A sitting posture recognition system based on sound signals

Abstract

The present invention discloses a sitting posture recognition system based on sound signals, and the specific steps are as follows: S1, data acquisition: calculate the energy spectrum density of the received sound signal; S2, data preprocessing: use the ensemble empirical mode decomposition algorithm to decompose the ESD signal into several intrinsic mode function components; normalize the IMF components, and use the principal component analysis method to transform and reorganize them into irrelevant score components; S3, signal extraction: select the score component based on the minimum signal-to-noise ratio to determine the boundary of the sitting posture signal; S4, sitting posture recognition: use the extracted sitting posture signal data to train the meta-learning algorithm model to achieve classification and recognition of different sitting postures. The present invention adopts the above-mentioned sitting posture recognition system based on sound signals, which can effectively recognize the sitting posture according to the mobile phone sound signal, and the average recognition accuracy of this system for different sitting postures can reach more than 89%.

Description

Technical Field

[0001] The present invention relates to the technical field of sitting posture detection systems, and in particular to a sitting posture recognition system based on sound signals. Background Art

[0002] Long-term bad sitting posture can lead to diseases such as lumbar muscle strain and even lumbar disc herniation, which seriously damages people's health. Therefore, an effective sitting posture monitoring system can reduce the possibility of disease. With the development of the Internet of Health Things (IoHT), some sitting posture recognition systems based on wearable devices or cameras have been proposed.

[0003] However, sensors installed on the body can cause inconvenience to users. Using cameras requires high energy consumption and users face the risk of privacy leakage. Existing smartphones all have microphones and speakers that can send and receive sound signals. Since sound signals have strong directionality and long propagation distance, it is possible to obtain fine-grained action information of users in a contactless manner. Summary of the invention

[0004] The purpose of the present invention is to provide a sitting posture recognition system based on sound signals, which can effectively recognize different sitting postures according to mobile phone sound signals, and the average recognition accuracy of this system for different sitting postures can reach more than 89%.

[0005] To achieve the above object, the present invention provides a sitting posture recognition system based on sound signals, and the specific steps are as follows:

[0006] S1. Data collection: Calculate the energy spectrum density of the received sound signal;

[0007] The energy spectral density (ESD) of the acoustic signal describes the distribution of energy with frequency, which can sense the information of different movements of surrounding objects. Therefore, the ESD of the received sound signal is calculated first, and then the human motion signal is analyzed and extracted through subsequent steps.

[0008] S2, data preprocessing: using the ensemble empirical mode decomposition algorithm to decompose the ESD signal into several intrinsic mode function components;

[0009] IMF={IMF1;IMF2;IMF3;…;IMF m}

[0010] The IMF components are normalized and transformed and reorganized into irrelevant score components using principal component analysis;

[0011] S3, signal extraction: select the score component based on the minimum signal-to-noise ratio and determine the boundary of the sitting signal;

[0012]

[0013] S4. Sitting posture recognition: Use the extracted sitting posture signal data to train the meta-learning algorithm model to achieve classification and recognition of different sitting postures. Among them, the different sitting posture signal data are classified and detected based on the model agnostic meta-learning (MAML) algorithm.

[0014] Preferably, in step S1, the ESD of the sound signal received by the microphone after the sound emitted by the mobile phone speaker is reflected is calculated.

[0015] Preferably, in step S2, the original aliased signal is decomposed and each component is processed to finally obtain effective sitting posture signal data.

[0016] In order to find effective components from the aliased ESD data to extract the sitting signal, the collected data is decomposed into multiple intrinsic mode function (IMF) components of different frequencies by using the ensemble empirical mode decomposition (EEMD) algorithm. The IMF components are then normalized and reorganized into a set of new uncorrelated score components by using the principal component analysis (PCA) algorithm.

[0017] Preferably, in step S3, according to the designed peak algorithm and sub-window merging algorithm, the score component based on the minimum signal-to-noise ratio is selected, the signal boundary of the sitting posture is determined, and the time series data is calculated and analyzed, and a sliding window is introduced to realize the 0 and 1 encoding of the valid sitting posture data signal.

[0018] A sitting posture recognition system based on mobile phone sound signals uses the speaker and microphone of the smartphone to send and receive signals, and judges the user's sitting posture by processing and analyzing the energy spectral density data of the sound signals and inputting them into a classification model.

[0019] Therefore, the present invention adopts the above-mentioned sitting posture recognition system based on sound signals, and obtains fine-grained motion waveforms to continuously monitor the user's sitting posture based on sound signals. The system is low-cost and non-invasive.

[0020] The sitting posture recognition system of the present invention includes the recognition of three types of actions: body tilt, leg crossing and leg shaking. Compared with the previous sitting posture recognition method based on wearable devices or cameras that require additional equipment, the present invention uses the built-in speakers and microphones of existing smart phones to send and receive sound signals to detect the user's sitting posture. Therefore, it is not limited by conditions such as lighting and effectively avoids the privacy leakage problem that may be caused by the camera. It is more universal and has important significance for user health monitoring.

[0021] The sitting posture recognition system of the present invention emits a sound signal through the speaker of a smart phone and receives the reflected waveform through a microphone to obtain the data of the user's sitting posture. Since the collected data contains noise information, the component most affected by the sitting posture is extracted based on the ensemble empirical mode decomposition (EEMD) algorithm and the principal component analysis (PCA) algorithm, and then the effective sitting posture data is selected and extracted using an innovative boundary determination algorithm for the component. Finally, the model-agnostic meta-learning (MAML) algorithm model is trained using a small sample to verify the effectiveness of the sitting posture recognition system based on the mobile phone sound signal. The experimental results show that the average recognition accuracy of this system for different sitting postures can reach more than 89%.

[0022] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 These are 6 incorrect sitting postures of the present invention;

[0024] Figure 2 It is a system framework diagram of the present invention;

[0025] Figure 3 It is the ESD calculated from the original signal;

[0026] Figure 4 It is the IMF component obtained after the EEMD algorithm decomposes the ESD signal;

[0027] Figure 5 It is the score component obtained by PCA transformation of IMF component;

[0028] Figure 6 is the component corresponding to the minimum SNR

[0029] Figure 7 It is the boundary of the sitting signal after processing;

[0030] Figure 8 is the recognition accuracy of different users’ sitting postures. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0032] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0033] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the main idea or essential features of the present invention. Therefore, from all points of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0034] In addition, it should be understood that although this specification is described according to the implementation modes, not every implementation mode includes only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art. These other implementation modes are also covered within the protection scope of the present invention.

[0035] It should also be understood that the specific embodiments described above are only used to explain the present invention, and the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, can make equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, which should be covered by the protection scope of the present invention / invention.

[0036] All terms (including technical terms or scientific terms) used in the present disclosure have the same meanings as those understood by ordinary technicians in the field to which the present disclosure belongs, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined in this article.

[0037] Technologies, methods, and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0038] The disclosed contents of the prior art documents cited in the present specification are incorporated into the present invention by reference in their entirety and are therefore part of the disclosure of the present invention.

[0039] The English and Chinese versions of the title of this article are as follows:

[0040] Energy Spectrum Density (ESD) of sound signals

[0041] Ensemble Empirical Mode Decomposition (EEMD)

[0042] Intrinsic Mode Function (IMF)

[0043] Principal Component Analysis (PCA)

[0044] Model-Agnostic Meta-Learning (MAML)

[0045] Embodiment 1

[0046] The present invention is a sitting posture recognition system based on sound signals, and its framework diagram is as follows Figure 2 The specific workflow is as follows:

[0047] S1. Calculate the ESD of the original signal received by the mobile phone microphone, such as Figure 3 shown.

[0048] S2. Perform EEMD decomposition on ESD data.

[0049] The frequency range of body movement is about 0.1Hz-2Hz, which is close to the frequency of low-frequency noise. These two signals will partially overlap in the frequency domain, and it is difficult to separate them directly using the traditional Fourier transform method. Through the adaptive time-frequency analysis method, namely the EEMD algorithm, the aliased signal X = {x1, x2, ..., x n} is decomposed into a series of non-overlapping IMF components, IMF = {IMF1; IMF2; IMF3; ...; IMF m}, where IMF1, IMF2, …, IMF m It is the IMF component from high frequency to low frequency, such as Figure 4 As shown. The number of IMF components m is calculated by m=log2n-1, where each IMF component contains local feature information of the original signal at different time scales. White noise and the number of decompositions play a decisive role in the decomposition performance. The number of decompositions needs to be large enough, generally more than 100. White noise needs to be added according to the standard deviation of the signal. While running the EEMD algorithm each time, the Fast Fourier Transform (FFT) is applied to calculate the instantaneous frequency of each component.

[0050] S3. Component transformation based on PCA. Since the signal-to-noise ratio (SNRs) of the IMF component is very low, the boundary of the sitting posture signal cannot be directly determined based on the traditional window function. Use the PCA algorithm to perform component transformation and find the component most affected by the sitting posture. The IMF component is converted into a set of linearly unrelated variables through orthogonal transformation, which is called score component, such as Figure 5 shown.

[0051] Before PCA processing, it is necessary to use the formula (where D represents the normalized multidimensional signal), normalize the IMF components to the range of [0, 1] to eliminate the influence of the dimensions of different IMF components on the PCA projection calculation. Then, calculate the eigenvectors of the covariance matrix of D and put them into the matrix Q. The score coefficient matrix of PCA is calculated by S = D × (D × Q).

[0052] S4. Signal boundary determination. Not all PCA components can be used to determine the boundary of the sitting signal. Sitting has a great influence on the score component of the PCA with the smallest signal-to-noise ratio. Therefore, the component corresponding to the minimum SNR is taken. Right now (in ) as the detection component to determine the sitting posture boundary, such as Figure 6 Then, wavelet denoising is used to After smoothing, the effective component obtained has a good signal-to-noise ratio, and the boundary of the sitting signal is determined based on the selected component.

[0053] If the window function method is used directly, many pseudo peaks will be detected, and it is difficult to determine the noise threshold. Because the slope of the peak is very different from the slope of the noise, the peak detection algorithm based on the sitting posture will also have many meaningless peaks caused by random noise that does not belong to the sitting posture.

[0054] To prevent false detection of sitting posture, an innovative peak detection algorithm is designed through the following process:

[0055] First, calculate the time series The average of the maximum values ​​of the left k nearest neighbors and the right k nearest neighbors in the middle finds 5 local maxima P as follows:

[0056]

[0057] Then, the mean and standard deviation (std) of P are calculated. If the difference between a data point in P and the mean is greater than δ multiplied by std, the point belongs to the actual peak. The larger δ is, the more conducive it is to detecting peaks with larger slopes. However, due to the continuity of human motion, some false positive peaks will be detected near the true peak, which will affect the detection performance. To solve this problem, all detected peak points greater than the mean are encoded as 1.

[0058]

[0059] At the same time, a sliding window with an overlap rate of 0% and a width of W is set to re-encode the signal. All points between the first point encoded as 1 and the last point in the window are also encoded as 1. The next point encoded as 1 outside the window is set as the new starting point, and repeated in this way until all points are fully encoded. To facilitate signal segmentation, the remaining unencoded points are encoded as 0. Although the pseudo peaks are eliminated, the encoded peaks only represent part of the sitting posture. Therefore, a sub-window merging algorithm with a width of W and an overlap rate of 50% is used to extract the complete sitting posture. Encoded components The upper sitting boundary can be determined by successive merging of subwindows, avoiding redundancy or truncation of the encoded signal.

[0060]

[0061] Among them, s i is the coding component Points on the left, I is the merge window. In each sub-window, only sampling points. Figure 7 As shown, the boundary of the sitting signal can be adaptively determined and extracted as the input for subsequent classification.

[0062] S5. Sitting posture recognition. The extracted sitting posture data is classified and detected using the MAML algorithm, and the model is updated and trained with a small number of samples. The results show that the sitting posture recognition accuracy can reach more than 89%. Figure 8 This proves that the sitting posture data collection and data processing method based on mobile phone sound signals of the present invention can effectively detect different sitting postures.

[0063] Therefore, the present invention adopts the above-mentioned sitting posture recognition system based on sound signals, which can effectively recognize the sitting posture according to the mobile phone sound signal, and the average recognition accuracy of this system for different sitting postures can reach more than 89%.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A sitting posture recognition system based on sound signals, characterized in that: The specific steps are as follows: S1. Data collection: Calculate the energy spectral density ESD of the sound signal received by the microphone after the sound emitted by the mobile phone speaker is reflected; S2, data preprocessing: using the ensemble empirical mode decomposition algorithm to decompose the ESD signal into several intrinsic mode function components; ； Normalize the IMF components, use the principal component analysis algorithm to transform the components, find out the components most affected by sitting posture, and transform the IMF components into a set of linearly unrelated variables through orthogonal transformation, which are called score components. S3, signal extraction: According to the designed peak algorithm and sub-window merging algorithm, select the score component based on the minimum signal-to-noise ratio , determine the signal boundary of the sitting posture, calculate and analyze the time series data, and introduce a sliding window to realize the 0 and 1 encoding of the valid sitting posture data signal; Specifically include: First, calculate the time series Center left k Nearest Neighbor and Right k Average of the maxima of the nearest neighbors, find 5 local maxima P , as shown below: ； Then, calculate P The mean value mean and standard deviation std; if P If the difference between the data point in and the mean is greater than or equal to δ multiplied by std, then the point is an actual peak point; the larger δ is, the more conducive it is to detecting peaks with larger slopes; all detected actual peak points greater than or equal to the mean are encoded as 1; ； S4. Sitting posture recognition: Use the extracted sitting posture signal data to train the meta-learning algorithm model to achieve classification and recognition of different sitting postures.

2. The sitting posture recognition system based on sound signals according to claim 1, characterized in that: In step S2, the original aliased signal is decomposed and each component is processed to finally obtain effective sitting posture signal data.

Images