CN112669568A - Multi-mode human body falling detection method - Google Patents

Abstract

The invention discloses a multi-mode human body falling detection method, which comprises the following steps: acquiring acceleration data of a detection object; judging the state of the detection object according to the acceleration data; collecting surface electromyographic signals of the detection object when the detection object is judged to be in a falling state; extracting electromyographic signal characteristics from the surface electromyographic signals; identifying the electromyographic signal characteristics through a support vector machine; and triggering an alarm system when the support vector machine identifies that the detected object is in a falling state. The invention provides a multi-mode human body falling detection method, which aims at solving the problems that the existing falling identification method is easily interfered by external noise to cause low identification accuracy and a support vector machine is greatly influenced by a punishment parameter c and a kernel function parameter g.

Description

A Multimodal Human Fall Detection Method

technical field

The invention belongs to the technical field of smart old-age care based on the Internet of Things and artificial intelligence, and in particular relates to a multi-mode human body fall detection method.

Background technique

As the problem of population aging becomes more and more serious, the physical and mental safety of the elderly has become one of the focuses of social concern. With the increase of age, the bone texture of the elderly becomes loose, and walking and falling have become an important factor that endangers the health of the elderly on a daily basis. The traditional fall monitoring and identification methods mainly use video images, sound or vibration, foot pressure sensors and other equipment to complete the fall monitoring and identification of the elderly. The identification model is simple, and the identification process is vague and uncheckable. In recent years, with the rise of wearable devices, the continuous improvement of three-axis acceleration sensing technology and electromyographic signal acquisition technology has provided better monitoring equipment to improve the recognition of falling behavior, and effectively ensured the recognition of falling behavior of the elderly. validity and reliability.

In the existing method, the three-axis acceleration sensing has a certain accuracy in judging the falling behavior, but because the three-dimensional angle acceleration reactor will be disturbed by external noise, it is easy to mislead the data measurement results during the system operation. Induction to judge the fall behavior is inaccurate and difficult to monitor.

There are many different methods for surface EMG recognition and feature extraction. The more common ones are neural network method and cluster analysis. Some progress has been made in surface EMG recognition, but they all have certain shortcomings. SVM (Support Vector Machine) is a supervised classifier, which shows many unique advantages in solving small sample, nonlinear and Gaussian pattern recognition, and can solve the theoretical guidance and learning of neural network structure without standard. However, it is difficult to determine the penalty parameter c and the kernel function parameter g of the algorithm.

SUMMARY OF THE INVENTION

The invention provides a multi-mode human body fall detection method, which adopts the following technical solutions:

A multi-modal human fall detection method includes the following steps:

Collect the acceleration data of the detection object;

Judging the state of the detection object according to the acceleration data;

When it is determined that the detection object is in a falling state, the surface EMG signal of the detection object is collected;

Extract the EMG signal features from the surface EMG signal;

Identify EMG features by support vector machine;

The alarm system is triggered when the support vector machine recognizes that the detection object is in a fall state.

Further, the acceleration data of the detection object is collected through a three-axis acceleration sensor.

Further, the acceleration data of the detection object collected by the three-axis acceleration sensor includes the acceleration values a _x , a _y , and _az of the X, Y, and Z axes.

Further, the specific method for judging the state of the detection object according to the acceleration data is:

Calculate the supine angle Pitch, the roll angle Roll and the left-right rotation angle Yaw according to the following formulas:

According to the comparison of the supine angle Pitch, the roll angle Roll and the left and right rotation angle Yaw with the corresponding normal values, it is determined whether the detected object falls.

Further, the supine angle Pitch, the rollover angle Roll and the left and right rotation angle Yaw are optimized through the Adelman filtering algorithm.

Further, the specific method for extracting the EMG signal features from the EMG signal is:

Preprocess the surface EMG signal;

The EMG signal features were extracted from the preprocessed surface EMG signal by wavelet packet permutation and combined entropy.

Further, the specific method for preprocessing the surface EMG signal is:

The information in the 16Hz-160Hz frequency band is extracted from the surface EMG signal.

Further, the specific method for extracting the EMG signal features from the preprocessed surface EMG signal through the wavelet packet permutation and combination entropy is as follows:

Perform five-layer wavelet packet decomposition on the surface EMG signal;

Using the permutation and combination entropy formula, the permutation and combination entropy of the reconstructed surface EMG signals in the nine low-frequency subspaces were calculated as the characteristics of the EMG signals.

Further, the penalty parameter c and the kernel function parameter g of the support vector machine are optimized by the particle swarm algorithm PSO.

Further, the specific method of optimizing the penalty parameter c and the kernel function parameter g of the support vector machine through the particle swarm algorithm PSO is as follows:

1) By optimizing the learning factors c ₁ , c ₂ and the weight coefficient, the position and velocity of the particles are initialized, and the initial setting of each particle is the initial best position;

2) Calculate the fitness of each particle, and the fitness of the particle is evaluated by K-fold cross-validation;

3) Update the speed and position of the particle according to the speed and position calculation formula of the PSO algorithm;

Among them, the speed and position of the PSO algorithm are calculated as follows:

v _ij (t+1)=wv _ij (t)+c ₁ r ₁ [p _ij -x _ij (t)]+c ₂ r ₂ [p _gj -x _ij (t)],

x _ij (t+1)=x _ij (t)+v _ij (t+1),

Among them, c ₁ , c ₁ are learning factors, r ₁ , r ₂ are random numbers in [0, 1], and w is a weight coefficient. p _ij represents the local optimal position of particle i in the jth dimension, p _gj represents the global optimal position of particle i in the jth dimension; v _ij (t) represents the flight speed of particle i in the jth dimension in the tth generation . x _ij (t) represents the current position of particle i in the j-th dimension in the t-th generation;

4) Check whether the termination condition is met. If so, map the optimal particle of the group to the optimal solution of the penalty parameter c and the kernel function parameter g of the support vector machine, otherwise go to step 2 and continue a new round of search.

The advantages of the present invention are that the provided multi-mode human body fall detection method is easy to be interfered by external noise, resulting in low recognition accuracy, and the support vector machine is affected by the penalty parameter c and the kernel function parameter g. For the larger problem, a multi-modal detection method is proposed, which integrates three-axis acceleration sensing and PSO optimization SVM to improve the accuracy of fall behavior recognition in the elderly.

Description of drawings

FIG. 1 is a schematic diagram of the multi-modal human body fall detection method of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

As shown in FIG. 1, a multi-mode human body fall detection method includes the following steps: S1: Acquire acceleration data of the detection object. S2: Determine the state of the detection object according to the acceleration data. S3: Collect the surface EMG signal of the detection object when it is determined that the detection object is in a falling state. S4: Extract the EMG signal features from the surface EMG signal. S5: Identify EMG signal features by support vector machine. S6: Trigger the alarm system when the support vector machine recognizes that the detection object is in a falling state. According to the above method, after the falling behavior is detected through the timely speed data, the falling behavior is re-identified and verified by the support vector machine, and the alarm system is triggered at this time. It is understandable that in order to cope with extreme conditions, the system is added to the manual control system, that is, the user can manually complete the alarm setting by himself. The above steps are described in detail below.

For step S1: the acceleration data of the detection object is collected.

In the present invention, the acceleration data of the detection object is collected by a three-axis acceleration sensor.

Specifically, the acceleration data of the detection object collected by the three-axis acceleration sensor includes the acceleration values a _x , a _y , and _az of the X, Y, and Z axes.

For step S2: determine the state of the detection object according to the acceleration data.

Specifically, the specific method for judging the state of the detection object according to the acceleration data is as follows:

Calculate the supine angle Pitch, the roll angle Roll and the left-right rotation angle Yaw according to the following formulas:

Then, according to the supine angle Pitch, the rollover angle Roll, and the left-right rotation angle Yaw, and the corresponding normal values, it is determined whether the detection object falls. First, monitor whether the lying angle Pitch is higher than the normal value. If it is not normal, continue to monitor whether the roll angle Roll and the left and right rotation angle Yaw are higher than the normal value. If it is not normal, the pre-alarm mode is enabled.

Since the three-axis accelerometer will be disturbed by external noise, it is easy to mislead the data measurement results during the operation of the system. In this application, the Adelman filtering algorithm is used to calculate the calculated supine angle Pitch, roll angle Roll and left-right rotation angle. Yaw for optimization.

For step S3: when it is determined that the detection object is in a falling state, the surface EMG signal of the detection object is collected.

When the lying angle Pitch, the rollover angle, and the left-right rotation angle Yaw greatly exceed the normal values, turn on the switch of the EMG signal acquisition instrument, enter the EMG acquisition behavior, and collect the user's surface EMG signal.

For step S4: extracting the EMG signal features from the surface EMG signal.

The specific method of extracting EMG signal features from EMG signal is as follows:

The surface EMG signal is preprocessed. Specifically, the information in the 16Hz-160Hz frequency band is extracted from the surface EMG signal, and high-frequency noise and unnecessary low-frequency information are filtered out.

The EMG signal features were extracted from the preprocessed surface EMG signal by wavelet packet permutation and combined entropy.

Specifically, five-layer wavelet packet decomposition is performed on the surface EMG signal. Using the permutation and combination entropy formula, the permutation and combination entropy of the reconstructed surface EMG signals in the nine low-frequency subspaces were calculated as the characteristics of the EMG signals.

Among them, the permutation and combination entropy formula is:

T为时间序列；x_i,x_i+1,…,x_i+n-1表示时间序列中连续的n个样本点，count表示序列中排列情况π出现的次数。in,

T is the time series; x _i , x _i+1 ,..., x _i+n-1 represent n consecutive sample points in the time series, and count represents the number of times the arrangement situation π occurs in the sequence.

For step S5: identify the EMG signal features through a support vector machine.

In this application, in order to improve the classification accuracy of the support vector machine SVM, the penalty parameter c and the kernel function parameter g of the support vector machine are optimized by the particle swarm algorithm PSO.

The specific method of optimizing the penalty parameter c and the kernel function parameter g of the support vector machine through the particle swarm algorithm PSO is as follows:

1. By optimizing the learning factors c ₁ , c ₂ and weight coefficients, initialize the position and velocity of the particle, and the initial setting of each particle is the initial best position.

2. Calculate the fitness of each particle, and the fitness of the particle is evaluated by K-fold cross-validation.

3. Update the velocity and position of the particle according to the velocity and position calculation formula of the PSO algorithm.

Among them, the speed and position of the PSO algorithm are calculated as follows:

v _ij (t+1)=wv _ij (t)+c ₁ r ₁ [p _ij -x _ij (t)]+c ₂ r ₂ [p _gj -x _ij (t)],

x _ij (t+1)=x _ij (t)+v _ij (t+1),

Among them, c ₁ , c ₁ are learning factors, r ₁ , r ₂ are random numbers in [0, 1], and w is a weight coefficient. p _ij represents the local optimal position of particle i in the jth dimension, and p _gj represents the global optimal position of particle i in the jth dimension. v _ij (t) represents the flying speed of particle i in the j-th dimension in the t-th generation. x _ij (t) represents the current position of particle i in the j-th dimension in the t-th generation.

4. Check whether the termination condition is met. If so, map the optimal particle of the group to the optimal solution of the penalty parameter c and the kernel function parameter g of the support vector machine, otherwise go to step 2 and continue a new round of search.

For step S6: the alarm system is triggered when the support vector machine recognizes that the detection object is in a falling state.

The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above-mentioned embodiments do not limit the present invention in any form, and all technical solutions obtained by means of equivalent replacement or equivalent transformation fall within the protection scope of the present invention.

Claims (10)

1. a multimodal human body fall detection method, is characterized in that, comprises the following steps: Collect the acceleration data of the detection object; Determine the state of the detection object according to the acceleration data; When it is determined that the detection object is in a falling state, the surface EMG signal of the detection object is collected; extracting the EMG signal features from the surface EMG signal; Identify the electromyographic signal features through a support vector machine; The alarm system is triggered when the support vector machine recognizes that the detection object is in a falling state. 2. The multimodal human body fall detection method according to claim 1, characterized in that, Acceleration data of the detected object is collected through a three-axis acceleration sensor. 3. The multimodal human body fall detection method according to claim 2, wherein, The acceleration data of the detection object collected by the three-axis acceleration sensor includes the acceleration values a _x , a _y , and _az of the X, Y, and Z axes. 4. The multimodal human body fall detection method according to claim 3, characterized in that, The specific method for judging the state of the detection object according to the acceleration data is as follows: Calculate the supine angle Pitch, the roll angle Roll and the left-right rotation angle Yaw according to the following formulas:

According to the comparison of the supine angle Pitch, the roll angle Roll, and the left-right rotation angle Yaw with the corresponding normal values, it is determined whether the detection object falls. 5. The multimodal human body fall detection method according to claim 4, wherein, The supine angle Pitch, the rollover angle Roll and the left-right rotation angle Yaw are optimized through an Adelman filter algorithm. 6. The multimodal human body fall detection method according to claim 1, wherein, The specific method for extracting the EMG signal features from the EMG signal is as follows: Preprocessing the surface EMG signal; The electromyographic signal features are extracted from the preprocessed surface electromyographic signals through wavelet packet permutation and combination entropy. 7. The multimodal human body fall detection method according to claim 6, wherein, The specific method for preprocessing the surface EMG signal is: Information in the frequency band of 16Hz-160Hz is extracted from the surface EMG signal. 8. The multimodal human body fall detection method according to claim 7, wherein, The specific method for extracting the EMG signal features from the preprocessed surface EMG signal through the wavelet packet permutation and combination entropy is: performing five-layer wavelet packet decomposition on the surface EMG signal; The permutation and combination entropy of the surface EMG signals reconstructed in nine low-frequency subspaces is calculated by using the permutation and combination entropy formula as the characteristics of the EMG signal. 9. The multimodal human fall detection method according to claim 1, wherein, The penalty parameter c and the kernel function parameter g of the support vector machine are optimized by the particle swarm algorithm PSO. 10. The multi-modal human fall detection method according to claim 9, wherein, The specific method for optimizing the penalty parameter c and the kernel function parameter g of the support vector machine through the particle swarm algorithm PSO is: 1) By optimizing the learning factors c ₁ , c ₂ and the weight coefficient, the position and velocity of the particles are initialized, and the initial setting of each particle is the initial best position; 2) Calculate the fitness of each particle, and the fitness of the particle is evaluated by K-fold cross-validation; 3) Update the speed and position of the particle according to the speed and position calculation formula of the PSO algorithm; Among them, the speed and position of the PSO algorithm are calculated as follows: v _ij (t+1)=wv _ij (t)+c ₁ r ₁ [p _ij -x _ij (t)]+c ₂ r ₂ [p _gj -x _ij (t)], x _ij (t+1)=x _ij (t)+v _ij (t+1), Among them, c ₁ , c ₁ are learning factors, r ₁ , r ₂ are random numbers in [0, 1], and w is a weight coefficient. p _ij represents the local optimal position of particle i in the jth dimension, p _gj represents the global optimal position of particle i in the jth dimension; v _ij (t) represents the flight speed of particle i in the jth dimension in the tth generation . x _ij (t) represents the current position of particle i in the j-th dimension in the t-th generation; 4) Check whether the termination condition is met. If so, map the optimal particle of the group to the optimal solution of the penalty parameter c and the kernel function parameter g of the support vector machine, otherwise turn to step 2 and continue a new round of search.

Images