An ultra-wideband three-dimensional indoor positioning method and device

Abstract

The application discloses a three-dimensional indoor positioning method and device based on ultra-wideband. The method comprises the following steps: obtaining an original ranging sequence between an ultra-wideband node to be positioned and a plurality of ultra-wideband anchor points, wherein the original ranging sequence comprises a plurality of ranging values and first three-dimensional space coordinate values composed of X, Y and Z coordinate values corresponding to the ranging values; performing outlier rejection on the plurality of ranging values according to a preset adaptive anti-disturbance feasible region interval and Z coordinate values to obtain an initial ranging sequence and determine an initial coarse settlement distance; obtaining a prior coarse three-dimensional space coordinate vector of the ultra-wideband node to be positioned through a preset weighted objective function and a limited memory quasi-Newton method with boundary constraints; and inputting the prior coarse three-dimensional space coordinate vector, second three-dimensional space coordinate values of the ultra-wideband anchor points and the initial coarse settlement distance into a model to obtain three-dimensional fine space coordinate values of the ultra-wideband node to be positioned, wherein the model comprises an attention mechanism vector capsule network, a distance perception gated multi-scale graph convolution network and a heterogeneous multi-scale feature fusion network.

Description

Technical Field

[0001] This invention relates to the field of indoor positioning technology, and in particular to an ultra-wideband three-dimensional indoor positioning method and device. Background Technology

[0002] In applications such as Industrial Internet of Things (IIoT), warehousing and logistics, and drone inspection, high-precision 3D positioning is the core foundation for ensuring autonomous equipment operation. Ultra-wideband (UWB) technology, with its strong penetration and resistance to multipath interference, has become the mainstream solution in this field. However, the deployment of UWB positioning anchor points is often limited to top-mounted installation or planar wiring. This quasi-coplanar topology, where base station heights tend to be uniform, makes the system's positioning resolution capability in the height dimension extremely vulnerable. Under such constrained geometric conditions, even centimeter-level interference in the underlying ranging can be structurally amplified, causing drastic meter-level jumps in Z-axis positioning results, potentially leading to system safety incidents or equipment downtime and failure.

[0003] Based on this, existing technologies address the aforementioned shortcomings through multi-source sensor fusion methods and data-driven methods based on traditional neural networks. Specifically, in multi-source sensor fusion schemes, high-precision inertial measurement units and other hardware are typically integrated into ultra-wideband tags. While this method can effectively stabilize altitude output, it significantly increases hardware costs and power consumption. Furthermore, in practical engineering, multi-source time synchronization is prone to errors, making it difficult to widely adopt in low-cost, low-power IoT tags. In contrast, data-driven methods based on traditional neural networks typically treat the X, Y, and Z dimensions of the ultra-wideband 3D coordinates with positioning as independent scalar targets for single fitting, severing the inherent strong geometric coupling between spatial coordinates. This black-box mapping, lacking a physical mechanism, struggles to extract explicit spatial feature representations when dealing with the nonlinear error amplification caused by high vertical geometric accuracy factors. Once there is a slight drift in the distribution of ambient noise, the model performance can easily degrade significantly or even fail, lacking industrial-grade generalization and robustness to cope with complex and variable working conditions, and failing to obtain highly accurate ultra-wideband 3D indoor positioning results. Summary of the Invention

[0004] The purpose of this invention is to provide an ultra-wideband three-dimensional indoor positioning method and apparatus to solve the problems of high cost, high power consumption and poor accuracy in the prior art when acquiring ultra-wideband three-dimensional indoor positioning.

[0005] To achieve the above objectives, in a first aspect, embodiments of the present invention provide an ultra-wideband three-dimensional indoor positioning method, comprising: Obtain the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value. Based on the preset adaptive anti-disturbance feasible region and the Z coordinate value, outliers are removed from multiple ranging values ​​to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points are determined respectively. The preset adaptive anti-disturbance feasible region is determined by the interquartile range algorithm based on multiple ranging values. For the original three-dimensional spatial coordinates of the ultra-wideband node to be located, the prior coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinates is obtained by using a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints; wherein, the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region; The prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance are input into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate value of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: attention mechanism vector capsule network, distance-aware gated multi-scale graph convolutional network, and heterogeneous multi-scale feature fusion network.

[0006] In one embodiment, the preset adaptive disturbance rejection feasible region is determined based on multiple ranging values ​​using a quartile distance algorithm, including: For each ultrawideband anchor point, multiple ranging values ​​are arranged in ascending order to determine the first quartile of the ranging value at position M and the third quartile of the ranging value at position N. Based on the first quartile, the third quartile, and the adaptive expansion coefficient, a preset adaptive disturbance rejection feasible region is determined.

[0007] In one embodiment, determining the preset adaptive disturbance rejection feasible region interval based on the first quartile, the third quartile, and the adaptive expansion coefficient includes: Calculate the difference between the third quartile and the first quartile to determine the interquartile range; Based on the first quartile, the adaptive expansion coefficient, and the interquartile range, determine the uplink and downlink thresholds; Based on the uplink and downlink thresholds, a preset adaptive disturbance rejection feasible region is constructed, which is defined by the following expression: ; in, Indicates the downlink range threshold. Indicates the uplink range threshold. Indicates the first quartile. Indicates the third quartile. This represents the adaptive expansion coefficient.

[0008] In one embodiment, the step of removing outliers from multiple ranging values ​​based on a preset adaptive anti-disturbance feasible region and the Z-coordinate value to obtain an initial ranging sequence includes: Determine whether the Z coordinate value is greater than or equal to the downlink region threshold and less than or equal to the uplink region threshold; If not, the ranging value corresponding to the Z coordinate value is determined to be an abnormal ranging value. The abnormal value is removed to obtain the initial ranging sequence.

[0009] In one embodiment, the original three-dimensional spatial coordinate values ​​include: original X coordinate values, original Y coordinate values, and original Z coordinate values. The process of obtaining the prior coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinate values ​​of the ultra-wideband node to be located, using a preset weighted objective function and a finite-memory quasi-Newton method with boundary constraints, includes: Within the feasible region constrained by the box, a pre-defined weighted objective function is constructed based on the soft constraint of layer height. For the original Z coordinate value, the prior coarse Z coordinate value is determined by minimizing the preset weighted objective function of the residual; Given the original X and Y coordinate values, a priori coarse X and Y coordinate values ​​are obtained using a finite-memory quasi-Newton method with boundary constraints and a preset coordinate estimation formula. Based on the prior coarse Z-coordinate value, prior coarse X-coordinate value, and prior coarse Y-coordinate value, the prior coarse three-dimensional space coordinate vector is obtained by splicing through the orthogonal basis space recombination operator.

[0010] In one embodiment, the step of inputting the prior coarse three-dimensional spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the three-dimensional fine spatial coordinate value positioning model to obtain the three-dimensional fine spatial coordinate value of the ultra-wideband node to be located includes: The prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point are input into the attention mechanism vector capsule network to obtain the three-axis independent vector capsule features of the ultra-wideband node to be located. The triaxial independent vector capsule features and multiple initial coarse calculation distances are input into a distance-aware gated multi-scale graph convolutional network to obtain distance-aware gated matrix features; wherein, the distance-aware gated multi-scale graph convolutional network includes: a graph convolution module and a distance-aware gated module; The distance-aware gating matrix features are input into a heterogeneous multi-scale feature fusion network to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located.

[0011] In one embodiment, the attention mechanism vector capsule network includes: a four-dimensional spatial projection module, a multi-head self-attention mechanism module, a nonlinear fully connected perceptron module, and an independent feature capsule extraction module. The step of inputting the prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point into the attention mechanism vector capsule network to obtain the three-axis independent vector capsule features of the ultra-wideband node to be located includes: The prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point are input into the four-dimensional spatial projection module. The spatial projection residual is processed by the four-dimensional spatial projection module to obtain four-dimensional local geometric features. The four-dimensional local geometric features are input into the multi-head self-attention mechanism module, and the multi-head self-attention mechanism module extracts attention features from the four-dimensional local geometric features to obtain multi-head self-attention features. The multi-head self-attention features are input into the nonlinear fully connected perceptron module, and the multi-head self-attention features are processed by the nonlinear fully connected perceptron module to obtain fully connected perceptron features. The fully connected sensing features are input into the independent feature capsule extraction module. The independent feature capsule extraction module performs independent feature extraction processing on each coordinate axis of the fully connected sensing features and performs activation processing through a nonlinear activation function to obtain three-axis independent vector capsule features.

[0012] In one embodiment, the distance-aware gated multi-scale graph convolutional network includes: a graph convolution module, a distance encoding module, and a distance-aware gated module. The step of inputting the triaxial independent vector capsule features and multiple initial coarse calculation distances into the distance-aware gated multi-scale graph convolutional network to obtain distance-aware gated matrix features includes: The triaxial independent vector capsule features are input into the graph convolution module, and the graph convolution module performs isotropic aggregation processing on the triaxial independent vector capsule features to obtain normalized Laplacian matrix features. Multiple initial coarse calculation distances are input into the distance encoding module, and distance encoding processing is performed by the distance encoding module to obtain multiple dynamic decay attention matrices; The normalized Laplacian matrix features and multiple dynamic decaying attention matrices are input into the distance-aware gating module, and gating processing is performed by the distance-aware gating module to obtain the distance-aware gating matrix features.

[0013] In one embodiment, the heterogeneous multi-scale feature fusion network includes: three parallel-connected graph convolutional network modules of different scales, a stitching module, and a residual fully connected regression head. The step of inputting the distance-aware gating matrix features into the heterogeneous multi-scale feature fusion network to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located includes: The distance-aware gated matrix features are input into three parallel graph convolutional network modules of different scales to obtain the first scale features, the second scale features, and the third scale features of three different scales. The first scale feature, the second scale feature, and the third scale feature are input into the stitching module. The stitching module performs stitching processing on the first scale feature, the second scale feature, and the third scale feature to obtain multi-scale stitched features. The multi-scale stitched features are input into the residual fully connected regression head, and the multi-scale stitched features are integrated through the residual fully connected regression head to obtain three-dimensional fine spatial coordinate values.

[0014] Secondly, embodiments of the present invention provide an ultra-wideband three-dimensional indoor positioning device, comprising: The original ranging sequence acquisition module is used to acquire the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value. The initial ranging acquisition module is used to remove outliers from multiple ranging values ​​based on a preset adaptive anti-disturbance feasible region and Z-coordinate values ​​to obtain an initial ranging sequence, and to determine the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points based on the initial ranging sequence; wherein, the preset adaptive anti-disturbance feasible region is determined based on multiple ranging values ​​using the interquartile range algorithm; The prior coarse 3D spatial coordinate vector acquisition module is used to obtain the prior coarse 3D spatial coordinate vector corresponding to the original 3D spatial coordinate values ​​of the ultra-wideband node to be located by means of a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints; wherein, the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region; The 3D fine spatial coordinate value acquisition module is used to input the prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate values ​​of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: an attention mechanism vector capsule network, a distance-aware gated multi-scale graph convolutional network, and a heterogeneous multi-scale feature fusion network. The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: This invention provides an ultra-wideband 3D indoor positioning method that obtains original ranging sequences between an ultra-wideband node to be positioned and multiple ultra-wideband anchor points. Each original ranging sequence includes multiple ranging values ​​and a first 3D spatial coordinate value corresponding to each ranging value, wherein the first 3D spatial coordinate value consists of an X-coordinate value, a Y-coordinate value, and a Z-coordinate value. Based on a preset adaptive anti-disturbance feasible region and the Z-coordinate value, outliers are removed from the multiple ranging values ​​to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances between the ultra-wideband node to be positioned and the multiple ultra-wideband anchor points are determined. The preset adaptive anti-disturbance feasible region is determined based on the multiple ranging values ​​using a quartile distance algorithm. This outlier removal based on the Z-coordinate value avoids the problem of drastic jumps and easy divergence in the Z-axis direction due to the quasi-coplanar topology of the base station. For the original 3D spatial coordinates of the ultra-wideband node to be located, a priori coarse 3D spatial coordinate vector is obtained by using a pre-defined weighted objective function and a finite-memory quasi-Newton method with boundary constraints. Since the pre-defined weighted objective function is constructed based on layer-height soft constraints within the box-constrained feasible region, it addresses the extremely flat non-convex characteristic of the Z-axis direction under the condition of weak vertical geometric accuracy factor when the ultra-wideband anchor points are quasi-coplanar. Directly solving the 3D spatial coordinates would easily lead to local minima, resulting in a highly avalanche-like divergence problem. Finally, the priori coarse 3D spatial coordinate vector, the second-dimensional spatial coordinates of each ultra-wideband anchor point, and the initial coarse calculation distance are input into the 3D fine spatial coordinate localization model to obtain the 3D fine spatial coordinates of the ultra-wideband node to be located. The 3D fine spatial coordinate localization model includes: an attention-based vector capsule network, a distance-aware gated multi-scale graph convolutional network, and a heterogeneous multi-scale feature fusion network. In this way, there is no need to integrate high-precision inertial measurement units and other hardware into the ultra-wideband tag, which reduces costs and power consumption. It also avoids the problem in existing technologies where the model treats the three-dimensional coordinate values ​​as independent scalar targets for single fitting, which severs the inherent strong geometric coupling relationship between spatial coordinates. This improves the accuracy of obtaining more precise three-dimensional spatial coordinate values. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 A flowchart illustrating an ultra-wideband three-dimensional indoor positioning method provided in an embodiment of the present invention; Figure 2 A schematic diagram of a three-dimensional fine spatial coordinate value positioning model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an ultra-wideband three-dimensional indoor positioning device provided in an embodiment of the present invention. Detailed Implementation

[0016] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0017] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0018] In this invention, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between the associated objects, indicating that three relationships can exist.

[0019] like Figure 1 As shown, Figure 1 This is a flowchart illustrating an ultra-wideband three-dimensional indoor positioning method provided in an embodiment of the present invention, which specifically includes the following steps: S10: Obtain the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively.

[0020] Each of the original ranging sequences includes multiple ranging values, which are obtained by measuring distances within a high-frequency sliding time window W of length T maintained for the communication link of the ultra-wideband anchor point. Based on this, the first three-dimensional spatial coordinate value corresponding to each ranging value is the three-dimensional spatial coordinate value of each high-frequency sliding time window W when collecting each ranging value. The first three-dimensional spatial coordinate value consists of X coordinate value, Y coordinate value and Z coordinate value.

[0021] Specifically, for the ultra-wideband node to be located and multiple ultra-wideband anchor points, the original ranging sequences corresponding to them are obtained. Each original ranging sequence includes: multiple ranging values ​​and the first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value.

[0022] S11: Based on the preset adaptive anti-disturbance feasible domain interval and Z coordinate value, outlier values ​​are removed from multiple ranging values ​​to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points are determined respectively.

[0023] The initial coarse calculation distance is used to provide a clean data base for subsequent high-precision calculations.

[0024] The preset adaptive disturbance rejection feasible region is determined based on multiple ranging values ​​using a quartile distance algorithm. This preset adaptive disturbance rejection feasible region can construct a dynamic, real-time security trust threshold for the underlying ranging data. In complex industrial environments, traditional fixed filtering thresholds are prone to failure due to environmental changes; however, the preset adaptive disturbance rejection feasible region can automatically scale according to the real-time distribution of the ambient noise, thereby accurately isolating the actual line-of-sight ranging signal from non-line-of-sight burst pulse interference caused by metal obstruction or multipath effects.

[0025] Optionally, based on the above embodiments, in some embodiments of the present invention, one way to obtain the preset adaptive anti-disturbance feasible region interval may be: S20: For each ultra-wideband anchor point, the multiple ranging values ​​are arranged in ascending order to determine the first quartile of the ranging value at position M and the third quartile of the ranging value at position N.

[0026] Where M is less than N, for example, M is the 25th of the total number of multiple ranging values ​​and N is the 75th of the total number of multiple ranging values, but not limited thereto. This invention is not specifically limited, and those skilled in the art can set it according to the actual situation.

[0027] Specifically, for the multiple ranging values ​​between the ultra-wideband node to be located and each ultra-wideband anchor point, the multiple ranging values ​​are arranged in ascending order, and the ranging value at the Mth position is taken as the first quartile, and the ranging value at the Nth position is taken as the third quartile.

[0028] S21: Determine the preset adaptive anti-disturbance feasible region interval based on the first quartile, the third quartile, and the adaptive expansion coefficient.

[0029] Optionally, based on the above embodiments, in some embodiments of the present invention, one implementation of S21 may be: S211: Calculate the difference between the third quartile and the first quartile to determine the interquartile range.

[0030] S212: Determine the uplink region threshold and the downlink region threshold based on the first quartile, the adaptive expansion coefficient, and the interquartile range.

[0031] S213: Construct a preset adaptive disturbance-resistant feasible region based on the uplink region threshold and the downlink region threshold.

[0032] Specifically, the difference between the third quartile and the first quartile is calculated, and the difference is used as the interquartile range. After obtaining the interquartile range, the uplink and downlink thresholds are determined based on the first quartile, the adaptive expansion coefficient, and the interquartile range. Finally, a preset adaptive anti-disturbance feasible region is constructed based on the uplink and downlink thresholds.

[0033] Optionally, based on the above embodiments, in some embodiments of the present invention, the preset adaptive disturbance rejection feasible region interval is defined by the following expression: ; in, Indicates the downlink range threshold. Indicates the uplink range threshold. Indicates the first quartile. Indicates the third quartile. This represents the adaptive spreading coefficient. The value of the adaptive spreading coefficient can be dynamically adjusted according to the signal-to-noise ratio of each ultra-wideband anchor signal, thereby achieving precise suppression of severe jump noise.

[0034] Specifically, based on the preset adaptive anti-disturbance feasible domain range and Z coordinate value, outliers are removed from multiple ranging values ​​between the ultra-wideband node to be located and each ultra-wideband anchor point to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances between the ultra-wideband node to be located and the multiple ultra-wideband anchor points are determined respectively.

[0035] Optionally, based on the above embodiments, since the first three-dimensional spatial coordinate value corresponding to each high-frequency sliding time window W is relatively stable in the horizontal X-axis and Y-axis directions, but prone to drastic changes in the Z-axis direction due to the quasi-coplanar topology of the base station, in some embodiments of the present invention, one way to obtain the initial ranging sequence by removing outliers from multiple ranging values ​​according to the preset adaptive anti-disturbance feasible region and the Z-coordinate value is as follows: S111: Determine whether the Z coordinate value is greater than or equal to the downlink region threshold and less than or equal to the uplink region threshold.

[0036] S112: If not, then determine that the ranging value corresponding to the Z coordinate value is an abnormal ranging value, remove the abnormal value from the ranging value, and obtain the initial ranging sequence.

[0037] Specifically, each Z-coordinate value is compared with the downlink and uplink thresholds to determine whether the Z-coordinate value is greater than or equal to the downlink threshold and less than or equal to the uplink threshold, i.e., whether the Z-coordinate value is within the preset adaptive anti-disturbance feasible region. If the Z-coordinate value is not greater than or equal to the downlink threshold, or if the Z-coordinate value is greater than the uplink threshold, i.e., the Z-coordinate value is not within the preset adaptive anti-disturbance feasible region, it indicates that the ranging value corresponding to the Z-coordinate value is an abnormal ranging value. This abnormal value is then removed, thereby achieving abnormal value removal processing for multiple ranging values ​​to obtain the initial ranging sequence.

[0038] Optionally, based on the above embodiments, in some embodiments of the present invention, the initial ranging sequence includes: multiple initial ranging values ​​after outlier removal from multiple sequencing values. Based on the initial ranging sequence, one implementation of determining the initial coarse settlement distances between the ultra-wideband node to be located and the multiple ultra-wideband anchor points can be: by performing smooth expectation calculation or mean pooling on the multiple initial ranging values ​​to obtain the initial coarse settlement distances between the ultra-wideband node to be located and the multiple ultra-wideband anchor points.

[0039] S12: For the original three-dimensional spatial coordinate values ​​of the ultra-wideband node to be located, the prior coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinate values ​​is obtained by using a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints.

[0040] The original three-dimensional spatial coordinate values ​​include: the original X coordinate value, the original Y coordinate value, and the original Z coordinate value.

[0041] The pre-defined weighted objective function is constructed based on soft constraints of layer height within the box-constrained feasible region. Soft constraints based on layer height within the box-constrained feasible region can be used to address the extremely flat, non-convex characteristics of the Z-axis (height) direction under the condition of weak Vertical Dilution of Precision (VDOP) when the anchor points are quasi-coplanar in ultra-wideband applications. Directly performing a three-dimensional joint solution can easily lead to local minima, resulting in a height avalanche divergence problem. The box-constrained feasible region is defined by the following expression: ,in, This represents the original three-dimensional spatial coordinate vector of the ultra-wideband node to be located. and These represent the lower bound vector (e.g., the ground) and upper bound vector (e.g., the height limited by the ceiling) of the coordinates that the ultra-wideband node to be located can reach in the actual physical space.

[0042] Optionally, based on the above embodiments, in some embodiments of the present invention, S12 may be implemented as follows: S121: Construct a pre-defined weighted objective function based on the layer height soft constraint within the feasible region of the box constraint.

[0043] The preset weighted objective function is defined by the following expression: ; in, This represents the vertical projection matrix. The purpose of the vertical projection matrix is ​​to extract and isolate the Z-axis (height) component from the preset weighted objective function so that an independent soft constraint penalty can be applied to it. This represents the Tikhonov regularization parameter, used to dynamically adjust the weight ratio between the ranging error penalty term and the floor height soft constraint term. The geometric adaptive weight of the i-th ultrawideband anchor point is calculated by coupling the reciprocal of the distance measurement value of the i-th ultrawideband anchor point with its relative height difference. High-quality anchor points that are closer in distance and have a greater physical height difference will be given a larger weight, thereby enhancing the dominant role of effective geometric features in the optimization process.

[0044] S122: For the original Z coordinate value, determine the prior coarse Z coordinate value by minimizing the preset weighted objective function of the residual.

[0045] Specifically, after constructing a pre-weighted objective function based on soft constraints of layer height within the box-constrained feasible region, the Z-axis is extracted as an independent search variable to address the high-dimensional divergence characteristics of the Z-coordinate values. A discrete layer height manifold space is defined. First, a coarse global search is performed with a large step size (0.1m) and a large soft constraint radius to locate the neighborhood of the minimum value. Then, a fine search for the extreme value is performed with a small step size (0.02m) to minimize the residual. Determine the prior coarse Z-coordinate value .

[0046] Optionally, based on the above embodiments, in some embodiments of the present invention, the prior coarse Z-coordinate value can be determined according to the formula. It is confirmed that, among them, Represents the high manifold space of discrete layers The height of a candidate layer is selected from the data. This indicates that the Z-axis height is assumed to be locked. The state of the weighted objective function is preset under the premise. This means that after the Z-axis is locked, only the original X-coordinate value and the original Y-coordinate value are allowed to change freely in order to find the minimum error at this time.

[0047] S123: For the original X coordinate value and the original Y coordinate value, the prior coarse X coordinate value and the prior coarse Y coordinate value are obtained by using the finite memory quasi-Newton method with boundary constraints and the preset coordinate estimation formula.

[0048] Optionally, based on the above embodiments, in some embodiments of the present invention, the preset coordinate estimation formula is defined by the following expression: ; in, This indicates the initial coarse calculation distance.

[0049] S124: Based on the prior coarse Z-coordinate value, prior coarse X-coordinate value, and prior coarse Y-coordinate value, the prior coarse three-dimensional space coordinate vector is obtained by splicing through the orthogonal basis space recombination operator.

[0050] Among them, the orthogonal basis space recombination operator is used to rigidly combine the unbiased horizontal coordinates obtained from the dimensionality reduction solution with the controlled discrete vertical coordinates in three-dimensional space. This provides an absolutely non-divergent physical coordinate system for subsequent feature extraction based on the three-dimensional fine spatial coordinate value localization model, thereby improving the accuracy of the three-dimensional fine spatial coordinate values ​​of the ultra-wideband node to be located. Specifically, after obtaining the prior coarse Z-coordinate, prior coarse X-coordinate, and prior coarse Y-coordinate values ​​of the ultra-wideband node to be located, the prior coarse Z-coordinate, prior coarse X-coordinate, and prior coarse Y-coordinate values ​​are spliced ​​together using an orthogonal basis space recombination operator to obtain a prior coarse three-dimensional space coordinate vector.

[0051] Optionally, based on the above embodiments, in some embodiments of the present invention, the prior coarse three-dimensional spatial coordinate vector is defined by the following expression: ; in, The feature selection matrix represents the horizontal plane and is used to extract and constrain the physical dimension of the two-dimensional horizontal space. These represent the absolutely unbiased, finely estimated X-axis and Y-axis coordinates obtained by completely releasing the horizontal plane search degrees of freedom and using a continuous optimization algorithm after removing the Z-axis divergence interference. This represents the feature selection matrix in the vertical direction, used to isolate features in the height dimension.

[0052] S13: Input the prior coarse three-dimensional spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the three-dimensional fine spatial coordinate value positioning model to obtain the three-dimensional fine spatial coordinate value of the ultra-wideband node to be located.

[0053] Optionally, based on the above embodiments, refer to Figure 2 As shown, the 3D fine-grained spatial coordinate localization model includes: an attention-based vector capsule network 10, a distance-aware gated multi-scale graph convolutional network 20, and a heterogeneous multi-scale feature fusion network 30. The attention-based vector capsule network 10 is used to acquire the three-axis independent vector capsule features of the ultra-wideband node to be localized. This completely breaks the entanglement between scalar features, achieving perfect decoupling of spatial features. The distance-aware gated multi-scale graph convolutional network 20 is used to acquire distance-aware gated matrix features. Through this method, it adaptively isolates far-field noise using the topological information transmission process, focusing on high-quality geometric associations to achieve spatial topological associations between samples and perform final high-precision residual correction. The heterogeneous multi-scale feature fusion network 30 is used to acquire the three-dimensional fine spatial coordinates of the ultra-wideband node to be located. Based on this, in some embodiments of the present invention, one implementation of S13 can be: S131: Input the prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point into the attention mechanism vector capsule network to obtain the three-axis independent vector capsule features of the ultra-wideband node to be located.

[0054] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the attention mechanism vector capsule network 10 includes: a four-dimensional space projection module, a multi-head self-attention mechanism module, a nonlinear fully connected perceptron module, and an independent feature capsule extraction module. Based on this, in some embodiments of the present invention, one implementation of S131 can be: S1311: Input the prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point into the four-dimensional spatial projection module, and perform spatial projection residual processing through the four-dimensional spatial projection module to obtain four-dimensional local geometric features.

[0055] Specifically, the prior coarse 3D spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point are input into the 4D spatial projection module. The 4D spatial projection module calculates the spatial projection residuals in each dimension to obtain the 4D local geometric features. .

[0056] S1312: Input the four-dimensional local geometric features into the multi-head self-attention mechanism module, and extract attention features from the four-dimensional local geometric features through the multi-head self-attention mechanism module to obtain multi-head self-attention features.

[0057] Specifically, four-dimensional local geometric features The input is fed into the multi-head self-attention mechanism module to extract high-dimensional hidden features. Subsequently, a multi-head self-attention mechanism is introduced to obtain global topological dependencies, generating multi-head self-attention features containing context-aware information. .

[0058] In this way, the limitation of isolated observation of a single ultrawideband anchor point is broken through the multi-head self-attention mechanism module, which forces the model to cross-validate the topological dependencies of multiple ultrawideband anchor points in a global view, and generates comprehensive features with context-aware capabilities.

[0059] S1313: Input the multi-head self-attention features into the nonlinear fully connected perceptron module, and perform fully connected processing on the multi-head self-attention features through the nonlinear fully connected perceptron module to obtain fully connected perceptron features.

[0060] Specifically, the multi-head self-attention features are input into the nonlinear fully connected perceptron module. Using the original ranging matrix determined based on the original ranging sequence, the nonlinear fully connected perceptron generates geometric precision factor sensing weights. Using this geometric precision factor to perceive weights Multi-head self-attention characteristics Perform Hadamard product element-wise recalibration to obtain fully connected perceptual features: .

[0061] S1314: Input the fully connected sensing features into the independent feature capsule extraction module, and perform independent feature extraction processing on each coordinate axis of the fully connected sensing features through the independent feature capsule extraction module, and perform activation processing through a nonlinear activation function to obtain three-axis independent vector capsule features.

[0062] Specifically, the fully connected sensing features are input into the independent feature capsule extraction module. The independent feature capsule extraction module performs independent feature extraction processing on the fully connected sensing features for each coordinate axis, obtaining independent intermediate feature variables along the three orthogonal axes X, Y, and Z. , The triaxial independent vector capsule features are obtained by performing activation processing using a nonlinear activation function. .

[0063] Optionally, based on the above embodiments, in some embodiments of the present invention, the triaxial independent vector capsule feature is defined by the following expression: ; S132: Input the triaxial independent vector capsule features and multiple initial coarse calculation distances into a distance-aware gated multi-scale graph convolutional network to obtain distance-aware gated matrix features.

[0064] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the distance-aware gated multi-scale graph convolutional network 20 includes: a graph convolution module, a distance encoding module, and a distance-aware gate module. Based on this, in some embodiments of the present invention, one implementation of S132 can be: S1321: Input the triaxial independent vector capsule features into the graph convolution module, and perform isotropic aggregation processing on the triaxial independent vector capsule features through the graph convolution module to obtain the normalized Laplacian matrix features.

[0065] Specifically, the triaxial independent vector capsule features are input into the graph convolution module. Based on the manifold assumption that triaxial independent vector capsule features with similar geometric states have an adjacency relationship in the feature space, the graph convolution module employs... The -NN algorithm dynamically calculates Euclidean distance similarity in the feature space, searches for neighboring triaxial independent vector capsule features, and establishes connecting edges to form a static topological graph. Further, to achieve the transformation from the static topological graph to a computer-analyzable algebraic tensor, it calculates and extracts the normalized Laplacian matrix features corresponding to this static topological graph.

[0066] S1322: Input multiple initial coarse calculation distances into the distance encoding module, and perform distance encoding processing through the distance encoding module to obtain multiple dynamic decay attention matrices.

[0067] Specifically, multiple initial coarse calculation distances are input into the distance encoding module, and the distance encoding module performs distance encoding processing on the multiple initial coarse calculation distances to obtain multiple dynamic decay attention matrices.

[0068] S1323: Input the normalized Laplacian matrix features and multiple dynamic decaying attention matrices into the distance-aware gating module, and perform gating processing through the distance-aware gating module to obtain the distance-aware gating matrix features.

[0069] Specifically, the obtained normalized Laplacian matrix features and multiple dynamic decaying attention matrices are input into the distance-aware gating module. The distance-aware gating module performs gating processing on the normalized Laplacian matrix features and multiple dynamic decaying attention matrices to obtain the distance-aware gating matrix features.

[0070] S133: Input the distance-aware gating matrix features into the heterogeneous multi-scale feature fusion network to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located.

[0071] Optionally, based on the above embodiments, continue to refer to... Figure 2 As shown, the heterogeneous multi-scale feature fusion network 30 includes: three graph convolutional network modules of different scales connected in parallel, a splicing module, and a residual fully connected regression head. For example, the three graph convolutional network modules of different scales can be: a 64-dimensional graph convolutional network module, a 32-dimensional graph convolutional network module, and a 16-dimensional graph convolutional network module. With three graph convolutional network modules of different scales, the small-scale graph convolutional network module captures fine-grained high-frequency fluctuation errors in the near field, while the large-scale graph convolutional network module captures low-frequency global system drift in the far field. Based on this, in some embodiments of the present invention, one implementation of S133 can be: S1331: Input the distance-aware gating matrix features into three parallel graph convolutional network modules of different scales to obtain the first scale features, the second scale features, and the third scale features of three different scales.

[0072] Specifically, the distance-aware gated matrix features are input into three parallel graph convolutional network modules of different scales. The distance-aware gated matrix features are then subjected to graph convolution operations by the three graph convolutional network modules of different scales to obtain the first-scale features, second-scale features, and third-scale features of three different scales.

[0073] S1332: Input the first scale feature, the second scale feature and the third scale feature into the stitching module, and perform stitching processing on the first scale feature, the second scale feature and the third scale feature through the stitching module to obtain multi-scale stitched features.

[0074] Specifically, the obtained first-scale features, second-scale features, and third-scale features are input into the stitching module, and the stitching module performs stitching processing on the first-scale features, second-scale features, and third-scale features to obtain multi-scale stitched features.

[0075] S1333: Input the multi-scale stitching features into the residual fully connected regression head, and integrate the multi-scale stitching features through the residual fully connected regression head to obtain three-dimensional fine spatial coordinate values.

[0076] Specifically, the multi-scale stitched features are input into the residual fully connected regression head, which integrates the multi-scale stitched features to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located.

[0077] Thus, the ultra-wideband 3D indoor positioning method provided in this embodiment obtains the original ranging sequences corresponding to the ultra-wideband node to be positioned and multiple ultra-wideband anchor points, respectively. Each original ranging sequence includes: multiple ranging values ​​and a first 3D spatial coordinate value corresponding to each ranging value. The first 3D spatial coordinate value consists of an X-coordinate value, a Y-coordinate value, and a Z-coordinate value. Based on a preset adaptive anti-disturbance feasible region and the Z-coordinate value, outliers are removed from the multiple ranging values ​​to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances corresponding to the ultra-wideband node to be positioned and the multiple ultra-wideband anchor points are determined. The preset adaptive anti-disturbance feasible region is determined based on the multiple ranging values ​​using the interquartile range algorithm. In this way, outlier removal from multiple ranging values ​​based on the Z-coordinate value avoids the problem of drastic jumps and easy divergence in the Z-axis direction due to the quasi-coplanar topology of the base station. For the original 3D spatial coordinates of the ultra-wideband node to be located, a priori coarse 3D spatial coordinate vector is obtained by using a pre-defined weighted objective function and a finite-memory quasi-Newton method with boundary constraints. Since the pre-defined weighted objective function is constructed based on layer-height soft constraints within the box-constrained feasible region, it addresses the extremely flat non-convex characteristic of the Z-axis direction under the condition of weak vertical geometric accuracy factor when the ultra-wideband anchor points are quasi-coplanar. Directly solving the 3D spatial coordinates would easily lead to local minima, resulting in a highly avalanche-like divergence problem. Finally, the priori coarse 3D spatial coordinate vector, the second-dimensional spatial coordinates of each ultra-wideband anchor point, and the initial coarse calculation distance are input into the 3D fine spatial coordinate localization model to obtain the 3D fine spatial coordinates of the ultra-wideband node to be located. The 3D fine spatial coordinate localization model includes: an attention-based vector capsule network, a distance-aware gated multi-scale graph convolutional network, and a heterogeneous multi-scale feature fusion network. In this way, there is no need to integrate high-precision inertial measurement units and other hardware into the ultra-wideband tag, which reduces costs and power consumption. It also avoids the problem in existing technologies where the model treats the three-dimensional coordinate values ​​as independent scalar targets for single fitting, which severs the inherent strong geometric coupling relationship between spatial coordinates. This improves the accuracy of obtaining more precise three-dimensional spatial coordinate values.

[0078] In one embodiment, such as Figure 3 As shown, Figure 3The schematic diagram of an ultra-wideband three-dimensional indoor positioning device provided in an embodiment of the present invention includes: an original ranging sequence acquisition module 10, an initial ranging acquisition module 11, a priori coarse three-dimensional spatial coordinate vector acquisition module 12, and a three-dimensional fine spatial coordinate value acquisition module 13.

[0079] The original ranging sequence acquisition module 10 is used to acquire the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value.

[0080] The initial ranging acquisition module 11 is used to remove outliers from multiple ranging values ​​based on a preset adaptive anti-disturbance feasible region interval and Z coordinate value to obtain an initial ranging sequence, and to determine the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points based on the initial ranging sequence; wherein, the preset adaptive anti-disturbance feasible region interval is determined by the interquartile range algorithm based on multiple ranging values.

[0081] The prior coarse three-dimensional spatial coordinate vector acquisition module 12 is used to obtain the prior coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinate values ​​of the ultra-wideband node to be located by means of a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints; wherein, the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region.

[0082] The 3D fine spatial coordinate value acquisition module 13 is used to input the prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate value of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: attention mechanism vector capsule network, distance-aware gated multi-scale graph convolutional network, and heterogeneous multi-scale feature fusion network.

[0083] In the above embodiments, the original ranging sequence acquisition module acquires the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points, respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value, wherein the first three-dimensional spatial coordinate value is composed of an X-coordinate value, a Y-coordinate value, and a Z-coordinate value. The initial ranging acquisition module removes outliers from the multiple ranging values ​​according to a preset adaptive anti-disturbance feasible region interval and the Z-coordinate value to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances corresponding to the ultra-wideband node to be located and the multiple ultra-wideband anchor points are determined respectively. The preset adaptive anti-disturbance feasible region interval is determined by the interquartile range algorithm based on the multiple ranging values. In this way, outlier removal from multiple ranging values ​​based on the Z-coordinate value avoids the problem of drastic jumps and easy divergence in the Z-axis direction due to the quasi-coplanar topology of the base station. The prior coarse 3D spatial coordinate vector acquisition module obtains the prior coarse 3D spatial coordinate vector corresponding to the original 3D spatial coordinate values ​​of the ultra-wideband node to be located by using a preset weighted objective function and a finite-memory quasi-Newton method with boundary constraints. Since the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region, it addresses the extremely flat non-convex characteristic of the Z-axis direction under the condition of weak vertical geometric accuracy factor when the ultra-wideband anchor points are quasi-coplanar. If the 3D spatial coordinate values ​​are directly solved jointly, it is easy to get trapped in local minima, which will lead to a high degree of avalanche divergence. The 3D fine spatial coordinate value acquisition module inputs the prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate value of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: an attention mechanism vector capsule network, a distance-aware gated multi-scale graph convolutional network, and a heterogeneous multi-scale feature fusion network. In this way, there is no need to integrate high-precision inertial measurement units and other hardware into the ultra-wideband tag, which reduces costs and power consumption. It also avoids the problem in existing technologies where the model treats the three-dimensional coordinate values ​​as independent scalar targets for single fitting, which severs the inherent strong geometric coupling relationship between spatial coordinates. This improves the accuracy of obtaining more precise three-dimensional spatial coordinate values.

[0084] Specific limitations regarding the ultra-wideband 3D indoor positioning device can be found in the limitations of the ultra-wideband 3D indoor positioning method described above, and will not be repeated here. Each module in the aforementioned server can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the corresponding operations of each module.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static random access memory (SRAM) and dynamic random access memory (DRAM), etc.

[0086] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0087] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely exemplary descriptions of the invention as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A method for ultra-wideband three-dimensional indoor positioning, characterized in that, include: Obtain the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value. Based on the preset adaptive anti-disturbance feasible region and the Z coordinate value, outliers are removed from multiple ranging values ​​to obtain an initial ranging sequence. Based on the initial ranging sequence, the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points are determined respectively. The preset adaptive anti-disturbance feasible region is determined by the interquartile range algorithm based on multiple ranging values. For the original three-dimensional spatial coordinates of the ultra-wideband node to be located, the prior coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinates is obtained by using a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints; wherein, the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region; The prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance are input into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate value of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: attention mechanism vector capsule network, distance-aware gated multi-scale graph convolutional network, and heterogeneous multi-scale feature fusion network.

2. The method according to claim 1, characterized in that, The preset adaptive anti-disturbance feasible region is determined based on multiple ranging values ​​using the interquartile range algorithm, including: For each ultrawideband anchor point, multiple ranging values ​​are arranged in ascending order to determine the first quartile of the ranging value at position M and the third quartile of the ranging value at position N. Based on the first quartile, the third quartile, and the adaptive expansion coefficient, a preset adaptive disturbance rejection feasible region is determined.

3. The method according to claim 2, characterized in that, The step of determining the preset adaptive disturbance rejection feasible region interval based on the first quartile, the third quartile, and the adaptive expansion coefficient includes: Calculate the difference between the third quartile and the first quartile to determine the interquartile range; Based on the first quartile, the adaptive expansion coefficient, and the interquartile range, determine the uplink and downlink thresholds; Based on the uplink and downlink thresholds, a preset adaptive disturbance rejection feasible region is constructed, which is defined by the following expression: ； in, Indicates the downlink range threshold. Indicates the uplink range threshold. Indicates the first quartile. Indicates the third quartile. This represents the adaptive expansion coefficient.

4. The method according to claim 3, characterized in that, The step involves removing outliers from multiple ranging values ​​based on a preset adaptive anti-disturbance feasible region and the Z-coordinate value to obtain an initial ranging sequence, including: Determine whether the Z coordinate value is greater than or equal to the downlink region threshold and less than or equal to the uplink region threshold; If not, the ranging value corresponding to the Z coordinate value is determined to be an abnormal ranging value. The abnormal value is removed to obtain the initial ranging sequence.

5. The method according to claim 4, characterized in that, The original three-dimensional spatial coordinate values ​​include: original X coordinate values, original Y coordinate values, and original Z coordinate values. For the original three-dimensional spatial coordinate values ​​of the ultra-wideband node to be located, a priori coarse three-dimensional spatial coordinate vector corresponding to the original three-dimensional spatial coordinate values ​​is obtained through a preset weighted objective function and a finite-memory quasi-Newton method with boundary constraints. This includes: Within the feasible region constrained by the box, a pre-defined weighted objective function is constructed based on the soft constraint of layer height. For the original Z coordinate value, the prior coarse Z coordinate value is determined by minimizing the preset weighted objective function of the residual; Given the original X and Y coordinate values, a priori coarse X and Y coordinate values ​​are obtained using a finite-memory quasi-Newton method with boundary constraints and a preset coordinate estimation formula. Based on the prior coarse Z-coordinate value, prior coarse X-coordinate value, and prior coarse Y-coordinate value, the prior coarse three-dimensional space coordinate vector is obtained by splicing through the orthogonal basis space recombination operator.

6. The method according to claim 5, characterized in that, The step of inputting the prior coarse three-dimensional spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the three-dimensional fine spatial coordinate value positioning model to obtain the three-dimensional fine spatial coordinate value of the ultra-wideband node to be located includes: The prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point are input into the attention mechanism vector capsule network to obtain the three-axis independent vector capsule features of the ultra-wideband node to be located. The triaxial independent vector capsule features and multiple initial coarse calculation distances are input into a distance-aware gated multi-scale graph convolutional network to obtain distance-aware gated matrix features; wherein, the distance-aware gated multi-scale graph convolutional network includes: a graph convolution module and a distance-aware gated module; The distance-aware gating matrix features are input into a heterogeneous multi-scale feature fusion network to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located.

7. The method according to claim 6, characterized in that, The attention mechanism vector capsule network includes: a four-dimensional spatial projection module, a multi-head self-attention mechanism module, a nonlinear fully connected perceptron module, and an independent feature capsule extraction module. The step of inputting the prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point into the attention mechanism vector capsule network to obtain the three-axis independent vector capsule features of the ultra-wideband node to be located includes: The prior coarse three-dimensional spatial coordinate vector and the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point are input into the four-dimensional spatial projection module. The spatial projection residual is processed by the four-dimensional spatial projection module to obtain four-dimensional local geometric features. The four-dimensional local geometric features are input into the multi-head self-attention mechanism module, and the multi-head self-attention mechanism module extracts attention features from the four-dimensional local geometric features to obtain multi-head self-attention features. The multi-head self-attention features are input into the nonlinear fully connected perceptron module, and the multi-head self-attention features are processed by the nonlinear fully connected perceptron module to obtain fully connected perceptron features. The fully connected sensing features are input into the independent feature capsule extraction module. The independent feature capsule extraction module performs independent feature extraction processing on each coordinate axis of the fully connected sensing features and performs activation processing through a nonlinear activation function to obtain three-axis independent vector capsule features.

8. The method according to claim 7, characterized in that, The distance-aware gated multi-scale graph convolutional network includes: a graph convolution module, a distance encoding module, and a distance-aware gated module. The step of inputting the three-axis independent vector capsule features and multiple initial coarse calculation distances into the distance-aware gated multi-scale graph convolutional network to obtain distance-aware gated matrix features includes: The triaxial independent vector capsule features are input into the graph convolution module, and the graph convolution module performs isotropic aggregation processing on the triaxial independent vector capsule features to obtain normalized Laplacian matrix features. Multiple initial coarse calculation distances are input into the distance encoding module, and distance encoding processing is performed by the distance encoding module to obtain multiple dynamic decay attention matrices; The normalized Laplacian matrix features and multiple dynamic decaying attention matrices are input into the distance-aware gating module, and gating processing is performed by the distance-aware gating module to obtain the distance-aware gating matrix features.

9. The method according to claim 8, characterized in that, The heterogeneous multi-scale feature fusion network includes: three parallel-connected graph convolutional network modules of different scales, a stitching module, and a residual fully connected regression head. The step of inputting the distance-aware gating matrix features into the heterogeneous multi-scale feature fusion network to obtain the three-dimensional fine spatial coordinates of the ultra-wideband node to be located includes: The distance-aware gated matrix features are input into three parallel graph convolutional network modules of different scales to obtain the first scale features, the second scale features, and the third scale features of three different scales. The first scale feature, the second scale feature, and the third scale feature are input into the stitching module. The stitching module performs stitching processing on the first scale feature, the second scale feature, and the third scale feature to obtain multi-scale stitched features. The multi-scale stitched features are input into the residual fully connected regression head, and the multi-scale stitched features are integrated through the residual fully connected regression head to obtain three-dimensional fine spatial coordinate values.

10. An ultra-wideband three-dimensional indoor positioning device, characterized in that, include: The original ranging sequence acquisition module is used to acquire the original ranging sequences corresponding to the ultra-wideband node to be located and multiple ultra-wideband anchor points respectively. Each original ranging sequence includes: multiple ranging values ​​and a first three-dimensional spatial coordinate value corresponding to each ranging value. The first three-dimensional spatial coordinate value is composed of X coordinate value, Y coordinate value and Z coordinate value. The initial ranging acquisition module is used to remove outliers from multiple ranging values ​​based on a preset adaptive anti-disturbance feasible region and Z-coordinate values ​​to obtain an initial ranging sequence, and to determine the initial coarse calculation distances between the ultra-wideband node to be located and multiple ultra-wideband anchor points based on the initial ranging sequence; wherein, the preset adaptive anti-disturbance feasible region is determined based on multiple ranging values ​​using the interquartile range algorithm; The prior coarse 3D spatial coordinate vector acquisition module is used to obtain the prior coarse 3D spatial coordinate vector corresponding to the original 3D spatial coordinate values ​​of the ultra-wideband node to be located by means of a preset weighted objective function and a finite memory quasi-Newton method with boundary constraints; wherein, the preset weighted objective function is constructed based on the layer height soft constraint within the box-constrained feasible region; The 3D fine spatial coordinate value acquisition module is used to input the prior coarse 3D spatial coordinate vector, the second-dimensional spatial coordinate value corresponding to each ultra-wideband anchor point, and the initial coarse calculation distance into the 3D fine spatial coordinate value localization model to obtain the 3D fine spatial coordinate value of the ultra-wideband node to be located. The 3D fine spatial coordinate value localization model includes: attention mechanism vector capsule network, distance-aware gated multi-scale graph convolutional network, and heterogeneous multi-scale feature fusion network.

Images