Pedestrian inertial positioning method and device based on adaptive sliding window and space-time network model

Abstract

The invention provides a pedestrian inertial positioning method and device based on an adaptive sliding window and a space-time network model, and relates to the technical field of inertial positioning. The method comprises: collecting inertial measurement unit (IMU) motion data, including acceleration and angular velocity information, of a pedestrian by a sensor device, and performing preprocessing, such as denoising, standardization and coordinate system rotation; the method comprises the following steps: dynamically segmenting pedestrian motion data into straight-going and corner state slices by using an adaptive sliding window technology, and classifying the straight-going and corner state slices by using a Transform-base classifier; using a space-time variance attention network to extract time and space features in different state motion segments, and fusing the two features; and training a model by using the real position coordinate information, and predicting a regression relationship between the input characteristics and the position information so as to realize accurate positioning of a pedestrian walking track. By applying the method provided by the embodiment of the invention, the pedestrian inertial positioning precision and stability can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of inertial navigation systems, and particularly aims at the application of inertial measurement units (IMUs) in indoor pedestrian navigation. Background Art

[0002] Traditional pedestrian inertial navigation systems mainly rely on inertial measurement units (IMUs), which are integrated electronic devices used to measure and report specific physical motion data of a system, including acceleration, angular velocity, and sometimes the direction around a magnetic field. These systems provide information on relative position and direction without the need for external references, making them extremely important in environments where wireless or GPS signals are limited. However, traditional IMU systems have some inherent defects. First, they are usually affected by error accumulation, especially during long-term operation. This is because the positioning information of IMU systems is obtained by integrating acceleration and angular velocity measurements, and these measurements themselves may contain noise and biases. Therefore, over time, these small errors accumulate, resulting in increasingly large position errors. Second, the motion patterns of pedestrians are variable and complex, making it more difficult to use traditional IMU systems for precise navigation. The motion characteristics of pedestrians, such as walking, turning, stopping, and starting, all require the system to be able to quickly and accurately identify and adapt to these changes. However, traditional systems often cannot effectively distinguish these different motion patterns, thereby affecting the overall navigation performance. To solve these problems, various attempts have been made in recent years, including using complex filtering algorithms and combining external sensor data (such as GPS or vision systems). These methods have improved the performance of traditional IMU systems to a certain extent, but there are still limitations, especially in indoor or other environments where GPS signals are limited. Summary of the Invention

[0003] The purpose of the embodiments of the present invention is to provide a pedestrian inertial positioning method and device to increase the accuracy and stability of the inertial positioning system. The specific technical solutions are as follows:

[0004] In a first aspect, the embodiments of the present invention provide a pedestrian inertial positioning method based on an adaptive sliding window and a spatio-temporal network model. The method includes:

[0005] Collect IMU motion data of pedestrians, including acceleration and angular velocity information, and preprocess the data, including denoising, normalization, and coordinate system rotation;

[0006] Use the adaptive sliding window corner detection technology to dynamically segment the IMU data of pedestrian motion into straight-line and corner mode slices;

[0007] Use the spatio-temporal variance attention network to extract the time and space features in different state motion segments and fuse the two features;

[0008] The ability to train a model using real position coordinate information to predict the regression relationship between input features and position information and to determine the pedestrian walking trajectory based on this relationship for pedestrian positioning.

[0009] In the first embodiment of the present invention, collecting the IMU motion data of pedestrians, including acceleration and angular velocity information, and preprocessing the data, including denoising, normalization, and coordinate system rotation, includes:

[0010] Obtaining the pedestrian walking IMU data through a sensor device, constructing an IMU data model. The 3D angular velocity (ω) and 3D acceleration (α) provided by the IMU are affected by biases and noises based on certain characteristics of the sensor, as shown in the following formula:

[0011]

[0012]

[0013] Where and are the true values measured by the gyroscope and accelerometer at time stamp t respectively, and are time-varying biases, and are noise values, usually following a zero-mean Gaussian distribution. Then, a coordinate system transformation is implemented on it to eliminate device differences. The direction at time t can be updated according to the following formula using the relative direction Ω(t) between two discrete times t and t - 1, where ω(t - 1) is the angular velocity of the object in the body frame relative to the navigation frame at time (t - 1).

[0014]

[0015]

[0016] can be used to rotate the measurement value x ∈ [ω, α] from the body coordinate system to the navigation coordinate system, represented by where represents the Hamiltonian product between two quaternions.

[0017] In the first embodiment of the present invention, using the adaptive sliding window corner detection technology to dynamically segment the pedestrian motion IMU data into straight-line and corner state slices includes:

[0018] The sliding window technique is adopted to continuously analyze the data stream of the inertial measurement unit (IMU), and the window size is automatically adjusted to adapt to the changes in the walking speed of pedestrians and the dynamic characteristics of turning angles. Multiple overlapping windows are utilized, each window covering 200 data points and moving with a fixed step size to ensure the integrity and continuity of the time series. The data of each window is processed and then input into a Transformer-based classifier, whose task is to determine whether the data points within the window belong to the straight walking or turning state. The output of the classifier is aggregated through a voting system to improve the robustness of the decision-making. If the data points within multiple overlapping windows are classified as the same state, then this state is considered the final decision. This voting mechanism can effectively reduce the classification errors that may be generated by individual windows. In addition, the network also includes a continuity detection mechanism for smoothing the classification results near the transition points. This mechanism judges whether the classification results need to be adjusted by setting thresholds to ensure that the transition between the straight walking and turning states is coherent. Without sacrificing the time series information, the turning points in the pedestrian trajectory are accurately identified and marked, thereby providing accurate state classification for the subsequent spatio-temporal variance change attention network and enhancing the generalization ability of the model in dealing with different walking patterns and complex paths.

[0019] In the first embodiment of the present invention, the extraction of temporal and spatial features in different state motion segments by using the spatio-temporal variance attention network and the fusion of the two features include:

[0020] First, for each slice, the overall variance is calculated according to the following formula:

[0021]

[0022] where μ represents the mean of the time steps used within each slice, and σ 2 is the calculated variance, which serves not only as a statistical measure to capture the data volatility within the slice but also provides key information on the motion changes within the slice for the network. This variance is then passed through a fully connected embedding layer f embed and converted into a variance embedding V e .

[0023] Meanwhile, the IMU data passes through a one-dimensional convolutional neural network through normalization and linear layers to learn the spatial representation, achieve spatial embedding, capture the spatial correlation between different sensor signals, and obtain the spatial feature S e . Temporal embedding uses a one-layer bidirectional LSTM model to mine temporal information and then adds the position encoding provided by a trainable neural network. These embedded features will provide rich context information for the model, helping to more accurately estimate the speed of pedestrians.

[0024] The encoder uses a variance-aware self-attention mechanism to encode the features output by the embedding layer. The spatial feature S e and the variance embedding V e are jointly input into the encoder. In this layer, the variance embedding is mainly used to adjust the attention weights and strengthen the model's attention to time steps with higher variability, which is achieved through variance-aware self-attention calculation, as shown in the following formula:

[0025]

[0026] where W Q , W K , W V are the transformation weight matrices for query, key, and value respectively, W var is the variance weight matrix, d k is the dimension of the key, and A var is the final variance-adjusted attention score. The encoder is stacked with multiple self-attention layers, and each layer includes a residual connection and layer normalization to ensure the effectiveness and stability of deep learning.

[0027] The decoder also includes a stack of multiple identical layers. In each layer, a masked self-attention sublayer is used to extract dependencies in the time dimension. The mask emphasizes that the output at timestamp t can only depend on the IMU samples before timestamp t. Next, the output of the encoder stack is passed through a multi-head attention sublayer to fuse spatial, variance, and time information into a single vector representation, and then through a fully connected feed-forward sublayer.

[0028] In the first embodiment of the present invention, the ability to train a model using real position coordinate information to predict the regression relationship between input features and position information and determine the pedestrian walking trajectory based on this relationship to achieve pedestrian positioning includes:

[0029] Input the pedestrian's IMU time series and its real position coordinates into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above pre-trained model, input the pedestrian IMU data in other scenarios to obtain the ability to achieve pedestrian positioning by obtaining the pedestrian walking trajectory.

[0030] In a second aspect, an embodiment of the present invention provides a pedestrian inertial positioning device based on an adaptive sliding window and a spatio-temporal network model. The device includes:

[0031] Data collection and preprocessing module: Collect the IMU motion data of the pedestrian, including acceleration and angular velocity information, and preprocess the data, including denoising, normalization, and coordinate system rotation;

[0032] Sliding window corner detection module: Use adaptive sliding window corner detection technology to dynamically segment the pedestrian motion IMU data into straight-line and corner mode slices;

[0033] Spatio-temporal variance feature fusion module: Extract time and space features in motion segments of different states using a spatio-temporal variance attention network, and fuse the two features.

[0034] Position prediction module: Train a model using real position coordinate information to predict the regression relationship between input features and position information, and determine the pedestrian walking trajectory based on this relationship to achieve the ability of pedestrian positioning.

[0035] In one embodiment of the present invention, the data collection and preprocessing module is specifically used for:

[0036] Acquire pedestrian walking IMU data through sensor devices, construct an IMU data model, perform coordinate system transformation on it, and eliminate device differences.

[0037] In one embodiment of the present invention, the sliding window corner detection module is specifically used for:

[0038] Use a Transformer-base classifier to classify the preprocessed pedestrian data, and use a voting mechanism and continuity detection to continuously analyze the IMU data stream and automatically adjust the window size. Finally, obtain the classified pedestrian motion segments.

[0039] In one embodiment of the present invention, the spatio-temporal variance feature fusion module is specifically used for:

[0040] The spatio-temporal variance attention network uses spatial features and variance embeddings as common inputs to the encoder, adjusts the attention weights through variance-aware self-attention calculation, and the output of the encoder passes through a multi-head attention sublayer to fuse spatial, variance, and time information into a single vector representation.

[0041] In one embodiment of the present invention, the position prediction module is specifically used for:

[0042] Input the pedestrian IMU time series and its real position coordinates into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above pre-trained model, input the pedestrian IMU data in other scenarios to obtain the pedestrian walking trajectory and achieve the ability of pedestrian positioning.

[0043] Advantageous effects of the embodiments of the present invention:

[0044] The present invention provides a pedestrian inertial positioning method and device based on an adaptive sliding window and a spatio-temporal network model. By combining the innovative adaptive sliding window technology and the spatio-temporal network, it overcomes the limitations of traditional IMU systems and significantly improves the accuracy and stability of the pedestrian navigation system. Brief Description of the Drawings

[0045] The accompanying drawings of this invention patent provide a detailed view of this inertial positioning system, aiming to assist the description text in more intuitively presenting the various components of the invention and their interrelationships. The overall architecture of the system, the working principles of key components, and the specific implementation of the data processing flow are shown in the accompanying drawings. Each drawing is a simplified schematic diagram, and the elements therein do not represent the actual quantity ratio or precise layout, but rather focus on clarifying the concepts and operation processes of the invention.

[0046] Figure 1 It is a schematic flow chart of a pedestrian inertial positioning method based on an adaptive sliding window and a spatio-temporal network model provided by an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of an adaptive sliding window corner detection network provided by an embodiment of the present invention;

[0048] Figure 3 It is a schematic diagram of a spatio-temporal variance attention network provided by an embodiment of the present invention;

[0049] Figure 4 It is a schematic flow chart of a pedestrian inertial positioning device based on an adaptive sliding window and a spatio-temporal network model provided by an embodiment of the present invention. Detailed implementation manners

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art based on the present invention belong to the scope of protection of the present invention.

[0051] Traditional inertial measurement unit (IMU) systems have inherent defects, mainly including error accumulation and difficulties in recognizing pedestrian motion patterns. These systems determine the position by integrating the measured values of acceleration and angular velocity, but these measured values may contain noise and biases, resulting in the gradual accumulation of errors over time, especially more significant during long-term operation. In addition, the motion characteristics of pedestrians are variable (such as walking, turning, stopping, and starting), and traditional IMU systems often cannot effectively distinguish these different motion patterns, affecting the navigation performance. To improve the performance, complex filtering algorithms and methods of combining external sensor data (such as GPS or vision systems) are adopted, which alleviate the problem to a certain extent. However, these solutions still have limitations in indoor or environments with limited GPS signals.

[0052] To solve the above problems, the embodiments of the present invention provide a pedestrian inertial positioning method and device based on an adaptive sliding window and a spatio-temporal network model, which will be specifically described below.

[0053] First, a pedestrian inertial positioning method based on an adaptive sliding window and a spatio-temporal network model provided by an embodiment of the present invention will be described.

[0054] See Figure 1 , which is a schematic flowchart of a pedestrian inertial positioning method based on an adaptive sliding window and a spatio-temporal network model provided by an embodiment of the present invention. This method is applied to an electronic device with computing capabilities. Exemplarily, this method is applied to a computer. The above method includes the following steps S101 to S104.

[0055] The present invention relates to a method for pedestrian inertial navigation, aiming to provide a navigation data generation model based on an adaptive sliding window and a spatio-temporal network. The following embodiments provide a specific application manner of the present invention to facilitate those skilled in the art to understand and implement the present invention. The specific implementation manners of the present invention are described in detail below:

[0056] Step S101: Collect the IMU motion data of the pedestrian, including acceleration and angular velocity information, and preprocess the data, including denoising, normalization, and coordinate system rotation.

[0057] Specifically, the IMU data of the pedestrian walking is obtained through a sensor device to construct an IMU data model. The 3D angular velocity (ω) and 3D acceleration (α) provided by the IMU are affected by biases and noises based on certain characteristics of the sensor, as shown in the following formula:

[0058]

[0059]

[0060] Among them, and are the true values measured by the gyroscope and accelerometer at time stamp t respectively, and are time-varying biases, and are noise values, usually following a zero-mean Gaussian distribution. Then, a coordinate system transformation is implemented to eliminate device differences. The direction at time t can be updated according to the following formula using the relative direction Ω(t) between two discrete times t and t - 1, where ω(t - 1) is the angular velocity of the object in the body frame relative to the navigation frame at time (t - 1).

[0061]

[0062]

[0063] It can be used to rotate the measurement value \(x\in[\omega,\alpha]\) from the body coordinate system to the navigation coordinate system, denoted by , where represents the Hamilton product between two quaternions.

[0064] Exemplarily, first, an inertial measurement unit (IMU) is used to collect the motion data of a pedestrian, including acceleration and angular velocity information. To ensure data quality, the collected data needs to be preprocessed, including denoising and normalization. In addition, the motion state of the pedestrian (such as going straight or turning) also needs to be accurately labeled. To further improve the accuracy and efficiency of data processing, the collected data will be segmented into fragments of 200 data points each, and slid with a step of 10 data points to ensure the continuity and integrity of the data.

[0065] Step S102: Use the adaptive sliding window corner detection technique to dynamically segment the IMU data of pedestrian motion into straight and corner mode slices.

[0066] Specifically, a Transformer-base classifier is used to classify the preprocessed pedestrian data, and a voting mechanism and continuity detection are used to continuously analyze the IMU data stream and automatically adjust the window size. Finally, the classified pedestrian motion segments are obtained.

[0067] Exemplarily, referring to Figure 2 , the preprocessed data segments (size 200, step 10) are input into the Transformer-based classifier. This step aims to effectively classify each data segment using the powerful learning ability of the Transformer to distinguish the pedestrian motion state (going straight or turning). The classified data segments will then undergo a voting mechanism to improve the accuracy and robustness of the classification. At this stage, the system will comprehensively consider the classification results of multiple adjacent data segments and determine the final state classification through the majority voting method. The system further implements continuity detection to ensure the logical continuity and consistency of the classification results between data segments. This step is achieved by analyzing the classification results of adjacent data segments and adjusting the classification if necessary to reduce misclassification and improve the overall robustness of the system. Throughout the process, the sliding window technique is used to continuously process the data stream to ensure that the generation and classification of data segments can dynamically adapt to the motion changes of the pedestrian.

[0068] Step S103: Use the spatio-temporal variance attention network to extract the time and space features in the motion segments of different states and fuse the two features.

[0069] Specifically, the spatio-temporal variance attention network uses spatial features and variance embeddings as the common input to the encoder. It calculates and adjusts the attention weights through variance-aware self-attention. The output of the encoder passes through a multi-head attention sub-layer to fuse spatial, variance, and temporal information into a single vector representation.

[0070] Exemplarily, refer to Figure 3 The network uses a spatial encoder and a self-attention mechanism to extract spatial features, uses LSTM and positional encoding to enhance time series analysis, and dynamically processes data uncertainty by taking variance as an input feature. The encoder of the network uses a variance-aware self-attention mechanism to encode the features output by the embedding layer, where spatial features and variance embeddings are input to the encoder simultaneously. The variance embedding is mainly used to adjust the attention weights and enhance the model's attention to time steps with higher variability.

[0071] Step S104: Use the real position coordinate information to train the model to predict the regression relationship between the input features and the position information, and determine the pedestrian walking trajectory based on this relationship to achieve the ability of pedestrian positioning.

[0072] Specifically, the pedestrian IMU time series and its real position coordinates are jointly input into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above pre-trained model, the pedestrian IMU data in other scenarios is input to obtain the pedestrian walking trajectory to achieve the ability of pedestrian positioning.

[0073] Corresponding to the aforementioned pedestrian inertial positioning method based on the adaptive sliding window and spatio-temporal network model, the embodiment of the present invention also provides a pedestrian inertial positioning device based on the adaptive sliding window and spatio-temporal network model.

[0074] Refer to Figure 4 , a schematic flowchart of a pedestrian inertial positioning device based on the adaptive sliding window and spatio-temporal network model provided by the embodiment of the present invention. The device is applied to an electronic device with computing capabilities. The above device includes:

[0075] The data collection and preprocessing module 401 is used to collect the IMU motion data of the pedestrian, including acceleration and angular velocity information, and preprocess the data, including denoising, standardization, and coordinate system rotation;

[0076] The sliding window corner detection network module 402 is used to dynamically segment the pedestrian motion IMU data into straight-line and corner mode slices by using the adaptive sliding window corner detection technology;

[0077] The spatio-temporal variance feature fusion module 403 is used to extract the temporal and spatial features in different state motion segments by using the spatio-temporal variance attention network and fuse the two features;

[0078] The position prediction module 404 has the ability to train a model using real position coordinate information to predict the regression relationship between input features and position information, and based on this relationship, determine the pedestrian walking trajectory to achieve pedestrian positioning.

[0079] In one embodiment of the present invention, the above-mentioned data collection and preprocessing module 401 is specifically configured to:

[0080] Obtain pedestrian walking IMU data through a sensor device, construct an IMU data model, perform coordinate system transformation on it, and eliminate device differences.

[0081] In one embodiment of the present invention, the above-mentioned sliding window corner detection network module 402 is specifically configured to:

[0082] Use a Transformer-base classifier to classify the preprocessed pedestrian data, and use a voting mechanism and continuity detection to continuously analyze the IMU data stream and automatically adjust the window size. Finally, obtain the classified pedestrian motion segments.

[0083] In one embodiment of the present invention, the above-mentioned spatio-temporal variance feature fusion module 403 is specifically configured to:

[0084] The spatio-temporal variance attention network uses spatial features and variance embeddings as common inputs to the encoder. The attention weights are adjusted through variance-aware self-attention calculation. The output of the encoder passes through a multi-head attention sublayer to fuse spatial, variance, and time information into a single vector representation.

[0085] In one embodiment of the present invention, the above-mentioned position prediction module 404 is specifically configured to:

[0086] Input the pedestrian IMU time series and its real position coordinates into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above-mentioned pre-trained model, input the pedestrian IMU data in other scenarios to obtain the pedestrian walking trajectory to achieve the ability of pedestrian positioning.

Claims

1. A pedestrian inertial positioning method based on an adaptive sliding window and a spatio-temporal network model, characterized in that It includes the following steps: Collect the IMU motion data of pedestrians, including acceleration and angular velocity information, and preprocess the data, including denoising, standardization, and coordinate system rotation; Use the adaptive sliding window corner detection technology to dynamically segment the IMU data of pedestrian motion into straight-line and corner mode slices; Use the spatio-temporal variance attention network to extract the time and space features in the motion segments of different states and fuse the two features; Use the real position coordinate information to train the model to predict the regression relationship between the input features and the position information, and determine the pedestrian walking trajectory based on this relationship to achieve the ability of pedestrian positioning.

2. The method according to claim 1, wherein The collection of the IMU motion data of pedestrians, including acceleration and angular velocity information, and the preprocessing of the data, including denoising, standardization, and coordinate system rotation, include: Obtain the IMU data of pedestrian walking through the sensor device, construct an IMU data model. The 3D angular velocity (ω) and 3D acceleration (α) provided by the IMU will be affected by the bias and noise based on certain characteristics of the sensor, as shown in the following formula: wherein, and are the true values measured by the gyroscope and the accelerometer at time stamp t, and are time-varying biases, and are noise values, which usually follow a zero-mean Gaussian distribution. Then, a coordinate system transformation is implemented on them to eliminate device differences. The orientation at time t can be updated according to the relative orientation Ω(t) between two discrete times t and t - 1 using the following formula, where ω(t - 1) is the angular velocity of the body in the body frame relative to the navigation frame at time (t - 1). Can be used to rotate a measurement value \(x\in[\omega,\alpha]\) from a body coordinate system to a navigation coordinate system, represented by where, represents the Hamilton product between two quaternions.

3. The method according to claim 1, wherein The use of the adaptive sliding window corner detection technology to dynamically segment the IMU data of pedestrian motion into straight-line and corner state slices includes: Use a Transformer-base classifier to classify the preprocessed pedestrian data, and use a voting mechanism and continuity detection to continuously analyze the IMU data stream and automatically adjust the window size. Finally, obtain the classified pedestrian motion segments.

4. The method according to claim 1, wherein The use of the spatio-temporal variance attention network to extract the time and space features in the motion segments of different states and fuse the two features includes: The spatio-temporal variance attention network uses the spatial features and variance embedding to jointly input the encoder, calculates and adjusts the attention weights through variance-aware self-attention. The output of the encoder passes through a multi-head attention sublayer to fuse the spatial, variance, and time information into a single vector representation. Among them, W Q , W K , W V are the separately query, key, and value conversion weight matrices, W var is the variance weight matrix, d k is the dimension of the key, and A var is the final variance-adjusted attention score.

5. The method according to claim 1, wherein The use of the real position coordinate information to train the model to predict the regression relationship between the input features and the position information, and determine the pedestrian walking trajectory based on this relationship to achieve the ability of pedestrian positioning, includes: Jointly input the pedestrian IMU time series and its real position coordinates into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above pre-trained model, input the IMU data of pedestrians in other scenarios to obtain the ability to determine the pedestrian walking trajectory and achieve pedestrian positioning.

6. A pedestrian inertial positioning device based on an adaptive sliding window and a spatio-temporal network model, characterized in that, The device includes: Data collection and preprocessing module: Collect the IMU motion data of pedestrians, including acceleration and angular velocity information, and preprocess the data, including denoising, standardization, and coordinate system rotation; Sliding window corner detection module: Use the adaptive sliding window corner detection technology to dynamically segment the IMU data of pedestrian motion into straight-line and corner mode slices; Spatio-temporal variance feature fusion module: Use the spatio-temporal variance attention network to extract the time and space features in the motion segments of different states and fuse the two features; Position prediction module: Use the real position coordinate information to train the model to predict the regression relationship between the input features and the position information, and determine the pedestrian walking trajectory based on this relationship to achieve the ability of pedestrian positioning.

7. The device according to claim 6, characterized in that, The data collection and preprocessing module is specifically used for: Obtain pedestrian walking IMU data through a sensor device, construct an IMU data model, perform coordinate system transformation on it, and eliminate device differences.

8. The device according to claim 6, characterized in that, The sliding window corner detection module is specifically used for: Using a Transformer-base classifier to classify the preprocessed pedestrian data, and using a voting mechanism and continuity detection to continuously analyze the IMU data stream and automatically adjust the window size. Finally, the classified pedestrian motion segments are obtained.

9. The device according to claim 6, characterized in that The spatio-temporal variance feature fusion module is specifically used for: The spatio-temporal variance attention network uses spatial features and variance embeddings as common inputs to the encoder. The attention weights are adjusted through variance-aware self-attention calculation. The output of the encoder passes through a multi-head attention sub-layer to fuse spatial, variance, and time information into a single vector representation.

10. The device according to claim 6, characterized in that, The data collection and preprocessing module, the sliding window corner detection module, the spatio-temporal variance feature fusion module, and the position prediction module are specifically used for: Input the pedestrian IMU time series and its true position coordinates into the positioning model of the present invention. The pre-trained model updates and adjusts the model parameters. After obtaining the above pre-trained model, input the pedestrian IMU data in other scenarios to obtain the pedestrian walking trajectory and achieve the ability of pedestrian positioning.

Images