect of the present invention, there is provided a method of recognizing a multi-gait step using a classifier matrix and a Boost classifier based on a surface electromyogram signal, wherein the gait step comprises three stages of stance and one stage of swing do.

As described above, according to the multi-gait step recognition method using the surface EMG signal-based boost classifier and the classification matrix according to the present invention, it is possible to classify the gait steps using only the surface electromyogram (sEMG) Is obtained.

FIG. 1 is a block diagram of a configuration of an overall system for performing a multi-gait step recognition method using a classification matrix and a Boost classifier based on a surface electromyogram signal according to an embodiment of the present invention.
FIG. 2 is a view showing a walking step of a human being used in the present invention. FIG.
3 is a flow chart illustrating a method of recognizing multiple walking steps using a surface EMG signal based boost classifier and a classification matrix according to an embodiment of the present invention.
4 is a flow block diagram illustrating the overall process of a multi-gait step recognition method using a surface electromyogram (sEMG) signal based boost classifier and a classification matrix of the present invention.
5 is a graph showing a walking operation and acquired data according to an embodiment of the present invention.
6 is a table showing a classification matrix according to an embodiment of the present invention.
7 is a view showing an attachment position of an EMG signal sensing electrode according to an embodiment of the present invention.
8 is a table showing the accuracy according to the window size of the heel according to the experiment of the present invention.
9 is a graph showing the accuracy according to the window size of the toe according to the experiment of the present invention.
10 is a graph comparing the accuracy of the supervisor method and the realtime method by the classifier for the heel and the toe according to the experiment of the present invention.
11 is a graph of the accuracy of classifying gait steps in a real-time manner according to an experiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings.

In the description of the present invention, the same parts are denoted by the same reference numerals, and repetitive description thereof will be omitted.

First, a configuration of an overall system for implementing a multi-gait step recognition method using a classification matrix and a surface electromyogram signal based boost classifier according to an embodiment of the present invention will be described with reference to FIG.

As shown in FIG. 1, the multi-gait step recognition method using the classification matrix and the surface EMG signal-based boost classifier according to the present invention receives the EMG signal 11 or the gait pressure signal 12, To the program system 30 on the computer terminal 20 that recognizes the program. That is, the surface EMG signal-based boost classifier and the multiple gait step recognition method using the classification matrix can be implemented as a program and installed in the computer terminal 20 and executed. A program installed in the computer terminal 20 can operate as a single program system 30. [ Here, the computer terminal 20 refers to a device having a computing function for processing a program.

Meanwhile, as another embodiment, the multi-gait step recognition method using the surface electromyogram signal based boost classifier and the classification matrix is constituted by a program and is composed of one electronic circuit such as an ASIC . Or as a dedicated terminal 30 for processing only recognition of the walking step. This is referred to as a surface electromyogram signal-based gait step recognition device 30. Other possible forms may also be practiced.

Next, a walking step (or a walking operation) to be recognized in the present invention will be described with reference to FIG.

As shown in FIG. 2, the walking stage of the human being is divided into four stages of stance (Stance) and four stages of swing (swing). The stance refers to the state where the foot of the leg which is a standard touches the ground, and the feet are apart from the ground.

As shown in FIG. 2, the stance phase is divided into an initial contact, a loading response, a mid stance, and a terminal stance depending on the joint angle and the position of the sole contacting the ground Respectively. The swinging period is divided into pre swing, initial swing, mid swing, and terminal swing according to the joint angle and the instantaneous speed. In the present invention, the initial contact (initial contact), the mid stance (last stance), and the last stance (terminal stance), which are three stages of stance which are the core of the walking step, Four steps are distinguished.

Next, a method of analyzing an EMG signal for practicing the present invention will be described.

The EMG signal can be used to analyze muscle activity using IEMG, Integrated EMG, Average Value, Peak, Mean Value, and Threshold. The IEMG integrates the measured signals until the end of a specific operation and converts it to a single feature value. Average Value and Peak are the feature values of the average and maximum values of the measured signals until the start and end of a specific operation. The Mean value extracts a value located at the center of the maximum and minimum values of the measured signal until the end of a specific operation. Finally, the Threshold is a method of executing or transmitting an instruction every time the reference is exceeded after defining an instruction according to muscle activity.

That is, if you set up 1.3mv in the upward direction and 2.6mv in the left direction, if the maximum value is 1.5mv when the fist is given, the upward direction command is transmitted. If 2.8mv is outputted, the left direction command is transmitted. The existing EMG signal analysis method was developed for measuring muscle activity. Therefore, the diet is simple and focused on activity. Therefore, there is a disadvantage in that the method of recognizing a pattern through a complex combination such as gait analysis has a low accuracy.

In recent years, EMG signal analysis methods using signal processing algorithms have been proposed for motion recognition of arms, walks, and backs. Typically, there are feature extraction algorithms using VAR (variance of EMG), wAMP (Willision amplitude), root mean square (RMS), mean absolute value (MAV), zero crossing (ZC) and sign slope change (SSC). The values are converted into feature values by using the variance, the mean of the amplitude, the effective value, the average of the absolute values, the number of repetitions, and the slope of the amplitude. In the present invention, six feature extraction algorithms, VAR, MAV, VAR, WAMP, ZC, and SSC, are used to convert the feature values.

A method of recognizing a multi-gait step using a surface electromyogram (sEMG) signal based boost classifier and a classification matrix according to an embodiment of the present invention will be described in detail with reference to FIG.

As shown in FIG. 3, the gait step classification method based on the EMG signal according to the embodiment of the present invention is divided into (a) training data generation step S10 and (b) operation identification step S20 Literature 5]. 4 is a block diagram illustrating a walking step recognition method according to the present invention.

First, the training data generation step (S10) will be described.

The first step (S11) of the training data generation step (S10) acquires the electromyogram signal of the muscle to be measured for each channel.

At this time, in order to generate the classifier according to the walking motion, two pressure sensors are additionally measured to confirm that the heel and the toe touch the ground. 5 is a graph showing a walking operation and acquired data.

The acquired data is supervised classification method. EMG signals are extracted and used as training data when the heel touches the ground and when it does not touch, when the toe touches the ground, and when it does not touch.

Supervised method is a method in which an experimenter or a trainer directly extracts a corresponding gait operation and inputs it to a training or classifier in order to obtain accurate results in a training or an experiment. That is, if training is performed on the acquired initial stance, the EMG signal corresponding to the initial stance is directly extracted using a pressure sensor, and then the EMG signal is trained. The classification is also extracted in the same way and input to the classifier to check whether the result is the initial stance.

In the second step S12, the feature value extraction step extracts feature values for each input data channel. In the present invention, a total of six feature extraction algorithms were used: VAR (variance of EMG), waveform length (WL), root mean square (RMS), mean absolute value (MAV) and zero crossing.

Data which is changed from EMG signal to feature value by using the feature extraction algorithm is converted into one-dimensional data through recombination [Non-Patent Document 5]. That is, N-dimensional data measured with N channels are converted into data composed of N x 6 characteristic values.

The final third step (S13) is to create a heel classifier and a toe classifier. Each classifier classifies when each part touches the ground and when it falls.

The classifier receives the EMG signal generated from each muscle using the feature extraction. Classifiers are classified using LDA (linear discriminant analysis). This is a linear discriminant analysis to compare training data and input feature values. That is, the classifier determines the walking phase using the input EMG signal. When the EMG signal is received, the feature is extracted, and the EMG signal is input to the classifier, the result of the gait step can be obtained.

The Heal classifier classifies the heel as 1 when it touches the ground and 0 as it falls. The toe sorter classifies whether or not the forefoot is touching the ground, and the contents are the same as the heal sorter. A classifier group is a bundle of heal classifiers and toe classifiers. That is, when the feature value is input, the result is obtained from the heal classifier, and the result is obtained from the toe classifier. Therefore, it is called a classifier group because it uses two classifiers.

Next, the operation identifying step S20 will be described.

The first step S21 of the operation identifying step extracts an EMG signal inputted in real time by a window size defined in advance, in order to extract an EMG signal.

The extraction of the window size is performed by extracting EMG signals input in real time as much as the window size, creating characteristic data, and performing classification using the extracted EMG signals. For example, assuming that a window size is 500 ms and 1000 pieces (1 KHz) of data are input per second, 500 to 500 window sizes are extracted in one step, and then feature extraction and classification are performed. Then, in 2 steps, the EMG signal is extracted from 2 to 501 by 1 ms, and feature extraction and classification are performed. Repeat this way continuously.

The second step S22 proceeds in the same manner as the training data generation step by the feature value extraction.

In the third step S23, the feature value obtained through the preceding step S22 is inputted to the two classifiers to output the result.

As the last step (S24), the data obtained from the two classifiers are finally finally identified through the Classification Matrix of FIG. 6 to the gait operation.

FIG. 4 shows a method of identifying the walking using the extracted classifier value, and the uppermost box in FIG. 4 is the EMG signal for the walking operation.

Next, the effects of the present invention through experiments will be described in detail.

First, the experimental environment and method for the effect of the present invention will be described.

BioPAC's MP150 and BM-EMG2 models were used for the experiments. The analysis program was developed using Matlab 2013. The EMG signals of the 4 channels (Rectus femoris, Vastus lateralis, Vastus medialis, Semitendinosus) and 2 pressure sensors (Heel, Toe) were obtained and the attachment positions are shown in FIG. 7 is a view showing an attachment position of an EMG signal sensing electrode according to an embodiment of the present invention.

A total of three subjects participated in the experiment, and each participant repeated a 50-step flat gait operation to obtain training data and input data. The accuracy of the gait recognizer was calculated using Equation (1).

[Equation 1]

In the experiment, input data was extracted by two different methods, supervised classification method and real - time classification method. The supervisory classification method extracts signals corresponding to the gait steps using a pressure sensor. The real-time classification method continuously extracts input data while moving according to a predefined window size while receiving a signal.

More specifically, the supervisory classification method extracts accurate data using the measured data (offline), and confirms that the results are appropriate. In this case, the window size is meaningless, and the corresponding EMG signal is extracted, and the characteristic value is created, and the result is put into the classifier. That is, the part corresponding to the initial stance of the measured EMG signal is extracted and inserted into the classifier to confirm that the result of initial stance is obtained.

The real-time classification method is a method of confirming whether the walking step is properly classified when the EMG signal is inputted (on-line method) in real time, not in the supervision classification offline mode. That is, if EMG signal is input at 1KHz, 1000 EMG signals per second will be input. At this time, since 1000 data includes 1 to 4 steps of walking, it is necessary to cut and input using window size. Accordingly, the input window size is extracted as much as the window size set for the condition, and the feature value is obtained and classified. At this time, as described above, if the classification is 500 ms, one step is equivalent to 1 to 500, 2 steps are 2 to 501, and 3 steps are 3 to 502 .

Experimental results by the above-described experimental environment and method will be described.

Experiments were conducted to determine the optimal window size for the Heel classifier and the Toe classifier during real-time classification. FIG. 8 shows the accuracy according to the window size of the heel, and FIG. 9 shows the accuracy according to the window size of the toe.

The mean value of the gait step obtained from the pressure sensor was extracted as the optimum window size regardless of the classifier in both the heel and the toe. Thus, the window size was 502 ms for the heel and the toe was 927 ms. Next, we compare the accuracy of supervisor type and realtime type classifier for heel and toe. Fig. 10 shows the experimental results thereof.

As described above, the window size is for distinguishing a walking step from an EMG signal input in real time in a real-time classification method. That is, the window size refers to an extraction area that is determined to classify a walking step by extracting a specific section.

On the other hand, when the training data is generated, it is known that the feature value is extracted by extracting the corresponding region. In other words, during the training process, the first stance is extracted by 250 ms, the second stance is extracted by 300 ms, and the third stance is extracted by 350 ms. The reason why the size of the window is extracted differently is that the size of the walking is different for each person in the walking step.

Each of the walking motions progresses sequentially in accordance with the walking step. For example, the walking progresses sequentially, such as 1-> 2-> 3-> 4-> 1-> 2-> 3-> 4. That is, there is no case in which the number 1 movement suddenly jumps to number 3. Therefore, if it is determined that the initial contact is the initial contact in the previous gait, the next is the mid stance. The contents in FIG. 2 indicate that the current contact is the mid stance if it is the initial contact. Accordingly, the window size of the corresponding walking operation is selected and used according to the walking step.

On the other hand, if a person performs a fixed walk, the window size is meaningless. However, people walk to areas within 1.1 ~ 1.4s of normal walking. Therefore, if the training is performed 20 times, if the average corresponding to the stance and the average corresponding to the swing are determined as the window size, the walking area may be equally lost or even more, . That is, it is not that a person moves fixedly like a robot. Therefore, when the EMG signal is inputted in real time, the present walking step can not be known. Therefore, the window size is determined by averaging. The current walking phase is known as the result of putting in the classifier. That is, it can be known at the end.

The supervisor method has a 13% higher accuracy than the real time method as a whole. The accuracy of the heel and the toe was about 6% higher than that of the heel regardless of the classifier. Overall SVM showed higher accuracy than LDA. SVM_Toe showed the highest accuracy with 94% when Supervisor and 79% when Real Time.

Next, we conducted experiments to classify walking steps in real time. 11 shows the result thereof.

In the present invention, a method of classifying the gait steps into four levels using a two-level classifier and a classification matrix based on EMG signals of the body has been described. Experimental results showed that the average time of the walking step was the window size for the real - time classification. Therefore, according to the present invention, it is possible to classify the walking step only by the EMG signal.

The invention made by the present inventors has been described concretely with reference to the embodiments. However, it is needless to say that the present invention is not limited to the embodiments, and that various changes can be made without departing from the gist of the present invention.

10: EMG signal 12: Walking pressure signal
20: computer terminal 30: walking step recognition device

Claims (9)

A multi-gait step recognition method using a surface EMG signal-based boost classifier and a classification matrix,
(a) receiving a training data of an EMG signal of a plurality of channels and generating a classifier for classifying the EMG signal;
(b) recognizing the walking operation with the classifier, the multi-step walking step recognition method using the surface EMG signal based boost classifier and the classification matrix.
The method of claim 1, wherein the step (a)
(a1) obtaining walking data for walking, and training data for N (N is a natural number of 2 or more) channels of EMG signals for the walking operation;
(a2) extracting characteristic values by M feature extraction algorithms for each channel (M is a natural number equal to or greater than 2) for each training data to obtain N × M training characteristic vectors; And
(a3) generating a classifier that uses the training feature vector as an input value and the walking operation as an output value using the training feature vectors, and generates a classifier based on the surface EMG signal-based boost classifier and the classification matrix Multiple walking step recognition method.
3. The method of claim 2, wherein step (b)
(b1) receiving an EMG signal of the actual walking of the N channels in real time and extracting EMG signal data of a predetermined window size;
(b2) extracting characteristic values of the actual walking data by the M feature extraction algorithms for each channel with respect to the EMG signal data of the actual walking data to obtain characteristic vectors of the N * M dimensional actual walking data; And
(b3) inputting the feature vector of the actual gait, and identifying the gait of the actual gait through the classifier, wherein the step of recognizing the gait of the actual gait includes the step of recognizing the gait step based on the surface EMG signal- Way.
The method of claim 3,
Wherein the walking operation is constituted by on / off heel operation and on / off heel operation, and the classifier comprises a heel classifier for classifying the forefoot operation and a heel classifier for classifying the heel operation. Based Boost Classifier and Multiple Gait Step Recognition Method Using Classification Matrix.
5. The method of claim 4,
In the step (b3), a classification matrix indicating four walking motions is formed by on / off of the forefoot operation and on / off of the heel action, and the results of the forefoot classifier and the heel classifier are transmitted through the classification matrix And identifying a gait operation based on the surface EMG signal.
6. The method of claim 5,
Wherein the walking motion data is acquired through a pressure sensor and a window size is extracted by averaging the time between motions for each of the forefoot motion and the heel motion extracted by the pressure sensor, Wherein the window size of the heel classifier is obtained independently of each other.
6. The method of claim 5,
The method of claim 1, wherein in step (a1), a window size is independently calculated for each of the four walking motions, and a multiple walking step recognition method using a classifier matrix and a surface EMG signal based boost classifier.
The method according to claim 6,
Wherein the window size of the walking movement is used according to a walking step that is sequentially performed in the step (b1), wherein the multi-step walking step recognition method uses the surface EMG signal based boost classifier and the classification matrix.
9. The method according to any one of claims 1 to 8,
Wherein the gait step comprises three steps of stance and one step of swing. The multi-gait step recognition method using the superfluous classifier based on the surface EMG signal and the classification matrix.

Images