An indoor positioning method and system based on multi-sensor fusion

Abstract

The present invention discloses an indoor positioning method and system based on multi-sensor fusion, which relates to the field of indoor positioning technology, and includes the following steps: deploying Wi-Fi transceiver equipment and various sensors; detecting the range of the Wi-Fi CSI area to be measured; refining the detection of the sub-area by the PIR sensor; calculating the centroid positioning coordinates; accurately positioning by ultrasonic ranging; fusing the positioning results of the sensors. The technical method of the present application deploys sensors, collects CSI information, voltage signals of PIR sensors, and ultrasonic signals, refines the personnel detection in the sub-area by a dual dynamic threshold method, performs weighted positioning based on the signal strength of the PIR sensor, and calculates the relative position of the personnel in the area. Combined with the centimeter-level precise ranging of the ultrasonic sensor, the precise two-dimensional coordinates of the personnel are obtained by triangulation. Finally, the results of PIR centroid positioning and ultrasonic ranging are fused by the Kalman filter algorithm to achieve real-time updating and dynamic tracking, and output high-precision personnel location information.

Description

Technical Field

[0001] The present invention relates to the technical field of indoor positioning, and particularly to an indoor positioning method and system based on multi-sensor fusion. Background Art

[0002] With the rapid development of fields such as smart home, security monitoring, and medical care, the demand for indoor personnel positioning is increasing. Indoor positioning technology refers to the detection and positioning of moving personnel inside a building. Compared with outdoor GPS positioning, the indoor environment is more complex, and signal propagation is interfered by obstacles such as walls. The positioning accuracy, stability, and applicability of traditional positioning means such as Bluetooth, infrared, or ultrasonic waves are all limited indoors. Therefore, realizing an indoor positioning technology with low cost, high precision, and not relying on portable terminal devices and tags, which is convenient for administrators to obtain personnel position information in real time, has become an important research direction at present.

[0003] Currently, researchers have proposed various methods for indoor positioning, especially in sensor fusion, to ensure positioning accuracy. For example, "An Indoor Positioning Method and System Based on the Hybrid of WiFi and Ultrasonic Waves". This method first obtains WiFi fingerprints and constructs a fingerprint database; then, uses a clustering algorithm to perform clustering analysis on the collected WiFi fingerprint data; finally, locks the target position within a single sub-region and combines ultrasonic ranging for fine positioning of the local region. This method relies on electronic tags and requires the construction of a WiFi fingerprint database. When the device layout changes, the fingerprint database needs to be updated continuously, increasing the maintenance cost, and relying only on ultrasonic waves is easily affected by the reflection interference of physical obstacles, affecting the positioning accuracy. Therefore, using a fusion algorithm can not only make up for the weaknesses of a single sensor but also optimize the positioning accuracy through dynamic adjustment. Summary of the Invention

[0004] The technical solution of the present invention to solve the above technical problems is to provide an indoor positioning method based on multi-sensor fusion, including the following steps:

[0005] Step 1, deploy Wi-Fi transceiver devices and various sensors: Deploy Wi-Fi signal transceiver devices, PIR sensors, and ultrasonic sensors in the area to be positioned. Among them, the Wi-Fi signal transceiver device is used to provide overall wireless signal coverage, the PIR sensor is used to detect human infrared radiation to sense personnel activities, and the ultrasonic sensor is used for ranging;

[0006] Step 2: Wi-Fi CSI test area range detection: collect CSI information, pre-process the CSI signal, extract the main change components of the CSI data, and perform maximum value normalization. Observe the frequency characteristics of signal amplitude changes through short-time Fourier transform, extract features of phase difference, build a training data set and train the SVM classification model to distinguish the activity status of people in the environment.

[0007] Step 3, detailed detection of the sub-area by the PIR sensor: according to the voltage signal of the PIR sensor, the voltage peak and climbing slope characteristics are extracted, and the double dynamic threshold method is used to judge the movement state of the person and determine the person's activities in the sub-area;

[0008] Step 4: Calculate the centroid coordinates: Based on the signal strength of multiple PIR sensors, calculate the relative position (x PIR ,y PIR );

[0009] Step 5: Ultrasonic ranging precise positioning: Using the known coordinates of the ultrasonic sensor and the measured distance value, the precise position of the person (x ultra ,y ultra );

[0010] Step 6, fusion sensor positioning results: obtain PIR centroid positioning coordinates and ultrasonic ranging coordinates, define state variables and observation vectors, perform state prediction and update through Kalman filtering, calculate Kalman gain, combine observation results with predicted values, correct the current coordinates and speed of the target, and update the error covariance matrix, and finally output the fused target position (x k ,y k ).

[0011] Furthermore, in step 1, a modular splicing method is used to deploy sensors and divide sub-areas so as to overlap and cover the detection areas of the sensors.

[0012] Furthermore, in step 2, the CSI information includes:

[0013] H(f i )=|H(f i )|e jθ(fi) ;

[0014] Among them, H(f i ) is the CSI value of the i-th subcarrier; |H(f i )| is the amplitude of CSI; θ(f i ) is the phase of CSI; f i is the frequency of the ith subcarrier;

[0015] Preprocess the CSI signal: Use the PCA (Principal Component Analysis) method to extract the main changing components in the CSI information. Decompose the data into multiple principal components through PCA. The formula is as follows:

[0016] X = UΣV T ;

[0017] Among them, X is the CSI data matrix; U and V are the eigenvector matrices respectively; T represents the transpose operation of the eigenvector matrix; Σ is the singular value matrix, representing the principal components of the signal;

[0018] Perform maximum normalization on the CSI amplitude:

[0019]

[0020] Among them, |H norm (f i ,t)| is the normalized CSI amplitude value, representing the result after normalization of the CSI amplitude at frequency f i and time t; |H(f i ,t)| is the original CSI amplitude value, representing the unnormalized CSI amplitude information obtained from the Wi-Fi device at frequency f i and time t; max(|H(f i ,t)|) is the maximum value of the CSI amplitude at all frequencies and time points of the current CSI data, serving as the benchmark for normalization;

[0021] Obtain the frequency characteristics of the signal amplitude change at different times through the Short-Time Fourier Transform (STFT); Convert the CSI amplitude signal from the time domain |H norm (f i ,t)| to the frequency domain S(f,t), using the formula:

[0022]

[0023] Among them, S(f,t) represents the CSI time-frequency characteristic value at frequency f and time t, used to reflect the change of the CSI amplitude with frequency; e -j2πfτ is the kernel function of the Fourier transform, where j is the imaginary unit, π is a mathematical constant, τ is the integration variable, and dτ represents the differential element of the integration;

[0024] Extract the phase difference characteristics;

[0025] Construct a training dataset using the extracted frequency characteristics of amplitude change and phase difference characteristics for training the classification model; Use the extracted CSI characteristics to train the SVM classifier; During the training process, use the labeled CSI characteristic data to distinguish the situations of "presence of human activity" and "absence of human activity"; The SVM classifier model is as follows:

[0026] f(x) = w T φ(x) + b;

[0027] Where f(x) represents the predicted output of the input x; x is the input feature, i.e., the amplitude change and phase difference features; w is the weight of the SVM model; φ(x) represents the feature vector obtained after the input x passes through the feature mapping function φ; b is the bias.

[0028] Furthermore, in step 3,

[0029] Set the voltage peak threshold: Dynamically adjust the detection threshold V of the voltage peak according to the environmental noise level and typical human activity signals th,peak ; When the voltage peak of the sensor exceeds the set threshold, it indicates that there is human activity in this area; the voltage peak determination condition is as follows:

[0030] V peak > V th,peak ;

[0031] Where V peak is the voltage peak output by the PIR sensor; V th,peak is the voltage peak threshold dynamically adjusted according to environmental conditions;

[0032] Set the ramp slope threshold: By sampling the voltage signal change rate of the sensor, calculate the ramp slope of the voltage:

[0033]

[0034] Where dV is the voltage increment; dt is the time increment;

[0035] Dynamically set the slope threshold α according to different exercise intensities th,slope ; When the slope of the voltage change exceeds this threshold, it indicates that there is a strong exercise state; the ramp slope determination condition is as follows:

[0036] α slope > α th,slope ;

[0037] Where α slope is the currently measured voltage ramp slope; α th,slope is the slope threshold dynamically adjusted according to exercise intensity;

[0038] Judge different types of human activities by combining these two dynamic thresholds of voltage peak and ramp slope; when the following conditions are met, it is determined that there are people moving in the sub-region:

[0039]

[0040] Further, in step 4, when a person enters the detection area of the PIR sensor, the PIR sensor outputs a signal strength S related to the distance of the person i , arrange n PIR sensors in a sub-region, and the sensor coordinates are (x i , y i ), where i = 1, 2,..., n;

[0041] Taking the signal strength of each PIR sensor as the weight, based on the spatial position of the sensor, calculate the relative position of the person through weighted average;

[0042] Use the weighted average method to estimate the centroid coordinates:

[0043]

[0044] where w i is the signal strength of the i-th PIR sensor; x PIR and y PIR are the centroid coordinates of the person in the sub-region, that is, the estimated relative position; x i and y i are the coordinates of the i-th sensor; ∑w i x i and ∑w i y i represent the results of summing the products of the coordinates x i and y i of each point and its weight w i ; ∑w i represents the sum of all weights and is a normalization factor;

[0045] The relative position (x PIR , y PIR ) of the person in the sub-region is obtained.

[0046] Further, in step 5, for the sub-region where the person is located, the coordinates of each ultrasonic sensor are denoted as (x i , y i ), where i = 1, 2, 3; use the triangulation method to calculate the precise coordinates of the person through the distance measurement values of multiple ultrasonic sensors; assume the actual position of the person is (x ultra , y ultra ), and solve the coordinates (x ultra , y ultra ) by combining the following equations:

[0047] The distance d1 obtained from ultrasonic sensor 1:

[0048]

[0049] Distance d2 obtained from the ultrasonic sensor 2:

[0050]

[0051] Distance d3 obtained from the ultrasonic sensor 3:

[0052]

[0053] Solve it by numerical methods (such as Newton iteration method or least squares method); the exact position (x ultra , y ultra ) of the person within the sub-region is obtained.

[0054] Furthermore, in step 6,

[0055] Define the state variable X k of the target as:

[0056]

[0057] where x k and y k are the two-dimensional coordinate positions of the target at time k; v x and v y are the velocities of the target in the x and y directions.

[0058] Define the observation vector as Z k :

[0059]

[0060] where and are the relative positions provided by the PIR centroid positioning at time k; and are the exact positions provided by the ultrasonic ranging at time k.

[0061] The prediction model of the Kalman filter calculates the target position and velocity at the current moment based on the state X k-1 at the previous moment;

[0062] Define the state transition equation as:

[0063] X k|k-1 = F·X k-1 + B·U k ;

[0064] where X k|k-1 is the state at the current moment predicted based on the state at the previous moment at time k; B is the control matrix, which can be ignored if there is no control input; U kis the control input vector, which can be ignored if there is no control input; F is the state transition matrix, assuming that the position changes smoothly over time:

[0065]

[0066] Calculate the Kalman gain: The Kalman gain K k is defined as:

[0067] K k = P k|k-1 H T (HP k|k-1 H T + R) -1 ;

[0068] where P k|k-1 is the predicted state error covariance matrix, representing the uncertainty of the predicted value at the current moment; R is the measurement noise covariance matrix; H is the observation matrix:

[0069]

[0070] State update: After calculating the Kalman gain, the state is updated by combining the observation result Z k with the predicted value X k|k-1 . The state update equation is:

[0071] X k = X k|k-1 + K k ·(Z k - H·X k|k-1 );

[0072] where Z k is the measurement result of the PIR and ultrasonic sensors; H·X k|k-1 is the observed value obtained from the prediction;

[0073] Update the state error covariance matrix P k . The error covariance update equation is:

[0074] P k = (I - K k H)P k|k-1 ;

[0075] where I is the identity matrix;

[0076] Output the fused target position (x k , y k ).

[0077] In order to solve the above technical problems, the present invention further proposes an indoor positioning system based on multi-sensor fusion, which is used to execute the indoor positioning method based on multi-sensor fusion as described above, comprising:

[0078] Divide the sub-areas and deploy sensor modules. Set the area of ​​the positioning area to S, and divide the area into n sub-areas according to the size of the area to be positioned.

[0079] The data acquisition module includes the Wi-Fi access point and the receiving device to collect the channel status information of the Wi-Fi signal, including the amplitude and phase changes of each subcarrier; the PIR sensor collects the infrared radiation intensity signal and outputs the voltage signal that changes with the movement of the person; the ultrasonic sensor transmits and receives the ultrasonic signal, measures the time difference of the reflected signal, and collects the signal reflecting the distance between the person and the sensor;

[0080] Data processing module, used to process the channel status information of Wi-Fi signals, voltage signals of PIR sensors, and ultrasonic signals;

[0081] The fusion algorithm module uses Kalman filtering to fuse the data of the PIR sensor and the ultrasonic sensor, defines the state variables and observation vectors, predicts and updates the state through Kalman filtering, calculates the Kalman gain, combines the observation results with the predicted values, corrects the current coordinates and speed of the target, and updates the error covariance matrix, and finally outputs the fused target position.

[0082] Compared with the prior art, this application has the following beneficial effects:

[0083] 1. The present invention provides an expandable multi-sensor joint layout strategy for indoor human body positioning. The modular layout strategy can be flexibly adjusted according to the actual situation of the scene to be tested. By overlapping the sensor detection area, the detection blind area is avoided, and the detection accuracy and reliability are improved. When the detection of a single sensor is biased or the signal is weakened, the system can still perform compensation detection through other sensors.

[0084] 2. The present invention provides a multi-sensor indoor positioning method from coarse to fine. The method realizes accurate positioning from coarse to fine, from large range to small area. It has good adaptability to environmental changes, ensures accurate tracking of moving targets, and can continuously correct the position of personnel. Compared with the traditional single Wi-Fi or PIR positioning method, this method can improve the accuracy of personnel positioning to within 5 cm.

[0085] 3. The present invention provides a multi-sensor indoor positioning system, which systematizes and modularizes the indoor positioning method, enabling it to be extended to practical applications in different scenarios. Its efficient layout strategy, precise positioning method, and reliable fusion algorithm enable the system to provide excellent detection and positioning services in different scenarios, providing strong support for intelligent management and security monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the structures shown in these drawings.

[0087] Figure 1 It is a flowchart of the steps of an indoor positioning method based on multi-sensor fusion according to the present invention;

[0088] Figure 2 It is a schematic diagram of the sensor layout of the indoor positioning method based on multi-sensor fusion according to the present invention;

[0089] Figure 3 It is a multi-sensor fusion flowchart of the indoor positioning method based on multi-sensor fusion according to the present invention;

[0090] Figure 4 It is a structural block diagram of the indoor positioning system based on multi-sensor fusion provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0091] The present invention proposes an indoor positioning method based on multi-sensor fusion, aiming to design an indoor personnel detection and positioning method combining Wi-Fi CSI, PIR sensors, and ultrasonic ranging.

[0092] The following will illustrate an indoor positioning method based on multi-sensor fusion proposed by the present invention in specific embodiments:

[0093] Embodiment 1:

[0094] In the technical solution of this embodiment, as Figure 1 shown, an indoor positioning method based on multi-sensor fusion includes the following steps:

[0095] Step 1, deploy Wi-Fi transceiver equipment and various sensors: deploy Wi-Fi signal transceiver equipment, PIR sensors and ultrasonic sensors in the area to be located, where the Wi-Fi signal transceiver equipment is used to provide overall wireless signal coverage, the PIR sensor is used to detect human infrared radiation to sense human activities, and the ultrasonic sensor is used to measure distance;

[0096] Step 2: Wi-Fi CSI test area range detection: collect CSI information, pre-process the CSI signal, extract the main change components of the CSI data, and perform maximum value normalization. Observe the frequency characteristics of signal amplitude changes through short-time Fourier transform, extract features of phase difference, build a training data set and train the SVM classification model to distinguish the activity status of people in the environment.

[0097] Step 3, detailed detection of the sub-area by the PIR sensor: according to the voltage signal of the PIR sensor, the voltage peak and climbing slope characteristics are extracted, and the double dynamic threshold method is used to judge the movement state of the person and determine the person's activities in the sub-area;

[0098] Step 4: Calculate the centroid coordinates: Based on the signal strength of multiple PIR sensors, calculate the relative position (x PIR ,y PIR );

[0099] Step 5: Ultrasonic ranging precise positioning: Using the known coordinates of the ultrasonic sensor and the measured distance value, the precise position of the person (x ultra ,y ultra );

[0100] Step 6, fusion sensor positioning results: obtain PIR centroid positioning coordinates and ultrasonic ranging coordinates, define state variables and observation vectors, perform state prediction and update through Kalman filtering, calculate Kalman gain, combine observation results with predicted values, correct the current coordinates and speed of the target, and update the error covariance matrix, and finally output the fused target position (x k ,y k ).

[0101] Furthermore, in step 1, a modular splicing method is used to deploy sensors and divide sub-areas so as to overlap and cover the detection areas of the sensors.

[0102] Furthermore, in step 2, the CSI information includes:

[0103]

[0104] Among them, H(f i ) is the CSI value of the i-th subcarrier; |H(fi ) | is the amplitude of the CSI; θ(f i ) is the phase of the CSI; f i is the frequency of the i-th subcarrier;

[0105] Preprocess the CSI signal: Use the PCA (Principal Component Analysis) method to extract the main varying components in the CSI information. Decompose the data into multiple principal components through PCA, and the formula is as follows:

[0106] X = UΣV T ;

[0107] Where X is the CSI data matrix; U and V are the eigenvector matrices respectively; T represents the transpose operation of the eigenvector matrix; Σ is the singular value matrix, representing the principal components of the signal;

[0108] Perform maximum normalization on the CSI amplitude:

[0109]

[0110] Where, |H norm (f i ,t)| is the normalized CSI amplitude value, representing the result after normalization of the CSI amplitude at frequency f i and time t; |H(f i ,t)| is the original CSI amplitude value, representing the unnormalized CSI amplitude information obtained from the Wi-Fi device at frequency f i and time t; max(|H(f i ,t)|) is the maximum value of the CSI amplitude at all frequencies and time points of the current CSI data, serving as the benchmark for normalization;

[0111] Obtain the frequency characteristics of the signal amplitude change at different moments through the Short-Time Fourier Transform (STFT); Convert the CSI amplitude signal from the time domain |H norm (f i ,t)| to the frequency domain S(f,t), using the formula:

[0112]

[0113] Where, S(f,t) represents the CSI time-frequency characteristic value at frequency f and time t, which is used to reflect the change of the CSI amplitude with frequency; e -j2πfτ is the kernel function of the Fourier transform, where j is the imaginary unit, π is a mathematical constant, τ is the integration variable, and dτ represents the differential element of the integration;

[0114] Extract the phase difference characteristics;

[0115] Construct a training dataset using the extracted amplitude change frequency features and phase difference features for training a classification model; train an SVM classifier using the extracted CSI features; during the training process, use the CSI feature data with annotations to distinguish between the situations of "human activity" and "no human activity"; the SVM classifier model is as follows:

[0116] f(x) = w T φ(x) + b;

[0117] Among them, f(x) represents the predicted output of the input x; x is the input feature, that is, the amplitude change and phase difference features; w is the weight of the SVM model; φ(x) represents the feature vector obtained after the input x passes through the feature mapping function φ; b is the bias.

[0118] Furthermore, in step 3,

[0119] Set the voltage peak threshold: Dynamically adjust the detection threshold V of the voltage peak according to the environmental noise level and typical human activity signals th,peak ; When the voltage peak of the sensor exceeds the set threshold, it indicates that there is human activity in this area; the voltage peak determination condition is as follows:

[0120] V peak > V th,peak ;

[0121] Among them, V peak is the voltage peak output by the PIR sensor; V th,peak is the voltage peak threshold dynamically adjusted according to environmental conditions;

[0122] Set the ramp slope threshold: By sampling the voltage signal change rate of the sensor, calculate the ramp slope of the voltage:

[0123]

[0124] Among them, dV is the voltage increment; dt is the time increment;

[0125] Dynamically set the slope threshold α according to different exercise intensities th,slope ; When the slope of the voltage change exceeds this threshold, it indicates a strong exercise state; the ramp slope determination condition is as follows:

[0126] α slope > α th,slope ;

[0127] Among them, α slope is the currently measured voltage ramp slope; α th,slope is the slope threshold dynamically adjusted according to the exercise intensity;

[0128] Determine different types of personnel activities by combining two dynamic thresholds, namely the voltage peak and the ramp slope; when the following conditions are met, it is determined that there are personnel activities in the sub-region:

[0129]

[0130] Furthermore, in step 4, when a person enters the detection area of the PIR sensor, the PIR sensor outputs a signal strength S related to the distance of the person i , arrange n PIR sensors in a sub-region, and the sensor coordinates are (x i , y i ), where i = 1, 2,..., n;

[0131] Take the signal strength of each PIR sensor as the weight, and based on the spatial position of the sensor, calculate the relative position of the person through weighted average;

[0132] Use the weighted average method to estimate the centroid coordinates:

[0133]

[0134] where, w i is the signal strength of the i-th PIR sensor; x PIR and y PIR are the centroid coordinates of the person in the sub-region, that is, the estimated relative position; x i and y i are the coordinates of the i-th sensor; ∑w i x i and ∑w i y i represent the results of summing the products of the coordinates x i and y i of each point and its weight w i ; ∑w i represents the sum of all weights and is the normalization factor;

[0135] The relative position (x PIR , y PIR ) of the person in the sub-region is obtained.

[0136] Furthermore, in step 5, for the sub-region where the person is located, the coordinates of each ultrasonic sensor are denoted as (x i , y i ), where i = 1, 2, 3; use the triangulation method to calculate the exact coordinates of the person through the distance measurement values of multiple ultrasonic sensors; let the actual position of the person be (x ultra , y ultra ), and solve the coordinates (x ultra , yultra ):

[0137] The distance d1 obtained from the ultrasonic sensor 1:

[0138]

[0139] The distance d2 obtained from the ultrasonic sensor 2:

[0140]

[0141] The distance d3 obtained from the ultrasonic sensor 3:

[0142]

[0143] Solve it by numerical methods (such as Newton iteration method or least squares method); the exact position (x ultra , y ultra ) of the person within the sub-region is obtained.

[0144] Furthermore, in step 6,

[0145] Define the state variable X of the target k as:

[0146]

[0147] where x k and y k are the two-dimensional coordinate positions of the target at time k; v x and v y are the speeds of the target in the x and y directions.

[0148] Define the observation vector as Z k :

[0149]

[0150] where, and are the relative positions provided by the PIR centroid positioning at time k; and are the exact positions provided by the ultrasonic ranging at time k.

[0151] The prediction model of the Kalman filter calculates the target position and speed at the current moment according to the state X at the previous moment k-1 ;

[0152] Define the state transition equation as:

[0153] X k|k-1 = F·X k-1 + B·U k ;

[0154] Among them, X k|k-1 is the state at the current moment predicted based on the state at the previous moment at time k; B is the control matrix, which can be ignored if there is no control input; U k is the control input vector, which can be ignored if there is no control input; F is the state transition matrix, assuming that the position changes smoothly over time:

[0155]

[0156] Calculate the Kalman gain: The Kalman gain K k is defined as:

[0157] K k = P k|k-1 H T (HP k|k-1 H T + R) -1 ;

[0158] Among them, P k|k-1 is the predicted state error covariance matrix, indicating the uncertainty of the predicted value at the current moment; R is the measurement noise covariance matrix; H is the observation matrix:

[0159]

[0160] State update: After calculating the Kalman gain, the state is updated by combining the observation result Z k with the predicted value X k|k-1 . The state update equation is:

[0161] X k = X k|k-1 + K k ·(Z k - H·X k|k-1 )

[0162] Among them, Z k is the measurement result of the PIR and ultrasonic sensors; H·X k|k-1 is the observed value obtained based on the prediction;

[0163] Update the state error covariance matrix P k , and the error covariance update equation is:

[0164] P k = (I - K k H)P k|k-1 ;

[0165] Among them, I is the identity matrix;

[0166] Output the fused target position (x k , y k ).

[0167] To solve the above technical problems, the present invention also proposes an indoor positioning system based on multi-sensor fusion for performing the indoor positioning method based on multi-sensor fusion as described above, including:

[0168] Divide the sub-regions and deploy the sensor modules. Set the area of the positioning region as S, and divide this region into n sub-regions according to the size of the area to be positioned;

[0169] Data acquisition module, including Wi-Fi access points and receiving devices to collect the channel state information of Wi-Fi signals, including the amplitude and phase changes of each sub-carrier; PIR sensors to collect infrared radiation intensity signals and output voltage signals that change with the movement of people; ultrasonic sensors to measure the time difference of the reflected signals by transmitting and receiving ultrasonic signals, and collect signals reflecting the distance between people and sensors;

[0170] Data processing module, used to process the channel state information of Wi-Fi signals, the voltage signals of PIR sensors, and ultrasonic signals;

[0171] Fusion algorithm module, using Kalman filtering to fuse the data of PIR sensors and ultrasonic sensors, define state variables and observation vectors, perform state prediction and update through Kalman filtering, calculate the Kalman gain, combine the observation results with the predicted values, correct the current coordinates and speeds of the target, and update the error covariance matrix, and finally output the fused target position.

[0172] Embodiment 2:

[0173] An indoor positioning method based on multi-sensor fusion includes the following steps:

[0174] Step 1, deploy Wi-Fi transceiver devices and each sensor:

[0175] First, it is necessary to deploy Wi-Fi signal transceiver devices, PIR sensors, and ultrasonic sensors in the area to be positioned. The deployment method of the sensors and the division of sub-regions are as Figure 2 shown. The Wi-Fi signal transceiver device is installed in the center of the area to be measured to provide overall wireless signal coverage. The black solid circles in the figure represent the positions of PIR sensors, and the dashed hollow circles represent the effective detection areas of PIR sensors. Since the detection radius of PIR sensors is affected by the installation height, in this embodiment, based on the typical indoor installation height and the models of common PIR sensors, the radius of this detection area is determined. The gray squares in the figure represent the installation positions of ultrasonic sensors. S1, S2, S3, and S4 represent four squares as the divided sub-regions, and the side length of each sub-region is equal to the detection radius of the PIR sensor. The black bold square represents a detection unit, which is composed of four sub-regions spliced together.

[0176] The basic detection unit adopts modular splicing, so that the detection area of ​​the sensor can be reasonably planned through overlapping coverage. This layout method can flexibly adjust the splicing according to the actual indoor shape and area of ​​the scene to be tested. At the same time, each additional basic detection unit can not only effectively expand the detection range, but also avoid adding additional sensors, thereby improving the utilization rate of each sensor. In addition, this method significantly reduces hardware costs.

[0177] Step 2: Wi-Fi CSI test area range detection:

[0178] Based on the layout of step 1, first, the Wi-Fi AP is used as the signal transmitter, and the receiving device collects CSI information in real time through the wireless network card to obtain the amplitude and phase of each subcarrier. CSI information is obtained from multiple subcarriers of the wireless device. For each subcarrier, CSI information includes:

[0179]

[0180] Among them, H(f i ) is the CSI value of the i-th subcarrier; |H(f i )| is the amplitude of CSI; θ(f i ) is the phase of CSI; f i is the frequency of the ith subcarrier.

[0181] Secondly, the CSI signal is preprocessed: the PCA principal component analysis method is used to extract the main change components in the CSI data, remove irrelevant noise components, decompose the data into multiple principal components through PCA, and select the first few principal components that can explain most of the signal changes. The formula is as follows:

[0182] X=UΣV T ;

[0183] Where X is the CSI data matrix; U and V are eigenvector matrices respectively; T represents the transpose operation of the eigenvector matrix; Σ is the singular value matrix, representing the principal components of the signal;

[0184] Third, perform maximum normalization on the CSI amplitude:

[0185]

[0186] Fourth, through the short-time Fourier transform STFT, the frequency characteristics of the signal amplitude changes at different times are observed. The frequency domain representation is achieved by converting the CSI amplitude signal from the time domain |H norm (f i ,t)|convert to the frequency domain S(f,t), using the formula:

[0187]

[0188] Among them, S(f,t) represents the CSI time-frequency eigenvalue at frequency f and time t, which is used to reflect the variation of the CSI amplitude with frequency.

[0189] Fifth, extract features from the phase difference to eliminate systematic noise.

[0190] Sixth, construct a training dataset using the amplitude change frequency features and phase difference features extracted in the previous steps for training a classification model. Each piece of data contains the amplitude change frequency features and phase difference features. Use the extracted CSI features to train an SVM classifier. During the training process, use the labeled CSI feature data to distinguish between the situations of "human activity" and "no human activity". The SVM classifier model is as follows:

[0191] f(x) = w T φ(x) + b;

[0192] Among them, f(x) represents the predicted output of the input x; x is the input feature, that is, the amplitude change and phase difference features; w is the weight of the SVM model; φ(x) represents the feature vector obtained after the input x passes through the feature mapping function φ; b is the bias.

[0193] The trained SVM model can classify human activities in the environment based on the CSI features extracted in real time. If the CSI features exceed a certain threshold or show significant amplitude and phase changes, the SVM model will output a judgment of "human activity".

[0194] Step 3, refined detection of sub-regions by PIR sensors:

[0195] The PIR sensor generates a voltage signal by sensing infrared radiation. The movement of the human body will cause changes in the voltage output of the PIR sensor. By collecting the voltage output of the PIR sensor, processing it, and extracting the features reflecting the intensity of human activity. Two key features of the voltage change are the voltage peak and the ramp slope.

[0196] To accurately detect the human motion states of different intensities, a dual dynamic threshold method is adopted. This method ensures that the sensor can adapt to different motion states and identify minor and intense motions by setting two dynamic thresholds, namely the voltage peak threshold and the ramp slope threshold.

[0197] Setting of the voltage peak threshold: Dynamically adjust the detection threshold V of the voltage peak according to the environmental noise level and typical human activity signals. th,peak When the voltage peak of the sensor exceeds the set threshold, it indicates that there is human activity in this area. The voltage peak judgment condition is as follows:

[0198] V peak >V th,peak ;

[0199] Among them, V peak is the peak voltage output by the PIR sensor; V th,peak is the peak voltage threshold dynamically adjusted according to environmental conditions.

[0200] Ramp slope threshold setting: By sampling the rate of change of the voltage signal of the sensor, calculate the ramp slope of the voltage:

[0201]

[0202] Among them, dV is the voltage increment; dt is the time increment.

[0203] Dynamically set the slope threshold α according to different exercise intensities th,slope . When the slope of the voltage change exceeds this threshold, it indicates a strong exercise state. The ramp slope determination condition is as follows:

[0204] α slope >α th,slope ;

[0205] Among them, α slope is the currently measured voltage ramp slope; α th,slope is the slope threshold dynamically adjusted according to exercise intensity.

[0206] Judge different types of personnel activities by combining these two dynamic thresholds of peak voltage and ramp slope. When the following conditions are met, it is determined that there are personnel activities in the sub-region:

[0207]

[0208] Step 4, calculate the centroid positioning coordinates:

[0209] When a person enters the detection area of the PIR sensor, the PIR sensor will output a signal intensity S related to the distance of the person i , usually the voltage output of the sensor. The closer the distance, the stronger the signal intensity, and the farther the distance, the weaker the signal intensity. Arrange n PIR sensors in a sub-region, and the sensor coordinates are (x i , y i ), where i = 1, 2,..., n. The detection areas of these sensors have a certain overlap and can detect personnel activities simultaneously.

[0210] Taking the signal strength of each PIR sensor as the weight, based on the spatial positions of these sensors, the relative position of the person is calculated through weighted averaging. Specifically, the greater the sensor signal strength, the closer the person is to the sensor, so the coordinates of this sensor contribute more to the final centroid position.

[0211] Estimate the centroid coordinates using the weighted averaging method:

[0212]

[0213] where w i is the signal strength of the i-th PIR sensor; x PIR and y PIR are the centroid coordinates of the person in the sub-region, that is, the estimated relative position; x i and y i are the coordinates of the i-th sensor; ∑w i x i and ∑w i y i represent the results of summing the products of the coordinates x i and y i of each point with their weight w i ; ∑w i represents the sum of all weights and is the normalization factor.

[0214] Through weighted averaging, the relative position (x PIR , y PIR ) of the person in the sub-region is obtained.

[0215] Step 5, Ultrasonic ranging for precise positioning:

[0216] For the sub-region where the person is located, the known coordinates of each ultrasonic sensor are denoted as (x i , y i ), where i = 1, 2, 3. The coordinates of the ultrasonic sensors are fixed and are usually pre-measured and stored when installing the sensors. Using the triangulation method, through the distance measurement values of multiple ultrasonic sensors, the precise coordinates of the person are calculated. Let the actual position of the person be (x ultra , y ultra ), and solve for the coordinates (x ultra , y ultra ) by combining the following system of equations:

[0217] The distance d1 obtained from ultrasonic sensor 1:

[0218]

[0219] The distance d2 obtained from ultrasonic sensor 2:

[0220]

[0221] The distance d3 obtained from the ultrasonic sensor 3:

[0222]

[0223] Since this is a system of non-linear equations, it is usually necessary to solve it by numerical methods (such as Newton's iteration method or least squares method).

[0224] By solving this system of equations, the system obtains the exact position (x ultra , y ultra ) of the person within the sub-region.

[0225] Step 6, fuse the sensor positioning results:

[0226] As Figure 3 shown, first, obtain the PIR centroid positioning coordinates obtained in Step 4 and the ultrasonic ranging coordinates obtained in Step 5, and ensure that the output data of the PIR and ultrasonic sensors have the same time stamp to ensure that the measurement results are based on the position of the person at the same moment. To this end, time stamps are added to the output data of each sensor to ensure data synchronization.

[0227] Second, define the state variables and the observation vector. The state variables are the core of the Kalman filter, which reflects the true motion state of the target, while the observation vector is the measurement value from the sensor used to correct the state prediction. The two need to cooperate with each other to complete the precise positioning of the target.

[0228] Define the state variables of the target to describe the motion state of the target, which usually includes the position and velocity, representing the motion state of the target. Define the state vector X k as:

[0229]

[0230] Define the observation vector Z k as:

[0231]

[0232] where, and are the relative positions provided by the PIR centroid positioning at time k; and are the exact positions provided by the ultrasonic ranging at time k.

[0233] Third, the prediction model. According to the state of the previous moment, estimate the target position and velocity at the current moment. To obtain a preliminary estimate at the current moment (without combining the actual measurement values).

[0234] Through the prediction model of Kalman filtering, the system predicts the position and velocity of the target at the current moment based on the previous state estimate. The prediction process assumes that the target motion is stationary (constant velocity motion model), and the position and velocity of the target are predicted through a linear state transition matrix:

[0235] Define the state transition equation as:

[0236] X k|k-1 = F·X k-1 + B·U k ;

[0237] Assume that the position changes smoothly over time:

[0238]

[0239] At this time, the predicted target position X k|k-1 is based on the estimated value of the previous state and does not incorporate the measurement results.

[0240] Fourth, calculate the Kalman gain. The state prediction is corrected by the measurement results of the PIR and ultrasonic sensors. The Kalman gain is used to balance the weight distribution between the predicted value and the measured value. The larger the Kalman gain, the more reliable the measurement data of the sensor, and thus the greater the impact of the measured value on the final positioning result.

[0241] The Kalman gain is defined as:

[0242] K k = P k|k-1 H T (HP k|k-1 H T + R) -1 ;

[0243] Where P k|k-1 is the predicted state error covariance matrix, representing the uncertainty of the predicted value at the current moment; R is the measurement noise covariance matrix, reflecting the uncertainty in the measurement results of the PIR and ultrasonic sensors; H is the observation matrix, describing how the measurement is mapped to the state space:

[0244]

[0245] Fifth, state update. After calculating the Kalman gain, the system combines the observation result Z k with the predicted value X k|k-1 to perform a state update and correct the current coordinates and velocity of the target. The final state Z k is the fused accurate position.

[0246] The state update equation is:

[0247] Xk = X k|k-1 + K k ·(Z k - H·X k|k-1 );

[0248] Wherein, Z k is the measurement result of the PIR and ultrasonic sensors; H·X k|k-1 is the observed value obtained according to the prediction.

[0249] Fifth, update the error covariance. The system updates the state error covariance matrix P k , updates the system's estimate of uncertainty, and ensures that the prediction for the next time step is more accurate. The smaller the covariance matrix, the higher the confidence level of the system in the prediction.

[0250] The error covariance update equation is:

[0251] P k = (I - K k H)P k|k-1 ;

[0252] This step is used to adjust the system's estimate of uncertainty in the next prediction, ensuring that the Kalman filter can continuously optimize the positioning accuracy in a dynamically changing environment.

[0253] Finally, output the fused accurate position. After the state prediction and update of the Kalman filter, the system finally outputs the accurate position (x k , y k ) of the person. This coordinate adapts to various changes in the dynamic environment by combining the rough information of PIR centroid positioning with the accurate result of ultrasonic ranging, and can provide high-precision and real-time updated person position information.

[0254] Embodiment 3:

[0255] An indoor positioning system based on multi-sensor fusion, as Figure 4 shown, includes dividing sub-regions and deploying sensor modules, data acquisition modules, data processing modules, fusion algorithm modules, and coordinate output.

[0256] In dividing sub-regions and deploying sensor modules, assuming that the area of the positioning region is S, the region is divided into n sub-regions according to the size of the area to be located. In this embodiment, taking the example of dividing the region to be measured into 4 sub-regions, the present invention will be described in detail. This layout method can be extended and spliced for application to larger regions to be located.

[0257] In the data acquisition module, the Wi-Fi access point and the receiving device collect the channel state information of the Wi-Fi signal, including the amplitude and phase changes of each subcarrier; the PIR sensor collects the infrared radiation intensity signal and outputs a voltage signal that varies with the movement of people; the ultrasonic sensor measures the time difference of the reflected ultrasonic signal by transmitting and receiving ultrasonic signals, and collects the signal reflecting the distance between the person and the sensor.

[0258] In the data processing module, the processing process of the Wi-Fi CSI data is as follows: First, the CSI signal is denoised by principal component analysis to extract the main changing components and remove the irrelevant noise; second, the CSI amplitude is normalized to make the signal consistent under different time and environmental conditions; finally, the CSI signal is transformed into the frequency domain, and the frequency characteristics of the CSI signal over time are extracted through spectral analysis. The processing process of the PIR signal is: By the double dynamic threshold method, the voltage peak value and the ramp slope of the PIR sensor are analyzed to detect human activities of different intensities. The processing process of the ultrasonic data is: By measuring the time difference of the ultrasonic wave propagation, the distance between the person and the sensor is calculated. The multiple ranging data are smoothed to reduce noise and errors.

[0259] In the fusion algorithm module, the Kalman filter is mainly used to fuse the data of the PIR sensor and the ultrasonic sensor, combining the centroid localization and the precise ranging results. First, the centroid position is calculated according to the PIR sensor signal strength to provide a rough position estimate of the person within the sub-region, and then the centroid coordinates are calculated by the weighted average method. Second, the ranging data of multiple ultrasonic sensors are used to calculate the precise position of the person by the triangulation method. During the Kalman filtering process, the state variables of the target are defined, including the position and speed, and the current position of the target is predicted through the prediction model based on the previous state estimates (including the position and speed). The rough centroid position provided by the PIR and the precise result of the ultrasonic ranging are fused through the Kalman gain. Combining the measurement results of different sensors, the state estimate is updated, and the fused precise coordinates are output.

[0260] After being processed by the fusion algorithm, the system finally outputs the precise coordinates of the person. Combining the PIR centroid localization and the ultrasonic ranging, the fused person position is output in the form of two-dimensional coordinates. And as the person moves, the system continuously updates the position information of the person to ensure that the system can provide the precise position of the person in real time.

[0261] Verification test:

[0262] To calculate the error by comparing the gap between the measured value and the true value, in this embodiment, the Euclidean distance formula is used to calculate the error between the measured value and the true value in the two-dimensional space. For a two-dimensional coordinate (x, y), assuming the true coordinates are (x t ,yt ), the measured coordinates are (x m , y m ), then the error E calculation formula is as follows:

[0263]

[0264] Table 1 shows the improvement in positioning accuracy and the reduction in positioning error achieved by this indoor positioning system.

[0265] Table 1

[0266]

[0267] By comparing this table, it can be clearly seen that the method of the present invention has a gradual improvement in the positioning accuracy of personnel, better shows the coordinate estimation of each step and the change trend of the positioning error, and helps to conduct a quantitative analysis of the improvement of the positioning accuracy of the entire system.

[0268] The indoor positioning method and system based on multi-sensor fusion achieve a passive indoor positioning technology through multi-level positioning and multi-sensor fusion, avoiding the process of the person to be located carrying an electronic tag. The feasibility of this method is further verified through the indoor positioning system. The comparison of the relevant indicators of the single one and the method proposed by the present invention is shown in Table 2:

[0269] Table 2

[0270]

[0271]

[0272] Through this comparison table, it can be clearly seen that the method of the present invention has a significant improvement in terms of positioning accuracy, response time, system robustness and adaptability compared with the prior art, and can better meet the actual application requirements of the indoor positioning system.

[0273] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An indoor positioning method based on multi-sensor fusion, characterized in that, The following steps are involved: Step 1, deploy Wi-Fi transceiver equipment and various sensors: deploy Wi-Fi signal transceiver equipment, PIR sensors and ultrasonic sensors in the area to be located, where the Wi-Fi signal transceiver equipment is used to provide overall wireless signal coverage, the PIR sensor is used to detect human infrared radiation to sense human activities, and the ultrasonic sensor is used to measure distance; Step 2: Wi-Fi CSI test area range detection: collect CSI information, pre-process the CSI signal, extract the main change components of the CSI data, and perform maximum value normalization. Observe the frequency characteristics of signal amplitude changes through short-time Fourier transform, extract features of phase difference, build a training data set and train the SVM classification model to distinguish the activity status of people in the environment. Step 3, detailed detection of the sub-area by the PIR sensor: according to the voltage signal of the PIR sensor, the voltage peak and climbing slope characteristics are extracted, and the double dynamic threshold method is used to judge the movement state of the person and determine the person's activities in the sub-area; Step 4, calculate the centroid positioning coordinates: According to the signal intensities of multiple PIR sensors, calculate the relative position (x PIR , y PIR ) of the person within the sub-region by the weighted average method; Step 5, Ultrasonic Ranging for Precise Positioning: Using the known coordinates of the ultrasonic sensor and the measured distance values, calculate the precise position (x ultra , y ultra ) of the person through triangulation; Step 6, fuse the sensor positioning results: Obtain the centroid positioning coordinates of the PIR and the ultrasonic ranging coordinates, define the state variables and the observation vector, perform state prediction and update through Kalman filtering, calculate the Kalman gain, combine the observation results with the predicted values, correct the current coordinates and speed of the target, and update the error covariance matrix, and finally output the fused target position (x k , y k ).

2. The indoor positioning method based on multi-sensor fusion according to claim 1, wherein, In step 1, a modular splicing method is used to deploy sensors and divide sub-areas so as to overlap the detection areas of the sensors.

3. The indoor positioning method based on multi-sensor fusion according to claim 1, characterized in that, In step 2, the CSI information includes: where, H(f i ) is the CSI value of the i-th subcarrier; |H(f i )| is the amplitude of the CSI; θ(f i ) is the phase of the CSI; f i is the frequency of the i-th subcarrier; Preprocess the CSI signal: Use the PCA principal component analysis method to extract the main change components in the CSI information, and decompose the data into multiple principal components through PCA. The formula is as follows: X = UΣV T ; Where X is the CSI data matrix; U and V are eigenvector matrices respectively; T represents the transpose operation of the eigenvector matrix; Σ is the singular value matrix, representing the principal components of the signal; Perform maximum normalization on the CSI amplitude: Among them, |H norm (f i , t)| is the normalized CSI amplitude value, representing the result of the normalization of the CSI amplitude at frequency f i and time t; |H(f i , t)| is the original CSI amplitude value, representing the unnormalized CSI amplitude information obtained from the Wi-Fi device at frequency f i and time t; max(|H(f i , t)|) is the maximum value of the CSI amplitude at all frequencies and time points of the current CSI data, serving as the benchmark for normalization; Obtain the frequency characteristics of the signal amplitude change at different times through the short-time Fourier transform STFT; convert the CSI amplitude signal from the time domain |H norm (f i ,t)| to the frequency domain S(f,t), using the formula: Among them, S(f, t) represents the CSI time-frequency eigenvalue at frequency f and time t, which is used to reflect the variation of the CSI amplitude with frequency; e -j2πfτ is the kernel function of the Fourier transform, where j is the imaginary unit, π is a mathematical constant, τ is the integration variable, and dτ represents the differential element of the integration; Extracting phase difference features; The training data set is constructed by extracting the amplitude change frequency features and phase difference features for training the classification model. The SVM classifier is trained using the extracted CSI features. During the training process, the labeled CSI feature data is used to distinguish "personnel activity" or "no personnel activity". The SVM classifier model is as follows: f(x) = w T φ(x) + b; Among them, f(x) represents the predicted output of input x; x is the input feature, namely the amplitude change and phase difference features; w is the weight of the SVM model; φ(x) represents the feature vector obtained after the input x passes through the feature mapping function φ; b is the deviation.

4. The indoor positioning method based on multi-sensor fusion according to claim 1, wherein, In step 3, Set the voltage peak threshold: Dynamically adjust the detection threshold V of the voltage peak according to the environmental noise level and typical human activity signals th,peak ; When the voltage peak of the sensor exceeds the set threshold, it indicates that there is human activity in this area; The voltage peak determination conditions are as follows: V peak >V th,peak ; Among them, V peak is the peak voltage output by the PIR sensor; V th,peak is the peak voltage threshold dynamically adjusted according to environmental conditions; Set the climbing slope threshold: Calculate the voltage climbing slope by sampling the voltage signal change rate of the sensor: Where dV is the voltage increment; dt is the time increment; Dynamically set the threshold α of the slope according to different exercise intensities th,slope ; When the slope of the voltage change exceeds the threshold, there is a strong exercise state; the determination condition for the climbing slope is as follows: α slope >α th,slope ； Among them, α slope is the currently measured voltage ramp slope; α th,slope is the slope threshold dynamically adjusted according to the exercise intensity; The two dynamic thresholds of voltage peak value and climbing slope are combined to judge the activity of personnel; when the following conditions are met, it is determined that there are personnel activities in the sub-area:

5. The indoor positioning method based on multi-sensor fusion according to claim 1, characterized in that In step 4, when a person enters the detection area of the PIR sensor, the PIR sensor outputs a signal strength S related to the distance of the person i , arrange n PIR sensors in a sub-region, and the sensor coordinates are (x PIRSi , y PIRSi ), where i = 1, 2,..., n; The signal strength of each PIR sensor is used as a weight, and the relative position of the person is calculated through weighted average based on the spatial position of the sensor; Estimate the centroid coordinates using a weighted average method: where, w i is the signal strength of the i-th PIR sensor; x PIR and y PIR are the centroid coordinates of the person in the sub-region, that is, the estimated relative position; x PIRSi and y PIRSi are the coordinates of the i-th sensor; ∑w i x PIRSi and ∑w i y PIRSi represent the results of summing the products of the coordinates x PIRSi and y PIRSi of each point with its weight w i ; ∑w i represents the sum of all weights and is the normalization factor; By means of weighted average, the relative position (x PIR , y PIR ) of the personnel within the sub-region is obtained.

6. The indoor positioning method based on multi-sensor fusion according to claim 1, characterized in that, In step 5, for the sub-region where the person is located, the coordinates of each ultrasonic sensor are denoted as (x UltraSi , y UltraSi ), where i = 1, 2, 3; using the triangulation method, the exact coordinates of the person are calculated through the distance measurement values of multiple ultrasonic sensors; let the actual position of the person be (x ultra , y ultra ), and the coordinates (x ultra , y ultra ) are solved by combining the following equations: The distance d1 obtained from ultrasonic sensor 1: The distance d2 obtained from ultrasonic sensor 2: The distance d3 obtained from ultrasonic sensor 3: Solve by Newton's iterative method or least squares method; the exact position (x ultra , y ultra ) of the person within the sub-region is obtained.

7. The indoor positioning method based on multi-sensor fusion according to claim 1, wherein In step 6, Define the state variable X of the target k as follows: where x k and y k are the two-dimensional coordinate positions of the target at time k; v x and v y are the velocities of the target in the x and y directions; Define the observation vector as Z k : Among them, and are the relative positions provided by the PIR centroid positioning at time k; and are the precise positions provided by the ultrasonic ranging at time k; The prediction model of the Kalman filter calculates the target position and velocity at the current moment based on the state X at the previous moment. k-1 Calculate the target position and velocity at the current moment; The state transfer equation is defined as: X k|k-1 = F·X k-1 + B·U k ; Among them, X k|k-1 is the current state predicted from the previous state at time k; B is the control matrix, which can be ignored if there is no control input; U k is the control input vector, which can be ignored if there is no control input; F is the state transition matrix, assuming that the position changes smoothly over time: Calculate the Kalman gain: The Kalman gain K k is defined as: K k = P k|k-1 H T (HP k|k-1 H T + R) -1 ; where P k|k-1 is the predicted state error covariance matrix, representing the uncertainty of the predicted value at the current moment; R is the measurement noise covariance matrix; H is the observation matrix: State update: After calculating the Kalman gain, the state is updated by combining the observation result Z k with the predicted value X k|k-1 The state update equation is as follows: X k = X k|k-1 + K k ·(Z k - H·X k|k-1 ); Among them, Z k is the measurement result of the PIR and ultrasonic sensors; H·X k|k-1 is the observed value obtained according to the prediction; Update the state error covariance matrix P k , and the error covariance update equation is as follows: P k =(I - K k H)P k|k-1 ; Where I is the identity matrix; The target position (x k , y k ) after output fusion.

8. An indoor positioning system based on multi-sensor fusion, which is used to execute the indoor positioning method based on multi-sensor fusion according to any one of claims 1 to 7, characterized in that, include: Divide the sub-regions and deploy the sensor modules. Set the area of the positioning region as S, and divide this region into n sub-regions according to the size of the area of the region to be positioned. The data acquisition module includes a Wi-Fi access point and a receiving device to collect the channel state information of the Wi-Fi signal, including the amplitude and phase changes of each sub-carrier; the PIR sensor collects the infrared radiation intensity signal and outputs a voltage signal that changes with the movement of people; the ultrasonic sensor measures the time difference of the reflected ultrasonic signal by transmitting and receiving ultrasonic signals, and collects the signal reflecting the distance between the person and the sensor. The data processing module is used to process the channel state information of the Wi-Fi signal, the voltage signal of the PIR sensor, and the ultrasonic signal. The fusion algorithm module uses the Kalman filter to fuse the data of the PIR sensor and the ultrasonic sensor. Define the state variables and the observation vector, perform state prediction and update through the Kalman filter, calculate the Kalman gain, combine the observation results with the predicted values, correct the current coordinates and speed of the target, and update the error covariance matrix. Finally, output the fused target position.

Images