Library indoor positioning method fusing robust tracking and channel state information

Abstract

The application provides a library indoor positioning method fusing robust dead reckoning and channel state information, comprising the following steps: step 1, data acquisition is carried out by using a mobile terminal sensor; step 2, a human body motion classifier based on a decision tree is constructed, user displacement and heading change are obtained according to the collected data of the mobile terminal sensor, and the user motion state is classified; step 3, for a specific motion state, a terminal position classifier which is not sensitive to the terminal placement mode is constructed, and the terminal posture is continuously tracked; step 4, according to the influence mechanism of different motion states on the human body heading, a corresponding optimal strategy robust heading estimation is designed and selected; step 5, step length estimation and gait detection are carried out according to the corresponding optimal strategy robust heading estimation, and a fusion positioning result is obtained. The application studies and constructs the correlation law of unrestricted terminal and human body motion, can adapt to the robust PDR model of a complex sensing environment, and effectively improves the indoor positioning precision.

Description

Technical Field

[0001] This invention relates to a library indoor positioning method that integrates robust trajectory estimation and channel state information, belonging to the field of indoor positioning technology. Background Technology

[0002] Currently, existing trajectory estimation algorithms do not consider adaptability to complex and ever-changing user motion states and terminal positions. Therefore, research on identifying user motion states or terminal positions is needed. This mainly involves two issues: First, they do not fundamentally consider the coupling relationship between terminal and human motion states, and the classification of motion states is often incomplete, rarely considering the terminal's own motion. Terminal position changes or hand tremors introduce additional acceleration and angular velocity signals, which may be mistaken for normal walking or turning movements, leading to false alarms or incorrect heading estimations. Second, existing methods only consider the classification and identification of user motion states or terminal positions separately, without considering both in a unified manner. Directly considering the combination classification of all motion states and terminal positions can result in hundreds of combinations, potentially leading to overfitting, and the training process is extremely cumbersome for users.

[0003] Track estimation suffers from cumulative errors, necessitating correction using absolute positioning methods. However, the complex sensing environment of libraries, such as bookshelves obstructing non-line-of-sight directions, presents challenges even with absolute positioning alone. Cross-positioning technology based on indoor Wi-Fi channel state information and angle measurement from smart terminals is a feasible indoor absolute positioning approach. Furthermore, combining this with relatively accurate CSI cross-positioning results to calibrate heading and step length can improve the reliability of indoor positioning. Summary of the Invention

[0004] This invention addresses the technical problem of course estimation algorithm failure caused by complex environments by proposing a library indoor positioning method that integrates robust track estimation and channel state information.

[0005] The technical solution adopted by the present invention to solve the above problems is as follows:

[0006] Step 1: Collect data using the mobile terminal's sensors;

[0007] Step 2: Construct a human motion classifier based on decision trees, obtain user displacement and heading changes based on data collected by mobile terminal sensors, and classify user motion states in combination.

[0008] Step 3: For specific motion states, construct a terminal position classifier that is insensitive to the terminal placement method, and continuously track the terminal posture;

[0009] Step 4: According to the influence mechanism of different motion states on the human body heading, the corresponding optimal strategy robust heading estimation is designed and selected;

[0010] Step 5: According to the corresponding optimal strategy robust heading estimation, step length estimation and gait detection are performed to obtain the fusion positioning result.

[0011] Optionally, the classification result of the user motion state in step 2 includes: human body horizontal displacement, human body vertical displacement, human body rotation, static standing, and terminal self-motion which does not cause human body displacement and rotation;

[0012] The human body horizontal displacement includes normal walking of the user;

[0013] The human body vertical displacement includes the user going up and down stairs, taking an escalator, and taking an elevator;

[0014] The terminal self-motion which does not cause human body displacement and rotation includes terminal position conversion and hand shaking.

[0015] Optionally, the step of classifying the user motion state in step 2 includes:

[0016] According to the user angular velocity and acceleration data obtained by the mobile terminal sensor, based on the acceleration and angular velocity signal mode, displacement and magnetic field change rule, it is judged whether the user's acceleration matches the elevator acceleration mode, if it matches, the user is in the elevator taking state;

[0017] If it does not match, it is judged whether the user acceleration variance is less than a first threshold value, if it is, it is judged whether the magnetic field variance is less than a second threshold value, if it is, the user is in the static standing state, if it is not, the user is in the escalator taking state;

[0018] If the user acceleration variance is greater than the first threshold value, it is judged whether the user vertical direction appears continuous positive or negative position change, if it appears, the user is in the up and down stairs state;

[0019] If it does not appear, it is judged whether the user horizontal direction has continuous displacement change, if it has, the user is in the normal walking state;

[0020] If it does not occur, it is judged whether the user horizontal direction angle change is greater than a third threshold value, if it is greater than, the user is in the turning state, if it is less than, the user is in the terminal self-motion state.

[0021] Optionally, the step of designing the corresponding optimal strategy robust heading estimation in step 4 includes:

[0022] Step 4.1: In the normal walking state of the user, the principal component of the horizontal acceleration horizontal plane component is extracted by principal component analysis, and the user heading in the terminal coordinate system is obtained by solving the ambiguity judgment;

[0023] Step 4.2: In the user turning around and going up and down the stairs state, the horizontal direction is obtained based on the terminal posture tracking module, the angular velocity of the horizontal direction is integrated to obtain the user heading change, and the absolute heading at the previous moment is combined to obtain the current user heading;

[0024] Step 4.3: In the case of user terminal motion alone, terminal motion and normal walking at the same time, the current user heading is obtained by weighted average adjacent normal gait heading;

[0025] Step 4.4: In the user riding the escalator and elevator state, the initial user heading is obtained according to the prior knowledge, the current user motion state is obtained through the motion state classifier output result, and the acceleration or magnetic field change rule is recognized at the specific moment as the landmark information to correct the initial user heading and the terminal posture.

[0026] Optionally, the step of obtaining the fusion positioning result in step 5 comprises:

[0027] Step 5.1: Step length estimation and gait detection are performed according to the corresponding optimal strategy robust heading estimation, a PDR model is established based on the step length estimation and gait detection result, and the indoor positioning and navigation trajectory of the pedestrian is calculated;

[0028] Step 5.2: Angle of arrival cross positioning based on channel state information is performed;

[0029] Step 5.3: The heading estimation result and step length estimation parameter are corrected based on the angle of arrival cross positioning;

[0030] Step 5.4: AOA precision confidence estimation and angle of arrival cross positioning precision estimation are performed, the channel state information estimation precision is derived, and the non-line-of-sight environment is identified;

[0031] Step 5.5: The weighting parameters of the CSI cross positioning result and the PDR model positioning result are set, and the fusion positioning result is obtained.

[0032] Optionally, the step of performing step length estimation and gait detection in step 5.1 comprises:

[0033] Step 5.1.1: The gait cycle and acceleration statistical value are used as input, the step length estimation value is used as output, and the step length estimation model is constructed in combination with the related fitting parameters;

[0034] Step 5.1.2: When the user is taking the elevator, escalator, going up or down the stairs, the step length estimation model does not work, the user's position is directly obtained from the corresponding landmark prior information, when the user turns around, the step length is set to zero, when the user is at the terminal, the step length estimation value is obtained by weighted average of step length estimation of adjacent normal walking, when the user is normal walking, according to the position of the terminal, different terminal acceleration models are selected as parameters to train the step length estimation model to obtain the step length estimation value;

[0035] Step 5.1.3: Different acceleration threshold and time interval threshold are set for different user motion states and terminal positions, the peak points in each gait cycle are detected, and the number of points whose peak value is greater than the acceleration threshold and whose adjacent peak time interval is greater than the time interval threshold is counted as the step number estimation result;

[0036] The expression of the step length estimation model is:

[0037]

[0038] In formula (1), f SL (·) is a fitting function, K1 and K2 are model parameters, a 竖直,i is the vertical direction acceleration sample, a 竖直极差 is the difference between the maximum and minimum values of the vertical direction acceleration sample, Na is the number of acceleration signal samples, T is the gait cycle, the vertical direction acceleration is obtained by projecting the acceleration signal vector in the vertical direction, and the vertical direction vector is provided by the terminal attitude tracking module.

[0039] Optionally, the step of correcting the heading estimation result and the step length estimation parameter in step 5.3 includes:

[0040] Step 5.3.1: An optimization objective function J1 is obtained based on the consistency of the walking distance in the corresponding time period and the cumulative step length distance estimated by the terminal speed estimation;

[0041] Step 5.3.2: An optimization objective function J2 is obtained based on the consistency of the distance between two high-precision positioning track points and the PDR displacement size;

[0042] Step 5.3.3: The objective functions J1 and J2 are weighted and optimized, the heading is calculated based on the channel state information angle of arrival cross positioning result, and the correction of the heading estimation result and the step length estimation parameter is completed;

[0043] The expression of the optimization objective function J1 is:

[0044]

[0045] In formula (2), f is the terminal speed estimation value corresponding to the mth step, and ΔTm is the duration of the mth step;

[0046] The expression of the optimization objective function J2 is:

[0047]

[0048] In formula (3), L n and L n-n0 are the user position vectors at n and n-n0, respectively, and are the user position vectors calculated by the displacement calculation formula; f SL (K1, K2) and ψ correspond to the step length and user heading at the moment, n0 is the time interval of the optimized trajectory, and N is the total length of the user trajectory participating in parameter correction.

[0049] Optionally, the step of deriving the channel state information estimation accuracy and identifying the non-line-of-sight environment in step 5.4 includes:

[0050] Step 5.4.1: Analyzing the angle of arrival accuracy based on the MUSIC algorithm, constructing an AOA accuracy confidence metric model, and obtaining an angle of arrival cross-location accuracy confidence estimate;

[0051] Step 5.4.2: Constructing a CSI cross-location confidence metric model based on the AOA Cramer-Rao bound formula, and obtaining a CSI cross-location accuracy confidence estimate σ c ;

[0052] Step 5.4.3: If the angle of arrival cross-location accuracy confidence estimate is less than a preset threshold, a non-line-of-sight situation occurs.

[0053] Optionally, the step of obtaining the fusion positioning result in step 5.5 includes:

[0054] Step 5.5.1: The PDR model combines the position coordinates of the previous step and the user displacement to calculate the current position of the user;

[0055] Step 5.5.2: Based on the single access point positioning of the channel state information, the accuracy confidence estimate of the current positioning is calculated according to the estimation accuracy of the single access point positioning of the previous step;

[0056] Step 5.5.3: The error variance values of the step length and heading estimation, and the error covariance matrix of the position variable estimation at the previous moment are given, and the unscented Kalman filter calculates the predicted value of the state variable and the error covariance matrix of the predicted value through unscented transformation. The square root of the trace of the covariance matrix is the PDR predicted position error σ p , and the fusion positioning result is obtained by parameter weighted fusion of the CSI cross-location and PDR.

[0057] The expression for calculating the current position of the user is:

[0058]

[0059] In formula (4), L i =(x i ,y i ) is the position vector of the user at the i-th step, only the two-dimensional horizontal position vector is considered here, and the vertical coordinate value is usually given by the prior information value of the floor on which the user is located, L i-1 is the position vector of the user at the previous step, SL i is the step length estimation at the i-th step, and ψ i is the heading estimation of the user in the given world coordinate system.

[0060] The expression for fusing the positioning results is:

[0061] T final =w c T CSI +w p T PDR (5)

[0062] In formula (5), T PDR is the predicted position of PDR, T CSI is the cross positioning result of CSI, w c and w p are weighting parameters, which are determined by the positioning error accuracy.

[0063] The beneficial effects of the present application are:

[0064] 1. The present application studies and constructs the correlation law of unrestricted terminal and human motion, and can adapt to the robust PDR model of complex sensing environment.

[0065] 2. The step length and heading calibration and fusion positioning method given by the present application can effectively improve the indoor positioning accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 is the flow chart of the library indoor positioning method of fusing the robust dead reckoning and channel state information provided by the present application;

[0067] Figure 2 is the schematic diagram of the motion state classifier based on the decision tree provided by the present application;

[0068] Figure 3 is the schematic diagram of the robust heading estimation algorithm of adapting to the complex sensing environment provided by the present application. DETAILED DESCRIPTION

[0069] In combination with Figures 1-3In the embodiment, AOA is angle of arrival, and CSI is channel state information, as shown in the following formula: Figure 1 The steps of the library indoor positioning method combining robust estimation of trajectory and channel state information in the embodiment include:

[0070] S1: mobile terminal sensor data acquisition;

[0071] S2: constructing a human motion state classifier based on a decision tree, and combining and classifying user motion states;

[0072] The classification results of the user motion state include: human horizontal displacement, human vertical displacement, human rotation, stationary standing, and terminal self-motion which does not cause human displacement and rotation.

[0073] The human horizontal displacement includes normal walking of the user.

[0074] The human vertical displacement includes the user going up and down stairs, taking an escalator, and taking an elevator.

[0075] The terminal self-motion which does not cause human displacement and rotation includes terminal position conversion and hand shaking

[0076] As shown in the following formula: Figure 2 In the embodiment, the user angular velocity and acceleration data are obtained by the mobile terminal sensor, and whether the acceleration of the user matches the acceleration mode of the elevator is judged based on the acceleration and angular velocity signal mode, displacement, and magnetic field change rule. If yes, the user is in the elevator taking state.

[0077] If no, whether the acceleration variance of the user is lower than a first threshold value is judged. If yes, whether the magnetic field variance is lower than a second threshold value is judged. If yes, the user is in the stationary standing state. If no, the user is in the escalator taking state.

[0078] If the acceleration variance of the user is greater than the first threshold value, whether the user appears continuous positive or negative position change in the vertical direction is judged. If yes, the user is in the up and down stairs state.

[0079] If no, whether the user appears continuous displacement change in the horizontal direction is judged. If yes, the user is in the normal walking state.

[0080] If no, whether the angle change of the user in the horizontal direction is greater than a third threshold value is judged. If yes, the user is in the turning state. If no, the user is in the terminal self-motion state.

[0081] S3: constructing a terminal position classifier which is not sensitive to the terminal placement mode for a specific motion state, and continuously tracking the terminal posture.

[0082] S301: Construct a terminal position state classifier that is insensitive to terminal placement, for example, when the user is standing, the terminal position conversion will not classify the motion state as a normal walking state or the like;

[0083] S302: Take the sensor signal vector amplitude, horizontal component value and vertical component value as the original input signal, extract and select the maximum value, minimum value, variance, median and the like of the three signals in a sliding window, and train and test the random forest multi-class classifier.

[0084] S4: According to the influence mechanism of different motion states on the heading of the human body, the corresponding optimal strategy robust heading estimation is designed and selected;

[0085] As shown in Figure 3 S401: In the normal walking state, the present application uses principal component analysis to extract the principal component of the acceleration horizontal component, and through defuzzification judgment, the walking direction in the terminal coordinate system can be obtained;

[0086] S402: For the two motion states of the user turning around and going up and down the stairs, the user heading change is calculated, and the absolute heading at the previous moment is added to obtain the current user heading. The user heading change is obtained by integrating the angular velocity in the horizontal direction, and the horizontal direction can be provided by the terminal attitude tracking module;

[0087] S403: When the terminal itself motion occurs alone or simultaneously with normal walking, the additional acceleration or attitude change will destroy the sensor signal rule relied on by the previous two methods, and at this time, the heading estimation is given by weighted average of the adjacent normal gait heading;

[0088] S404: For the motion state of taking the escalator / elevator, the user heading estimation is obtained according to prior knowledge, such as going up and down the escalator and entering and exiting the elevator, the user heading is generally limited to a very small range. The motion state classifier output result can be combined with the acceleration or magnetic field change rule to identify the specific moment as a landmark information to correct the user heading and the terminal attitude.

[0089] S5: According to the corresponding optimal strategy robust heading estimation, step length estimation and gait detection are performed to obtain the fusion positioning result;

[0090] S501: According to the corresponding optimal strategy robust heading estimation, step length estimation and gait detection are performed, and a PDR model is established based on the step length estimation and gait detection result to calculate the indoor positioning and navigation trajectory of the pedestrian in the library;

[0091] S50101: Step length estimation method uses gait cycle, acceleration statistics as input, combined with related fitting parameters, to construct a step length fitting nonlinear function model, the nonlinear function model is as follows:

[0092]

[0093] In formula (1), f SL (·) is a fitting function, K1 and K2 are model parameters, a 竖直,i is a vertical direction acceleration sample, a 竖直极差 is the difference between the maximum and minimum values of the vertical direction acceleration sample, Na is the number of acceleration signal samples, T is the gait cycle, the vertical direction acceleration is obtained by projecting the acceleration signal vector in the vertical direction, and the vertical direction vector is provided by the terminal attitude tracking module.

[0094] S50102: In order to adapt to complex sensing environment, the step length estimation model needs to be trained for specific terminal position and motion state. When the user takes the elevator, escalator, goes up and down the stairs, the step length estimation model does not work, and the user position is directly obtained from the corresponding landmark prior information; When the user turns around, the step length can usually be set to zero or a small value; When the terminal itself moves, the step length estimation can be obtained by weighted average of the step length estimation of the adjacent normal walking, because the terminal acceleration mode is different, such as the acceleration amplitude and variance of the terminal placed in the pocket is much larger than that of the handheld terminal, the step length estimation model parameters also need to be trained respectively for different terminal positions under normal walking;

[0095] S50103: Gait detection uses peak detection method to estimate the number of steps. The peak detection method uses the characteristic that the acceleration peak point is generated when the foot lands during normal walking to detect the peak point in each gait cycle and take the number of peak points as the step number estimation result. Peak detection needs to set two thresholds: acceleration threshold and time threshold. The peak point must be greater than a certain acceleration threshold to ensure that the point is a true acceleration peak; The time interval between adjacent peak values must be greater than a certain time threshold to eliminate false detection points.

[0096] S50104: Different acceleration and time interval thresholds need to be set for different user motion states and terminal positions. Compared with the step length estimation parameter setting, the peak detection algorithm has stronger robustness, and the tolerance of the setting of its acceleration and time interval threshold parameters is larger, and it is easier to obtain through data training.

[0097] S502: Perform Angle of Arrival (AOA) cross positioning based on Channel State Information (CSI);

[0098] S503: Correct the indoor positioning heading estimation result and step length estimation parameter by using the CSI cross positioning result;

[0099] S50301: Give an optimization objective function J1 by using the consistency of the walking distance in the corresponding time period and the cumulative step length distance estimated by the terminal speed;

[0100]

[0101] In formula (2), is the terminal speed estimation value corresponding to the mth step, and ΔT m is the duration of the mth step, is the step length estimation of the mth step, m1 is the number of steps involved in the segmented optimization, and M is the total number of steps involved in the step length estimation model parameter calibration;

[0102] S50302: Give an optimization objective function J2 by using the consistency of the distance between two high-precision positioning trajectory points and the PDR displacement size:

[0103]

[0104] In formula (3), L n and L n-n0 are the user position vectors at time n and n-n0, respectively, and are the user position vectors calculated by the displacement calculation formula; f SL (K1, K2) and ψ are the step length and user heading at the corresponding time, n0 is the time interval of the optimized trajectory, and N is the total length of the user trajectory involved in the parameter correction. Since the estimation deviation of the absolute heading does not affect the displacement size between two time points, the displacement size between two time points with a short interval time is mainly determined by the step length estimation. Therefore, when optimizing the step length parameter by trajectory distance consistency, it is feasible to only consider the influence of the step length estimation;

[0105] S50303: When the terminal speed estimation and trajectory optimization can be used at the same time, the objective functions J1 and J2 are weighted and optimized;

[0106] S50304: Calculate the heading by using the CSI cross positioning result, thereby correcting the indoor positioning heading estimation result.

[0107] S504: Perform positioning accuracy confidence estimation and AOA estimation accuracy, derive the CSI estimation accuracy, and identify the non-line-of-sight environment;

[0108] S50401: The AOA estimation accuracy analysis of the MUSIC algorithm includes: the MUSIC algorithm mainly uses the orthogonality of the signal subspace and the noise subspace for direction finding, and its expression is as follows:

[0109] F(θ)=a H (θ)GG H a(θ)=0(4)

[0110] Equation (4), where a is the steering vector, θ is the AOA parameter to be estimated, and G is the noise subspace.

[0111] MUSIC algorithm angle estimation result at the minimum point of the function F(θ), that is, satisfying Through formula derivation, it can be concluded that the angle estimation mean square error of the MUSIC algorithm can be represented as:

[0112]

[0113] In equation (5), N represents the number of snapshots, n represents the number of sources, and m represents the number of array elements. k represents the kth eigenvalue greater than σ 2 in the sample covariance matrix, s k represents the eigenvector corresponding to the eigenvalue λ k , and g k represents the kth eigenvector that spans the noise subspace. The derivative of the vector a(θ) with respect to θ is the vector d(θ).

[0114] S50402: The CSI confidence measurement model (positioning error covariance matrix) is given by the following equation:

[0115] CRB(L)=FIM -1 (L)(6)

[0116] In equation (6), FIM is the Fisher information matrix. Assuming that the observation vector satisfies the Gaussian probability density distribution,

[0117]

[0118] In equation (7),

[0119]

[0120] S50403: If the AOA estimation accuracy exceeds the preset threshold, it is considered that the non-line-of-sight situation occurs.

[0121] S505: Set the weighting parameters of two positioning results and give the fusion positioning result.

[0122] S50501: The PDR model combines the position coordinates of the previous step and the user displacement to calculate the current position of the user:

[0123]

[0124] In formula (9), L i = (x i , y i ) is the i-th step user position vector, only two-dimensional horizontal position vector is considered here, the vertical coordinate value is usually given by the prior information value of the floor where it is located, L i-1 is the user position vector of the last step, SL i is the step length estimation of the i-th step, ψ i is the user heading estimation under the given world coordinate system, qd i = [SL i ψ i ] T , is the nonlinear displacement function of the user with respect to the vector qd i . Given the error variance values of the step length and heading estimation and the error covariance matrix of the last time position variable estimation, the UKF calculates the predicted value of the state variable and the prediction error covariance matrix through the unscented transformation, and the square root of the trace of the covariance matrix is the PDR predicted position error σ p ;

[0125] S50502: Fuse the CSI cross positioning and PDR through parameter weighting. The final positioning result should be

[0126] T final = w c T CSI + w p T PDR (10)

[0127] In formula (10), T PDR is the PDR predicted position, T CSI is the CSI cross positioning result, w c and w p are weighting parameters, which are determined by the positioning error accuracy.

[0128] The present application carries out indoor positioning test, indoor positioning test scene and result are as follows: indoor positioning test scene is built in Heilongjiang Province Traditional Chinese Medicine University library, scene range is about 100m*70m, spans two floors, there are three positioning sites available in the scene range on average; the terminal equipment used in the experiment is Xiaomi 10 smart phone, the equipment is equipped with gyroscope, Wi-Fi antenna, magnetic force sensor, acceleration sensor and other experimental related equipment, the inertial sensor and magnetic force sensor embedded in the mobile phone are set to use 50Hz sampling frequency, and the CSI data refresh frequency is 1Hz. Before the experiment, the smart phone is subjected to "8" movement and static placement, respectively calibrating the magnetic force sensor and gyroscope sensor, and eliminating the zero offset error in data processing. The pedestrian holds the smart phone, goes upstairs through the stairs and downstairs through the elevator, crosses two floor areas, and collects about 500 positioning and sensor data of positioning trajectory points. Each positioning trajectory point corresponds to a pedestrian gait, and the experiment is repeated 10 times, respectively giving the indoor positioning statistical results of the CSI cross positioning method and the fusion positioning method proposed in the present application, as shown in Table 1, verifying the effectiveness of the fusion robust track prediction and channel state information positioning method proposed in the present application.

[0129] Table 1

[0130] CSI cross-location method The fusion positioning method proposed by the present application Mean absolute error 3.6 meters 1.9 meters 90% error 5.2 meters 3.4 meters

[0131] The above is only the preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the present application, and any simple modification, equivalent replacement and improvement of the above embodiments within the scope of the present application are still within the protection scope of the present application.

Claims

1. A library indoor positioning method fusing robust tracking and channel state information, characterized in that, The steps of the fusion robust dead reckoning and channel state information library indoor positioning method include: Step 1: data acquisition using mobile terminal sensors; Step 2: constructing a human motion classifier based on a decision tree, obtaining user displacement and heading change according to the collected data of the mobile terminal sensors, and combining and classifying the user motion state; Step 3: constructing a terminal position classifier that is insensitive to terminal placement for a specific motion state, and continuously tracking the terminal posture; Step 4: designing and selecting a corresponding optimal strategy robust heading estimation according to the influence mechanism of different motion states on the human heading; Step 5: step length estimation and gait detection according to the corresponding optimal strategy robust heading estimation to obtain a fusion positioning result; The step of obtaining the fusion positioning result in step 5 includes: Step 5.1: step length estimation and gait detection according to the corresponding optimal strategy robust heading estimation, establishing a PDR model based on the step length estimation and gait detection results, and calculating the library indoor positioning and navigation trajectory of the pedestrian; Step 5.2: performing angle of arrival cross positioning based on channel state information; Step 5.3: correcting the heading estimation result and step length estimation parameter based on the angle of arrival cross positioning; Step 5.4: performing AOA precision confidence estimation and angle of arrival cross positioning precision estimation, deriving the channel state information estimation precision, and identifying the non-line-of-sight environment; Step 5.5: setting the weighting parameters of the CSI cross positioning result and the PDR model positioning result to obtain a fusion positioning result; The step of obtaining the fusion positioning result in step 5.5 includes: Step 5.5.1: calculating the current position of the user based on the PDR model and the position coordinates and user displacement of the previous step; Step 5.5.2: calculating the precision confidence estimation of the current positioning based on the single access point positioning estimation precision of the previous step; Step 5.5.3: Given the error variance of the step length and the heading estimation and the error covariance matrix of the position variable estimation at the last time, the unscented Kalman filter calculates the predicted value of the state variable and the error covariance matrix of the predicted value by unscented transformation, and the square root of the trace of the covariance matrix is the PDR predicted position error The fusion positioning result is obtained by fusing the CSI cross positioning and the PDR through parameter weighting. The expression for calculating the current position of the user is: （4） In equation (4), is the step is the step , is a nonlinear displacement function of the user with respect to the vector is the user position vector of the previous step, is the user position vector of the previous step, is the step length estimate of the step is the step length estimate of the step is the user heading estimate in the given world coordinate system; The expression for the fusion positioning result is: （5） In formula (5), is the PDR predicted position, is the CSI cross-locating result, and is a weighting parameter determined by the accuracy of the positioning error. 2.The library-based indoor positioning method of fusing RSS and CSI in accordance with claim 1, wherein, The classification results of the user motion state in step 2 include: human horizontal displacement, human vertical displacement, human rotation, stationary standing, and terminal self-motion that does not cause human displacement and rotation; The human horizontal displacement includes normal walking of the user; The human vertical displacement includes the user going up and down stairs, taking an escalator, and taking an elevator; The terminal self-motion that does not cause human displacement and rotation includes terminal position conversion and hand shaking. 3.The library-based indoor positioning method of fusing RSS and CSI in accordance with claim 1, wherein, The step of classifying the user motion state in step 2 includes: According to the user angular velocity and acceleration data obtained by the mobile terminal sensors, based on the acceleration and angular velocity signal mode, displacement, and magnetic field change law, it is judged whether the user acceleration matches the elevator acceleration mode, if it matches, the user is in the elevator taking state; If it does not match, it is judged whether the user acceleration variance is lower than a first threshold value, if it is, it is judged whether the magnetic field variance is lower than a second threshold value, if it is, the user is in the stationary standing state, if it is not, the user is in the escalator taking state; If the user acceleration variance is greater than the first threshold value, it is judged whether there is a continuous positive or negative position change in the vertical direction of the user, if there is, the user is in the up and down stairs state; If not, it is judged whether the user has a continuous displacement change in the horizontal direction, and if so, the user is in a normal walking state; If not, it is judged whether the angle change of the user in the horizontal direction is greater than a third threshold, and if so, the user is in a turning state, and if not, the user is in a terminal self-motion state. 4.The library-based indoor positioning method of fusing RSS and CSI in accordance with claim 1, wherein, The steps of designing the corresponding optimal strategy robust heading estimation in step 4 include: Step 4.1: In the normal walking state of the user, the principal component of the horizontal component of the horizontal acceleration is extracted by principal component analysis, and the user's heading in the terminal coordinate system is obtained by solving the ambiguity judgment; Step 4.2: In the user's turning and stair climbing state, the horizontal plane direction is obtained based on the terminal attitude tracking module, the angular velocity of the horizontal plane direction is integrated to obtain the user's heading change, and the current user's heading is obtained in combination with the absolute heading at the previous moment; Step 4.3: In the case of user terminal self-motion alone or terminal self-motion and normal walking occurring simultaneously, the current user's heading is obtained by weighted average of the adjacent normal gait heading; Step 4.4: In the user's escalator and elevator state, the initial user's heading is obtained according to the prior knowledge, the current user's motion state is obtained through the output of the motion state classifier, and the specific moment is identified as the landmark information to correct the initial user's heading and the terminal attitude. 5.The library-based indoor positioning method of fusing RSS and CSI in accordance with claim 1, wherein, The steps of step 5.1 for step length estimation and gait detection include: Step 5.1.1: The step length estimation model is constructed by using the gait cycle and acceleration statistical value as input, the step length estimation value as output, and combining the related fitting parameters; Step 5.1.2: When the user is taking the elevator, escalator, or going up and down the stairs, the step length estimation model does not work, the user's position is directly obtained from the corresponding landmark prior information, the step length is set to zero when the user turns, the step length estimation value is obtained by weighted average of the step length estimation of the adjacent normal walking when the user's terminal is self-motion, and the step length estimation value is obtained by training the step length estimation model with different terminal acceleration models as parameters according to the position of the terminal when the user is in normal walking state; Step 5.1.3: Different acceleration threshold values and time interval threshold values are set for different user motion states and terminal positions, the peak points in each gait cycle are detected, and the number of points with peak value greater than the acceleration threshold value and adjacent peak time interval greater than the time interval threshold value is counted as the step number estimation result; The expression of the step length estimation model is: （1） In equation (1), is the fitting function, and is the model parameter, is the vertical acceleration sample, is the difference between the maximum and minimum values of the vertical acceleration sample, is the number of acceleration signal samples, is the gait cycle, the vertical acceleration is obtained by projecting the acceleration signal vector in the vertical direction, and the vertical direction vector is provided by the terminal attitude tracking module. 6.The library-based indoor positioning method of fusing RSS and CSI in a robust way according to claim 5, wherein, The steps of step 5.3 for correcting the heading estimation result and step length estimation parameters include: Step 5.3.1: An optimization objective function J1 is obtained based on the consistency of the walking distance in the corresponding time period and the cumulative step length distance estimated by the terminal speed estimation; Step 5.3.2: An optimization objective function J2 is obtained based on the consistency of the distance between two high-precision positioning track points and the PDR displacement size; Step 5.3.3: The target functions J1 and J2 are weighted and optimized, the heading is calculated based on the channel state information to cross-locate the results, and the correction of the heading estimation result and the step length estimation parameter is completed; The expression of the optimization objective function J1 is: （2） In formula (2), is the terminal rate estimation value corresponding to the step is the step duration, is the step duration, is the step duration, is the step length estimation of the step is the step length estimation of the step is the number of steps involved in the piecewise optimization, is the total number of steps involved in the step length estimation model parameter calibration; The expression of the optimization objective function J2 is: （3） In formula (3), and They are respectively and User location vector at any given time, and This is the user position vector calculated using the displacement calculation formula; and For the step size and user heading at the corresponding time, To optimize the time interval of the trajectory, The total duration of the user trajectory involved in parameter correction. 7.The library-based indoor positioning method of fusing RSS and CSI in a robust manner according to claim 1, wherein, The step of deriving channel state information estimation accuracy and identifying non-line-of-sight environment in step 5.4 includes: Step 5.4.1: analyzing the angle of arrival accuracy based on the MUSIC algorithm, constructing an AOA accuracy confidence metric model, and obtaining an angle of arrival confidence estimate; Step 5.4.2: Constructing the CSI cross-locating confidence metric model based on the AOA Cramer-Rao bound formula to obtain the CSI cross-locating accuracy confidence estimation ; Step 5.4.3: if the angle of arrival cross-locating accuracy confidence estimate is less than a preset threshold, a non-line-of-sight situation occurs.

Images