A highly sensitive human body motion detection system based on optical fiber sensing technology
Through fiber optic sensing technology combined with deep learning and neural network, high sensitivity and accuracy of human body motion detection in complex environments is achieved, and the performance stability and sensitivity of existing systems in complex environments is solved, and the immediacy is met. High demand for physical motion detection.
Patent Information
- Application Number
- CN202410112776.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-26
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-01-26
AI Technical Summary
The existing high-sensitivity human body motion detection system based on fiber optic sensing technology has poor performance stability and low sensitivity in complex environments, and is prone to false alarms and missed reports. It also has the generalization performance of physiological characteristics and diverse body motion modes of different individuals. Restricted, it cannot meet the real-time physical detection needs with high immediacy.
It adopts fiber optic sensing array module, optical signal acquisition and processing module, human body motion analysis module, sensitivity adjustment module and system monitoring and alarm module, combined with deep learning and neural network technology, and captures tiny optical changes through fiber optic sensors to extract human features. The body motion detection algorithm based on Yolov3 algorithm and deep neural network is used to identify complex body motion patterns, and the sensitivity is adjusted adaptively through the neural network to realize the intelligence and adaptability of the system.
Improves the robustness and detection accuracy of the system, overcomes the impact of light changes, background fluctuations and appearance changes on body motion detection, enhances the sensitivity to slight changes, and ensures stable and accurate monitoring in complex environments. Meets the application needs of high immediacy.
Smart Images

Figure CN118021287B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of optical fiber sensing, deep learning and adaptive neural networks, and specifically to a high-sensitivity human body motion detection system based on optical fiber sensing technology. Background Art
[0002] Fiber optic sensing technology uses optical fibers as sensors, aiming to capture tiny optical changes through fiber optic sensors and achieve high-sensitivity detection of specific physical quantities in the environment. By arranging fiber optic sensor arrays, the high-sensitivity human motion detection system based on fiber optic sensing technology can perceive tiny optical changes in real time, providing a more accurate monitoring method for human body motion. Deep learning technology is a computational model based on neural networks, which aims to solve complex nonlinear relationships and pattern recognition tasks through multi-level abstraction and learning. In the human body motion detection system, deep learning technology automatically learns and extracts abstract features of human motion through a large amount of data training models, so that the system can more accurately identify and classify different body motion patterns, so that the body motion detection system can adapt to various complex scenarios and changes, and improve the intelligence level of the high-sensitivity human body motion detection system based on fiber optic sensing technology.
[0003] Adaptive neural network technology is an advanced method that can dynamically adjust system parameters. It aims to solve the problem that the sensitivity of traditional body motion detection systems is difficult to adjust under different environments and usage conditions. By introducing adaptive neural networks, the high-sensitivity human body motion detection system based on fiber optic sensing technology has the ability to automatically learn and adapt to environmental changes. The system can automatically adjust the weights and structural parameters of the neural network according to changes in external inputs, user behaviors and environmental conditions to achieve rapid adaptation and learning of new information. By integrating adaptive neural network technology, the system can adjust the sensitivity of fiber optic sensors in real time according to environmental changes and user needs, allowing the system to adapt to different scenarios more flexibly, ensuring high sensitivity in various body motion monitoring applications, while reducing the need for manual adjustments by users and improving the autonomy and adaptability of the system.
[0004] However, the existing high-sensitivity human motion detection systems based on fiber optic sensing technology have problems with poor performance stability and low sensitivity when facing complex environments. This includes problems such as false alarms and missed alarms caused by interference with the system under conditions of large temperature fluctuations and strong light. The system also has limited generalization performance for the physiological characteristics and diverse motion patterns of different individuals, and the system's real-time motion detection and response capabilities cannot meet the needs of applications with higher immediacy. Summary of the invention
[0005] The purpose of the present invention is to provide a high-sensitivity human body motion detection system based on fiber optic sensing technology, so as to solve the problems of poor system performance stability and low sensitivity of the existing high-sensitivity human body motion detection system based on fiber optic sensing technology proposed in the above background technology when facing complex environments, including the problem that the system is disturbed and produces false alarms and missed alarms when the temperature fluctuates greatly and the light is strong, and the system has limited generalization performance for the physiological characteristics and diverse body motion patterns of different individuals, and the system's real-time body motion detection and response capabilities cannot meet the application requirements with higher immediacy.
[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a highly sensitive human body motion detection system based on optical fiber sensing technology, comprising an optical fiber sensing array module, an optical signal acquisition and processing module, a human body motion analysis module, a sensitivity adjustment module, a system monitoring and alarm module, and a user interface and notification module, characterized in that: the optical fiber sensing array module is used to layout and configure optical fiber sensors to build an all-round perception detection system, the optical signal acquisition and processing module is used to acquire optical signals transmitted through optical fiber sensors, and perform efficient signal processing, including noise suppression, filtering, and data preprocessing, to ensure that the information extracted from the sensor is accurate and reliable, and to provide a clear data basis for subsequent analysis, the human body motion analysis module comprises a human feature extraction unit and a human body motion detection unit, the human feature extraction unit is used to extract relevant human features from the optical signal collected by the optical fiber sensor, including motion patterns and Physiological indicators provide rich information for subsequent body motion detection and analysis. The human body motion detection unit proposes a body motion detection algorithm based on deep learning to learn and identify complex human motion patterns, including body movements in bed and physiological characteristics of heart rate and respiratory rate, so as to achieve accurate detection of different body movements. The sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on a neural network to identify relevant characteristics of human activities. According to the optical signals collected in real time, combined with neural network technology, it dynamically learns and adapts to human body motion patterns in different environments, and realizes intelligent adjustment of the system for human body motion detection. The system monitoring and alarm module is used to monitor human body motion information in real time, monitor the system operation status, and trigger an alarm when an abnormality and emergency are detected. The user interface and notification module is used for real-time interaction between the system and the user, and is responsible for sending notifications and alarm information to the user to ensure that the user can quickly learn about important events monitored by the system.
[0007] Preferably, the fiber optic sensor array module captures tiny optical changes in the surrounding environment in real time by arranging a high-sensitivity fiber optic sensor array, including the selection of fiber optic sensors, layout design, and connection with the system, to ensure effective perception of optical signal changes caused by human body movement.
[0008] Preferably, the optical signal acquisition and processing module acquires a variety of optical signals obtained by the optical fiber sensor array module, and performs preprocessing operations on the acquired optical signals to remove noise and interference, ensuring the system's high-sensitivity detection of tiny physiological changes, thereby providing reliable data support for subsequent human body movement analysis.
[0009] Preferably, the human body motion analysis module includes a human feature extraction unit, which extracts key features from complex signals, identifies and extracts human body motion and physiological characteristics, including movement patterns and physiological indicators, and efficiently captures physiological signals related to human body motion, providing a representative and discriminative feature set for subsequent body motion analysis.
[0010] Preferably, the human body motion analysis module includes a human body motion detection unit, which proposes a body motion detection algorithm based on deep learning. By analyzing a large amount of real-time data and combining time series information and spatial features, it can realize the recognition of multiple body motion detections and distinguish different human body motion patterns, thereby ensuring the system's efficient monitoring and accurate classification of body motion changes.
[0011] Specifically, the body motion detection algorithm based on deep learning is as follows: First, the human body information collected by the optical fiber sensor is directly divided into non-intersecting small squares through the Yolov3 algorithm, and then the target whose center point is located in the small square is predicted. The Yolov3 prediction target border calculation formula is specifically expressed as:
[0012] u(x)=σ(t x )+c(x)+α xy ·ΔT+β xy ΔI
[0013] u(y)=σ(t y )+c(y)+α xy ·ΔT+β xy ΔI
[0014] u(v)=exp(t v )R v +α xy ·ΔT+β xy ΔI
[0015] u(b)=exp(t b1 +t b2 )R b +α xy ·ΔT+β xy ΔI
[0016] Where u(·) represents the target bounding box function, x and y represent the center coordinates of the target bounding box, representing the position of the detected target in the image, v represents the position parameter of the target bounding box, which is used to adjust the prior box to accurately match the position of the detected target, b represents the size parameter of the target bounding box, which is used to adjust the size of the prior box to adapt to the size of the detected target, σ(·) represents the Sigmoid activation function, and t x With t y They represent the offset parameters of the center coordinates of the predicted bounding box relative to the x and y center coordinates of the small square, t v Represents the offset parameter of the predicted bounding box position relative to the prior position, t b1 With t b2 They represent the offset parameters of the width and height of the predicted bounding box relative to the prior width, exp(·) represents the exponential function, c(·) represents the center coordinate function of the divided small squares, and α xy represents the weight factor for the temperature change at the x and y positions, ΔT represents the temperature change, β xy represents the weight factor of the illumination change at the x and y positions, ΔI represents the illumination change, R ν With R b Respectively represent the prior size of the position and size of the divided small squares. The Yolov3 algorithm is used to accurately detect the position and bounding box of the human target from the input information, and provide accurate human target position information for the subsequent body motion detection key point extraction and deep learning classifier, thereby improving the robustness and performance of the entire body motion detection system. Secondly, a deep neural network model is constructed to extract features from the input information to realize the extraction of key points of the human skeleton. The constructed deep neural network model includes 16 convolutional layers, 5 pooling layers and 3 fully connected layers. The specific formula of the convolutional layer is expressed as:
[0017] Z [l] =W [l] *A [l-1] +b [l]
[0018] Among them, Z [l] represents the convolution output function of the lth layer of the deep neural network, W [l] Represents the convolution kernel weight matrix of the lth layer of the deep neural network, A [l-1] represents the activation output function of the l-1th layer of the deep neural network, b [l] Represents the bias term of the lth layer, * represents the convolution operation, which helps detect edge features in human skeleton information. Further, the feature information after the convolution layer is flattened into a one-dimensional vector, classified through the fully connected layer, and learns the high-level representation of the input data, and finally outputs the classification of the input by the network. The specific formula of the pooling layer is expressed as:
[0019]
[0020]
[0021] in, Indicates that the maximum pooling operation selects the maximum value in each local area as the output function, Indicates that the average pooling operation calculates the average value in each local area as the output function, i, j respectively represent the spatial dimension index of the output in different directions, k represents the index of the feature channel, representing different feature channels, s represents the stride of the pooling window, m and n represent the spatial dimension index in the pooling window, which is used to traverse the local area. By selecting the appropriate pooling operation, the size of the feature information is reduced, the amount of calculation is reduced, and the deep learning neural network is more adaptable to the spatial changes of the input information. Finally, the feature information is divided into two branch outputs through the deep neural network model. The deep neural network model predicts the partial affinity field between the key point confidence map and the key point, thereby overcoming the influence of illumination changes, background fluctuations and appearance changes on body motion detection. The calculation formula of the key point confidence map is expressed as:
[0022]
[0023] in, represents the confidence of the dth feature information of the qth person at position t, var represents the variance of the Gaussian distribution, t d,q represents the actual position of the dth feature information of the qth person, d represents the index of the feature information through the deep neural network model, q represents the different individual indexes of body motion detection, and the partial affinity field calculation formula between key points is expressed as:
[0024]
[0025] in, Represents the vector of the partial affinity field at position t, representing the directional relationship between key points, x d1,q With x d2,q They represent the coordinates of feature information d1 and d2 at the qth person, respectively. ||·||2 represents the two-norm operation. When certain conditions are met, t is judged to be on the limb of the qth person. Take vector Otherwise, the default value is 0. The body motion detection algorithm based on deep learning provides powerful classification and recognition capabilities for the high-sensitivity human body motion detection system based on fiber optic sensing technology, which can improve the generalization performance and detection accuracy of the system. By accurately capturing the key points of the human skeleton and target detection, the algorithm effectively solves the impact of lighting changes, background fluctuations and appearance changes on body motion detection, and provides reliable technical support for the stable operation and accurate monitoring of the system.
[0026] Preferably, the sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on a neural network, uses a neural network model to learn the changing rules of body motion patterns, and ensures high sensitivity to tiny changes in human body motion in different scenarios by continuously optimizing model parameters and dynamically adjusting the perception sensitivity of the optical fiber sensor.
[0027] Specifically, the sensitivity adaptive adjustment algorithm based on neural network is as follows: First, a neural network model is established to realize the adaptive dynamic adjustment of the high-sensitivity human body motion detection system of optical fiber sensing technology. The neural network model is constructed by combining the radial basis function network with the feedforward neural network. The mapping formula from the input layer to the hidden layer of the radial basis function network is expressed as follows:
[0028]
[0029] Among them, φ i (t) represents the output of the i-th radial basis function, which represents the activation degree of the input signal x(t) for the radial basis function. i represents the index of the number of radial basis functions. x(t) represents the detection result from the human body motion analysis module as the input of the neural network. i represents the center of the hidden layer neuron, which is used to determine the position of each basis function in the input space, σ i The width parameter of the radial basis function determines the influence range of the basis function in the input space. exp represents the exponential function, and ||·|| represents the Euclidean distance. The specific formula for mapping the hidden layer to the output layer of the feedforward neural network is expressed as follows:
[0030]
[0031] Among them, L f represents the output of the feedforward neural network, represents the approximation of the nonlinear part of the system after radial basis function and weight adjustment, f[y(t-1)] represents the nonlinear function of the function y(t-1) fed back to the input of the overall system, y(t-1) represents the final output function of the sensitivity adjustment module during the t-1th iteration, y(t) represents the final output function of the sensitivity adjustment module during the tth iteration, w(t) represents the weight function of the overall neural network during the tth iteration, t represents the iteration number index of the neural network model, w i Represents the weight connecting the hidden layer and the output layer, which is used to adjust the influence of each radial basis function on the system output. n represents the total number of radial basis functions. At this time, the specific formula of the output of the adaptive dynamic control model based on the neural network is expressed as follows:
[0032]
[0033] Where k(t) represents the output signal of the adaptive dynamic control model based on the neural network, which is used to adjust the input signal of the sensitivity adjustment module to achieve adaptive dynamic adjustment, σ represents the damping ratio of the system, ρ represents the natural frequency of the system, and G t Represents the gain parameter of the t-th iteration of the neural network model. By integrating the radial basis function network and the feedforward neural network, an adaptive dynamic control model based on a neural network is constructed to better adapt to the complex nonlinear relationship of the body motion detection system. Secondly, by adopting the online learning rule of the σ adjustment mechanism, σ represents the adjustment speed used to adjust the weight relative to the target value. σ is obtained through the Lyapunov stability theory. According to the real-time feedback control error e(t) of the system, the weight of the neural network is adjusted through e(t), so that the neural network approaches the nonlinear part. The specific formula for weight update of the adaptive dynamic control model based on the neural network is expressed as:
[0034]
[0035] Where Δw represents the rate of change of the weight function w(t) during the t-th iteration, Ψ represents a diagonal matrix, whose elements can be adjusted to change the sensitivity of the neural network to errors, K represents a diagonal positive definite matrix, which is used to introduce damping to prevent over-adjustment of the algorithm, and w i * The expected target value of the weight of the radial basis function is expressed. Lyapunov stability theory is used to analyze the current stability conditions of the system and find the appropriate σ. Considering the sensitivity of the system, the size of σ is adjusted so that the system can adapt to external changes faster while maintaining stability. The learning error equation of the adaptive dynamic control model based on the neural network is specifically expressed as:
[0036] Δ[y(t),w(t)]=L f[y(t)] [y(t),w(t)]-f[y(t)]
[0037] Among them, Δ[y(t),w(t)] represents the learning error of the neural network, that is, the difference between the approximate nonlinear part f[y(t)] of the neural network and the actual nonlinear part. Finally, a control error evaluation method is proposed to associate the neural network learning error with the control error for the analysis of algorithm performance. The control error of the system is evaluated by the convolution method, and the control error is associated with the neural network learning error. The feedback control error e(t) of the high-sensitivity human body motion detection system based on optical fiber sensing technology is specifically expressed as:
[0038]
[0039] Among them, g(t) represents the unit impulse response of the system, * represents the convolution operation, g(t-τ) represents the unit impulse response of the system to the instantaneous input reaction, Δ[y(τ),w(τ)] represents the learning error between the expected output and the actual output of the neural network, τ represents the integral variable, and the upper bound estimation formula of the control error is further expressed as:
[0040]
[0041] Among them, ||e(t)|| ∞ represents the infinite norm of e(t) in the interval [0,T], T represents the total number of iterations of the adaptive dynamic control model based on neural network, Δ T [y(t),w(t)] represents the truncated part of Δ[y(t),w(t)] on the interval [0,T], that is, the part where the number of iterations exceeds T is truncated, ||Δ T [y(t),w(t)]∣∣ ∞ Indicates Δ T The infinite norm of [y(t),w(t)] on the interval [0,T] is obtained through convolution operation. The relationship between the control error e(t) and the neural network learning error Δ[y(t),w(t)] is obtained. The upper bound estimation of the control error can provide a quantitative measure of the system performance, which helps to improve the stability and robustness of the system in the time domain and frequency domain. The sensitivity adaptive adjustment algorithm based on the neural network realizes the intelligent adaptation of the high-sensitivity human motion detection system by combining the neural network with the fiber optic sensing technology. The adaptability and stability of the algorithm can ensure the reliability of the system in various complex operating environments, which helps to provide an effective means for realizing accurate, intelligent and robust human motion monitoring.
[0042] Preferably, the system monitoring and alarm module monitors the operating status of the entire system, including the connection status of the fiber optic sensor, the stability of the optical signal acquisition, and the feedback results of body motion detection, and sets reasonable alarm thresholds to achieve real-time monitoring of abnormal body movements and system failures, and trigger an alarm mechanism to notify system users.
[0043] Preferably, the user interface and notification module enables the user to monitor body movement information, adjust system parameters, and send alarm information to the user through notification, thereby ensuring that the user can learn about relevant body movement monitoring results in a timely manner by providing an interactive interface between the system and the user.
[0044] Compared with the prior art, the present invention has the following beneficial effects:
[0045] 1. The human body motion detection unit proposes a body motion detection algorithm based on deep learning. The algorithm accurately extracts the key points of the human skeleton through human posture estimation technology, and combines the tiny optical changes captured by the fiber optic sensor with human motion. The algorithm predicts the joint confidence map and the partial affinity field between the joints, and matches the feature information through a combinatorial optimization algorithm that solves the task allocation problem in polynomial time. It can effectively obtain the key information of the human skeleton. Secondly, the algorithm realizes accurate detection of human targets by combining the Yolov3 algorithm. Compared with traditional methods, the algorithm overcomes the performance stability problem of fiber optic sensing technology in extreme environments, including temperature changes and strong light conditions. By using the advanced feature extraction of the deep learning model, the algorithm can better adapt to different backgrounds, lighting conditions and changes in human posture. By accurately extracting the key points of the human skeleton, the algorithm It can significantly eliminate the negative impact of illumination changes, background fluctuations and appearance changes on human motion detection, thereby improving the robustness of the system. In addition, through the fast detection speed and accuracy of the Yolov3 algorithm and the high real-time performance of the human posture estimation technology, the system can realize real-time monitoring of human motion, providing users with a fast and accurate monitoring solution. The real-time performance of the algorithm is of great significance for rapid response and real-time alarm in emergency situations. In general, the body motion detection algorithm based on deep learning provides a powerful classification and recognition capability for the high-sensitivity human motion detection system based on fiber optic sensing technology, which can improve the generalization performance and detection accuracy of the system. By accurately capturing the key points of the human skeleton and detecting targets, the algorithm effectively solves the impact of illumination changes, background fluctuations and appearance changes on body motion detection, providing reliable technical support for the stable operation and accurate monitoring of the system.
[0046] 2. Sensitivity adjustment module A sensitivity adaptive adjustment algorithm based on neural network is proposed. This algorithm uses the body motion detection analysis results from the human body motion analysis module. Through neural network technology, radial basis function network and feedforward neural network are used to adaptively process and analyze the human body motion results detected by the system in real time. By building an adaptive control structure in the neural network model, the algorithm can dynamically adjust the controller parameters to compensate for the nonlinear characteristics existing in the fiber optic sensing technology. The use of neural networks enables the system to dynamically adjust the learning error and update the system parameters in real time according to the error between the system output and the reference model. Through Lyapunov stability theory, a stable controller parameter adjustment mechanism is constructed, so that the entire body motion detection system can adaptively adjust the sensitivity under different environments and conditions, and effectively cope with complex human body motion changes. This algorithm helps to improve the system's sensitivity to small changes, enabling the system to The algorithm can more accurately capture the subtle signals of human body movement. Through comprehensive consideration of neural network learning errors, it can more intelligently adapt to different scenarios and different human body movement characteristics, thereby providing more accurate and reliable human body movement monitoring results. In addition, the algorithm considers the influence of neural network learning errors on system control errors to ensure that the control error converges to zero asymptotically under the influence of neural network learning errors, so that the system can cope with errors and fluctuations in practical applications, enhance the reliability of the system in various scenarios, and make the system more robust in actual operation. In summary, the sensitivity adaptive adjustment algorithm based on neural network realizes the intelligent adaptation of high-sensitivity human body movement detection system by combining neural network with fiber optic sensing technology. The adaptability and stability of the algorithm can ensure the reliability of the system in various complex operating environments, and help provide an effective means to achieve accurate, intelligent and robust human body movement monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.
[0048] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION
[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] See also Figure 1 The present invention provides a high-sensitivity human body motion detection system based on optical fiber sensing technology, including an optical fiber sensing array module, an optical signal acquisition and processing module, a human body motion analysis module, a sensitivity adjustment module, a system monitoring and alarm module, and a user interface and notification module. The optical fiber sensing array module is used to layout and configure optical fiber sensors to build an all-round perception detection system. The optical signal acquisition and processing module is used to obtain optical signals transmitted by optical fiber sensors and perform efficient signal processing, including noise suppression, filtering, and data preprocessing, to ensure that the information extracted from the sensor is accurate and reliable, and to provide a clear data basis for subsequent analysis. The human body motion analysis module includes a human feature extraction unit and a human body motion detection unit. The human feature extraction unit is used to extract relevant human features from the optical signal collected by the optical fiber sensor, including motion patterns and physiological indicators, for subsequent body The motion detection and analysis provide rich information. The human motion detection unit proposes a motion detection algorithm based on deep learning to learn and identify complex human motion patterns, including body movements in bed and the physiological characteristics of heart rate and respiratory rate, so as to achieve accurate detection of different body movements. The sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on a neural network to identify the relevant characteristics of human activities. According to the optical signals collected in real time, combined with neural network technology, it dynamically learns and adapts to the human motion patterns in different environments, and realizes the intelligent adjustment of the system for human motion detection. The system monitoring and alarm module is used to monitor human motion information in real time, monitor the system operation status, and trigger alarms when abnormalities and emergencies are detected. The user interface and notification module is used for real-time interaction between the system and the user, and is responsible for sending notifications and alarm information to the user to ensure that the user can quickly learn about important events monitored by the system.
[0051] See also Figure 1 ,Furthermore, the fiber optic sensing array module captures the tiny optical changes in the surrounding environment in real time by arranging a high-sensitivity fiber optic sensor array, including the selection of fiber optic sensors, ,layout design and connection with the system, ensuring the effective perception of the optical signal changes caused by human ,body movement.
[0052] See also Figure 1 ,Furthermore, the optical signal acquisition and processing module collects a variety of optical ,signals obtained by the fiber optic sensor array module, and performs ,preprocessing operations on the collected optical signals to remove noise and interference, ,ensuring the system’s high-sensitivity detection of tiny physiological changes, ,thus providing reliable data support for subsequent human ,body motion analysis.
[0053] See also Figure 1Furthermore, the human body motion analysis module includes a human feature extraction unit, which extracts key features from complex signals, identifies and extracts human body motion and physiological characteristics, including movement patterns and physiological indicators, and efficiently captures physiological signals related to human body motion, providing a representative and discriminative feature set for subsequent body motion analysis.
[0054] See also Figure 1 Furthermore, the human body motion analysis module includes a human body motion detection unit. The human body motion detection unit proposes a body motion detection algorithm based on deep learning. By analyzing a large amount of real-time data and combining time series information and spatial features, it can realize the recognition of multiple body motion detections and distinguish different human body motion patterns, ensuring the system's efficient monitoring and accurate classification of body motion changes.
[0055] See also Figure 1 ,Furthermore, the body motion detection algorithm based on deep learning is as follows: First, the human body information collected by the optical fiber sensor is directly divided into non-intersecting small squares through the Yolov3 algorithm, and then the target whose center point is located in the small square is predicted. The calculation formula of the Yolov3 predicted target border is specifically expressed as:
[0056] u(x)=σ(t x )+c(x)+α xy ·ΔT+β xy ΔI
[0057] u(y)=σ(t y )+c(y)+α xy ·ΔT+β xy ΔI
[0058] u(v)=exp(t v )R v +α xy ·ΔT+β xy ΔI
[0059] u(b)=exp(t b1 +t b2 )R b +α xy ·ΔT+β xy ΔI
[0060] Where u(·) represents the target bounding box function, x and y represent the center coordinates of the target bounding box, representing the position of the detected target in the image, v represents the position parameter of the target bounding box, which is used to adjust the prior box to accurately match the position of the detected target, b represents the size parameter of the target bounding box, which is used to adjust the size of the prior box to adapt to the size of the detected target, σ(·) represents the Sigmoid activation function, and t x With ty They represent the offset parameters of the center coordinates of the predicted bounding box relative to the x and y center coordinates of the small square, t v Represents the offset parameter of the predicted bounding box position relative to the prior position, t b1 With t b2 They represent the offset parameters of the width and height of the predicted bounding box relative to the prior width, exp(·) represents the exponential function, c(·) represents the center coordinate function of the divided small squares, and α xy represents the weight factor for the temperature change at the x and y positions, ΔT represents the temperature change, β xy represents the weight factor of the illumination change at the x and y positions, ΔI represents the illumination change, R ν With R b Respectively represent the prior size of the position and size of the divided small squares. The Yolov3 algorithm is used to accurately detect the position and bounding box of the human target from the input information, and provide accurate human target position information for the subsequent body motion detection key point extraction and deep learning classifier, thereby improving the robustness and performance of the entire body motion detection system. Secondly, a deep neural network model is constructed to extract features from the input information to realize the extraction of key points of the human skeleton. The constructed deep neural network model includes 16 convolutional layers, 5 pooling layers and 3 fully connected layers. The specific formula of the convolutional layer is expressed as:
[0061] Z [l] =W [l] *A [l-1] +b [l]
[0062] Among them, Z [l] represents the convolution output function of the lth layer of the deep neural network, W [l] Represents the convolution kernel weight matrix of the lth layer of the deep neural network, A [l-1] represents the activation output function of the l-1th layer of the deep neural network, b [l] Represents the bias term of the lth layer, * represents the convolution operation, which helps detect edge features in human skeleton information. Further, the feature information after the convolution layer is flattened into a one-dimensional vector, classified through the fully connected layer, and learns the high-level representation of the input data, and finally outputs the classification of the input by the network. The specific formula of the pooling layer is expressed as:
[0063]
[0064] in, Indicates that the maximum pooling operation selects the maximum value in each local area as the output function, Indicates that the average pooling operation calculates the average value in each local area as the output function, i, j respectively represent the spatial dimension index of the output in different directions, k represents the index of the feature channel, representing different feature channels, s represents the stride of the pooling window, m and n represent the spatial dimension index in the pooling window, which is used to traverse the local area. By selecting the appropriate pooling operation, the size of the feature information is reduced, the amount of calculation is reduced, and the deep learning neural network is more adaptable to the spatial changes of the input information. Finally, the feature information is divided into two branch outputs through the deep neural network model. The deep neural network model predicts the partial affinity field between the key point confidence map and the key point, thereby overcoming the influence of illumination changes, background fluctuations and appearance changes on body motion detection. The calculation formula of the key point confidence map is expressed as:
[0065]
[0066] in, represents the confidence of the dth feature information of the qth person at position t, var represents the variance of the Gaussian distribution, t d,q represents the actual position of the dth feature information of the qth person, d represents the index of the feature information through the deep neural network model, q represents the different individual indexes of body motion detection, and the partial affinity field calculation formula between key points is expressed as:
[0067]
[0068] in, Represents the vector of the partial affinity field at position t, representing the directional relationship between key points, x d1,q With x d2,q They represent the coordinates of feature information d1 and d2 at the qth person, respectively. ||·||2 represents the two-norm operation. When certain conditions are met, t is judged to be on the limb of the qth person. Take vector Otherwise, the default value is 0. The body motion detection algorithm based on deep learning provides powerful classification and recognition capabilities for the high-sensitivity human body motion detection system based on fiber optic sensing technology, which can improve the generalization performance and detection accuracy of the system. By accurately capturing the key points of the human skeleton and target detection, the algorithm effectively solves the impact of lighting changes, background fluctuations and appearance changes on body motion detection, and provides reliable technical support for the stable operation and accurate monitoring of the system.
[0069] See also Figure 1 ,Furthermore, the sensitivity adjustment module proposes a sensitivity adaptive ,adjustment algorithm based on neural network, which uses the neural network model to learn ,the changing rules of body motion patterns, and ensures high sensitivity to tiny ,changes in human body motion in different scenarios by continuously ,optimizing the model parameters and dynamically adjusting the perception sensitivity of the ,fiber fiber sensor.
[0070] See also Figure 1 ,Furthermore, the sensitivity adaptive adjustment algorithm based on neural network is as follows: First, a neural network model is established to realize the adaptive dynamic adjustment of the high-sensitivity human body motion detection system of optical fiber sensing technology. The neural network model is constructed by combining the radial basis function network with the feedforward neural network. The specific formula for mapping the input layer to the hidden layer of the radial basis function network is expressed as:
[0071]
[0072] Among them, φ i (t) represents the output of the i-th radial basis function, which represents the activation degree of the input signal x(t) for the radial basis function. i represents the index of the number of radial basis functions. x(t) represents the detection result from the human body motion analysis module as the input of the neural network. i represents the center of the hidden layer neuron, which is used to determine the position of each basis function in the input space, σ i The width parameter of the radial basis function determines the influence range of the basis function in the input space. exp represents the exponential function, and ||·|| represents the Euclidean distance. The specific formula for mapping the hidden layer to the output layer of the feedforward neural network is expressed as follows:
[0073]
[0074] Among them, L f represents the output of the feedforward neural network, represents the approximation of the nonlinear part of the system after radial basis function and weight adjustment, f[y(t-1)] represents the nonlinear function of the function y(t-1) fed back to the input of the overall system, y(t-1) represents the final output function of the sensitivity adjustment module during the t-1th iteration, y(t) represents the final output function of the sensitivity adjustment module during the tth iteration, w(t) represents the weight function of the overall neural network during the tth iteration, t represents the iteration number index of the neural network model, w i Represents the weight connecting the hidden layer and the output layer, which is used to adjust the influence of each radial basis function on the system output. n represents the total number of radial basis functions. At this time, the specific formula of the output of the adaptive dynamic control model based on the neural network is expressed as follows:
[0075]
[0076] Where k(t) represents the output signal of the adaptive dynamic control model based on the neural network, which is used to adjust the input signal of the sensitivity adjustment module to achieve adaptive dynamic adjustment, σ represents the damping ratio of the system, ρ represents the natural frequency of the system, and G tRepresents the gain parameter of the t-th iteration of the neural network model. By integrating the radial basis function network and the feedforward neural network, an adaptive dynamic control model based on a neural network is constructed to better adapt to the complex nonlinear relationship of the body motion detection system. Secondly, by adopting the online learning rule of the σ adjustment mechanism, σ represents the adjustment speed used to adjust the weight relative to the target value. σ is obtained through the Lyapunov stability theory. According to the real-time feedback control error e(t) of the system, the weight of the neural network is adjusted through e(t), so that the neural network approaches the nonlinear part. The specific formula for weight update of the adaptive dynamic control model based on the neural network is expressed as:
[0077]
[0078] Where Δw represents the rate of change of the weight function w(t) during the t-th iteration, Ψ represents a diagonal matrix, whose elements can be adjusted to change the sensitivity of the neural network to errors, K represents a diagonal positive definite matrix, which is used to introduce damping to prevent over-adjustment of the algorithm, and w i * The expected target value of the weight of the radial basis function is expressed. Lyapunov stability theory is used to analyze the current stability conditions of the system and find the appropriate σ. Considering the sensitivity of the system, the size of σ is adjusted so that the system can adapt to external changes faster while maintaining stability. The learning error equation of the adaptive dynamic control model based on the neural network is specifically expressed as:
[0079] Δ[y(t),w(t)]=L f[y(t)] [y(t),w(t)]-f[y(t)]
[0080] Among them, Δ[y(t),w(t)] represents the learning error of the neural network, that is, the difference between the approximate nonlinear part f[y(t)] of the neural network and the actual nonlinear part. Finally, a control error evaluation method is proposed to associate the neural network learning error with the control error for the analysis of algorithm performance. The control error of the system is evaluated by the convolution method, and the control error is associated with the neural network learning error. The feedback control error e(t) of the high-sensitivity human body motion detection system based on optical fiber sensing technology is specifically expressed as:
[0081]
[0082] Among them, g(t) represents the unit impulse response of the system, * represents the convolution operation, g(t-τ) represents the unit impulse response of the system to the instantaneous input reaction, Δ[y(τ),w(τ)] represents the learning error between the expected output and the actual output of the neural network, τ represents the integral variable, and the upper bound estimation formula of the control error is further expressed as:
[0083]
[0084] Among them, ||e(t)|| ∞ represents the infinite norm of e(t) in the interval [0,T], T represents the total number of iterations of the adaptive dynamic control model based on neural network, Δ T [y(t),w(t)] represents the truncated part of Δ[y(t),w(t)] on the interval [0,T], that is, the part where the number of iterations exceeds T is truncated, ||Δ T [y(t),w(t)]∣∣ ∞ Indicates Δ T The infinite norm of [y(t),w(t)] on the interval [0,T] is obtained through convolution operation. The relationship between the control error e(t) and the neural network learning error Δ[y(t),w(t)] is obtained. The upper bound estimation of the control error can provide a quantitative measure of the system performance, which helps to improve the stability and robustness of the system in the time domain and frequency domain. The sensitivity adaptive adjustment algorithm based on the neural network realizes the intelligent adaptation of the high-sensitivity human motion detection system by combining the neural network with the fiber optic sensing technology. The adaptability and stability of the algorithm can ensure the reliability of the system in various complex operating environments, which helps to provide an effective means for realizing accurate, intelligent and robust human motion monitoring.
[0085] See also Figure 1 ,Furthermore, the system monitoring and alarm module monitors the ,operating status of the entire system, including the connection status of the ,optical fiber sensor, the stability of the optical signal acquisition and the ,feedback results of body motion detection, and sets reasonable ,alarm thresholds, to achieve real-time monitoring of abnormal body ,movements and system failures, and triggers the alarm mechanism to notify ,the system users.
[0086] See also Figure 1 ,Furthermore, the user interface and notification module enables the ,user to monitor body motion information, adjust system parameters, and send alarm information to the ,user via notification method, ensuring that the user can learn of the ,relevant body motion monitoring results in a timely manner.
[0087] When used specifically, first, the fiber optic sensor array module constructs a comprehensive perception and detection system by laying out and configuring the fiber optic sensors. Then, the optical signal acquisition and processing module acquires the optical signals transmitted by the fiber optic sensors and performs efficient signal processing, including noise suppression, filtering, and data preprocessing, to ensure that the information extracted from the sensors is accurate and reliable, and to provide a clear data basis for subsequent analysis. Subsequently, the human body motion analysis module includes a human feature extraction unit and a human body motion detection unit. The human feature extraction unit extracts relevant human features from the optical signals collected by the fiber optic sensors, including motion patterns and physiological indicators, to provide rich information for subsequent body motion detection and analysis. The human body motion detection unit proposes a body motion detection algorithm based on deep learning for learning. It can also identify complex human movement patterns, including body movements in bed and the physiological characteristics of heart rate and respiratory rate, so as to accurately detect different body movements. Secondly, the sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on neural network to identify the relevant characteristics of human activities. According to the optical signals collected in real time, combined with neural network technology, it can dynamically learn and adapt to the human movement patterns in different environments, and realize the intelligent adjustment of the system for human movement detection. Finally, the system monitoring and alarm module monitors human movement information in real time, monitors the system operation status, and triggers alarms when abnormalities and emergencies are detected. The user interface and notification module realizes real-time interaction between the system and the user, and is responsible for sending notifications and alarm information to the user to ensure that the user can quickly learn about important events monitored by the system.
[0088] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A highly sensitive human body motion detection system based on optical fiber sensing technology, comprising an optical fiber sensing array module, an optical signal acquisition and processing module, a human body motion analysis module, a sensitivity adjustment module, and a system monitoring and alarm module, characterized in that: The fiber optic sensor array module is used to layout and configure fiber optic sensors to build an all-round perception detection system. The optical signal acquisition and processing module is used to obtain optical signals transmitted by the fiber optic sensors and perform efficient signal processing, including noise suppression, filtering and data preprocessing, to ensure that the information extracted from the sensor is accurate and reliable, and to provide a clear data basis for subsequent analysis. The human body motion analysis module includes a human body feature extraction unit and a human body motion detection unit. The human body feature extraction unit is used to extract relevant human body features from the optical signals collected by the fiber optic sensors, including movement patterns and physiological indicators, to provide rich information for subsequent body motion detection and analysis. The human body motion detection unit proposes a body motion detection algorithm based on deep learning for learning and identifying complex human body motion patterns, including body motion in bed and physiological characteristics of heart rate and respiratory rate, so as to achieve accurate detection of different body motions. The sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on a neural network for identifying relevant characteristics of human activities. According to the optical signals collected in real time, combined with the neural network technology, it dynamically learns and adapts to the human body motion patterns in different environments, so as to achieve intelligent adjustment of the system for human body motion detection. The system monitoring and alarm module is used to monitor human body motion information in real time, monitor the system operation status, and trigger an alarm when abnormalities and emergencies are detected. The sensitivity adaptive adjustment algorithm based on neural network is specifically as follows: First, a neural network model is established to realize the adaptive dynamic adjustment of the high-sensitivity human body motion detection system of optical fiber sensing technology. The neural network model is constructed by combining the radial basis function network with the feedforward neural network. The specific formula for mapping the input layer to the hidden layer of the radial basis function network is expressed as: Among them, φ i (t) represents the output of the i-th radial basis function, which represents the activation degree of the input signal x(t) for the radial basis function. i represents the index of the number of radial basis functions. x(t) represents the detection result from the human body motion analysis module as the input of the neural network. i represents the center of the hidden layer neuron, which is used to determine the position of each basis function in the input space, σ i The width parameter of the radial basis function determines the influence range of the basis function in the input space. exp represents the exponential function, and ||·|| represents the Euclidean distance. The specific formula for mapping the hidden layer to the output layer of the feedforward neural network is expressed as follows: Where L represents the output of the feedforward neural network, represents the approximation of the nonlinear part of the system after radial basis function and weight adjustment, f[y(t)] represents the nonlinear function of the function y(t) fed back to the input of the overall system, y(t) represents the final output function of the sensitivity adjustment module during the tth iteration, w(t) represents the weight function of the overall neural network during the tth iteration, and w i Represents the weight connecting the hidden layer and the output layer, which is used to adjust the influence of each radial basis function on the system output. n represents the total number of radial basis functions. At this time, the specific formula of the output of the adaptive dynamic control model based on the neural network is expressed as follows: Where k(t) represents the output signal of the adaptive dynamic control model based on the neural network, which is used to adjust the input signal of the sensitivity adjustment module to achieve adaptive dynamic adjustment, σ represents the damping ratio of the system, ρ represents the natural frequency of the system, and G t represents the gain parameter of the t-th iteration of the neural network model. By integrating the radial basis function network and the feedforward neural network, an adaptive dynamic control model based on the neural network is constructed to better adapt to the complex nonlinear relationship of the body motion detection system. Secondly, by adopting The online learning rules of the adjustment mechanism, Represents the speed of adjusting the weight relative to the target value, obtained through Lyapunov stability theory According to the real-time feedback control error e(t) of the system, the weight of the neural network is adjusted through e(t) so that the neural network approaches the nonlinear part. The specific formula for weight update of the adaptive dynamic control model based on the neural network is expressed as: Where Δw represents the rate of change of the weight function w(t) during the t-th iteration, Ψ represents a diagonal matrix, whose elements can be adjusted to change the sensitivity of the neural network to errors, K represents a diagonal positive definite matrix, which is used to introduce damping to prevent over-adjustment of the algorithm, and w i * Represents the expected target value of the weight of the radial basis function. Using Lyapunov stability theory, analyze the current stability conditions of the system and find the appropriate Considering the sensitivity of the system, adjust The size of enables the system to adapt to external changes faster while maintaining stability. The learning error equation of the adaptive dynamic control model based on the neural network is specifically expressed as: Δ[y(t),w(t)]=L f[y(t)] [y(t),w(t)]-f[y(t)] Among them, Δ[y(t),w(t)] represents the learning error of the neural network, that is, the difference between the approximate nonlinear part f[y(t)] of the neural network and the actual nonlinear part. Finally, a control error evaluation method is proposed to associate the neural network learning error with the control error for the analysis of algorithm performance. The control error of the system is evaluated by the convolution method, and the control error is associated with the neural network learning error. The feedback control error e(t) of the high-sensitivity human body motion detection system based on optical fiber sensing technology is specifically expressed as: Among them, g(t) represents the unit impulse response of the system, * represents the convolution operation, g(t-τ) represents the unit impulse response of the system to the instantaneous input reaction, Δ[y(τ),w(τ)] represents the learning error between the expected output and the actual output of the neural network, τ represents the integral variable, and the upper bound estimation formula for the control error is expressed as: Among them, ||e(t)|| ∞ represents the infinite norm of e(t) in the interval [0,T], T represents the total number of iterations of the adaptive dynamic control model based on neural network, Δ T [y(t),w(t)] represents the truncated part of Δ[y(t),w(t)] on the interval [0,T], that is, the part where the number of iterations exceeds T is truncated, ||Δ T [y(t),w(t)]∣∣ ∞ Indicates Δ T The infinite norm of [y(t),w(t)] on the interval [0,T].
2. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The optical fiber sensing array module captures tiny optical changes in the surrounding environment in real time by arranging a high-sensitivity optical fiber sensor array, including the selection of optical fiber sensors, layout design, and connection with the system, ensuring effective perception of optical signal changes caused by human body movement.
3. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The optical signal acquisition and processing module acquires a variety of optical signals obtained by the optical fiber sensor array module, and performs preprocessing operations on the acquired optical signals to remove noise and interference, ensuring the system's high-sensitivity detection of tiny physiological changes, thereby providing reliable data support for subsequent human body movement analysis.
4. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The human body motion analysis module includes a human feature extraction unit, which extracts key features from complex signals, identifies and extracts human body motion and physiological characteristics, including movement patterns and physiological indicators, and efficiently captures physiological signals related to human body motion, providing a representative and discriminative feature set for subsequent body motion analysis.
5. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The human body motion analysis module includes a human body motion detection unit. The human body motion detection unit proposes a body motion detection algorithm based on deep learning. By analyzing a large amount of real-time data and combining time series information and spatial features, it realizes the recognition of multiple body motion detections and distinguishes different human body motion patterns, ensuring the system's efficient monitoring and accurate classification of body motion changes.
6. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The calculation formula of the partial affinity field between key points is expressed as: in, Represents the vector of the partial affinity field at position t, representing the directional relationship between key points, x d1,q With x d2,q They represent the coordinates of feature information d1 and d2 in the qth person, respectively. ||·||2 represents the two-norm operation. When the condition t∈q is satisfied, t is judged to be on the limb of the qth person. Take vector Otherwise, the default value is 0.
7. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The sensitivity adjustment module proposes a sensitivity adaptive adjustment algorithm based on a neural network, uses a neural network model to learn the changing rules of body motion patterns, continuously optimizes model parameters and dynamically adjusts the perception sensitivity of the optical fiber sensor, ensuring high sensitivity to tiny changes in human body motion in different scenarios.
8. A highly sensitive human body motion detection system based on optical fiber sensing technology as claimed in claim 1, characterized in that: The system monitoring and alarm module monitors the operating status of the entire system, including the connection status of the optical fiber sensor, the stability of the optical signal acquisition, and the feedback results of body motion detection. By setting reasonable alarm thresholds, it realizes real-time monitoring of abnormal body movements and system failures, and triggers the alarm mechanism to notify system users.
