Phi-OTDR (Optical Time Domain Reflectometer)-based user falling optical fiber sensing monitoring method and system

Abstract

The invention discloses a phi-OTDR-based user falling optical fiber sensing monitoring method and system, and the method comprises the steps: carrying out the preprocessing of collected original optical fiber vibration signals which comprise a walking signal, a falling signal, a falling signal during walking, and a human static signal, and obtaining a target optical fiber vibration signal; feature extraction is carried out from three dimensions of a time domain, a frequency domain and a time-frequency domain, and primary features are obtained; screening the primary features through orthogonality analysis and difference analysis to obtain first-level features, and screening the first-level features through a genetic algorithm to obtain second-level features; screening the second-level features through a mutual information evaluation function to obtain target features; respectively inputting the target features into a plurality of behavior recognition models to obtain a result whether the user falls down or not, and when a falling-down result is obtained, obtaining falling-down position information of the user from the target optical fiber vibration signal; and triggering alarm information based on the tumble result and the tumble position information, and sending the alarm information to a terminal of a monitoring person of the user.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber sensing, and in particular to, but is not limited to, a method and system for monitoring user falls based on optical fiber sensing using a φ-OTDR. Background Art

[0002] The global population is aging at a significant pace, with a large and growing number of elderly people. The phenomenon of "empty nesters" is becoming increasingly common. As elderly people's physical functions decline, falls have become a major factor leading to injury and even death, posing a serious threat to their safety.

[0003] Related technologies use smart bracelets, video surveillance, and smart flooring to monitor the health of the elderly. However, smart bracelets have a high false alarm rate and poor environmental adaptability. Sensor data is easily interfered with during strenuous exercise or when the device is not securely worn. Video surveillance carries a high risk of privacy leakage and is subject to environmental restrictions. Smart flooring is not only expensive to install and difficult to maintain, but also poorly adaptable to humid environments.

[0004] Therefore, how to accurately and efficiently monitor the safety of the elderly has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the embodiments of the present invention provide a method to at least solve the problem that the related technology cannot accurately and efficiently monitor the safety of the elderly.

[0006] According to a first aspect of an embodiment of the present invention, a method for monitoring a user falling down by optical fiber sensing based on φ-OTDR is provided, comprising: Preprocessing the collected original optical fiber vibration signal to obtain a target optical fiber vibration signal, wherein the original optical fiber signal includes a walking signal, a falling signal, a falling-while-walking signal, and a person-stationary signal; Extract the target optical fiber vibration signal from three dimensions: time domain, frequency domain and time-frequency domain to obtain primary features; Screening the primary features through orthogonality analysis and difference analysis to obtain first-level features, and screening the first-level features through genetic algorithm to obtain second-level features; The second-level features are screened by mutual information evaluation function to obtain target features; Inputting the target features into multiple behavior recognition models respectively to obtain target behavior results of the user, wherein the target behavior results include a fall result and a no-fall result, and when the fall result is obtained, obtaining the fall position information of the user from the target optical fiber vibration signal; An alarm message is triggered based on the fall result and the fall location information, and the alarm message is sent to a terminal of a person monitoring the user.

[0007] According to a second aspect of an embodiment of the present invention, a user fall optical fiber sensing monitoring system based on φ-OTDR is provided, comprising: A sensing fiber network, a vibration signal positioning module, a monitoring center, and a receiving terminal for the monitor. The monitoring center includes a signal acquisition module, a signal processing module, a feature extraction module, a feature screening module, a pattern recognition module, and a monitoring center core. The signal acquisition module is used to collect the original optical fiber vibration signal collected by the sensing optical fiber network, wherein the original optical fiber signal includes a walking signal, a falling signal, a falling-while-walking signal, and a human stationary signal; The signal processing module is used to pre-process the collected original optical fiber vibration signal to obtain the target optical fiber vibration signal; The feature extraction module is used to extract features of the target optical fiber vibration signal from three dimensions: time domain, frequency domain and time-frequency domain to obtain primary features; The feature screening module is used to screen the primary features through orthogonality analysis and difference analysis to obtain first-level features, and screen the first-level features through genetic algorithm to obtain second-level features; and screen the second-level features through mutual information evaluation function to obtain target features; The pattern recognition module is used to input the target features into multiple behavior recognition models respectively to obtain the user's target behavior results, wherein the target behavior results include a fall result and a no-fall result; The vibration signal positioning module is used to obtain the fall position information of the user from the target optical fiber vibration signal when the fall result is obtained; A monitoring center is configured to generate an alarm message based on the fall result and the fall location information, and send the fall result and the fall location information to a terminal of a monitoring person of the user via the alarm message; The monitoring center core is used to coordinate data transmission between the sensing optical fiber network, the vibration signal positioning module, and the monitoring center.

[0008] According to the solution provided by the embodiment of the present invention, the collected original optical fiber vibration signal is preprocessed to obtain a target optical fiber vibration signal, which includes a walking signal, a falling signal, a falling signal while walking, and a human stationary signal; the target optical fiber vibration signal is feature extracted from three dimensions: time domain, frequency domain, and time-frequency domain to obtain primary features; the primary features are screened by orthogonality analysis and difference analysis to obtain first-level features, and the first-level features are screened by a genetic algorithm to obtain second-level features; the second-level features are screened by a mutual information evaluation function to obtain target features; the target features are input into a variety of behavior recognition models respectively to obtain the user's target behavior results, which include the results of having fallen and the results of not having fallen, and when the result of having fallen is obtained, the user's fall position information is obtained from the optical fiber vibration signal; an alarm message is triggered based on the result of having fallen and the fall position information, and the alarm message is sent to the terminal of the user's monitoring person. In this method, multiple signals are collected to comprehensively cover the complex situations in actual scenarios. An efficient high-dimensional feature matrix screening algorithm is used to obtain a large amount of useful information, fully tap the potential value of the signal, and process the target features through multiple behavior recognition models, making the result of judging whether the user has fallen more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 A flowchart of a method for monitoring a user's fall using optical fiber sensing based on φ-OTDR according to an embodiment of the present invention is provided; Figure 2 A flow chart of the logical relationship of time-domain-frequency-time-frequency domain feature fusion provided by an embodiment of the present invention; Figure 3 A technical roadmap for the feature comprehensive scoring system provided by the embodiment of the present invention; Figure 4 A schematic diagram of the basic structure of an optical fiber communication system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0011] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0012] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described here can be implemented in an order other than that illustrated or described here.

[0013] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by those skilled in the art in the art to which the embodiments of the present invention pertain. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless specifically defined as herein, should not be interpreted in an idealized or overly formal sense.

[0014] Figure 1 This is a flow chart of a φ-OTDR-based optical fiber sensing method for monitoring a user's fall, provided in an embodiment of the present invention. The φ-OTDR-based optical fiber sensing method for monitoring a user's fall, provided in an embodiment of the present invention, can be performed by an electronic device, such as a computer or server. The user herein can be an elderly person, particularly one in a bathroom or restroom.

[0015] like Figure 1 As shown, the user fall optical fiber sensing monitoring method based on φ-OTDR includes: S101. Preprocess the collected original optical fiber vibration signal to obtain a target optical fiber vibration signal, where the original optical fiber signal includes a walking signal, a falling signal, a falling-while-walking signal, and a human stationary signal.

[0016] In an embodiment of the present invention, the original optical fiber vibration signal includes four major types of signals, namely walking signals, falling signals, falling signals while walking, and human stationary signals. Walking signals are divided into single-person walking and multi-person walking, falling signals are divided into sitting falls, side falls, and lying falls, and human stationary signals are divided into environmental white noise, long-term standing, and long-term sitting. The collected original optical fiber vibration signal can be pre-processed by removing the DC component by subtracting the corresponding signal mean from the original optical fiber vibration signal, and then denoising using the five-point sliding average method to obtain the target optical fiber vibration signal. Removing the DC component can make the frequency distribution trend more obvious, and denoising can reduce the power frequency and acquisition interference, thereby improving the signal-to-noise ratio.

[0017] S102 , extracting features of the target optical fiber vibration signal from three dimensions: time domain, frequency domain, and time-frequency domain to obtain primary features.

[0018] In an embodiment of the present invention, 15 types of common features such as maximum value and minimum value are extracted in the time domain; discrete Fourier transform, power spectrum analysis, inverse spectrum analysis and other methods are used in the frequency domain to add 6 types of eigenvalues on the basis of 15 types of features in the time domain; wavelet transform and wavelet packet decomposition are mainly used in the time-frequency domain, and singular value decomposition is performed on them to extract features, and finally a total of 141 types of features are extracted.

[0019] like Figure 2 As shown, Figure 2 A flow chart of the logical relationship of time domain-frequency domain-time frequency domain feature fusion provided by the embodiment of the present invention. Figure 2 In this study, features of the original optical fiber signal are extracted from the time, frequency, and time-frequency domains. This yields 15 time-domain features, 85 frequency-domain features (including those derived from various analytical methods such as discrete Fourier transform), and 43 time-frequency features (extracted using methods such as wavelet transform). After removing two invalid features, a dataset of 141 valid features is fused. In the time domain, basic statistical features such as maximum, minimum, peak-to-peak value, variance, standard deviation, and mean are precisely extracted, along with parameters reflecting signal waveform characteristics such as kurtosis, skewness, impulse factor, crest factor, shape factor, and margin factor, capturing the signal's temporal variation. In the frequency domain, key features such as frequency standard deviation, frequency variance, frequency mean square, center of gravity frequency, frequency root mean square (RMS), and occupied bandwidth are extracted using methods such as discrete Fourier transform, power spectrum analysis, and cepstrum analysis to explore the signal's frequency distribution and energy characteristics. In the time-frequency domain, wavelet transform and wavelet packet decomposition techniques, combined with singular value decomposition and other methods, are used to extract wavelet energy, singular values, singular value entropy and other features, comprehensively capturing the dynamic changes of the original optical fiber vibration signal at different time and frequency scales for supervised learning. Figure 2Its significance lies in solving the problem of feature extraction in the field of fiber optic sensing and comprehensively restoring signal characteristics through multi-domain fusion; improving the model fitting effect, and subsequently analyzing based on multiple behavior recognition models, the area under the curve reaches 0.9899; promoting research and application development, and providing high-quality data support for subsequent feature selection and pattern recognition, which has important application value in multiple fields.

[0020] S103. Screening the primary features through orthogonality analysis and difference analysis to obtain first-level features, and screening the first-level features through genetic algorithm to obtain second-level features.

[0021] In an embodiment of the present invention, the extracted primary features undergo a three-level screening process. The first stage uses orthogonality analysis and variance analysis for preliminary screening. The means and variances of different eigenvalues of similar signals are compared to determine the degree of similarity, and overly similar eigenvalues are removed. For example, statistical analysis is performed on similar eigenvalues of user behavior signals to remove highly similar features. Different types of similar eigenvalues are compared to determine whether there are sets of eigenvalues that can completely distinguish all features. Such eigenvalue combinations are identified by setting a threshold and screening the eigenvalues. Following orthogonality and variance analysis, eigenvalues identified as similar or redundant are removed to obtain the first-level features. The second stage involves eigenvalue selection using a genetic algorithm. The first-level features after the first stage of preliminary screening are encoded, typically using binary encoding, with each feature represented as one or more bits in a gene string, forming a chromosome. The maximum number of iterations is set to 100, the population size is set to 50, and a certain number of chromosomes are randomly generated to form the initial population. The fitness value of each chromosome is calculated based on the performance of the feature subset on the classification task (e.g., classification accuracy), with higher values indicating better performance. Based on the fitness value, individuals are selected from the population using methods such as roulette wheel selection. Individuals with higher fitness are more likely to be selected into the next generation. The selected individuals exchange some of their chromosome genes at a certain crossover probability to generate new individuals. Some genes of the individuals are mutated with a small mutation probability to explore the search space. It is determined whether the preset termination conditions are met (such as reaching the maximum number of iterations). If not, the fitness calculation step is returned. If it is met, the iteration is stopped to obtain the second-level features.

[0022] S104: Filter the second-level features using the mutual information evaluation function to obtain target features.

[0023] In embodiments of the present invention, the mutual information evaluation function is a measure of the degree of interdependence between variables. In feature screening, the mutual information evaluation function can be used to assess the correlation between input features and the target variable, thereby helping to select the most valuable features for predicting the target. The third level is mutual information evaluation function screening, which calculates the mutual information score between each feature in the second-level features and the target variable (behavior pattern category), sorts them by score, and selects the feature with the highest score as the target feature.

[0024] Among them, the comparison of the three-level screening results with the original data set and the results of other feature screening algorithms shows that the three-level screening algorithm has advantages in accuracy, processing time, universality and flexibility. The average classification accuracy reaches 93.01%, which can be used in the field of smart elderly care to identify behaviors such as falls of elderly people indoors.

[0025] S105. Input the target features into multiple behavior recognition models respectively to obtain the user's target behavior results, which include the results of having fallen and the results of not having fallen. When the result of having fallen is obtained, the user's fall position information is obtained from the target optical fiber vibration signal.

[0026] In an embodiment of the present invention, the weight ratios set for each of the multiple behavior recognition models generally refer to the process of assigning importance to different features or models. The target features are input into the multiple behavior recognition models respectively to obtain their respective output results. According to the weight ratios corresponding to the respective behavior recognition models, the respective output results are combined to obtain the final target behavior result of the user. The target behavior result includes determining whether the user has fallen or not. When it is recognized that the user has fallen, the user's fall position information is obtained from the target optical fiber vibration signal.

[0027] Among them, various behavior recognition models can include machine learning classifiers such as support vector machines (SVM), K-nearest neighbor (KNN), random forests (RF), and extreme random trees (ET). Samples are divided into training samples and test samples (test samples account for 33%). Recognition accuracy and processing time are used as evaluation indicators, and the classifier effect is evaluated through a confusion matrix.

[0028] The performance of the aforementioned algorithms was compared in terms of processing time, accuracy, universality, and flexibility, and the optimal algorithm or combination of algorithms was selected to improve the algorithm's processing capabilities for complex signals and pattern recognition accuracy. Furthermore, the data processing process was optimized to reduce computational steps, improve algorithm efficiency, reduce recognition time, and enable rapid response to elderly behavior. The target features were processed using SVM, random forest, and MLP learning models. The ROC curve was used to analyze the model fitting performance of the target features compared to a single analysis domain dataset. The results showed that the area under the curve of the target features reached 0.9899, providing comprehensive and reliable data support for subsequent analysis.

[0029] S106: triggering an alarm message based on the fall result and the fall location information, and sending the alarm message to a terminal of a person monitoring the user.

[0030] In an embodiment of the present invention, the user's fall location information may include absolute location, relative location, environmental information, and health status associations, etc. Based on the fall result and fall location information, an alarm mechanism is automatically activated to generate an alarm message. The alarm message, including the fall result and fall location information, is sent in real time to the user's monitoring terminal so that the user can receive timely assistance.

[0031] The system automatically saves relevant data about the fall, including alarm screenshots and video recordings. This information can be used for post-event analysis to help understand the incident and serve as necessary evidence or reference.

[0032] It can be understood that in an embodiment of the present invention, the collected original optical fiber vibration signal is preprocessed to obtain a target optical fiber vibration signal, and the original optical fiber signal includes a walking signal, a fall signal, a fall signal while walking, and a human stationary signal; the target optical fiber vibration signal is feature extracted from three dimensions: time domain, frequency domain, and time-frequency domain to obtain primary features; the primary features are screened by orthogonality analysis and difference analysis to obtain first-level features, and the first-level features are screened by genetic algorithm to obtain second-level features; the second-level features are screened by mutual information evaluation function to obtain target features; the target features are input into multiple behavior recognition models respectively to obtain the user's target behavior results, and the target behavior results include the result of having fallen and the result of not having fallen, and when the result of having fallen is obtained, the user's fall position information is obtained from the target optical fiber vibration signal; based on the result of having fallen and the fall position information, an alarm message is triggered, and the alarm message is sent to the terminal of the user's monitoring person. In this method, multiple signals are collected to comprehensively cover the complex situations in actual scenarios. An efficient high-dimensional feature matrix screening algorithm is used to obtain a large amount of useful information, fully tap the potential value of the signal, and process the target features through multiple behavior recognition models, making the result of judging whether the user has fallen more accurate.

[0033] In an embodiment of the present invention, the aforementioned processing method was used to extract 295 sets of data on four behavioral patterns: climbing, shaking, knocking, and stationary. Repeating the aforementioned feature extraction, screening, and pattern recognition steps, the results showed an accuracy rate of 98.45% in perimeter security, with a processing time as low as 5.35 milliseconds per pass and a maximum sensitivity of 98.43%, demonstrating the algorithm's universality and generalization capabilities.

[0034] In some embodiments of the present invention, S102 further includes S201 to S205, which are explained through the following steps.

[0035] S201. Input the primary features into an evaluation function scoring model to obtain a first score for each feature in the primary features. The evaluation function scoring model includes three evaluation functions: Spearman correlation coefficient, distance correlation coefficient, and ridge regression.

[0036] In some embodiments of the present invention, the primary features are input into an evaluation function scoring model, which uses three evaluation functions, namely, Spearman correlation coefficient, distance correlation coefficient, and ridge regression, to score each feature in the primary features to obtain a first score.

[0037] S202: Input the primary features into the base learning model to obtain the accuracy, processing time, and sensitivity of each feature, and obtain the second score of each feature through the accuracy, processing time, and sensitivity.

[0038] In some embodiments of the present invention, primary features are input into a base learning model, and the accuracy, processing time, and sensitivity of each feature are calculated. These three metrics are then used to score each feature at the learning model level to obtain a second score. The base learning model may include, but is not limited to, a decision tree, a neural network, and a support vector machine.

[0039] S203 : Screen the primary features based on the first score and the second score to obtain multiple feature subsets.

[0040] S204: Input multiple feature subsets into the base learning model to obtain their corresponding accuracy rates, and use the accuracy rates as weights.

[0041] In some embodiments of the present invention, multi-class feature screening is performed on the primary features based on the first score and the second score to obtain a multi-class feature subset. The respective accuracy rate refers to the proportion of instances correctly classified by the base learning model on the multi-class feature subset. The multi-class feature subset is input into the base learning model to obtain the corresponding accuracy rate, and the corresponding accuracy rate is used as the respective weight.

[0042] S205 : Normalize the first score and the second score to the same scale to obtain a target first score and a target second score, and obtain a target feature through the target first score and the target second score.

[0043] In some embodiments of the present invention, a specific normalization method is used to normalize the first score and the second score to a scale range of 0-1 to obtain a target first score and a target second score, and finally the target feature is obtained by weighted summing the target first score, the target second score and the weight.

[0044] In some embodiments of the present invention, S205 can be implemented through S2051, which is explained through the following steps.

[0045] S2051. Perform weighted summation on the target first score and the target second score by weight to obtain the final score of each feature in the primary features; and sort each feature from high to low according to the final score, and select the preset features as the target features.

[0046] In some embodiments of the present invention, the target first score and the target second score are weighted and summed to obtain a final score for each feature in the primary features. The features are then sorted from high to low based on the final score. After the sorting is complete, the top K features are selected and used as target features.

[0047] like Figure 3 As shown, Figure 3 The technical roadmap of the feature comprehensive scoring system provided by the embodiment of the present invention shows two levels of feature scoring: selecting a base score model based on the evaluation function (Spearman correlation coefficient, distance correlation coefficient, ridge regression) and the learning model (accuracy, processing time, sensitivity), as well as the steps of normalization and soft voting, to further optimize feature screening and improve system performance. Figure 3 In this model, the feature dataset is composed of primary features. Each primary feature is scored using the three evaluation functions in the scoring model: the Spearman correlation coefficient, the distance correlation coefficient, and the ridge regression. The primary features are then fed into the base learning model to obtain a single feature: accuracy, processing time, and sensitivity. Accuracy, processing time, and sensitivity are used to calculate the score for each feature, ultimately yielding six sets of scores: score 1, score 2, score 3, score 4, score 5, and score 6. Feature selection is performed within the primary features based on scores 1 to 6. Specifically, the top K features in each scoring category are selected as subsets, resulting in six feature subsets. These six feature subsets are then fed into the base learning model to obtain their corresponding accuracy, which is then used as the weight. In the second layer, the six scores are normalized to the same scale. For example, scores 1 to 6 are compressed to a scale of 0 to 1, yielding six normalized scores. The normalized scores 1 to 6 are weighted and summed by weight to obtain the final score of each feature in the primary features. Each feature is sorted from high to low according to the final score, and the top K features are selected as the target features.

[0048] In the embodiment of the present invention, S203 can be implemented through S2031 to S2034, which is explained through the following steps.

[0049] S2031, taking the first preset features of the primary features as subset 1 according to score 1, and taking the first preset features of the primary features as subset 2 according to score 2; S2032, taking the first preset features of the primary features as subset 3 according to the score 3, and taking the first preset features of the primary features as subset 4 according to the score 4; S2033, taking the first preset features of the primary features as subset 5 according to the score 5, and taking the first preset features of the primary features as subset 6 according to the score 6; S2034. Combine subset 1, subset 2, subset 3, subset 4, subset 5, and subset 6 to obtain 6 types of feature subsets.

[0050] In some embodiments of the present invention, the first score includes score 1 after scoring by the Spearman correlation coefficient, score 2 after scoring by the distance correlation coefficient, and score 3 after scoring by ridge regression. The second score includes score 4 after scoring by accuracy, score 5 after scoring by processing time, and score 6 after scoring by sensitivity. The primary features are screened according to the six scores to obtain six feature subsets.

[0051] In some embodiments of the present invention, inputting the target features into a plurality of behavior recognition models in S105 to obtain the user's target behavior results can be achieved through S1051 to S1052, which is explained in the following steps.

[0052] S1051. Input the target features into multiple behavior recognition models respectively to obtain the corresponding behavior results.

[0053] S1052. Obtain target behavior results from the behavior results according to the weight ratios set for each of the multiple behavior recognition models.

[0054] In some embodiments of the present invention, the weights assigned to various behavior recognition models are generally referred to as the process of assigning importance to different features or models. The target features are input into the various behavior recognition models to obtain their corresponding behavior results. The corresponding behavior results and their corresponding weights are then combined and calculated to obtain the final target behavior result.

[0055] In some embodiments of the present invention, S106 may be implemented through S1061 to S1063, which is described in the following steps.

[0056] S1061. Obtain the user's precise location information based on the target optical fiber vibration signal, and construct a simulated space map through the precise location information.

[0057] In some embodiments of the present invention, the precise location information, i.e., the specific location of the elderly person and their movement, is determined by analyzing the captured and processed target optical fiber vibration signal. A simulated spatial map is constructed based on the obtained precise location information to reflect the elderly person's activity range and path in a specific environment.

[0058] S1062: Optimize the simulated space map based on point conversion and curve fitting algorithms to obtain a walking route map for the user.

[0059] In some embodiments of the present invention, point transformation is used to adjust the coordinate system or reference point, and curve fitting technology is applied to optimize and smooth the simulated space map to generate a more accurate and coherent walking route map.

[0060] S1063. Obtain the user's walking trajectory updated in real time through a time series analysis algorithm and a walking route map, and obtain the user's fall location information through the walking trajectory.

[0061] In some embodiments of the present invention, the refined walking route map is combined with a time series analysis algorithm to update the elderly person's walking trajectory in real time based on data from different time periods, more accurately reflecting their actual walking patterns and behavior. The resulting walking trajectory can be used to accurately locate the fall location so that necessary assistance can be provided in a timely manner.

[0062] In an embodiment of the present invention, a user fall optical fiber sensing monitoring system based on φ-OTDR is proposed in the embodiment of the present invention, comprising: a sensing optical fiber network, a vibration signal positioning module, a monitoring center and a receiving terminal for the monitored person, wherein the monitoring center comprises a signal acquisition module, a signal processing module, a feature extraction module, a feature screening module, a pattern recognition module and a monitoring center core; wherein, The signal acquisition module is used to collect the original optical fiber vibration signals collected by the sensor optical fiber network. The original optical fiber signals include walking signals, falling signals, falling signals while walking signals, and human stillness signals; A signal processing module is used to pre-process the collected original optical fiber vibration signal to obtain the target optical fiber vibration signal; The feature extraction module is used to extract the features of the target optical fiber vibration signal from three dimensions: time domain, frequency domain and time-frequency domain to obtain primary features; The feature screening module is used to screen the primary features through orthogonality analysis and difference analysis to obtain the first-level features, and to screen the first-level features through genetic algorithm to obtain the second-level features; and to screen the second-level features through mutual information evaluation function to obtain the target features; The pattern recognition module is used to input the target features into multiple behavior recognition models to obtain the user's target behavior results, including the results of falling and the results of not falling; A vibration signal positioning module is used to obtain the user's fall location information from the target optical fiber vibration signal when a fall result is obtained; The core of the monitoring center is used to coordinate the data transmission between the sensing fiber optic network, vibration signal positioning module, and monitoring centers, and generate alarm information based on the fall results and fall location information, and send the fall results and fall location information to the user's monitoring terminal through the alarm information.

[0063] In an embodiment of the present invention, a preset length of armored single-mode optical fiber can be selected within the user's spatial range. This fiber maintains system stability and adaptability while maintaining a relatively low cost. A grid-like ground-laying method is adopted to reasonably control laying costs while increasing the number of fiber disturbance points and improving signal acquisition sensitivity, facilitating large-scale deployment in various scenarios, particularly in homes for elderly individuals living alone and nursing homes. Experiments have shown that the unit area laying cost is one-third of that of traditional smart flooring. This means that when determining whether a user has fallen, the present invention can utilize a low-cost, large-scale deployment system architecture, further deploying a φ-OTDR system and acquiring optical fiber signals from armored single-mode optical fibers laid out in a grid-like pattern within the user's living space. The grid-like ground-laying of sensing optical fibers increases the number of fiber disturbance points, comprehensively capturing changes in optical fiber signals caused by various behaviors, and obtaining rich and accurate raw signals.

[0064] Example 1 See also Figure 4 As shown, Figure 4 This diagram illustrates the basic structure of a fiber-optic communication system according to an embodiment of the present invention. When using the signal acquisition module for signal acquisition, the φ-OTDR system in the acquisition module consists of an ultra-narrow linewidth laser, an acousto-optic modulator, an EDFA (erbium-doped fiber amplifier), a circulator, the optical fiber under test, a photodetector, and a PC-based signal processing unit.

[0065] The ultra-narrow linewidth laser outputs lasers with a central wavelength of 1550nm and a linewidth of less than 3KHz, providing a highly coherent light source for the system; The signal generator generates an electrical signal to drive the acousto-optic modulator to modulate the laser; The modulated laser is amplified by EDFA to compensate for the loss in the optical fiber transmission; The circulator guides the transmission of optical signals. The amplified light enters from port 1 and is input into the optical fiber to be tested through port 2. The backscattered light in the optical fiber returns from port 2 and is output to the photodetector through port 3. The optical fiber under test is the sensing element. External vibrations and other effects can cause phase shifts in the light within it. External events are monitored by detecting backscattered light. In the experiment, a single-mode, single-core armored optical fiber was used, and a 2km serpentine lay was used across an 80 square meter open space. The photodetector converts the optical signal into an electrical signal, which is convenient for subsequent processing on the PC side; PC-side signal processing performs operations such as amplification, filtering, denoising, and feature extraction on electrical signals to identify external events.

[0066] Using the aforementioned system, 2 km of single-mode, single-core armored optical fiber was laid in a serpentine pattern across an 80-square-meter open space, covering an area of 57.6 square meters, with a unit laying rate of 3.68 meters per square meter. Prior to implementing the φ-OTDR-based user fall monitoring system, four types of signals were collected in the aforementioned environment: walking, falling, walking + falling, and ambient sound. Using adults simulating elderly behavior, the system was segmented by gender, height, weight, and time period, generating 2,242 sets of data to verify the system's accuracy. The φ-OTDR system boasts advantages such as strong electromagnetic interference resistance, waterproofness, corrosion resistance, excellent insulation, and low cost, making it adaptable to complex and specialized environments such as bathrooms and toilets, demonstrating broad application potential in behavioral monitoring. Experiments have shown that the system improves signal-to-noise ratio by 40% under 50Hz power frequency interference, significantly outperforming traditional solutions.

[0067] Example 2: An experimental platform was constructed based on the principles of the φ-OTDR system, utilizing components such as the Koheras BASIK X10 ultra-narrow linewidth laser and the KG-BPR series 350M photodetector. Parameters such as the laser operating wavelength was set to 1550nm and the pulse width to 10ns, achieving a spatial resolution of approximately 1m and a sensing distance exceeding 1km. An indoor area, such as a 6m x 4.8m area, was used to lay 1.5mm armored single-mode optical fiber in a grid pattern. This grid-like layout increased the fiber perturbation point density to 5 points per square meter, a threefold increase compared to traditional straight-line laying, significantly enhancing signal sensitivity. The total length was approximately 253.8m. Prior to implementing the φ-OTDR-based user fall monitoring system, signals from four behavioral patterns were collected under the aforementioned conditions: normal environment, walking, falling, and falling while walking. Adults simulated elderly behavior, accounting for factors such as weight, gait, and posture. A total of 1947 sets of valid data were collected to verify the system's accuracy.

[0068] The φ-OTDR-based fiber-optic fall monitoring system features tightly integrated modules that work together, from signal acquisition and processing to feature extraction, screening, and pattern recognition, to form a complete and efficient elderly fall monitoring process. By optimizing data transmission and processing logic between modules and adopting a parallel processing architecture, unnecessary hardware configuration and computing resource waste are reduced. This system achieves precise monitoring while reducing overall system costs and keeping recognition latency to under 0.5 seconds, providing strong support for large-scale deployment.

[0069] Those skilled in the art will appreciate that the units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present invention.

[0070] The above implementation methods are only used to illustrate the embodiments of the present invention, and are not intended to limit the embodiments of the present invention. Ordinary technicians in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of patent protection of the embodiments of the present invention should be defined by the claims.

Claims

1. A method for monitoring user falls based on optical fiber sensing based on φ-OTDR, characterized in that: include: Preprocessing the collected original optical fiber vibration signal to obtain a target optical fiber vibration signal, wherein the original optical fiber vibration signal includes a walking signal, a falling signal, a falling-while-walking signal, and a human stationary signal; Extract the target optical fiber vibration signal from three dimensions: time domain, frequency domain and time-frequency domain to obtain primary features; Screening the primary features through orthogonality analysis and difference analysis to obtain first-level features, and screening the first-level features through genetic algorithm to obtain second-level features; The second-level features are screened by mutual information evaluation function to obtain target features; Inputting the target features into multiple behavior recognition models respectively to obtain target behavior results of the user, wherein the target behavior results include a fall result and a no-fall result, and when the fall result is obtained, obtaining the fall position information of the user from the target optical fiber vibration signal; An alarm message is triggered based on the fall result and the fall location information, and the alarm message is sent to a terminal of a person monitoring the user.

2. The method according to claim 1, characterized in that After extracting the features of the target optical fiber vibration signal from the three dimensions of time domain, frequency domain and time-frequency domain to obtain the primary features, the method further includes: Inputting the primary features into an evaluation function scoring model to obtain a first score for each feature in the primary features, wherein the evaluation function scoring model includes three evaluation functions: Spearman correlation coefficient, distance correlation coefficient, and ridge regression; Inputting the primary features into a base learning model to obtain the accuracy, processing time, and sensitivity of each feature, and obtaining a second score for each feature based on the accuracy, processing time, and sensitivity; Perform feature screening on the primary features based on the first score and the second score to obtain multiple feature subsets; Inputting the multi-class feature subsets into the base learning model to obtain the corresponding accuracy rates of the respective features, and using the accuracy rates as the corresponding weights of the respective features; The first score and the second score are normalized to the same scale to obtain a target first score and a target second score, and a target feature is obtained through the target first score, the target second score and the weight.

3. The method according to claim 2, characterized in that The acquiring the target feature by using the target first score, the target second score and the weight includes: The target first score and the target second score are weightedly summed by the weights to obtain the final score of each feature in the primary features; and each feature is sorted from high to low according to the final score, and the preset features are selected as the target features.

4. The method according to claim 2, characterized in that The first score includes score 1 after the Spearman correlation coefficient score, score 2 after the distance correlation coefficient score, and score 3 after the ridge regression score. The second score includes score 4 after the accuracy score, score 5 after the processing time score, and score 6 after the sensitivity score. The feature screening is performed in the primary features based on the first score and the second score to obtain multiple feature subsets, including: According to the score 1, the first preset features of the primary features are taken as subset 1, and according to the score 2, the first preset features of the primary features are taken as subset 2; According to the score 3, the first preset features of the primary features are taken as subset 3, and according to the score 4, the first preset features of the primary features are taken as subset 4; According to the score 5, the first preset features of the primary features are taken as subset 5, and according to the score 6, the first preset features of the primary features are taken as subset 6; The subset 1, subset 2, subset 3, subset 4, subset 5 and subset 6 are combined to obtain 6 types of feature subsets.

5. The method according to claim 1, wherein The target features are input into multiple behavior recognition models to obtain the user's target behavior results, including: Inputting the target features into the multiple behavior recognition models respectively to obtain corresponding behavior results; The target behavior result is obtained from the behavior results according to the weight ratios set for each of the multiple behavior recognition models.

6. The method according to claim 1, characterized in that The obtaining the fall position information of the user from the target optical fiber vibration signal includes: Acquire the precise location information of the user based on the target optical fiber vibration signal, and construct a simulated space map through the precise location information; Optimizing the simulated space map based on point conversion and curve fitting algorithms to obtain a walking route map of the user; The user's walking trajectory updated in real time is obtained through a time series analysis algorithm and the walking route map, and the user's fall location information is obtained through the walking trajectory.

7. A user fall optical fiber sensing monitoring system based on φ-OTDR, characterized in that: include: A sensing fiber network, a vibration signal positioning module, a monitoring center, and a receiving terminal for the monitor. The monitoring center includes a signal acquisition module, a signal processing module, a feature extraction module, a feature screening module, a pattern recognition module, and a monitoring center core. The signal acquisition module is used to collect the original optical fiber vibration signal collected by the sensing optical fiber network, wherein the original optical fiber signal includes a walking signal, a falling signal, a falling-while-walking signal, and a person-stationary signal; The signal processing module is used to pre-process the collected original optical fiber vibration signal to obtain the target optical fiber vibration signal; The feature extraction module is used to extract features of the target optical fiber vibration signal from three dimensions: time domain, frequency domain and time-frequency domain to obtain primary features; The feature screening module is used to screen the primary features through orthogonality analysis and difference analysis to obtain first-level features, and to screen the first-level features through genetic algorithm to obtain second-level features; The second-level features are screened by mutual information evaluation function to obtain target features; The pattern recognition module is used to input the target features into multiple behavior recognition models respectively to obtain the user's target behavior results, wherein the target behavior results include a fall result and a no-fall result; The vibration signal positioning module is used to obtain the fall position information of the user from the target optical fiber vibration signal when the fall result is obtained; The monitoring center core is used to coordinate the data transmission between the sensing fiber optic network, the vibration signal positioning module, and the monitoring center, and generate alarm information based on the fall result and the fall location information, and send the fall result and the fall location information to the terminal of the user's monitor through the alarm information.

Images