Step length estimation method and system of shoulder-hung IMU (Inertial Measurement Unit), and storage medium

Abstract

The invention discloses a shoulder-hung IMU step length estimation method and system, and a storage medium, and the method comprises the steps: S1, obtaining the inertial data of each step according to the three-axis acceleration of a shoulder inertial sensor and the measurement data of a gyroscope; s2, according to the three-axis acceleration of the foot inertial sensor and the measurement data of the gyroscope, obtaining step length data of each step of the pedestrian through mechanical arrangement; and S3, constructing a step length estimation model according to the shoulder inertial data and the step length through the residual neural network. By adopting the technical scheme of the invention, the stability and reliability of indoor navigation positioning are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of indoor positioning, and particularly relates to a step length estimation method and system for a shoulder-mounted IMU, and a storage medium. Background Art

[0002] In the field of indoor positioning, due to the characteristics of the need to receive satellite signals, the Global Navigation Satellite System (GNSS) is difficult to achieve effective positioning inside buildings with shielding or complex structures. This technical shortcoming leads to significant deficiencies in satellite-dependent positioning solutions in indoor scenarios and is difficult to meet the requirements of high-precision location services.

[0003] As a key technology for pedestrian trajectory tracking, gait dead reckoning can construct a dynamic positioning model based on step length information by working in cooperation with an Inertial Measurement Unit (IMU) in an indoor environment lacking external positioning references. The current mainstream methods generally use IMU devices bound to the feet and utilize the Zero Velocity Update (ZUPT) technology to periodically calibrate sensor errors. This technology resets the IMU state parameters by identifying the stationary moment when the foot lands during the gait cycle, thereby achieving accurate step length estimation. However, due to the differences in the motion characteristics of the wearing position of the shoulder-mounted IMU, it is impossible to stably obtain a stationary state that meets the ZUPT conditions, resulting in systematic deviations in its gait parameter calculation. This technical defect severely restricts the application efficiency of the shoulder-mounted IMU in indoor positioning and urgently needs to break through the existing bottleneck through algorithm innovation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a step length estimation method and system for a shoulder-mounted IMU, and a storage medium, which can improve the accuracy and practicability of the shoulder IMU in an indoor positioning system by combining a residual neural network and an innovative step length prediction algorithm.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A step length estimation method based on a shoulder-mounted IMU, comprising:

[0007] Step S1, obtaining inertial data for each step according to the measurement data of the three-axis acceleration and gyroscope of the shoulder inertial sensor;

[0008] Step S2, obtaining the step length data of each step of the pedestrian through mechanical arrangement according to the measurement data of the three-axis acceleration and gyroscope of the foot inertial sensor;

[0009] Step S3, performing step length estimation through a residual neural network according to the shoulder inertial data and the step length.

[0010] Preferably, in step S2, the calculation formula for the step length data is:

[0011]

[0012] SL = ΔP n

[0013] wherein, is the attitude rotation matrix, V n is the velocity vector in the n system, a b is the acceleration vector measured in the b system, and are the angular velocity of the i system relative to the e system and the angular velocity of the e system relative to the n system, g n is the gravitational acceleration, P n is the position.

[0014] Preferably, in step S3, a step length estimation model is constructed by a residual neural network according to the shoulder inertial data and the step length. Among them, the deep residual network includes: an initial convolutional layer, three residual modules, a global pooling layer, and a fully connected layer; the initial convolution performs primary feature extraction, each residual module contains two convolutional layers, and each layer is followed by batch normalization and ReLU activation, and the input and the convolutional output are superimposed through a skip connection to achieve residual learning; the global average pooling uses adaptive pooling to compress the feature dimension, and the fully connected layer maps the features to the target output.

[0015] The present invention also provides a step length estimation device based on a shoulder-mounted IMU, including:

[0016] A first calculation module, configured to obtain inertial data for each step according to the measurement data of the three-axis acceleration of the shoulder inertial sensor and the gyroscope;

[0017] A second calculation module, configured to obtain the step length data of each step of the pedestrian through mechanical arrangement according to the measurement data of the three-axis acceleration of the foot inertial sensor and the gyroscope;

[0018] A third calculation module, configured to perform step length estimation according to the shoulder inertial data and the step length through a residual neural network.

[0019] Preferably, the calculation formula for the step length of the second calculation module is:

[0020]

[0021] SL = ΔP n

[0022] wherein, is the attitude rotation matrix, V n is the velocity vector in the n system, a b is the acceleration vector measured in the b system, and are the angular velocity of the i - system relative to the e - system and the angular velocity of the e - system relative to the n - system, g n is the acceleration due to gravity, P n is the position.

[0023] Preferably, the third calculation module is used to construct a step - length estimation model according to the shoulder inertial data and the step length through a residual neural network. The deep residual network includes: an initial convolutional layer, three residual modules, a global pooling layer, and a fully - connected layer; the initial convolution performs primary feature extraction. Each residual module contains two convolutional layers, and after each layer, batch normalization and ReLU activation are performed, and residual learning is realized by superimposing the input and the convolutional output through a skip connection; the global average pooling uses adaptive pooling to compress the feature dimension, and the fully - connected layer maps the features to the target output.

[0024] The present invention also provides a step - length estimation system based on a shoulder - mounted IMU, including: a memory and a processor. A computer program is stored on the memory and run by the processor. When the computer program is run by the processor, it executes the step - length estimation method based on the shoulder - mounted IMU.

[0025] The present invention also provides a storage medium. A computer program is stored on the storage medium. When the computer program runs, it executes the shoulder - mounted autonomous positioning method.

[0026] The present invention has the following beneficial effects:

[0027] 1. The modeling method based on the residual neural network (ResNet), through its unique cross - layer connection structure and residual learning module, can realize multi - level feature analysis of IMU motion signals, effectively capture the non - linear mapping relationship in gait dynamics. This deep feature extraction mechanism not only improves the quantization accuracy of step - length prediction but also significantly reduces the cumulative error in traditional models through the gradient optimization mechanism.

[0028] 2. The residual neural network, relying on its multi - scale feature fusion ability, can adaptively analyze the dynamic feature differences under different step frequencies, step spans, and motion patterns. The skip connection design in the network structure enhances the sensitivity of the model to changes in walking speed. Through the collaborative modeling of time - domain and space - domain features, robust estimation of cross - scenario gait parameters is realized. This dynamic adaptation mechanism enables the model to maintain stable performance under complex motion states and expands the application boundary of wearable devices.

[0029] 3. The residual architecture, through the parameter sharing strategy and feature reuse mechanism, can still establish a gait model with high generalization ability under the condition of limited training samples. This characteristic enables the model to converge quickly and maintain sub - meter - level estimation accuracy when training on small - to - medium - scale data sets, greatly improving the feasibility of engineering deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0031] Figure 1 This is a flowchart of the step length estimation method based on a shoulder-mounted IMU according to an embodiment of the present invention;

[0032] Figure 2 This is the structure diagram of the residual neural network provided by an embodiment of the present invention;

[0033] Figure 3 This is the trajectory diagram of the step length estimation method based on a shoulder-mounted IMU provided by an embodiment of the present invention. Detailed implementation manners

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the 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 of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0035] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the drawings and specific implementation manners.

[0036] Embodiment 1:

[0037] As Figure 1 shown, an embodiment of the present invention provides a step length estimation method based on a shoulder-mounted IMU, including:

[0038] Step S1: According to the triaxial acceleration of the shoulder inertial sensor and the measurement data of the gyroscope, inertial data for each step is sorted out;

[0039] Step S2: According to the triaxial acceleration of the foot inertial sensor and the measurement data of the gyroscope, through mechanical arrangement, the step length data of each step of the pedestrian is obtained;

[0040] Step S3: A step length estimation model is constructed through a residual neural network based on the shoulder inertial data and the step length.

[0041] As an implementation manner of an embodiment of the present invention, in step S1, according to the peak discrimination method, the position of each step is identified, and then the acceleration and angular velocity data of each step are listed separately to sort out the shoulder inertial data.

[0042] As an implementation manner of an embodiment of the present invention, in step S2, the original data of the accelerometer and gyroscope output by the foot inertial sensor at a certain moment are respectively denoted as and Then the measurement data at a certain sampling point can be expressed as:

[0043]

[0044] where k represents the sampling point at time k, and x k represents the six-dimensional data measured by the inertial sensor at this moment.

[0045] Furthermore, in the said step S1, the pedestrian position data is obtained specifically as follows:

[0046] Define b as the vehicle coordinate system and n as the navigation coordinate system;

[0047] Construct the rotation matrix from the b system to the n system Satisfy

[0048] Combined with the sampling of the inertial sensor at time k, the output of the inertial navigation system at time k is calculated by the following formula:

[0049]

[0050] where, and respectively represent the rotation matrix, velocity matrix and position matrix at time k, and respectively represent the acceleration in the n system and the angular rate in the b system at time k, [·] × is the skew-symmetric matrix of the vector, T is the step period, and SL is the step length of a single step

[0051] As an implementation manner of an embodiment of the present invention, in step S3, a step length estimation model is constructed by a residual neural network according to the shoulder inertial data and the step length. Specifically as follows:

[0052] First, the sorted shoulder inertial data of each step is used as the input, and the corresponding step length is used as the output to train the residual neural network and construct a regression model.

[0053] The deep residual network architecture of the embodiment of the present invention has a specific structure as Figure 2As shown in the figure, it includes four parts: an initial convolutional layer, three residual modules, a global pooling layer, and a fully connected layer. The initial convolution performs primary feature extraction. Each residual module contains two convolutional layers, and after each layer, batch normalization and ReLU activation are connected. Through skip connections, the input and the convolutional output are superimposed to achieve residual learning. Global average pooling uses adaptive pooling to compress the feature dimension, and the fully connected layer maps the features to the target output. Between the modules, 1×1 convolutions are used to adjust the channel dimension to ensure the feasibility of feature addition. This structure enhances the feature expression ability while alleviating the problem of gradient attenuation.

[0054] Based on the gradient stability characteristics of the deep residual network architecture, the present invention constructs an end-to-end regression model through a multi-layer feature extraction mechanism to realize the non-linear mapping between the time-series data of the shoulder inertial sensor and the human step length parameters. This innovative method effectively suppresses the gradient anomaly phenomenon by using the cross-layer connection structure of the residual module, optimizes the feature propagation path by combining batch normalization and activation functions, and realizes the compression coding of multi-dimensional motion features through the adaptive pooling layer. Finally, a robust step length prediction model is constructed. This technical solution significantly improves the calculation accuracy of the inertial navigation system, and the obtained pedestrian trajectory is as Figure 3 shown, providing reliable technical support for the autonomous positioning function of the wearable device.

[0055] Embodiment 2:

[0056] The embodiment of the present invention also provides a step length estimation device based on a shoulder-mounted IMU, including:

[0057] The first calculation module is used to obtain the inertial data of each step according to the measurement data of the three-axis acceleration and gyroscope of the shoulder inertial sensor;

[0058] The second calculation module is used to obtain the step length data of each step of the pedestrian through mechanical arrangement according to the measurement data of the three-axis acceleration and gyroscope of the foot inertial sensor;

[0059] The third calculation module is used to estimate the step length through a residual neural network according to the shoulder inertial data and the step length.

[0060] As an implementation manner of the embodiment of the present invention, the calculation formula for the step length of the second calculation module is:

[0061]

[0062] SL = ΔP n

[0063] Where is the attitude rotation matrix, V n is the velocity vector in the n system, a b is the acceleration vector measured in the b system, and are the angular velocity of the i - system relative to the e - system and the angular velocity of the e - system relative to the n - system, g n is the gravitational acceleration, P n is the position.

[0064] As an implementation manner of an embodiment of the present invention, the third calculation module is used to construct a step - length estimation model according to the shoulder inertial data and the step length through a residual neural network. Among them, the deep residual network includes: an initial convolutional layer, three residual modules, a global pooling layer, and a fully - connected layer; the initial convolution performs primary feature extraction. Each residual module contains two convolutional layers, and each layer is followed by batch normalization and ReLU activation, and the input and the convolutional output are superimposed through a skip connection to achieve residual learning; the global average pooling uses adaptive pooling to compress the feature dimension, and the fully - connected layer maps the features to the target output.

[0065] Embodiment 3:

[0066] The embodiment of the present invention further provides a shoulder - mounted autonomous positioning system, including: a memory and a processor. A computer program is stored on the memory and run by the processor. When the computer program is run by the processor, it executes the step - length estimation method based on the shoulder - mounted IMU.

[0067] Embodiment 4:

[0068] The embodiment of the present invention further provides a storage medium. A computer program is stored on the storage medium. When the computer program runs, it executes the step - length estimation method based on the shoulder - mounted IMU.

[0069] The above - described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solution of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A step length estimation method based on a shoulder-mounted IMU, characterized in that: include: Step S1, obtaining inertial data of each step according to the measurement data of the three-axis acceleration of the shoulder inertial sensor and the gyroscope; Step S2: obtaining the step length data of each step of the pedestrian through mechanical arrangement according to the measurement data of the three-axis acceleration of the foot inertial sensor and the gyroscope; Step S3: Estimate the step length through a residual neural network based on the shoulder inertia data and the step length.

2. The step length estimation method based on the shoulder-mounted IMU as claimed in claim 1, characterized in that: In step S2, the calculation formula of the step length data is: SL=ΔP n in, is the attitude rotation matrix, V n is the velocity vector in the n system, a b is the acceleration vector measured in frame b, and is the angular velocity of system i relative to system e and the angular velocity of system e relative to system n, g n is the acceleration due to gravity, P n It's the location.

3. The step length estimation method based on the shoulder-mounted IMU as claimed in claim 2, characterized in that: In step S3, a step length estimation model is constructed according to the shoulder inertia data and step length through a residual neural network, wherein the deep residual network includes: an initial convolution layer, three residual modules, a global pooling layer and a fully connected layer; the initial convolution performs primary feature extraction, and each residual module contains two convolution layers, each layer is followed by batch normalization and ReLU activation, and the input and convolution output are superimposed through jump connections to realize residual learning; global average pooling uses adaptive pooling to compress feature dimensions, and the fully connected layer maps features to the target output.

4. A step length estimation device based on a shoulder-mounted IMU, characterized in that: include: A first calculation module is used to obtain inertial data of each step according to the three-axis acceleration of the shoulder inertial sensor and the measurement data of the gyroscope; The second calculation module is used to obtain the step length data of each step of the pedestrian through mechanical arrangement based on the three-axis acceleration of the foot inertial sensor and the measurement data of the gyroscope; The third calculation module is used to estimate the step length through a residual neural network based on the shoulder inertia data and the step length.

5. The step length estimation method based on the shoulder-mounted IMU as claimed in claim 4, characterized in that: The calculation formula of the step length of the second calculation module is: SL=ΔP n in, is the attitude rotation matrix, V n is the velocity vector in the n system, a b is the acceleration vector measured in frame b, and is the angular velocity of system i relative to system e and the angular velocity of system e relative to system n, g n is the acceleration due to gravity, P n It's the location.

6. The step length estimation method based on the shoulder-mounted IMU as claimed in claim 5, characterized in that: The third computing module is used to build a step length estimation model based on shoulder inertia data and step length through a residual neural network, where the deep residual network includes: an initial convolution layer, three residual modules, a global pooling layer, and a fully connected layer; the initial convolution performs primary feature extraction, and each residual module contains two convolution layers, each layer is followed by batch normalization and ReLU activation, and the input and convolution output are superimposed through jump connections to achieve residual learning; global average pooling uses adaptive pooling to compress feature dimensions, and the fully connected layer maps features to the target output.

7. A step length estimation system based on a shoulder-mounted IMU, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the step length estimation method based on the shoulder-mounted IMU as described in any one of claims 1 to 3 is executed.

8. A storage medium, characterized in that: The storage medium stores a computer program, which executes the shoulder-mounted autonomous positioning method according to any one of claims 1 to 3 when running.

Images