Multimodal data fusion indoor positioning method and system based on RMLICT

Abstract

The invention provides an indoor positioning method and an indoor positioning system for multi-modal data fusion based on RMLICT. The method comprises the following steps: firstly, in a mapping stage, constructing a fingerprint map from unbalanced crowdsourcing trajectory data; in the off-line stage, track data enhancement, dimension reduction and clustering processing are carried out. In the on-line positioning stage, direction matching filtering is firstly carried out, and then the mapping relation between the RSS signal and the single-step motion is established by using enhanced particle filtering. According to the method, by optimizing data preprocessing and adopting an advanced particle filtering algorithm, the positioning precision and robustness are improved, the real-time performance of the system is remarkably improved, and the method is suitable for various complex indoor environments such as scenes with serious multi-path effects and signal shielding.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning technology, and in particular to an indoor positioning method and system based on multimodal data fusion of RMLICT, which is suitable for high-precision real-time positioning in complex nonlinear and non-Gaussian environments. Background Art

[0002] Indoor positioning has broad application prospects in intelligent navigation, security monitoring, asset management and other fields. However, traditional indoor positioning methods, such as those based on Kalman Filter (KF), are limited by their assumptions about linear Gaussian models, and their performance deteriorates in complex nonlinear and non-Gaussian environments. Particle Filter (PF), as a non-parametric filter based on Monte Carlo method, can effectively handle nonlinear and non-Gaussian problems, but it still has the problem of high computational complexity under high-dimensional state space and real-time computing requirements.

[0003] At the same time, the construction of fingerprint maps usually relies on high-quality, evenly distributed data. However, in practical applications, it is costly and difficult to obtain evenly distributed high-quality data. Therefore, a method is needed to effectively construct fingerprint maps from unbalanced crowdsourced trajectory data. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the present invention provides an indoor positioning method based on multimodal data fusion of RMLICT. The method introduces a mapping phase, uses the RMLICT method to build a fingerprint map from unbalanced crowdsourced trajectory data, and performs trajectory data enhancement, principal component analysis (PCA) dimensionality reduction, and K-means clustering in the offline phase, and finally applies an enhanced particle filter algorithm in the online phase for high-precision real-time positioning.

[0005] Through the above technical scheme, the present invention realizes the construction of high-precision fingerprint maps from unbalanced crowdsourced trajectory data, optimizes the data preprocessing process, and adopts an enhanced particle filtering algorithm to achieve high-precision, real-time indoor positioning in complex nonlinear and non-Gaussian environments, with good robustness and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 Schematic diagram of the overall architecture of the indoor positioning system of the present invention.

[0007] Figure 2 This is a flowchart of the mapping phase.

[0008] Figure 3 Flowchart of the offline phase.

[0009] Figure 4Schematic diagram of the state prediction and update process of the particle filter.

[0010] Figure 5 The experimental results of four algorithms, TMM, TMS, PF-TMS and RMLICT of the present invention, are shown in FIG. DETAILED DESCRIPTION

[0011] The present invention is further analyzed below in conjunction with specific embodiments.

[0012] Reference Figure 1 The embodiment of the present invention provides an indoor positioning system including the following main modules:

[0013] The wireless signal receiving module is responsible for receiving and measuring the Wi-Fi signal strength (RSS) data of the mobile device.

[0014] The inertial measurement unit (IMU) is responsible for obtaining inertial data such as acceleration and angular velocity of the mobile device.

[0015] Mapping module, responsible for building extended fingerprint maps.

[0016] The offline stage module is responsible for extracting the user's movement trajectory data in the indoor environment according to the extended fingerprint map for clustering and feature extraction.

[0017] The online positioning module is responsible for filtering the real-time collected RSS data and IMU data to achieve positioning.

[0018] The embodiment of the present invention also provides an indoor positioning method based on multimodal data fusion of RMLICT, including a mapping stage, an offline stage, and an online positioning stage;

[0019] The mapping phase is as follows Figure 1 Specifically:

[0020] Step S101, vertex extraction:

[0021] Obtain historically collected RSS (received signal strength) data, use the HDBSCAN algorithm (Hierarchical Density-Based Spatial Clustering of Applications with Noise, density-based hierarchical clustering algorithm) to cluster the unbalanced crowdsourced RSS data, extract the cluster center of each cluster from the clustering results and construct the vertex set of the fingerprint map, which represents the typical RSS features of different locations. HDBSCAN can automatically determine the number of clusters and adapt to changes in data density.

[0022] Assume that the RSS dataset is where r iis the i-th RSS vector. Through the HDBSCAN algorithm, {r i}Clustering is performed to obtain K cluster centers These cluster centers serve as the vertex sets of the fingerprint map.

[0023] Step S102: edge set extraction:

[0024] Obtain historically collected IMU data, extract edge sets between vertices, and verify the physical connectivity between vertices by analyzing the displacement and direction data in the edge sets, thereby capturing the physical space change relationship between vertices and establishing the positional relationship between positions; specifically:

[0025] Let the timestamp corresponding to the i-th vertex be t i , IMU data is in is the acceleration, is the angular velocity;

[0026] Through pedestrian dead reckoning (PDR), the displacement Δs and direction θ of the pedestrian between two vertices are calculated:

[0027] Δs=v·Δt

[0028] θ=θ t-1 +ω·Δt

[0029] Where v is the walking speed, ω is the angular velocity, and Δt is the time difference;

[0030] According to the displacement Δs and direction θ, the edge set {E ij}, where E ij Represents the connectivity between vertex i and vertex j;

[0031] Step S103: Fingerprint map construction:

[0032] Based on the extracted vertex set and edge set, the multidimensional scaling (MDS) method is used to map the high-dimensional RSS space to the two-dimensional physical space to construct a two-dimensional fingerprint map that accurately reflects the actual physical space; through reference point mapping, the two-dimensional fingerprint map is mapped to the actual physical space to ensure that the spatial position of the fingerprint map is consistent with the real environment, thereby obtaining an extended fingerprint map.

[0033] Let the distance matrix be D, where D ij =|c i -c j |. Obtain the coordinates of the vertex in two-dimensional space by solving the following optimization problem

[0034]

[0035] The offline phase aims to optimize data and enhance the robustness and computational efficiency of the model. Figure 2 Specifically:

[0036] Step S201: trajectory data enhancement:

[0037] The user's movement trajectory data in the indoor environment is extracted based on the extended fingerprint map, which increases the diversity of training data and enhances the generalization ability of the model; the movement trajectory data is enhanced;

[0038] The data enhancement adopts reversal, translation and rotation operations;

[0039] Let the original trajectory be The enhanced trajectory is where p′ t =p T-t+1 .

[0040] Step S202: Data dimensionality reduction processing:

[0041] The trajectory data after data enhancement is converted into a high-dimensional feature vector, and then the principal component analysis (PCA) is used to reduce the dimension of the trajectory data after data enhancement to reduce the redundancy of the data and the computational complexity;

[0042] Step S203: Clustering processing:

[0043] Cluster the trajectory data after dimension reduction to obtain K′ groups. The cluster centers are used as representative features and can be used in subsequent model training and particle filter algorithms to improve the real-time performance of the system.

[0044] Assume the original RSS data is R∈R N×M , N is the number of samples, M is the RSS dimension. Through PCA, R is projected into the d-dimensional subspace to obtain the reduced-dimensional data R'∈R N×d .

[0045] In the online positioning stage, an enhanced particle filter algorithm is used to combine the RSS and IMU data collected in real time to achieve high-precision positioning. Specifically:

[0046] Step S301: real-time data preprocessing:

[0047] User equipment collects RSS data in real time t} and IMU data {a t ,ω t}, and then perform direction matching filtering on the RSS data and IMU data; specifically:

[0048] Assume the actual direction of motion is θ actual , the expected direction is θ expected ; Calculate the direction difference Δθ=|θ actual -θ expected |; When cos(Δθ)<α, the data is considered abnormal and filtered;

[0049] Step S302: perform the following steps on the pre-processed real-time data: Figure 4 Filter processing to achieve positioning:

[0050] S3021, the pre-processed RSS sequence {r t} and IMU data {a t ,ω t}, calculate motion vectors through time synchronization, walking detection, direction estimation and data fusion Specifically:

[0051] First, the RSS and IMU data are interpolated or resampled so that they correspond to the same timestamp t. t Identify the walking cycle and determine the start and end time of each step by detecting the peak or zero crossing point of the acceleration signal. Then estimate the stride length:

[0052]

[0053] Among them, k represents the empirical coefficient, which is obtained through experiments or calibration, and a max and a min Indicates the maximum and minimum acceleration in the current step.

[0054] Then use the angular velocity data ω t Calculate the user's heading angle θ t , the direction update formula is:

[0055] θ t =θ t-1 +ω t ·Δt

[0056] Among them, θ t-1 Represents the direction angle at the previous moment, ω t represents the angular velocity data at time t, and Δt represents the sampling time interval.

[0057] According to the stride s t and direction θ t , calculate the displacement increment Δp for each step t The displacement increment formula is as follows:

[0058]

[0059] Finally, the displacement increment Δpt Accumulate and get the current movement direction as follows:

[0060]

[0061] S3022, Status Prediction

[0062] Initialize N s Particles Evenly distributed in the state space; using the PDR algorithm, the position of the particle at the next moment is predicted based on the IMU data:

[0063]

[0064] in, is the velocity estimate of particle i, R(θ i (t-1)) is based on the direction estimate θ of particle i at time t-1 i (t-1), a(t) is the acceleration vector collected by the IMU at time t, and Δt is the sampling frequency interval of the data;

[0065] S3023, particle position correction

[0066] Combined with the preprocessed RSS data, the particle position is corrected; the observation likelihood of the particle is calculated:

[0067]

[0068] S3024, weight update and resampling

[0069] Calculate particle weights:

[0070]

[0071] Where R t is the covariance of the observation noise, represents the updated weight of particle i at time t,

[0072] is the actual observation position;

[0073] Normalized weights:

[0074]

[0075] Resample according to the weight distribution, retain high-weight particles and discard low-weight particles;

[0076] S3025, Position Estimation

[0077] Estimate the current position by weighted averaging:

[0078]

[0079] S3026, Iteration Update

[0080] Repeat steps S3022-S3025, combining real-time RSS and IMU data to gradually approach the user's actual position.

[0081] like Figure 5 The cumulative distribution function (CDF) of different algorithms on location errors is shown in Figure 1, including the experimental results of four algorithms: TMM, TMS, PF-TMS and RMLICT. This experiment aims to compare the performance of different algorithms in reducing positioning errors. As can be seen from the figure, RMLICT has the best effect, indicating that its algorithm can significantly reduce positioning errors in terms of location accuracy, which is better than the other three algorithms. Therefore, RMLICT is suitable for scenarios that require high-precision positioning. PF-TMS and TMM are second, and they can also provide relatively stable accuracy in practical applications, but are slightly inferior to RMLICT.

[0082] The present invention provides an unsupervised indoor positioning method and system. By constructing a fingerprint map from unbalanced crowdsourced data in the mapping stage, optimizing the data preprocessing process in the offline stage, and using an enhanced particle filter algorithm in the online positioning stage, high-precision and real-time indoor positioning is achieved. The method exhibits good robustness and positioning accuracy in complex nonlinear and non-Gaussian environments, has broad application prospects, and can meet the needs of various indoor positioning applications.

Claims

1. An indoor positioning method based on multimodal data fusion of RMLICT, characterized in that: The method includes a mapping phase, an offline phase, and an online positioning phase; The mapping stage is specifically: Step S101, vertex extraction: Obtain historically collected RSS data, cluster the RSS data, extract the cluster center of each cluster from the clustering results and construct the vertex set of the fingerprint map; Step S102: edge set extraction: Obtain historically collected IMU data, extract edge sets between vertices, and verify the physical connectivity between vertices by analyzing the displacement and direction data in the edge sets, thereby capturing the physical spatial change relationship between vertices and establishing the positional relationship between positions; Step S103: constructing an extended fingerprint map based on the extracted vertex set and edge set; The offline phase is specifically: Step S201: trajectory data enhancement: Extract the user's movement trajectory data in the indoor environment based on the extended fingerprint map and perform data enhancement; Step S202, performing data dimension reduction processing on the trajectory data after data enhancement; Step S203: cluster the trajectory data after dimension reduction to obtain K′ groups, and use the cluster center as a representative feature; The online positioning stage is specifically: Step S301: real-time data preprocessing: Collect RSS data and IMU data in real time, and then perform direction matching filtering on the RSS data and IMU data; Step S302: Filter the preprocessed real-time data to achieve positioning.

2. The method according to claim 1, characterized in that Step S102 specifically includes: Let the timestamp corresponding to the i-th vertex be t i , IMU data is {a ti ,ω ti }, where a ti is the acceleration, ω ti is the angular velocity; The PDR is calculated by walking motion, and the displacement Δs and direction θ between two vertices are calculated: Δs=v·Δt θ=θ t-1 +ω·Δt Where v is the walking speed, ω is the angular velocity, and Δt is the time difference; According to the displacement Δs and direction θ, the edge set {E ij }, where E ij Represents the connectivity between vertex i and vertex j.

3. The method according to claim 1, characterized in that The data enhancement in step S201 uses reversal, translation, and rotation operations.

4. The method according to claim 1, characterized in that Step S301 specifically includes: Assume the actual direction of motion is θ actual , the expected direction is θ expected ; Calculate the direction difference Δθ=|θ actual -θ expected |; When cos(Δθ)<α, the data is considered abnormal and is filtered.

5. The method according to claim 1, characterized in that Step S302 specifically includes: S3021, the pre-processed RSS sequence {r t } and IMU data {a t ,ω t }, calculate motion vectors through time synchronization, walking detection, direction estimation and data fusion S3022, Status Prediction Initialize N s Particles Evenly distributed in the state space; using the PDR algorithm, the position of the particle at the next moment is predicted based on the IMU data: in, is the velocity estimate of particle i, R(θ i (t-1)) is based on the direction estimate θ of particle o at time t-1 i (t-1), a(t) is the acceleration vector collected by the IMU at time t, and Δt is the sampling frequency interval of the data; S3023, particle position correction Combined with the preprocessed RSS data, the particle position is corrected; the observation likelihood of the particle is calculated: S3024, weight update and resampling Calculate particle weights: Where R t is the covariance of the observation noise, represents the updated weight of particle i at time t, is the actual observation position; Normalized weights: Resample according to the weight distribution, retain high-weight particles and discard low-weight particles; S3025, Position Estimation Estimate the current position by weighted averaging: S3026, Iteration Update: Repeat steps S3022-S3025, combining real-time RSS and IMU data to gradually approach the user's actual position.

6. An indoor positioning system implementing the method according to any one of claims 1 to 5, characterized in that include: Mapping module, responsible for building the extended fingerprint map; The offline stage module is responsible for extracting the user's movement trajectory data in the indoor environment according to the extended fingerprint map for clustering and feature extraction; The online positioning module is responsible for filtering the real-time collected RSS data and IMU data to achieve positioning.

7. An electronic device, characterized in that: The invention comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the method according to any one of claims 1 to 5.

8. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the method described in any one of claims 1 to 5.

Images