Indoor heading estimation method, device, equipment and medium

Abstract

The present invention provides an indoor heading estimation method, device and medium, wherein the method includes grouping IMU data to obtain local features, integrating them with preset additional features, inputting the local features into a deep learning model to obtain correlation to update the additional features, and finally reducing the dimension of the integrated features to obtain the heading estimation result. The method can carefully mine the local features of the data and capture specific information in different regions; update the additional features through correlation to achieve global feature extraction and enrich feature representation; the dimensionality reduction process can reduce the computational complexity, reduce the training and inference time, remove redundant information and noise, and improve the generalization and robustness of the model. Compared with the existing technology, it can better adapt to the complexity of the indoor environment and improve the accuracy and reliability of heading estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of indoor heading estimation, and specifically provides an indoor heading estimation method, device, equipment and medium. Background Art

[0002] With the continuous development of science and technology, the demand for indoor positioning and navigation is growing. Accurate indoor heading estimation is crucial for many application areas such as indoor navigation, robot autonomous movement, virtual reality, etc.

[0003] However, existing heading estimation methods often have difficulty in fully exploring the potential features of data when processing data. For example, for inertial measurement unit (IMU) data, traditional methods may simply use the overall data for analysis, while ignoring the specific information of different local areas in the data, resulting in the inability to accurately capture various subtle changes and unique patterns in indoor environments, thereby reducing the accuracy of heading estimation.

[0004] Therefore, the art needs a new indoor heading estimation solution to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects, the present invention is proposed to provide a solution or at least partially solve the problem of low accuracy and poor reliability of heading estimation.

[0006] In a first aspect, the present invention provides an indoor heading estimation method, comprising: grouping IMU data to obtain local features of each group of data; integrating the local features of each group of data with preset additional features to obtain integrated features; inputting the local features of each group of data into a deep learning model to obtain the correlation between the local features, and updating the preset additional features based on the correlation; performing dimensionality reduction processing on the integrated features to obtain a heading estimation result.

[0007] In a technical solution of the above-mentioned indoor heading estimation method, the local features of each group of data are input into a deep learning model to obtain the correlation between the local features, and the process of updating the preset additional features based on the correlation includes at least: inputting the local features of each group of data into a deep learning model to obtain the correlation and correlation features between the local features; and updating the preset additional features based on the correlation between the local features.

[0008] In a technical solution of the above-mentioned indoor heading estimation method, the initial state of the preset additional feature is a trainable vector, and the initial state is not a feature of a specific local area in the image; the process of updating the preset additional feature based on the correlation between the local features includes: obtaining a maximum correlation feature, the maximum correlation feature is the feature with the greatest correlation between the local features; adding the maximum correlation feature to the preset additional feature to obtain an updated additional feature.

[0009] In a technical solution of the above-mentioned indoor heading estimation method, the deep learning model includes at least a multi-head attention mechanism; the local features of each group of data are input into the deep learning model, and the process of obtaining the correlation and correlation characteristics between the local features includes: inputting the local features of each group of data into the multi-head attention mechanism to obtain the correlation and correlation characteristics between the local features.

[0010] In a technical solution of the above-mentioned indoor heading estimation method, the process of obtaining the maximum correlation feature includes: sorting the correlation features between each local feature according to size based on the Softmax function in the multi-head attention mechanism to obtain the maximum correlation feature.

[0011] In a technical solution of the above-mentioned indoor heading estimation method, the process of grouping IMU data and obtaining local features of each group of data includes: processing the acquired IMU data, and the processing process at least includes: performing a linear expansion embedding operation on the acquired IMU data; slicing the data after the linear expansion embedding operation; and extracting local features of each part of the slice.

[0012] In a technical solution of the above-mentioned indoor heading estimation method, the deep learning model also includes a first additive normalization layer, a feedforward layer, and a second additive normalization layer; after updating the preset additional features based on the correlation, it includes: inputting the maximum correlation feature into the first additive normalization layer to obtain a first calculation result; inputting the first calculation result into the feedforward layer to obtain a second calculation result; inputting the first calculation result and the second calculation result into the second additive normalization layer for processing.

[0013] In a technical solution of the above-mentioned indoor heading estimation method, the feed-forward layer is composed of two fully connected layers, the first layer uses a ReLU activation function, and the second layer does not use any activation function.

[0014] In a second aspect, an electronic device includes a processor and a storage device, the storage device is suitable for storing multiple program codes, and the program codes are suitable for being loaded and run by the processor to execute the indoor heading estimation method of any one of the technical solutions of the above-mentioned indoor heading estimation method.

[0015] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of program codes, wherein the program codes are suitable for being loaded and run by a processor to execute the indoor heading estimation method of any one of the technical solutions of the indoor heading estimation method described above.

[0016] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0017] In the technical scheme of the present invention, the present invention provides an indoor heading estimation method, including: grouping IMU data to obtain local features of each group of data; integrating the local features of each group of data with preset additional features to obtain integrated features; inputting the local features of each group of data into a deep learning model, obtaining the correlation between each local feature, and updating the preset additional features based on the correlation; performing dimensionality reduction processing on the integrated features to obtain a heading estimation result. Compared with the prior art, the indoor heading estimation method provided by the present invention has the following beneficial effects: by grouping IMU data, the local features of each group of data can be carefully mined, so that the method can capture specific information of different local areas in the data; obtaining the correlation between each local feature, and updating the preset additional features based on the correlation, at this time, the additional features can be used as an aggregator of global features in the entire input vector, that is, the global features are obtained, and at this time, on the basis of the local correlation features, global feature extraction is added to further enrich the feature representation; the dimensionality reduction processing of the integrated features can reduce the computational complexity and reduce the training and reasoning time of the model on the one hand, and on the other hand, it can remove redundant information and noise, highlight key features, and improve the generalization ability and robustness of the model. Therefore, this solution can better adapt to the complexity of the indoor environment and improve the accuracy and reliability of heading estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0019] Figure 1 is a schematic flow chart of main steps of an indoor heading estimation method according to an embodiment of the present invention;

[0020] Figure 2 It is a flowchart of a process of inputting local features of each group of data into a deep learning model, obtaining correlations between the local features, and updating the preset additional features based on the correlations according to an embodiment of the present invention;

[0021] Figure 3 It is a schematic diagram of a flow chart of updating a preset additional feature based on the correlation between various local features according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0023] Embodiment 1

[0024] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of an indoor heading estimation method according to an embodiment of the present invention. Figure 1 As shown, the indoor heading estimation method in the embodiment of the present invention mainly includes the following steps S1 to S4.

[0025] Step S1, grouping the IMU data and obtaining local features of each group of data;

[0026] In this embodiment, grouping the IMU data can divide the overall data into multiple small parts so that the characteristics of each part can be analyzed in more detail. By obtaining the local features of each group of data, specific information of different local areas in the data can be captured, which may reflect different motion states, environmental changes, etc. Specifically, grouping can be performed according to factors such as time, space, and sensor dimensions. For example, the IMU data over a period of time is divided according to a fixed time window, or the data of different sensors are grouped separately. Then for each group of data, a feature extraction algorithm can be used to obtain local features, such as calculating statistical features such as mean, variance, standard deviation, or using deep learning methods to automatically extract high-level features.

[0027] Step S2: Integrate the local features of each set of data with preset additional features to obtain integrated features.

[0028] In this embodiment, the preset additional features are a set of features that are preset before heading estimation. Integrating them with local features can enrich feature representation and provide more information for heading estimation. Specifically, local features and additional features can be combined together to form integrated features by feature concatenation, weighted fusion, etc. For example, the local feature vector and the additional feature vector are directly concatenated, or different weights are assigned to them according to the importance of different features and then fused.

[0029] Step S3: inputting the local features of each set of data into a deep learning model, obtaining the correlation between the local features, and updating the preset additional features based on the correlation;

[0030] In this embodiment, by inputting local features into the deep learning model, the correlation between each local feature, that is, the global feature, can be analyzed. Based on this correlation, the preset additional features can be updated to better reflect the actual situation of the data and the relationship between the features. The deep learning model can be a model with a multi-head attention mechanism, a convolutional neural network and other structures. The model outputs correlation features by learning the relationship between local features, and then determines which local features contribute more to the update of additional features based on the correlation features, and then adjusts and updates the additional features accordingly.

[0031] Step S4: performing dimensionality reduction processing on the integrated features to obtain a heading estimation result.

[0032] In this embodiment, dimensionality reduction processing can reduce the dimension of features, on the one hand, reduce the computational complexity, speed up the training and reasoning speed of the model; on the other hand, it can remove redundant information and noise, highlight key features, and improve the generalization ability and robustness of the model. Dimensionality reduction algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA) can be used to process the integrated features. The reduced-dimensional features can be input into a heading estimation model, such as a regression model, a classification model, etc., to obtain the final heading estimation result. The integrated features can also be directly input into a fully connected layer, which performs linear transformation and nonlinear activation on the input integrated features, thereby mapping the high-dimensional integrated features to a new feature space. The output of the fully connected layer is then obtained by forward propagation calculation. This output is usually a vector representing the network's predicted probability or estimated value for different headings.

[0033] In one embodiment, Figure 2 As shown, step S3, inputting the local features of each set of data into the deep learning model, obtaining the correlation between the local features, and updating the preset additional features based on the correlation at least includes:

[0034] Step S31: input the local features of each group of data into a deep learning model to obtain the correlation and correlation features between the local features;

[0035] In this embodiment, the local features extracted from each set of data are provided as input to the deep learning model. This deep learning model is pre-trained or designed to process these local features and analyze the relationship between them. The model can identify the correlation between each local feature by processing the input local features. This correlation can be a numerical degree of association or a specific pattern or relationship. At the same time, the model also outputs correlation features, which are a specific representation of the correlation between local features. The correlation features can be new vectors, matrices or other data structures, which can more intuitively reflect the relationship between local features.

[0036] Step S32: updating the preset additional features based on the correlation between the local features.

[0037] In this embodiment, the correlation between local features reflects the inherent connection between different parts of the data. By using these correlations to update the preset additional features, the additional features can more accurately reflect the actual situation of the data. For example, if certain local features always appear at the same time or have a strong correlation under certain circumstances, incorporating this correlation into the update of the additional features can enable the additional features to better capture this pattern, thereby improving the accuracy of the features.

[0038] In one embodiment, the initial state of the preset additional feature is a trainable vector, and the initial state is not a feature of a specific local area in the image;

[0039] like Figure 3 As shown, step S32, the process of updating the preset additional features based on the correlation between the local features includes:

[0040] Step S321, obtaining a maximum correlation feature, wherein the maximum correlation feature is the feature with the maximum correlation among the local features;

[0041] In this embodiment, during this process, the feature with the greatest correlation, namely the “maximum correlation feature”, is determined. This feature represents the strongest correlation relationship between local features.

[0042] Step S322: Add the maximum correlation feature to the preset additional feature to obtain an updated additional feature.

[0043] In this embodiment, the strongest correlation information between local features is integrated into the additional features, so that the additional features can better reflect the actual situation of the data and the relationship between the local features.

[0044] Of course, the updating process can be performed based on the size of the correlation, direction or specific rules.

[0045] In one embodiment, the deep learning model includes at least a multi-head attention mechanism; the local features of each group of data are input into the deep learning model, and the process of obtaining the correlation and correlation characteristics between the local features includes: inputting the local features of each group of data into the multi-head attention mechanism to obtain the correlation and correlation characteristics between the local features.

[0046] In this embodiment, the local features extracted from each group of data are input into the multi-head attention mechanism. These local features can be various features extracted from the inertial measurement unit (IMU) data, such as acceleration, angular velocity, and other features in different local areas. In addition, the use of a multi-head attention mechanism can better capture the correlation of different aspects, and can also adapt to complex indoor environments and improve the performance and generalization ability of the model. Specifically, in the attention mechanism, the input data is usually converted into three vectors of Query, Key, and Value. Q*K is used to calculate the correlation between vectors. Specifically, for each Query vector, it performs a dot product operation with all Key vectors to obtain a correlation score, which represents the similarity or correlation between the current Query vector and the local features corresponding to each Key vector. Because the Q vector interacts with all K vectors, it is equivalent to querying within the range of the entire input data, thereby capturing the global correlation between the local features. Then divide by the square root of the dimension, then perform a softmax operation, and finally multiply by V to obtain a feature extraction work of attention. In this process, the Q*K process is actually the process of obtaining global features. Therefore, through the multi-head attention mechanism, global features can be obtained from different angles and subspaces, further enriching the representation of correlation features.

[0047] In addition, the deep learning model can also be a model based on a neural network architecture, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a model structure specifically designed for feature correlation analysis.

[0048] In one embodiment, the process of obtaining the maximum correlation feature includes: sorting the correlation features between each local feature according to size based on the Softmax function in the multi-head attention mechanism to obtain the maximum correlation feature.

[0049] In this embodiment, the Softmax function converts the value of the correlation feature into a probability distribution, so that the correlation between each local feature can be expressed in the form of probability.

[0050] In one embodiment, step S1, grouping IMU data and obtaining local features of each group of data includes:

[0051] Step S11, processing the acquired IMU data, wherein the processing process at least includes: performing a linear expansion embedding operation on the acquired IMU data;

[0052] In this embodiment, the linear expansion embedding operation of the original IMU data can increase the dimension and expression ability of the data. This operation can map the original data to a higher dimensional space, making the potential features in the data easier to extract and analyze.

[0053] Step S12, slicing the data after the linear expansion embedding operation;

[0054] In this embodiment, the purpose of slicing the data after the linear expansion embedding operation is to divide the data into multiple small parts, each of which can be analyzed and processed separately. This slicing process can help the model focus on the local structure and characteristics of the data in more detail. For example, the data can be sliced ​​into multiple continuous time periods according to the time series, or the data can be sliced ​​into multiple regions according to the spatial position. Each slice can be regarded as a local data group.

[0055] In addition, when processing time series data, the data can be grouped according to time windows, and the data in each time window can be regarded as a local data group.

[0056] Step S13: extracting local features of each slice.

[0057] In this embodiment, the local features of each slice are extracted to extract features that can reflect the characteristics of the data in each slice. These local features can be statistical features, frequency features, time domain features or other types of features, depending on the nature of the data and the requirements of the task.

[0058] In one embodiment, the deep learning model also includes a first additive normalization layer, a feedforward layer, and a second additive normalization layer; after updating the preset additional features based on the correlation, it includes: inputting the maximum correlation feature into the first additive normalization layer to obtain a first calculation result; inputting the first calculation result into the feedforward layer to obtain a second calculation result; inputting the first calculation result and the second calculation result into the second additive normalization layer for processing.

[0059] In this embodiment, when the maximum correlation feature is input into the first additive normalization layer, the layer will perform specific processing on this feature to obtain a first calculation result. This processing may include normalizing the feature so that the numerical distribution of the feature is more stable, which is convenient for subsequent calculations and processing. Then, the first calculation result is input into the feedforward layer. The feedforward layer usually performs a complex nonlinear transformation on the input, and extracts a more advanced feature representation through a series of weights and activation functions to obtain a second calculation result. Inputting the first calculation result and the second calculation result into the second additive normalization layer enables the model to better transfer gradients during back propagation, which can improve the model's parameter update efficiency and avoid the problem of gradient disappearance.

[0060] In one embodiment, the feed-forward layer consists of two fully connected layers, the first layer uses a ReLU activation function, and the second layer does not use any activation function.

[0061] In this embodiment, the first fully connected layer performs a linear transformation on the input data and maps the input features to a new feature space. By adjusting the weights and biases of the fully connected layer, complex patterns and relationships in the input data can be learned. The second fully connected layer does not use any activation function, which facilitates direct results in the next step.

[0062] Embodiment 2

[0063] The present invention also provides an electronic device. In an embodiment of a device according to the present invention, the device includes a processor and a storage device, the storage device can be configured to store a program for executing the indoor heading estimation method of the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the indoor heading estimation method of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0064] Embodiment 3

[0065] The present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the indoor heading estimation method of the above-mentioned method embodiment, and the program can be loaded and run by a processor to implement the above-mentioned indoor heading estimation method. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0066] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the original technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for indoor heading estimation, characterized in that: include: The IMU data are grouped to obtain local features of each group of data. Specifically, the acquired IMU data are processed, and the processing process at least includes: performing a linear expansion embedding operation on the acquired IMU data; performing a slicing process on the data after the linear expansion embedding operation; and extracting local features of each part of the slice; Integrate the local features of each set of data with preset additional features to obtain integrated features; wherein the initial state of the preset additional features is a trainable vector, and the initial state is not a feature of a specific local area in the image; Inputting the local features of each group of data into a deep learning model, obtaining the correlation between each local feature, and updating the preset additional features based on the correlation, specifically includes: inputting the local features of each group of data into a deep learning model, obtaining the correlation and correlation features between each local feature; updating the preset additional features based on the correlation between each local feature; wherein the process of updating the preset additional features based on the correlation between each local feature includes: obtaining a maximum correlation feature, the maximum correlation feature is the feature with the maximum correlation between each local feature; adding the maximum correlation feature to the preset additional features to obtain an updated additional feature; The integrated features are subjected to dimensionality reduction processing to obtain a heading estimation result.

2. The method according to claim 1, characterized in that The deep learning model at least includes a multi-head attention mechanism; The process of inputting the local features of each set of data into the deep learning model and obtaining the correlation between the local features and the correlation features includes: The local features of each group of data are input into the multi-head attention mechanism to obtain the correlation and correlation features between the local features.

3. The method according to claim 2, characterized in that The process of obtaining the most relevant features includes: Based on the Softmax function in the multi-head attention mechanism, the correlation features between local features are sorted according to size to obtain the maximum correlation feature.

4. The method according to claim 3, characterized in that The deep learning model further includes a first additive normalization layer, a feedforward layer, and a second additive normalization layer; and updating the preset additional features based on the correlation includes: Inputting the maximum correlation feature into the first additive normalization layer to obtain a first calculation result; The first calculation result is input into the feedforward layer to obtain a second calculation result; the first calculation result and the second calculation result are input into the second additive normalization layer for processing.

5. The method according to claim 4, characterized in that The feed-forward layer consists of two fully connected layers, the first layer uses the ReLU activation function, and the second layer does not use any activation function.

6. An electronic device comprising a processor and a storage device, the storage device being adapted to store a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the indoor heading estimation method according to any one of claims 1 to 5.

7. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the indoor heading estimation method according to any one of claims 1 to 5.

Images