KR101705082B1 - A Gait Phase Recognition method based on EMG Signal for Stairs Ascending and Stairs Descending - Google Patents

Abstract

The present invention relates to a method of recognizing a gait step based on an EMG signal for an up and down stairs, comprising the steps of: (a) training signals of a plurality of muscle EMG signals measured during an uphill or downhill walk; Receiving physical data of a knee angle; (b) determining a walking step using the physical data; (c) extracting features from the EMG signal; (d) generating a classifier that takes the extracted features as input and outputs the determined gait steps; And (e) recognizing a walking step of the uphill or downhill walk using the generated classifier.
According to the above-described method, the stepping step can be more accurately classified by classifying the uphill and downhill walking using only the EMG signal.

Description

[0001] The present invention relates to an EMG signal-based walking step recognition method for an up-down stairs,

The present invention relates to a method of recognizing a gait step based on an EMG signal for an up / down stair walking, in which an up-and-down stairs gait is divided into four steps using only an EMG signal without using a physical sensor.

In the present invention, the EMG signal is calculated by RMS, VAR, MAV, SSC, ZC, and WAMP characteristics, and the gait step is recognized by the LDA (Linear Discriminant Analysis) classifier. In the training step, And a gait step recognition step for generating a gait step range according to an EMG signal for an up / down stairs walk.

Recently, research is being actively conducted to analyze human walking characteristics. Walking such as fast walking, slow walking, stair climbing, and stair descending have different degrees of muscle and muscle activity depending on each walking mode [Non-Patent Document 1]. In particular, the upward and downward stairs have different gait characteristics from those of the flat because they need to keep walking and maintaining the body balance and center of gravity. In addition, since the stair walking requires a lot of movement of the knee joint and the ground reaction force, a lot of lower leg strength is needed compared to the flat walking. Therefore, there is a growing need for power prostheses and muscle aids for stair-climbing for severed patients and those with insufficient leg strength.

Walking steps should be well classified to naturally drive power legs for stairs. Conventional studies on classification of stairway walking steps using power legs are classified through physical sensors (pressure, acceleration, angle, motion camera, etc.) [Non-Patent Document 2-4]. This method can affect the prosthetic weight because the sensor must always be mounted on the power leg. In addition, it has a disadvantage in that a physical sensor is used instead of walking according to the user's intention, and is reproduced only at a pre-trained speed, so that a person can receive a feeling of being attracted to a motorist's prosthesis.

[Non-Patent Document 1] HJ. Cho, K-H. Kang, H-C. Park, M-S. Kang, Y-S. Choi, T-K. Kim, S-J. Yoon, "Effect of Inclined Backward and Forward Walking Training on Muscle Strength and Electromyographic Activity ", Journal of the Society of Living Environment System of Korea, Vol.16, No.2, pp186-193, April 2009. [Non-Patent Document 2] J-Y. Jung, Y-S. Yang, Y-W. Won, JJ Kim, "Development of Wireless Ambulatory Measurement System based on Inertial Sensors for Gait Analysis and its Application for Diagnosis on Elderly People with Diabetes Mellitus", Journal of the Institute of Electronics Engineers of Korea, Vol. 48, No. 2, pp38-46, March 2011. [Non-Patent Document 3] J-Y. Lee, K-J. Lee, Y-H. Kim, S-H. Lee, S-W. Park, "Development of Gait Analysis Algorithm for Hemiplegic Patients Based on Accelerometry ", Journal of the Institute of Electronics Engineers of Korea, Vol.41, No.4, pp.231-240, July 2004. [Non-patent Document 4] M-G. Chae, J-Y. Jung, C-J. Park, I-H. Jang, H-S. Park, "Gait Phases Classification using Joint Angle and Ground Reaction Force: Application of Backpropagation Neural Networks ", Journal of Institute of Control & Robotics and Systems, Vol. 18, No. 7, pp. 644-649, July 2012. [Non-Patent Document 5] J-H. Ryu, D-H. Kim, "sEMG Signal Based Gait Phase Recognition Method for Selecting Features and Channels Adaptively ", Journal of The Rehabilitation Engineering and Assistive Technology Society, Vol. 7, No. 2, pp. 19-26, December 2013. [Non-Patent Document 6] Alison C. Novak, Samantha M. Reid, Patrick A. Costigan, Brenda Brouwer, "Extension of Stair Negotiation in the Elderly", Lower Extremity Review, October 2010 [Non-Patent Document 7] H-S. Cho, Y-H. Chang, J-C Ryu, M-S. Mun, C-B. Kim, "Analysis of Stair Walking Characteristics for the Development of Exoskeletal Walking Assist Robot", Journal of The Rehabilitation Engineering and Assistive Technology Society of Korea, Vol.6, No.2, pp.15-22, December 2012. [Non-Patent Document 8] E-S. Kim, "Trajectory Generation Schemes for Bipedal Ascending and Descending Stairs using Genetic Algorithm (GA) ", University of Dong-A, February 2010. [Non-Patent Document 9] H. Huang, T.A. Kuiken and R.D. Lipschutz, " A Strategy for Identifying Locomotion Modes Using Surface Electromyography, "The Journal of Biomedical Engineering, IEEE Transactions on, Vol. 56, No. 1, pp. 65-73, January 2009. [Non-Patent Document 10] A. Phinyomark, S. Hirunviriya, C. Limsakul, P. Phukttaranont, "Evaluation of EMG Feature Extraction for Hand Movement Recognition Based on Euclidean Distance and Standard Deviation", in Proc. of ECTI-CON Conf. on 2010 International Conference, pp. 856-860, Chiang Mai, Thailand, May 2010. [Non-Patent Document 11] Boris I. Prilutsky, Ludmila N. Petrova, Leonid M. Raitsin, "Comparison of Mechanical Energy Expenditure of Joint Moments and Muscle Force During Human Locomotion ", The Journal of Biomechanics, Vol. , pp. 405-415, April 1996. [Non-Patent Document 12] S-H. Kim, J-H. Ryu, D-H. Kim, "Gait phase classification for staging walking using Feature Extraction and Muscle selection based on EMG Signals ", in Proc. of IEEK Conf. on Summer Conference, Vol. 37, No. 1, pp. 1053-1056, Jeju, Korea, June 2014. [Non-Patent Document 13] A. Phinomark, P. Phukpattaranont, C. Limsakul, "Feature reduction and selection for EMG signal classification", The Journal of Expert Systems with Applications, Vol. 39, No. 8, pp. 7420-7431 , June 2012.

An object of the present invention is to solve the above problems, and it is an object of the present invention to provide a method and apparatus for calculating an electromyogram signal by RMS, VAR, MAV, SSC, ZC and WAMP characteristics and recognizing a walking step through an LDA (Linear Discriminant Analysis) The present invention provides a method for recognizing a gait step based on an EMG signal for an up / down stairs walk, which generates a gait step range according to a change in a knee angle using an AHRS (Attitude & Heading Reference System) sensor.

Particularly, the object of the present invention is to classify the steps of uphill and downhill stairs using EMG signals only. It is possible to classify the range of walking steps by using the stairway walking observation of a normal person, The present invention provides a method of classifying the walking steps of up and down stairs into four steps using only the EMG signal, generating a walking step range according to the knee angle change using the AHRS sensor in the training step.

In order to achieve the above object, the present invention relates to a method for recognizing a gait step based on an EMG signal for up and down stairs, comprising: (a) training signals of a plurality of muscle EMG signals measured during an uphill or downhill walk; Receiving physical data of the measured sole pressure and the knee angle; (b) determining a walking step using the physical data; (c) extracting features from the EMG signal; (d) generating a classifier that takes the extracted features as input and outputs the determined gait steps; And (e) recognizing the walking step of the uphill or downhill walking using the generated classifier.

Further, the present invention provides an EMG signal-based walking step recognition method for an up / down stairs walking, wherein in step (b), the steps of the walking step are divided into the stance and the swinging step, .

Further, the present invention provides a method of recognizing a gait step based on an EMG signal for an up / down stair walking, wherein in step (b), in case of an on-going gait, the step is divided into three phases of a stance phase and one phase of a swing phase, A weight acceptance step, a pull-up step, and a forward continuance step, and the weight acceptance step determines that the sole pressure is at a predetermined point between the initial value and the maximum value And the pull-up step is performed until the angle of the knee becomes perpendicular to the ground after the weight acceptance.

According to another aspect of the present invention, there is provided a method of recognizing a gait step based on an EMG signal for an up / down stair walking, wherein in step (b), a weight acceptance step is performed such that a voltage of a pressure is increased from an initial value to a threshold value Threshold As shown in FIG.

[Equation 1]

I _init and i _max are the initial and maximum values of the plantar pressure, respectively.

According to another aspect of the present invention, there is provided a method for recognizing a gait step based on an EMG signal for an up / down stair walking, wherein in step (b), in the case of a downward gait, a weight acceptance of a stance, Leg pull-through, and Fool placement. The knee angle was measured while standing on the reference leg only after the start of the stance gait. Weight acceptance and controlled lowering are distinguished based on the time point at which the floor is lowest. After the start of the walking, the sole pressure is changed from the initial value to the voltage value at which the variance T _var of the window size is the lowest value y _range is divided into leg pull throughs.

Further, the present invention is characterized in that the y- _range is a voltage value in which T _var of the following equation (2) becomes the lowest section, in the electromyogram signal-based gait step recognition method for walking up and down stairs.

[Equation 2]

Where p is the window size, m is the average, and n is the number of samples.

In the step (a), the foot pressure of the pedestrian is measured by a pressure sensor, and the pedestrian's knee angle is measured by an AHRS (Attitude & Heading Reference) System) sensor.

The present invention also provides a method of recognizing a gait step based on an EMG signal for an up and down stairs, comprising the steps of: (a) detecting a Vastus medialis, a Vastus lateralis of the thigh of a pedestrian during an on- , Rectus femoris and Gastrocnemius are used in the lower limbs and Semitendinous of the thighs of the pedestrian when walking downhill and the gastrocnemius of the calf are used .

As described above, according to the EMG signal-based gait phase recognition method for walking up and down stairs according to the present invention, it is possible to more accurately classify the stair gait steps by classifying uphill and downhill gait using only EMG signals .

Particularly, according to the experiment result of the present invention, the accuracy of the previous study was 58.5% on the uphill walking and 35.3% on the downhill walking, whereas the method according to the present invention showed an average of 85.6% on the uphill walking and 69.5% on the downhill walking Recognition rate.

1 is a block diagram of a configuration of an overall system for implementing a walking step recognition method based on an EMG signal for an up / down stairs walking according to an embodiment of the present invention.
2 is a block diagram illustrating a method of recognizing an EMG signal pattern according to the present invention.
FIG. 3 is an exemplary view showing four steps of human walking in a staircase used in the present invention. FIG. 3 (A) is an example of a step 4 of an ascent and a step 4 of a downhill walk.
FIG. 4 is a flowchart illustrating a method of recognizing a gait step based on an EMG signal for an up / down stairs walk according to an embodiment of the present invention.
5 is an overall block diagram of a stair-step walking step classification in accordance with an embodiment of the present invention.
FIG. 6 is a graph illustrating a knee angle pattern using AHRS in a stair walking according to an embodiment of the present invention. FIG.
7 is a knee angle graph for classifying an uphill walking step according to an angle of a knee according to an embodiment of the present invention.
FIG. 8 is a table showing the range of the step of the uphill walking according to the knee angle according to the embodiment of the present invention. FIG.
9 is a knee angle graph for the stepping down step according to the knee angle according to an embodiment of the present invention.
FIG. 10 is a table showing the range of a downhill walking step according to a knee angle according to an embodiment of the present invention; FIG.
11 is a view showing an electrode attaching position and an experimental environment according to an experiment of the present invention.
12 is a diagram of experimental equipment and systems in accordance with the experiments of the present invention.
FIG. 13 is a graph showing recognition rates of individual muscles using a single feature according to the experiment of the present invention, wherein (a) graph.
FIG. 14 is a graph showing recognition rates of individual muscles using a plurality of features according to an experiment of the present invention. FIG. 14 (a) Recognition rate graph.
15 is a graph showing a comparison of recognition rates of the inventive method according to the 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 an EMG signal-based gait step recognition method for an up / down stairs walking according to an embodiment of the present invention will be described with reference to FIG.

As shown in FIG. 1, the EMG signal-based gait step recognition method for gait according to the present invention receives training data 11 or electromyogram signal 12 and selects characteristics and channels in an uphill or downhill walk Or to a program system 30 on a computer terminal 20 that recognizes a walking phase. That is, the EMG signal-based walking step recognition method for the up / down stairs walking can be implemented by 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. [

Meanwhile, as another embodiment, the EMG signal-based walking step recognition method for the up / down stairs walking may be implemented by a single electronic circuit such as an ASIC (on-demand semiconductor) in addition to the general purpose computer. Or a dedicated terminal 30 dedicated to only the recognition of the gait step in an uphill or downhill stair walking. Particularly, the electronic circuit as described above can be used as a control device for controlling the power regulator or as a part of the control device. This is called an electromyogram signal-based gait phase classifier 30. Other possible forms may also be practiced.

Next, a walking step recognizing method according to the present invention will be described with reference to FIG.

A walking step recognition method using an EMG signal is shown in Fig. The electromyogram signal obtained from the bio-signal collecting device is removed by a low-pass filter (LPF) or a band pass filter. After a feature extraction process that calculates parameters characteristic of each signal, it is classified through a classifier.

Next, the stair walking step to be classified according to the present invention will be described with reference to FIG.

In the present invention, the uphill and downhill stairs are classified into four stages. It uses knee angle and center of gravity.

In the biomechanics, the human walking stage is divided into a stance phase, which means that the foot of the leg determined by the standard touches the ground, and a swing phase, which means that the foot of the standard leg is separated from the ground. Generally, stairs are divided into uphill and downhill.

3 (A), as shown in FIG. 3 (A), the uphill walking includes a weight acceptance when the reference leg touches the floor of the stairs, a pull-up stage when the stairs are loaded with the force, Forward continuance when the foot is perpendicular to the ground and Foot clearance when the foot is removed from the ground.

As shown in FIG. 3 (B), the downward walking is a step of weight acceptance when the reference leg is put on the stairs, a controlled lowering step when the reference leg knee is bent most, A leg pull-through step, and a fool placement step when the wire is pulled away from the wire. [Non-Patent Document 6]

In addition, we can classify the stair step by using various physical sensors. In the study of classifying the walking phase through the motion camera, 8 infrared cameras and reflection markers are used to distinguish the walking phase through the joint torque pattern, the bending direction of the angle of the joint, and the power change rate [Non Patent Document 7]. In addition, a stepping walk is classified using an ankle joint angle [Non-Patent Document 8].

On the other hand, research has been conducted to recognize a walking mode using only EMGs without using a physical sensor [Non-Patent Document 9]. In this study, seven walking modes such as stepping down, climbing, flat walking, and obstacle walking were classified based on the front and heel of the sole.

A walking step recognition method based on an EMG signal for an up / down stairs walking according to an embodiment of the present invention will be described in detail with reference to FIG.

In the present invention, the step 4 is divided into four steps using only the EMG signal, and the recognition rate of the gait step for each individual muscle is calculated. The walking stage is classified into four stages, which is the key to stair walking, to be useful for power control. That is, uphill walking is defined as weight acceptance, pull-up, forward continuance, and foot clearance. Downhill walking is defined as weight acceptance, controlled lowering, leg pull through, and fool placement.

As shown in FIG. 4, the EMG signal-based gait step recognition method for walking up and down stairs includes steps of receiving a training electromyogram signal and physical data (S10), determining a walking step with input physical data (S20) (S30) of extracting the feature value from the electromyogram signal (S30), learning the classifier in the determined gait step with the extracted feature value, and recognizing the gait step using the learned classifier (S50) It is divided.

Steps S10 to S40 are steps of learning the range of the walking step to generate a classifier. Training data for training, that is, an electromyogram signal and physical data measured when the pedestrian is walking (S10).

At this time, the pressure sensor and the AHRS sensor are used in the training phase to learn the range of the walking step. The pressure sensor and the AHRS sensor are physical sensors that measure the foot pressure and knee angle of a pedestrian, respectively, and the measured data is called physical data. The range of the walking step is determined according to the plantar pressure and the knee angle (S20).

The EMG signals of the four channels were analyzed using RMS (Root mean square), VAR (Variance), WAMP (Willison amplitude), MAV (Mean Absolute), SSC (Slope sign change) and ZC (Zero crossing) (Non-Patent Document 10) (S30). Previously, four muscle and six feature extraction algorithms were used to generate training data groups with 24 values per step.

Next, the classifier is generated by learning the feature value and the physical data (S40). That is, the characteristic values are input to the classifier, and the walking step determined by the physical data is used as the output of the classifier to learn the classifier. At this time, the classifier uses LDA (Linear Discriminant Analysis).

The walking step is recognized using the learned classifier (S50). The newly inputted EMG signals are calculated in the same manner as the training process, entered into a previously trained classifier group, and recognized by the LDA (Linear Discriminant Analysis).

FIG. 5 shows the overall flow of the feature extraction and step-by-step walking classification method according to an embodiment of the present invention. As shown in FIG. 5, in the training process, physical data measured by using a pressure sensor and an AHRS sensor are used, whereas physical data is not used during the actual recognition process. In both the training process and the actual recognition process, the EMG signal is measured in the pedestrian's muscles, the feature value is extracted, and the classifier is applied as the feature value.

Hereinafter, the detailed operation will be described in detail.

First, the data collecting step will be described.

The total number of channels is six. The pressure sensor and the Attitude & Heading Reference System (AHRS) sensor are used only during the training phase, and the walking phase is classified only by the EMG signal.

The pressure sensor is attached to the sole to separate the stance and the swinging, and the AHRS sensor is worn in front of the knee knee and used to segment the walking step according to the knee angle change. The knee angle is defined as 90 ° when the reference leg is perpendicular to the ground, and has an angle value in the range of 60 ° to 110 °. The change of the knee angle according to the walking of the human stairs shows a constant pattern as shown in Fig.

Muscles use the thigh muscles commonly used in walking and the calf muscles often used in stair walking. In order to utilize EMG signals generated at each step, muscles in which muscle contraction and relaxation cross each other at the gait step were selected [Non-Patent Document 11]. Vastus medialis, Vastus lateralis, Rectus femoris, and Gastrocnemius were used for the ascent of the thighs, and when using the downward gait, (Semitendinous), external beam, and calyx gastrocnemius.

Next, the classification of the uphill walking step will be described.

Uphill walking is divided into three stages of stance phase and one stage of swing phase. The knee angle graph in one cycle of walking is shown in FIG.

The weight acceptance step is defined as the period from the start of the stance gait to the moment when the reference leg loads the stairs. As shown in Equation (1), a certain time point R is determined between the initial value and the maximum value of the AHRS sensor voltage to define a threshold value.

[Equation 1]

In this case, the initial value i _init of the voltage indicates the start of walking, and the maximum value i _max indicates the time when the knee angle becomes perpendicular to the ground. At a certain point in time R is the moment when the reference leg loads the stairs. As a result of observation, R was found to correspond to the point of 30% of the section where the walking starts and the knee angle is perpendicular to the ground.

From the weight acceptance, the section is set to the pull-up phase until the knee is perpendicular to the ground, and the period from this moment until the end of the stance is defined as the forward continuance. Finally, the foot step is defined as foot clearance, in which the foot falls from the floor of the stairs.

The table of FIG. 8 is a table showing a result of observing the first ascent of the subject by classifying the first ascent of the patient, using the method according to the present invention as a representative example of learning the walking step.

Next, the classification of the downward walking step will be described.

Fig. 9 shows a knee angle graph of one cycle of downhill walking. The stance of the downward gait is divided into two stages based on the point at which the knee angle becomes the lowest in the T time that supports the stair ground only with the reference leg after the start of the gait. The total time of the stance phase was 730 ms on average, and the time required to support the stair surface only with the reference leg was 350 ms or less in this time, and T was set to 350 ms. Therefore, the weight acceptance is set to the minimum value of the AHRS sensor voltage within the time point T after the start of downhill walking. The controlled lowering step is defined as the point at which the stance ends.

The downhill gait cycle divides the AHRS sensor voltage from the initial value to y _range by the leg pull through stage. The range of y _range is obtained using equation (2).

&Quot; (2) "

Equation (2) represents the dispersion of the AHRS sensor voltage value, p is the window size, m is the average, and n is the number of samples. The window size means that a certain period of time is set among the entire samples corresponding to the entire signal.

Here, the window size is defined as the window size. Since the variance is obtained only during the leg pull-through step during the entire downhill walking step, the window size is determined as the interval of the samples corresponding to the leg pull through step. For example, assuming that there are 1500 samples in the total signal sample and 350 samples in the leg pull through stage, the window size (p) is 350 and n is 1, 2, 3, 4, ... , 299, and 300, respectively.

In Equation (2), n is defined as an index of samples, that is, an observation coefficient (1, 2, 3, 4, ..., 299, 300) of a sample within a window size as described above. m is the voltage value average of the AHRS sensors. This means the average within the window size. T time means time corresponding to y _range voltage value.

T _var is the variance of the input signal by arbitrary interval (= window size). For example, if the sampling rate is 1 KHz, that is, assuming that the window size is 10 ms in the entire sample in which 1000 samples are input per second, the variance of 100 samples at a time is defined as T _var .

The voltage value at which T _var calculated through Equation (2) becomes the lowest section is designated as y _range . The lowest interval was observed to have a variance of typically 2 × 10 ^-6 . That is, inde The goal of finding the most lowest value among the calculated dispersion in the window size, the value is usually 2 × 10 results observed ^was confirmed a value of ^6. The range in which the AHRS voltage value is 2 × 10 ^-6 among all the samples is y _range . That is, T _var is the set of variances calculated within the window size, and y _range is the lowest value among the variances calculated.

Finally, from the y _range to the end of the swinging period, it is determined as the fool placement step.

The table of FIG. 10 is a representative example of learning the walking step using the method according to the present invention, and is a result of classifying the first downhill walking of the subject 1 time.

Next, the feature extraction and classification step will be described.

Feature extraction is a process of calculating characteristic values that can be represented for pattern recognition. There are time domain, frequency domain and frequency-time domain techniques for feature extraction. The frequency domain is useful when examining muscle fatigue. Since the time domain is computationally efficient, it is frequently used in a walking recognition system using EMG [Non-Patent Document 10].

In the present invention, root mean square (RMS), variance of EMG, WAMP, mean absolute value (MAV), sign slope change (SSC), and zero crossing . RMS is related to muscle contraction at constant force and non-fatigue state, VAR is the mean square of standard deviation and is used to determine the force of the EMG signal. WAMP represents the number of times when the EMG signal size change is above the threshold value. MAV is mainly used to extract muscle activity and strength and increases as muscle strength increases. ZC represents the number of times that the orthogonal arc crosses the zero potential, and SSC represents the number of times the slope of the electromyogram signal waveform is changed [Non-Patent Document 10].

Feature values are recognized using the LDA classifier. LDA classifies classes into linear classes in a way that maximizes the ratio between class-to-class variance and intra-class variance.

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

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

Four normal subjects walk up and down stairs 50 steps each. 25 steps are used for training data, and 25 steps are used for input data. The steps used in the experiment are as shown in Fig. 11, and the length of the stairs is 145 cm in width, 27 cm in height, and 16 cm in height. The pace is an average of 1500ms per step.

Figure 12 shows the constructed experimental equipment and system. Two BIOPAC BN-EMG, two STM32F4 DISCOVERY, FLEXIFORCE A201 pressure sensor and E2BOX EBIMU-9DOFV2 AHRS sensor were used as the equipment. First, the EMG signal from the BN-EMG is transmitted to the PC via Ethernet. At the same time, the pressure sensor and the AHRS sensor are attached to the collecting terminal (STM32F4), and pressure and angle data are transmitted to the main board (STM32F4). The main board synchronizes this data with the EMG signal and sends it to the PC.

Next, experimental results according to the experiment of the present invention will be described.

First, we explain the results of recognition rate of walking step for individual muscle.

Experiments were compared with the case of using all feature extraction algorithms (single feature) and the case of using two feature extraction algorithms (multiple features) with high accuracy. The plurality of features is a method of classifying using the two feature values when recognizing the walking phase using the LDA classifier. This makes it possible to obtain a larger amount of class information by reducing the difference of the feature space by combining the feature techniques having similar information about the EMG signal among the time domain feature extraction techniques [Non-Patent Document 13]. The characteristics were tested in combination with RMS + VAR, MAV + SSC and WAMP + ZC.

FIG. 13 is a graph showing the recognition rate of the uphill / downhill walking steps per the muscle of the subject using a single characteristic by the stance / diagonal period. As a result of the experiment, the recognition rate of muscle was different among the four subjects because the muscle development and activity were different. In addition, it was confirmed that the recognition rate of gait step for each muscle was large for each subject when using a single feature.

On the other hand, FIG. 14 is a figure showing the recognition rate of the step-up / down-step walking step by the subject using the plurality of features by the stance / ergometer. As a result of observation, when a plurality of features are used differently from the single feature of Fig. 13, the recognition rate of the gait step is further improved. In addition, it was confirmed that the recognition rate variation was small for each subject by using characteristic features with similar class information without using all the features.

Next, the results of comparing the recognition rate of the prior art and the method according to the present invention will be described.

Unlike the existing research using only the physical sensor, only the gait range is generated using the sensor in the training step and the recognition rate of the gait step is calculated using only the EMG signal in the recognition step. We confirmed the recognition rates of the walking steps of the two methods in order to examine the range of muscle training and appropriate training data used in the stairs through previous research and the present invention.

First, previous studies used four channels of the thigh muscles [Non-Patent Document 12]. The training data were classified into 2 stages of stance and 2 stages of uphill walking. Downward walking was reduced to weight acceptance and forward continuance. In the present invention, three channels of the thigh muscles and one channel of the calf muscles were used. The range of the training data was classified into 3 stages in the ascending phase and 1 stage in the irregular phase, and the Forward continuance and Controlled lowering in the downhill were reduced to 4 stages respectively.

Fig. 15 is the recognition rate of the walking step obtained by the two methods. In the previous study, the average of uphill walking (deviation) was 58.5% (± 17.63), and the average of downhill walking (deviation) was 35.3% (± 1.41) [Non-Patent Document 12]. When using the single feature of the proposed method, uphill walking was 76.1% (± 6.80) and downhill walking was 61.5% (± 9.01). When multiple features are used, 85.6% (± 5.19) of uphill walking and 69.5% (± 5.40) of downhill walking are recognized.

The recognition rate is calculated as the recognition rate of the walking step for the entire muscle of the subjects.

As a result of observing the recognition rate of the walking step, it can be confirmed that the method of the present invention is ahead of the recognition rate of the walking step of the previous study as a whole. In addition, we were able to analyze the muscles used in the stairs and the range of training data generated. However, uphill Gait 2 showed a higher recognition rate of the gait step than the method of the present invention by about 1.2%. However, as a result of statistical analysis using the one-way ANOVA, the preceding study showed an F-ratio of 44.77, a F-rejection of 6.591, and a p-value of 0.001 (significance level 0.05); The method 1 of the present invention showed an F ratio of 7.424, a F-rejection value of 3.490, and a p-value of 0.004. Therefore, the results of the previous study and the proposed method 1 were not statistically significant. On the other hand, the method 2 of the present invention proved statistically significant with an F ratio of 1.635, a F-rejection value of 3.490, and a p-value of 0.233 (significance level 0.05). In downhill walking, the F ratio was smaller than the F-rejection value and the p-value was less than 0.05 in all three methods.

According to the present invention, a method of classifying uphill and downhill stairs into four steps by EMG signals alone is proposed. As a result, by analyzing the average recognition rate of individual gait steps, we identified the muscles that helped each gait phase and those that did not. Also, it was confirmed that the recognition rate was increased about 9% in uphill walking and 8% in downhill walking compared to single feature using feature combinations. Finally, we examined the recognition rate by differentiating the range of the stance gait and the gait used, and analyzed the range of training data and the appropriate training data.

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: training data 12: EMG signal
20: Computer terminal 30: Stair walking step recognition device

Claims (8)

A method for recognizing a gait phase based on an EMG signal for up and down stairs,
(a) receiving a training signal of an electromyogram signal of a plurality of muscles measured at the time of an uphill or downhill walk, and physical data of a foot pressure and a knee angle measured at the time of walking;
(b) determining a walking step using the physical data;
(c) extracting features from the EMG signal;
(d) generating a classifier that takes the extracted features as input and outputs the determined gait steps; And
(e) recognizing a walking step of the uphill or downhill walk using the generated classifier,
The method of claim 1, wherein, in step (b), the stance group and the swing group of the walking step are divided by the sole pressure, and the walking step is subdivided according to the knee angle.
delete The method according to claim 1,
In step (b), in case of uphill walking, it is divided into three stages of stance phase and one stage of swing phase, and the stance phase includes weight acceptance, pull-up, continuance "),
The weight acceptance step is a step until the foot pressure reaches a predetermined point between the initial value and the maximum value. In the pull-up step, the angle of the knee after the weight acceptance Wherein the step of recognizing the gait step is based on an EMG signal for the up and down stairs.
The method of claim 3,
In the step (b), the weight acceptance step determines that the voltage of the pressure is from the initial value to the threshold value Threshold of the following equation (1). The step of recognizing the walking step based on the EMG signal for the up / .
[Equation 1]

I _init and i _max are the initial and maximum values of the plantar pressure, respectively.
The method according to claim 1,
In step (b), in case of downhill walking, the weight acceptance of the stance, controlled lowering of the stance, leg pull through of the swinging leg, Fool placement) are judged by dividing into four steps,
The weight acceptance and the control lowering are classified based on the point at which the knee angle becomes the lowest while the stance gait starts to be supported only with the reference leg. After the start of the gait walking, To a _range of y _range which is a voltage value at which the variance T _var of the window size is the lowest, is divided into leg pull through. The method of recognizing the walking step based on the EMG signal for the up / down stairs walking.
6. The method of claim 5,
_Wherein the y _range is a voltage value at which T _var in the following equation (2) becomes the lowest interval. The method of recognizing the walking step based on the EMG signal for the up / down stairs walking.
[Equation 2]

Where p is the window size, m is the average, and n is the number of samples.
The method according to claim 1,
In step (a), the pedestrian's foot pressure is measured by a pressure sensor, and the pedestrian's knee angle is measured by an AHRS (Attitude & Heading Reference System) sensor. Step recognition method.
The method according to claim 1,
In step (a), when the patient is on an uphill walk, the Vastus medialis, Vastus lateralis, Rectus femoris, and Gastrocnemius of the thigh of the pedestrian are used, Wherein the pedestrian's inner thigh, the semitendinous, the outer limb, and the calyx are used as the pedestrian's thighs.

Images