An indoor positioning method and system based on multi-modal data fusion of RMLICT

Abstract

The application provides an indoor positioning method and system based on multi-modal data fusion of RMLICT. First, in the mapping stage, a fingerprint map is constructed from unbalanced crowd-sourced trajectory data. In the offline stage, trajectory data enhancement, dimension reduction and clustering processing are performed. In the online positioning stage, direction matching filtering is performed first, and then enhanced particle filtering is used to establish the mapping relationship between the RSS signal and single-step motion. This method optimizes data preprocessing and uses advanced particle filtering algorithms, which not only improves the positioning accuracy and robustness, but also significantly improves the real-time performance of the system, and is suitable for various complex indoor environments, such as scenes with severe multipath effect and signal obstruction.

Description

Technical Field

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

[0002] Indoor positioning has broad application prospects in fields such as intelligent navigation, security monitoring, and asset management. However, traditional indoor positioning methods, such as those based on the Kalman filter (KF), suffer from performance degradation in complex nonlinear and non-Gaussian environments due to their assumption of a linear Gaussian model. Particle filters (PF), as a non-parametric filter based on the Monte Carlo method, can effectively handle nonlinear and non-Gaussian problems, but still suffer from high computational complexity under high-dimensional state spaces and real-time computing requirements.

[0003] Meanwhile, the construction of fingerprint maps typically relies on high-quality, evenly distributed data. However, in practical applications, obtaining evenly distributed, high-quality data is costly and difficult. Therefore, a method is needed to effectively construct fingerprint maps from imbalanced crowdsourced trajectory data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an indoor positioning method based on RMLICT multimodal data fusion. This method introduces a mapping phase, using the RMLICT method to construct a fingerprint map from imbalanced crowdsourced trajectory data. In the offline phase, trajectory data augmentation, principal component analysis (PCA) dimensionality reduction, and K-means clustering are performed. Finally, in the online phase, an enhanced particle filter algorithm is applied for high-precision real-time positioning.

[0005] Through the above technical solutions, this invention realizes the construction of a high-precision fingerprint map from unbalanced crowdsourced trajectory data, optimizes the data preprocessing process, and adopts an enhanced particle filter algorithm to achieve high-precision, real-time indoor positioning in complex nonlinear and non-Gaussian environments, with good robustness and broad application prospects. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the overall architecture of the indoor positioning system of the present invention.

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

[0008] Figure 3 This is a flowchart for the offline phase.

[0009] Figure 4This is a schematic diagram of the state prediction and update process of a particle filter.

[0010] Figure 5 Experimental results for four algorithms: TMM, TMS, PF-TMS, and the RMLICT algorithm of this invention. Detailed Implementation

[0011] The present invention will be further analyzed below with reference to specific embodiments.

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

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

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

[0015] The mapping module is responsible for building an expanded fingerprint map.

[0016] The offline phase module is responsible for clustering and extracting features from the user's movement trajectory data in the indoor environment based on the extended fingerprint map.

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

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

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

[0020] Step S101, Vertex Extraction:

[0021] Historically collected RSS (Received Signal Strength) data is acquired, and the imbalanced crowdsourced RSS data is clustered using the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) algorithm. The cluster centers of each cluster are extracted from the clustering results, and a vertex set of a fingerprint map is constructed to represent typical RSS features at different locations. HDBSCAN can automatically determine the number of clusters, adapting to changes in data density.

[0022] Let the RSS dataset be Where r iLet {r} be the i-th RSS vector. Using the HDBSCAN algorithm, {r} i Clustering was performed to obtain K cluster centers. These cluster centers serve as the vertex set of the fingerprint map.

[0023] Step S102, Edge Set Extraction:

[0024] By acquiring historically collected IMU data, extracting edge sets between vertices, and analyzing the displacement and orientation data within these edge sets, the physical connectivity between vertices is verified, thereby capturing the physical spatial changes between vertices and establishing positional relationships between locations; specifically:

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

[0026] Calculate the displacement Δs and direction θ of a pedestrian between two vertices using pedestrian dead reckoning (PDR):

[0027] Δs=v·Δt

[0028] θ=θ t-1 +ω·Δt

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

[0030] Based on the displacement Δs and direction θ, establish the edge set {E} between vertices. ij}, where E ij This indicates the connectivity between vertex i and vertex j;

[0031] Step S103, Fingerprint Map Construction:

[0032] Based on the extracted vertex and edge sets, the Multidimensional Scaling (MDS) method is used to map the high-dimensional RSS space to a two-dimensional physical space to construct a two-dimensional fingerprint map that accurately reflects the actual physical space. By mapping reference points, the two-dimensional fingerprint map is mapped to the actual physical space to ensure that the spatial location 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 The coordinates of the vertices in two-dimensional space are obtained 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, such as... Figure 2 Specifically:

[0036] Step S201, Trajectory Data Augmentation:

[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; data augmentation is performed on the movement trajectory data.

[0038] The data augmentation employs inversion, 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 augmented trajectory data is converted into high-dimensional feature vectors, and then principal component analysis (PCA) is used to reduce the dimensionality of the augmented trajectory data to reduce data redundancy and computational complexity.

[0042] Step S203, Clustering Process:

[0043] Clustering is performed on the dimensionality-reduced trajectory data to obtain K′ groups. The cluster centers serve 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] Let the original RSS data be R∈R N×M N is the number of samples, and M is the RSS dimension. Using PCA, R is projected onto a d-dimensional subspace to obtain the dimensionality-reduced data R'∈R. N×d .

[0045] During the online positioning phase, an enhanced particle filter algorithm is employed, combined with real-time collected RSS and IMU data, 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 Then, direction matching filtering is performed on the RSS and IMU data; specifically:

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

[0049] Step S302: Perform the following steps on the preprocessed real-time data: Figure 4 Filtering is used to achieve localization.

[0050] S3021, The preprocessed RSS sequence {r t} and IMU data {a t ,ω t The motion vector is calculated through time synchronization, walking detection, direction estimation, and data fusion. Specifically:

[0051] First, interpolate or resample the RSS and IMU data to make them correspond at the same timestamp t. Then, use the acceleration data {a} t The walking cycle is identified by detecting the peak or zero-crossing point of the acceleration signal to determine the start and end times of each step. This allows for the estimation of stride length.

[0052]

[0053] Where k represents an empirical coefficient, obtained through experimentation or calibration, and a max and a min This represents the maximum and minimum acceleration values ​​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] Where, θ t-1 ω represents the direction angle at the previous moment. t This represents the angular velocity data at time t, where Δt represents the sampling time interval.

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

[0058]

[0059] Finally, the displacement increment Δpt Accumulate to obtain the current motion direction. as follows:

[0060]

[0061] S3022, State Prediction

[0062] Initialize N s Particles The particles are uniformly distributed in the state space; using the PDR algorithm, the position of the particles at the next time step is predicted based on IMU data.

[0063]

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

[0065] S3023, Particle Position Correction

[0066] Based on the preprocessed RSS data, the particle positions are corrected; the observed likelihood of the particles is calculated.

[0067]

[0068] S3024, Weight Update and Resampling

[0069] Calculate particle weights:

[0070]

[0071] Where R t To observe the covariance of the noise, This represents the updated weight of particle i at time t.

[0072] This refers to the actual observation location;

[0073] Normalized weights:

[0074]

[0075] Resampling is performed based on the weight distribution, retaining high-weight particles and discarding low-weight particles;

[0076] S3025, Location Estimation

[0077] Estimate the current position using a weighted average:

[0078]

[0079] S3026, Iterative Update

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

[0081] like Figure 5 The figure shows the cumulative distribution function (CDF) of different algorithms for location errors, specifically 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 performs best, indicating that its algorithm significantly reduces positioning errors in terms of accuracy, outperforming the other three algorithms. Therefore, RMLICT is suitable for scenarios requiring high-precision positioning. PF-TMS and TMM are next best; they also provide relatively stable accuracy in practical applications, but slightly inferior to RMLICT.

[0082] This invention provides an unsupervised indoor positioning method and system. By constructing a fingerprint map from imbalanced crowdsourced data during the mapping phase, optimizing the data preprocessing process offline, and employing an enhanced particle filter algorithm in the online positioning phase, high-precision, real-time indoor positioning is achieved. This 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 RMLICT multimodal data fusion, characterized in that, This method includes a mapping phase, an offline phase, and an online positioning phase; The mapping stage specifically includes: Step S101, Vertex Extraction: Acquire 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: By acquiring historically collected IMU data, extracting edge sets between vertices, and analyzing the displacement and orientation data in the edge sets, the physical connectivity between vertices is verified, thereby capturing the physical spatial change relationship between vertices and establishing the positional relationship between locations. Step S103: Based on the extracted vertex set and edge set, construct an extended fingerprint map; The offline phase specifically refers to: Step S201, Trajectory Data Augmentation: Extract user movement trajectory data in the indoor environment based on the extended fingerprint map and perform data augmentation; Step S202: Perform dimensionality reduction processing on the augmented trajectory data; Step S203: Cluster the dimensionality-reduced trajectory data to obtain... Each group is divided into groups, with the cluster center serving as a representative feature. The online positioning stage specifically includes: Step S301, Real-time data preprocessing: Real-time acquisition of RSS and IMU data, followed by orientation matching and filtering of the RSS and IMU data; Step S302: Filter the preprocessed real-time data to achieve positioning; specifically: S3021. The preprocessed RSS sequence and IMU data Motion vectors are calculated through time synchronization, walking detection, direction estimation, and data fusion. ; S3022, State Prediction initialization Particles The particles are uniformly distributed in the state space; using the PDR algorithm, the position of the particles in the next time step is predicted based on IMU data. in, For particles Speed ​​estimation Based on particles At any moment Direction estimation The rotation matrix, Is it the IMU at any time? The collected acceleration vector, The time interval between data sampling frequencies; S3023, Particle Position Correction Based on the preprocessed RSS data, the particle positions are corrected; the observed likelihood of the particles is calculated. S3024, Weight Update and Resampling Calculate particle weights: in To observe the covariance of the noise, Representative particles At any moment The updated weights This refers to the actual observation location; Normalized weights: Resampling is performed based on the weight distribution, retaining high-weight particles and discarding low-weight particles; S3025, Location Estimation Estimate the current position using a weighted average: S3026, Iterative Update: Repeat steps S3022-S3025, combining real-time RSS and IMU data to gradually approximate the user's actual location.

2. The method according to claim 1, characterized in that, Step S102 is specifically as follows: Let the first The timestamps corresponding to each vertex are IMU data is ,in For acceleration, Angular velocity; The PDR is derived from walking motion, and the displacement between the two vertices is calculated. and direction : in, Walking speed, Angular velocity, For time difference; According to displacement and direction Establish the edge set between vertices ,in Represents vertices and vertex The connectivity between them.

3. The method according to claim 1, characterized in that, The data augmentation described in step S201 employs reverse, translation, and rotation operations.

4. The method according to claim 1, characterized in that, Step S301 specifically involves: Let the actual direction of motion be The expected direction is ; Calculation direction difference ;when If the data is deemed abnormal, it will be filtered out.

5. An indoor positioning system implementing the method of any one of claims 1-4, characterized in that, include: The mapping module is responsible for building an expanded fingerprint map; The offline phase module is responsible for clustering and extracting features from the user's movement trajectory data in the indoor environment based on the extended fingerprint map; The online positioning module is responsible for filtering and processing the real-time collected RSS and IMU data to achieve positioning.

6. An electronic device, characterized in that, It includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the method of any one of claims 1-4.

7. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores machine-executable instructions that, when invoked and executed by a processor, cause the processor to implement the method of any one of claims 1-4.

Images