Indoor and outdoor integrated mobile positioning precision optimization system integrating multiple sensors

Abstract

The invention relates to the field of positioning precision optimization, in particular to a multi-sensor integrated indoor and outdoor integrated mobile positioning precision optimization system, which comprises a multi-core SVM (Support Vector Machine) region classification module, a data processing module and a data processing module, generating probability values of the mobile device in an indoor area, an outdoor area and an indoor and outdoor boundary area through a multi-core SVM (Support Vector Machine) model based on a multi-core adaptive sensor decision function according to the obtained GPS data, geomagnetic data and IMU (Inertial Measurement Unit) sensor data of the mobile device; the neural network multi-sensor weight prediction module is used for generating a GPS sensor weight, a geomagnetic sensor weight and an IMU sensor weight through a sensor confidence model based on a convolutional neural network architecture; and the indoor and outdoor mobile positioning optimization module performs weighting calculation to generate regional optimization positioning data of the mobile equipment. According to the indoor and outdoor positions and the confidence coefficient of the multiple sensors, the complementary advantages of the sensors are played to the maximum extent, and precision optimization of indoor and outdoor integrated mobile positioning is achieved.

Description

Technical Field

[0001] This invention relates to the field of positioning accuracy optimization, and in particular to an integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors. Background Technology

[0002] With the rapid development of mobile internet technology, integrated indoor and outdoor mobile positioning has become a core foundation for fields such as intelligent navigation, emergency rescue, IoT device management and control, and smart park management. Its positioning accuracy directly determines the service quality of related applications.

[0003] In practical applications, single-sensor positioning solutions have significant limitations. While the Global Positioning System (GPS) provides high positioning accuracy in open outdoor environments, its signal is easily attenuated indoors due to blockages and multipath effects. Inertial sensors such as geomagnetic sensors, accelerometers, and gyroscopes can achieve short-term positioning in signal-free environments, but they suffer from cumulative errors that increase over time, leading to excessive positioning deviations with long-term use.

[0004] To overcome the limitations of single sensors, multi-sensor fusion positioning has become a mainstream research direction. While existing fusion schemes introduce classification mechanisms for indoor, outdoor, and indoor-outdoor boundary areas, traditional simple classification models struggle to adapt to the multi-scale characteristics of multi-sensor data, resulting in low classification accuracy for complex transitional areas like indoor-outdoor boundaries. Furthermore, current multi-sensor positioning data fusion methods lack the ability to dynamically predict sensor data confidence levels and cannot adjust fusion weights by combining real-time data features with indoor and outdoor area characteristics. This prevents multi-sensor data from fully leveraging its complementary advantages in integrated indoor-outdoor mobile scenarios, hindering the achievement of high-precision and stable positioning in such scenarios.

[0005] Therefore, how to accurately identify the indoor and outdoor areas where the device is located, and how to predict and calculate the reliability of multi-sensor positioning in order to achieve multi-sensor positioning accuracy fusion optimization for complex indoor and outdoor integrated areas are currently technical problems that need to be solved. Summary of the Invention

[0006] To address this, the present invention provides an indoor-outdoor integrated mobile positioning accuracy optimization system that integrates multiple sensors. It uses a multi-core SVM to accurately classify and identify indoor and outdoor areas, and a neural network to combine scene probability and real-time feature prediction of multiple sensors to calculate the multi-sensor weights that reflect the reliability of multi-sensor positioning. Based on the multi-sensor weights, it performs dynamic and accurate fusion calculation of multi-source positioning data, thereby achieving multi-sensor positioning accuracy fusion optimization for complex indoor-outdoor integrated areas.

[0007] To achieve the above objectives, this invention proposes an integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors, comprising: The multi-core SVM region classification module is used to process the acquired GPS data, geomagnetic data and IMU sensor data of the mobile device through a multi-core SVM model based on a multi-core adaptive sensor decision function to generate probability values ​​of the mobile device being located in an indoor area, an outdoor area and an indoor-outdoor boundary area. The neural network multi-sensor weight prediction module is used to perform regression prediction on the probability values, GPS data, geomagnetic data and IMU sensor data through a sensor confidence model based on a convolutional neural network architecture to generate GPS sensor weights, geomagnetic sensor weights and IMU sensor weights. The indoor and outdoor mobile positioning optimization module is used to perform weighted calculations on the multi-sensor positioning data based on the GPS sensor weight, geomagnetic sensor weight, and IMU sensor weight to generate regional optimized positioning data for the mobile device.

[0008] Furthermore, the multi-core SVM region classification module includes: The boundary function construction unit is used to calculate the GPS boundary feature function, geomagnetic stability feature function, and IMU motion activity feature function based on the GPS data, geomagnetic data, and IMU sensor data, respectively; to construct the GPS boundary kernel function based on the GPS boundary feature function of the GPS sample data and the calculated data, and the standard deviation of the GPS sample data; to construct the geomagnetic boundary kernel function based on the geomagnetic stability feature function of the geomagnetic training samples and the calculated samples, and the standard deviation of the geomagnetic training samples; and to construct the IMU motion boundary function based on the IMU motion activity feature function of the IMU sensor sample data and the calculated data, and the standard deviation of the IMU sensor sample data.

[0009] The decision function construction unit is used to calculate GPS kernel weights based on GPS boundary feature functions, calculate geomagnetic kernel weights based on the geomagnetic matching confidence and geomagnetic field strength fluctuation variance of the geomagnetic data, calculate IMU sensor weights based on the heading angle drift of the IMU sensor data, and after weighted summation of the GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary function based on the GPS kernel weights, geomagnetic kernel weights, and IMU sensor weights, multiply by the radial basis function to construct a comprehensive kernel function, and construct the multi-core adaptive sensor decision function based on the comprehensive kernel function. The region classification calculation unit is used to generate the probability value by passing the multi-core adaptive sensor decision function through the Pratt probability calibration algorithm.

[0010] Furthermore, the boundary function construction unit includes: The GPS boundary feature function construction sub-unit is used to construct the GPS boundary feature function based on the current number of locked satellites, the maximum number of locked satellites, and the GPS horizontal accuracy factor of geomagnetic data. A geomagnetic stability feature function construction sub-unit is used to construct the geomagnetic stability feature function based on the cosine similarity of geomagnetic vectors at adjacent times within a time window of geomagnetic data.

[0011] Furthermore, the boundary function construction unit further includes: The multidimensional activity calculation subunit is used to calculate the acceleration deviation value based on the difference between the magnitude of the acceleration amplitude and the gravitational acceleration within the time window of the IMU sensor data, and to calculate the acceleration fluctuation value based on the ratio of the variance of the acceleration magnitude to the maximum calibration variance within the time window of the IMU sensor data. The IMU motion activity feature function construction sub-unit is used to perform a weighted summation of the acceleration deviation value, the acceleration fluctuation value, and the step frequency data within the time window of the IMU sensor data to generate the IMU motion activity feature function.

[0012] Furthermore, the multi-core SVM region classification module also includes: The training optimization unit is used to construct a combined loss function based on the basic classification loss term and the boundary enhancement loss term, and to train and optimize the multi-core SVM model using the combined loss function with the GPS sample data, geomagnetic training samples and IMU sensor sample data.

[0013] Furthermore, the training optimization unit includes: The boundary enhancement loss term construction subunit is used to construct the boundary enhancement loss term by multiplying and summing the sample pair's true label and the integrated kernel function substituted into the sample pair, after passing through the classification consistency loss function, with the GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary function substituted into the sample pair.

[0014] Furthermore, the neural network multi-sensor weight prediction module includes: A multi-branch temporal feature extraction unit is used to pass the probability values, GPS data, geomagnetic data and IMU sensor data through a convolution branch respectively to generate temporal features of probability values, features of GPS data, features of geomagnetic data and features of IMU sensor data; A multi-head cross-modal attention fusion unit is used to generate fused features by passing the probability value temporal features, GPS data features, geomagnetic data features and IMU sensor data features through a multi-head cross-modal attention mechanism; A multi-scale output mapping unit is used to pass the fused features through a residual mapping head to generate the GPS sensor weights, geomagnetic sensor weights, and IMU sensor weights; The sensor confidence model includes convolutional branches, a multi-head cross-modal attention mechanism, and a residual mapping head.

[0015] Furthermore, the multi-scale output mapping unit includes: An output mapping subunit is used to pass the fused features through the output mapping layer to generate initial mapping features; The residual calculation subunit is used to sequentially pass the initial mapping features through a first convolution operation, a first batch of normalization, a first activation function, a second convolution operation, a second batch of normalization, and a dropout layer to generate residual features; The residual connection output subunit is used to perform a residual connection between the initial mapping features and the residual features, and then generate the GPS sensor weights, geomagnetic sensor weights and IMU sensor weights through a second activation function. The residual mapping head includes an output mapping layer, a first convolution operation, a first batch normalization, a first activation function, a second convolution operation, a second batch normalization, a dropout layer, and a second activation function.

[0016] Furthermore, the neural network multi-sensor weight prediction module also includes: The prediction training unit is used to optimize and train the sensor confidence model using the mean squared error loss function.

[0017] Furthermore, the indoor and outdoor mobile positioning optimization module is used to generate the optimized positioning data for the region by weighted summation of the multi-sensor positioning data based on the GPS sensor weight, geomagnetic sensor weight, and IMU sensor weight, and then by normalization calculation.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention, through a multi-core SVM region classification module based on multi-scale adaptive sensor kernel functions, fully mines scene association information from multi-dimensional feature vectors of GPS, geomagnetism, and device motion, and can accurately distinguish and identify indoor, outdoor, and indoor-outdoor boundary areas. Through a confidence model of a convolutional neural network architecture combined with scene probability and real-time features of multiple sensors, it performs dynamic regression prediction of weights, which can intelligently adjust the weights of each sensor according to changes in indoor and outdoor areas. This enables a smooth transition of sensor weights in indoor-outdoor boundary areas based on their location and the confidence of multi-sensor data, avoiding the scene adaptation defects of fixed weight fusion schemes, maximizing the complementary advantages of each sensor, and achieving seamless connection and accuracy optimization of integrated indoor and outdoor mobile positioning.

[0019] In particular, this invention constructs a multi-core fusion architecture of GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary kernel function through a multi-core SVM region classification module, and optimizes the model training by combining it with boundary enhancement loss term based on classification consistency loss. At the same time, it realizes the probabilistic output of region classification results through Pratt probability calibration algorithm, which effectively solves the problem of fuzzy determination of indoor, outdoor, and indoor-outdoor boundary regions.

[0020] In particular, the neural network multi-sensor weight prediction of this invention adopts a multi-branch temporal feature extraction unit, which configures independent convolutional branches for probability values, GPS data, geomagnetic data, and IMU sensor data respectively. This can capture the temporal dynamic features of different types of data in a targeted manner, avoiding the loss of modal information caused by a single feature extraction branch. Through multi-head cross-modal attention fusion, it can perform fine weight allocation and correlation modeling on the temporal features of probability values, GPS data features, geomagnetic data features, and IMU sensor data features extracted from multiple branches. The multi-scale output mapping unit adopts a residual mapping head architecture, which solves the gradient vanishing problem in deep neural network training, and effectively suppresses model overfitting through dropout layers. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of the integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors, according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the multi-core SVM model of the integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors, according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating the sensor confidence model of the integrated indoor and outdoor mobile positioning accuracy optimization system that integrates multiple sensors, according to an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0025] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] like Figures 1 to 4 As shown, this invention provides an indoor-outdoor integrated mobile positioning accuracy optimization system that integrates multiple sensors. It uses a multi-core SVM to accurately classify and identify indoor and outdoor areas, and a neural network to combine scene probability and real-time feature prediction of multiple sensors to calculate the multi-sensor weights that reflect the reliability of multi-sensor positioning. Based on the multi-sensor weights, it performs dynamic and accurate fusion calculation of multi-source positioning data, thereby realizing the fusion optimization of multi-sensor positioning accuracy in complex indoor-outdoor integrated areas.

[0027] like Figure 1 and 2 As shown, this embodiment proposes an indoor-outdoor integrated mobile positioning accuracy optimization system that integrates multiple sensors, including: The multi-core SVM region classification module is used to process the acquired GPS data, geomagnetic data and IMU sensor data of the mobile device through a multi-core SVM model based on a multi-core adaptive sensor decision function to generate probability values ​​of the mobile device being located in an indoor area, an outdoor area and an indoor-outdoor boundary area. The neural network multi-sensor weight prediction module is used to perform regression prediction on the probability values, GPS data, geomagnetic data and IMU sensor data through a sensor confidence model based on a convolutional neural network architecture to generate GPS sensor weights, geomagnetic sensor weights and IMU sensor weights. The indoor and outdoor mobile positioning optimization module is used to perform weighted calculations on the multi-sensor positioning data based on the GPS sensor weight, geomagnetic sensor weight, and IMU sensor weight to generate regional optimized positioning data for the mobile device.

[0028] like Figure 3 As shown, the multi-core SVM region classification module further includes: The boundary function construction unit is used to calculate the GPS boundary feature function, geomagnetic stability feature function, and IMU motion activity feature function based on the GPS data, geomagnetic data, and IMU sensor data, respectively; to construct the GPS boundary kernel function based on the GPS sample data and the calculated data, and the standard deviation of the GPS sample data; to construct the geomagnetic boundary kernel function based on the geomagnetic stability feature function based on the geomagnetic training samples and the calculated samples, and the standard deviation of the geomagnetic training samples; and to construct the IMU motion boundary function based on the IMU sensor sample data and the calculated data, and the standard deviation of the IMU sensor sample data. The decision function construction unit is used to calculate GPS kernel weights based on GPS boundary feature functions, calculate geomagnetic kernel weights based on the geomagnetic matching confidence and geomagnetic field strength fluctuation variance of the geomagnetic data, calculate IMU sensor weights based on the heading angle drift of the IMU sensor data, and after weighted summation of the GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary function based on the GPS kernel weights, geomagnetic kernel weights, and IMU sensor weights, multiply by the radial basis function to construct a comprehensive kernel function, and construct the multi-core adaptive sensor decision function based on the comprehensive kernel function. The region classification calculation unit is used to generate the probability value by passing the multi-core adaptive sensor decision function through the Pratt probability calibration algorithm.

[0029] like Figure 3 As shown, the boundary function construction unit further includes: The GPS boundary feature function construction sub-unit is used to construct the GPS boundary feature function based on the current number of locked satellites, the maximum number of locked satellites, and the GPS horizontal accuracy factor of geomagnetic data. A geomagnetic stability feature function construction sub-unit is used to construct the geomagnetic stability feature function based on the cosine similarity of geomagnetic vectors at adjacent times within a time window of geomagnetic data.

[0030] Furthermore, the boundary function construction unit further includes: The multidimensional activity calculation subunit is used to calculate the acceleration deviation value based on the difference between the magnitude of the acceleration amplitude and the gravitational acceleration within the time window of the IMU sensor data, and to calculate the acceleration fluctuation value based on the ratio of the variance of the acceleration magnitude to the maximum calibration variance within the time window of the IMU sensor data. The IMU motion activity feature function construction sub-unit is used to perform a weighted summation of the acceleration deviation value, the acceleration fluctuation value, and the step frequency data within the time window of the IMU sensor data to generate the IMU motion activity feature function.

[0031] Specifically, the process of generating the GPS boundary kernel function, the geomagnetic boundary kernel function, and the IMU motion boundary function can be represented as follows: ; ; In the formula, These represent the GPS boundary characteristic function, the geomagnetic boundary kernel function, and the IMU motion boundary function, respectively. These represent the current number of satellites locked to the mobile device and the maximum number of satellites locked to it, respectively. The maximum number of satellites locked to it is preferably 12. The GPS horizontal dilution of precision (DIP) is a core indicator of the geometric accuracy of GPS positioning, determined by the distribution of satellites. This represents the total number of time steps within the time window, preferably 30. These represent the geomagnetic vectors at adjacent moments within the time window of the geomagnetic data. They represent their modulus lengths, The cosine similarity of geomagnetic vectors at adjacent times within a time window of geomagnetic data. These represent the acceleration deviation value, the acceleration fluctuation value, and the step frequency data within the time window of the IMU sensor data, respectively. , Let these represent the magnitude of the acceleration amplitude and the gravitational acceleration, respectively. The magnitude of the acceleration amplitude is preferably the square root of the sum of the squares of the three-dimensional acceleration vectors, i.e., the total magnitude of the three-dimensional acceleration vectors. These represent the variance of the acceleration modulus and the maximum calibration variance, respectively. The maximum calibration variance is calculated by normalizing the statistical maximum variance of common motion scenarios. For example, the initial variance before normalization for walking is 5, and the initial variance before normalization for running is 7. These represent the total number of steps within the time window and the duration of the time window, respectively. These represent the weighting values, preferably 0.4, 0.4, and 0.2. These represent the GPS boundary kernel function, the geomagnetic boundary kernel function, and the IMU motion boundary function, respectively. These represent the training samples, the input real-time GPS data, geomagnetic data, and IMU sensor data, respectively. These represent the standard deviations of the GPS sample data, the geomagnetic training samples, and the IMU sensor sample data, respectively.

[0032] Specifically, the process of constructing the multi-core adaptive sensor decision function can be expressed as follows: ; In the formula, These represent the GPS boundary characteristic function, the initial geomagnetic weight, and the initial motion weight, respectively. These represent the variance of geomagnetic field intensity fluctuation and the geomagnetic matching confidence, respectively. The variance of geomagnetic field intensity fluctuation is preferably the mean of the standard deviations of each component of the three-dimensional geomagnetic vector within 5 seconds. The geomagnetic matching confidence is the similarity between the current geomagnetic feature vector and the corresponding location features in the pre-stored geomagnetic fingerprint database. The yaw angle drift of the gyroscope, representing the data from the IMU sensor, is the absolute value of the deviation between the yaw angle obtained by integrating the gyroscope data over 15 seconds and the geomagnetically corrected yaw angle. This represents the GPS kernel weight, geomagnetic kernel weight, or IMU sensor weight, determined by m being GPS, mag, or motion. The preferred attenuation coefficient is 0.9. This represents the synthesis kernel function. Represents the radial basis kernel function. This represents the GPS boundary kernel function, the geomagnetic boundary kernel function, or the IMU motion boundary function. This represents the decision function of a multi-core adaptive sensor. Let represent the Lagrange multiplier of the i-th support vector with respect to the k-th probability value, the true class label, and the bias term, respectively.

[0033] Specifically, the process of generating the probability value using the Pratt probability calibration algorithm can be represented as follows: ; In the formula, This represents the probability value that the mobile device is located in an indoor area, an outdoor area, or an indoor-outdoor boundary area. k=1 corresponds to the mobile device being located in an indoor area, k=2 corresponds to the mobile device being located in an outdoor area, and k=3 corresponds to the mobile device being located in an indoor-outdoor boundary area. This represents the sigmoid function. Let represent the scaling calibration parameter for the k-th class and the offset calibration parameter for the k-th class, respectively. These are specific parameters obtained by the Pratt algorithm through fitting training samples with real region labels. This represents the decision function of the aforementioned multi-core adaptive sensor.

[0034] Furthermore, the multi-core SVM region classification module also includes: The training optimization unit is used to construct a combined loss function based on the basic classification loss term and the boundary enhancement loss term, and to train and optimize the multi-core SVM model using the combined loss function with the GPS sample data, geomagnetic training samples and IMU sensor sample data.

[0035] Furthermore, the training optimization unit includes: The boundary enhancement loss term construction subunit is used to construct the boundary enhancement loss term by multiplying and summing the sample pair's true label and the integrated kernel function substituted into the sample pair, after passing through the classification consistency loss function, with the GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary function substituted into the sample pair.

[0036] Specifically, the combination loss function can be expressed as: ; In the formula, These represent the boundary enhancement loss term, the combined loss function, and the basic classification loss term, respectively. This represents the GPS boundary kernel function, geomagnetic boundary kernel function, or IMU motion boundary function substituted into the sample pair, determined by m as GPS, mag, or motion. This represents the classification consistency loss function, which is primarily used to measure the true labels of sample pairs. , With the synthesis kernel function Consistency of similarity The true region probability of the k-th probability value of the i-th sample is determined by researchers through calibration. This represents the decision function of a multi-core adaptive sensor. This represents the auxiliary loss coefficient, which is preferably 0.03.

[0037] like Figure 4 As shown, the neural network multi-sensor weight prediction module further includes: The multi-branch temporal feature extraction unit is used to pass the probability values, GPS data, geomagnetic data and IMU sensor data through convolutional branches to generate temporal features of probability values, GPS data features, geomagnetic data features and IMU sensor data features, respectively. A multi-head cross-modal attention fusion unit is used to generate fused features by passing the probability value temporal features, GPS data features, geomagnetic data features and IMU sensor data features through a multi-head cross-modal attention mechanism; A multi-scale output mapping unit is used to pass the fused features through a residual mapping head to generate the GPS sensor weights, geomagnetic sensor weights, and IMU sensor weights; The sensor confidence model includes convolutional branches, a multi-head cross-modal attention mechanism, and a residual mapping head.

[0038] Specifically, the process of generating fused features can be represented as: ; In the formula, The probability values ​​of the g-th branch represent the temporal characteristics, GPS data characteristics, geomagnetic data characteristics, and IMU sensor data characteristics. Indicates the convolution branch, Indicates global pooling. This represents probability values, GPS data, geomagnetic data, and IMU sensor data. This represents a convolution operation, with the following values ​​from the inside out: hidden layer dimension 0, kernel size 5, edge padding parameter 2; hidden layer dimension 0, kernel size 3, edge padding parameter 1; hidden layer dimension 2, kernel size 3, edge padding parameter 1. Representing the query vector, key vector, and value vector of a multi-head cross-modal attention mechanism. Represents the learnable mapping matrix. This indicates a multi-head cross-modal attention mechanism. Indicates the initial fusion features, Indicates average pooling. This indicates the fusion feature.

[0039] like Figure 4 As shown, the multi-scale output mapping unit further includes: An output mapping subunit is used to pass the fused features through the output mapping layer to generate initial mapping features; The residual calculation subunit is used to sequentially pass the initial mapping features through a first convolution operation, a first batch of normalization, a first activation function, a second convolution operation, a second batch of normalization, and a dropout layer to generate residual features; The residual connection output subunit is used to perform a residual connection between the initial mapping features and the residual features, and then generate the GPS sensor weights, geomagnetic sensor weights and IMU sensor weights through a second activation function. The residual mapping head includes an output mapping layer, a first convolution operation, a first batch normalization, a first activation function, a second convolution operation, a second batch normalization, a dropout layer, and a second activation function.

[0040] Specifically, the process of generating the GPS sensor weights, geomagnetic sensor weights, and IMU sensor weights can be expressed as follows: ; In the formula, Indicates fusion features, Indicates the output mapping layer. Indicates the initial mapping features, This represents the first convolution operation, the first batch of normalization, and the first activation function. This indicates the second convolution operation and the second batch normalization. Indicates the discarded layer. This represents the residual connection and the second activation function. This represents the weights of the GPS sensor, the geomagnetic sensor, and the IMU sensor.

[0041] Furthermore, the neural network multi-sensor weight prediction module also includes: The prediction training unit is used to optimize and train the sensor confidence model using the mean squared error loss function.

[0042] In particular, the prediction training unit uses the mean squared error loss function to optimize the sensor confidence model, enabling the model to converge quickly to the optimal solution and ensuring that the predicted values ​​of GPS sensor weights, geomagnetic sensor weights, and IMU sensor weights are highly matched with the actual scenario requirements.

[0043] Furthermore, the indoor and outdoor mobile positioning optimization module is used to generate the optimized positioning data for the region by weighted summation of the multi-sensor positioning data based on the GPS sensor weight, geomagnetic sensor weight, and IMU sensor weight, and then by normalization calculation.

[0044] Specifically, the multi-sensor positioning data includes GPS positioning data, geomagnetic positioning data, and PDR positioning data.

[0045] In this embodiment, a multi-core SVM region classification module, based on a multi-scale adaptive sensor kernel function, fully mines scene association information from multi-dimensional feature vectors of GPS, geomagnetism, and device motion. This allows for accurate differentiation and identification of indoor, outdoor, and indoor / outdoor boundary areas. A confidence model using a convolutional neural network architecture, combined with scene probability and real-time features from multiple sensors, performs dynamic weight regression prediction. This intelligently adjusts the weights of each sensor according to changes in indoor / outdoor areas, achieving a smooth transition of sensor weights based on the location and confidence level of multi-sensor data in indoor / outdoor boundary areas. This avoids the scene adaptation defects of fixed-weight fusion schemes, maximizes the complementary advantages of each sensor, and achieves seamless integration and accuracy optimization of indoor / outdoor integrated mobile positioning. The multi-core SVM region classification module constructs a multi-core fusion architecture using GPS boundary kernel function, geomagnetic boundary kernel function, and IMU motion boundary kernel function. Combined with boundary enhancement loss term optimization model training based on classification consistency loss, and the Pratt probability calibration algorithm, the region classification results are output probabilistically, effectively solving the problem of fuzzy determination of indoor, outdoor, and indoor / outdoor boundary areas. The neural network multi-sensor weight prediction uses a multi-branch temporal feature extraction unit, which configures independent convolutional branches for probability values, GPS data, geomagnetic data, and IMU sensor data. This enables targeted capture of the temporal dynamic features of different types of data, avoiding the loss of modal information caused by a single feature extraction branch. Through multi-head cross-modal attention fusion, it can perform fine-grained weight allocation and correlation modeling for the temporal features of probability values, GPS data features, geomagnetic data features, and IMU sensor data features extracted from multiple branches. The multi-scale output mapping unit adopts a residual mapping head architecture, which solves the gradient vanishing problem in deep neural network training, and effectively suppresses model overfitting through dropout layers.

[0046] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0047] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fusion multi-sensor integrated indoor and outdoor mobile positioning precision optimization system, characterized in that, The method comprises the following steps: a multi-core SVM area classification module is used to generate probability values of a mobile device being located in an indoor area, an outdoor area and an indoor-outdoor boundary area by a multi-core SVM model based on a multi-core adaptive sensor decision function, through GPS data, geomagnetic data and IMU sensor data of the mobile device; a neural network multi-sensor weight prediction module is used to generate GPS sensor weights, geomagnetic sensor weights and IMU sensor weights by a sensor confidence model based on a convolutional neural network architecture, through regression prediction of the probability values, the GPS data, the geomagnetic data and the IMU sensor data; an indoor-outdoor mobile positioning optimization module is used to generate area optimization positioning data of the mobile device by weighted calculation of multi-sensor positioning data based on the GPS sensor weights, the geomagnetic sensor weights and the IMU sensor weights.

2. The integrated indoor and outdoor mobile positioning precision optimization system of claim 1, wherein, The multi-core SVM area classification module comprises: a boundary function construction unit is used to calculate a GPS boundary feature function, a geomagnetic stability feature function and an IMU motion activity degree feature function based on the GPS data, the geomagnetic data and the IMU sensor data respectively, to construct a GPS boundary kernel function based on the GPS boundary feature function of GPS sample data and calculation data and a standard deviation of GPS sample data, to construct a geomagnetic boundary kernel function based on the geomagnetic stability feature function of geomagnetic training samples and calculation samples and a standard deviation of the geomagnetic training samples, and to construct an IMU motion boundary function based on the IMU motion activity degree feature function of IMU sensor sample data and calculation data and a standard deviation of the IMU sensor sample data; a decision function construction unit is used to calculate a GPS kernel weight based on the GPS boundary feature function, to calculate a geomagnetic kernel weight based on geomagnetic matching confidence and geomagnetic field intensity fluctuation variance of the geomagnetic data, to calculate an IMU sensor weight based on a heading angle drift amount of the IMU sensor data, to multiply a radial basis kernel function after weighted summation of the GPS boundary kernel function, the geomagnetic boundary kernel function and the IMU motion boundary function based on the GPS kernel weight, the geomagnetic kernel weight and the IMU sensor weight, so as to construct a comprehensive kernel function, and to construct the multi-core adaptive sensor decision function based on the comprehensive kernel function; an area classification calculation unit is used to generate the probability values by Platt probability calibration algorithm through the multi-core adaptive sensor decision function. 3.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 2, characterized in that, The boundary function construction unit comprises: a GPS boundary feature function construction subunit is used to construct the GPS boundary feature function based on a current number of locked satellites, a maximum number of locked satellites and a GPS horizontal dilution of precision of the geomagnetic data; a geomagnetic stability feature function construction subunit is used to construct the geomagnetic stability feature function based on cosine similarity of geomagnetic vectors at adjacent time points in a time window of the geomagnetic data. 4.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 2, wherein, The boundary function construction unit further comprises: The multi-dimensional activity degree calculation subunit is configured to calculate an acceleration deviation value based on a difference between a modulus of an acceleration amplitude and a gravity acceleration within a time window of IMU sensor data, and calculate an acceleration fluctuation value based on a ratio between a variance of an acceleration modulus and a maximum calibration variance within the time window of the IMU sensor data; The IMU motion activity degree feature function construction subunit is configured to perform weighted summation on the acceleration deviation value, the acceleration fluctuation value, and step frequency data within the time window of the IMU sensor data, to generate the IMU motion activity degree feature function. 5.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 2, wherein, The multi-kernel SVM area classification module further includes: The training optimization unit is configured to construct a combined loss function based on a basic classification loss term and a boundary enhancement loss term, and perform training optimization on the multi-kernel SVM model by using the GPS sample data, the geomagnetic training sample, and the IMU sensor sample data through the combined loss function. 6.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 5, wherein, The training optimization unit includes: The boundary enhancement loss term construction subunit is configured to multiply and sum, after passing through a classification consistency loss function, the sample pair true label and the comprehensive kernel function of the sample pair, with a GPS boundary kernel function, a geomagnetic boundary kernel function, and an IMU motion boundary function of the sample pair, to construct the boundary enhancement loss term. 7.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 1, wherein, The neural network multi-sensor weight prediction module includes: The multi-branch time sequence feature extraction unit is configured to pass the probability value, GPS data, geomagnetic data, and IMU sensor data through a convolution branch respectively, to generate probability value time sequence features, GPS data features, geomagnetic data features, and IMU sensor data features; The multi-head cross-modal attention fusion unit is configured to pass the probability value time sequence features, the GPS data features, the geomagnetic data features, and the IMU sensor data features through a multi-head cross-modal attention mechanism, to generate fusion features; The multi-scale output mapping unit is configured to pass the fusion features through a residual mapping head, to generate the GPS sensor weight, the geomagnetic sensor weight, and the IMU sensor weight. The sensor confidence model includes a convolution branch, a multi-head cross-modal attention mechanism, and a residual mapping head. 8.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to claim 7, wherein, The multi-scale output mapping unit includes: The output mapping subunit is configured to pass the fusion features through an output mapping layer, to generate initial mapping features; The residual calculation subunit is configured to pass the initial mapping features through a first convolution operation, a first batch normalization, a first activation function, a second convolution operation, a second batch normalization, and a dropout layer in sequence, to generate residual features; The residual connection output subunit is configured to perform residual connection on the initial mapping features and the residual features, and pass the result through a second activation function, to generate the GPS sensor weight, the geomagnetic sensor weight, and the IMU sensor weight. The residual mapping head includes an output mapping layer, a first convolution operation, a first batch normalization, a first activation function, a second convolution operation, a second batch normalization, a dropout layer, and a second activation function. 9.The integrated indoor and outdoor mobile positioning precision optimization system with fusion multi-sensor according to any one of claims 1 to 8, characterized in that, The neural network multi-sensor weight prediction module further includes: The prediction training unit is configured to perform optimization training on the sensor confidence model through a mean square error loss function.

10. The integrated indoor and outdoor mobile positioning accuracy optimization system with fusion multi-sensor according to any one of claims 1 to 8, characterized in that, The indoor and outdoor mobile positioning optimization module is used to generate the optimized positioning data for the region by weighted summation of multi-sensor positioning data based on the weights of the GPS sensor, the geomagnetic sensor, and the IMU sensor, and then by normalization calculation.

Images