Underground personnel positioning method based on step length estimation, computer equipment and medium thereof

Abstract

The present invention discloses a method for positioning underground personnel based on step length estimation, a computer device and a medium thereof, wherein the positioning method comprises: using a gait detection model to obtain the historical step length, the number of steps N and the direction data of the target personnel; using the historical step length as a training set to train an improved step length estimation model, the improved step length estimation model is used to output the step length estimation value when the target personnel walks; inputting the step length estimation value, the number of steps N and the direction data into the positioning model, and outputting the coordinate position of the target personnel. The present invention relies on the sensing equipment carried by the target personnel to obtain the gait information of the target personnel, predicts the step length change of the target personnel through the improved step length estimation model, and calculates the coordinate position of the target personnel using the positioning model, which not only saves the equipment cost, but also the positioning result can meet the requirements, and is more convenient in practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of underground positioning, and in particular to an underground personnel positioning method based on step length estimation, a computer device and a medium thereof. Background Art

[0002] In the coal mining industry, since most workers work underground, gas, coal dust explosions, and roof collapse accidents occur from time to time. Therefore, personnel safety has always been the focus of the enterprise. Accurate positioning of personnel can achieve good personnel management and work scheduling management, timely discover safety hazards, and ensure personnel safety.

[0003] At present, the common method of locating underground personnel is based on UWB positioning technology, which realizes positioning by transmitting and receiving ultra-wideband pulse signals. The supporting facilities of UWB positioning technology mainly include UWB positioning base stations and UWB tags. The UWB positioning base stations are responsible for communicating with UWB tags and transmitting data back to the background. UWB tags can be worn on people or objects and are located by communicating with base stations. This positioning method requires the deployment of many positioning base stations underground, which is not only costly but also difficult. In addition, if the signal transmission and reception of one or some positioning base stations fails, the positioning results will be biased. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides an underground personnel positioning method, computer equipment and medium based on step length estimation, which obtains the positioning result of the personnel by calculating the step length and direction of the personnel's walking, while improving the accuracy of the positioning result and reducing the dependence on the positioning base station.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a method for positioning underground personnel based on step length estimation, comprising:

[0006] S1. Use the gait detection model to obtain the historical step length, number of steps N and direction data of the target person. ;

[0007] S2. Using the historical step length as a training set to train an improved step length estimation model, the improved step length estimation model is used to output the step length estimation value of the target person when walking ;

[0008] S3, the step length estimation value , number of steps N and direction data Input the positioning model and output the coordinate position of the target person;

[0009] The improved step length estimation model comprises: an input unit, an estimation unit and an output unit.

[0010] The processing of the input unit includes:

[0011] Extract the step length feature after preprocessing the input inertial navigation data ,

[0012] Calculate step length characteristics The attention weight , Q, K, and T represent query value, key value, and value matrix respectively; Represents the dimension of the key;

[0013] Calculate the attention weight of the vertical component , the vertical component includes the vertical acceleration and angular velocity, Q V , K V , T V Respectively represent the query value, key value, and value matrix in the vertical direction; Indicates the dimension of the vertical direction key;

[0014] The step length feature The attention weight and the attention weight of the vertical component Fusion , α and β represent weight coefficients;

[0015] The weighted step length feature is normalized and converted to a value of [0,1] As input to the estimation unit;

[0016] The estimation unit is based on the input value Perform iterative updates until the output state Satisfy requirements; output status Represents the state information output at time t, including the step length feature;

[0017] The output unit outputs the state Output step size estimate after denormalization , represents the weight matrix of the output unit, Represents the bias term.

[0018] Furthermore, the gait detection model obtains direction data include:

[0019] Calculate the current rotation posture of the target person , is the real part, which represents the cosine value of the rotation angle, represents the three rotation axes in three-dimensional space, represents the imaginary unit, ;

[0020] Rotation posture , angular velocity ; , , Respectively represent the angular velocities of the three axes output by the gyroscope;

[0021] Updated rotation attitude , Indicates sampling time;

[0022] Direction data With rotation posture The relationship is: ,in, , , represents a unit vector;

[0023] Normalize the rotation attitude formula , Represents the normalized rotation posture;

[0024] The rotation matrix is:

[0025] ,

[0026] Direction data The calculation formula is: .

[0027] Furthermore, the index list marked during step detection is used to determine the time steps of the target person's steps and landing;

[0028] Calculate the features at each time step The attention weight , the weighted feature of the time step is ;

[0029] Using weighted features Step length feature Correction yields:

[0030] , represents the corrected step length feature, t represents the time, k represents the total number of steps before time t, n=1, 2, ..., k; Represents the step feature weight coefficient before t time;

[0031] The corrected step length feature Input into the training model.

[0032] Furthermore, the gait detection model performs step count detection including:

[0033] A walking cycle of the target person includes: the foot is gradually lifted, the body moves forward, and the foot gradually lands; when the foot is gradually lifted, the accelerometer detects a positive acceleration; when the foot gradually lands, the accelerometer detects a negative acceleration; that is, the acceleration data collected by the accelerometer changes roughly in a sinusoidal period after filtering; the filtered acceleration data is marked with maximum and minimum values;

[0034] Initialize the pedometer to 0, traverse all the maximum points, and look for the minimum point behind the current maximum point; if there are no other extreme points between the current maximum point and the minimum point found, the interval is a valid step, the counter is increased by 1, and the step interval is added to the index list; otherwise, skip the interval.

[0035] Furthermore, the positioning model outputs the coordinate position of the target person, including:

[0036] Assume the starting coordinate position of the target person is , the starting direction is , the direction of the target person changes to ; Then every time the target person takes a step, the direction data is updated to , k represents the kth step taken by the target person; the coordinate position is updated to , ;

[0037] When k=N, the position update and direction update are completed, and the final coordinate position of the target person is The total walking distance is .

[0038] Furthermore, the estimation unit is based on the input value Iterate and update the output state The process includes:

[0039] The estimation unit will normalize the value Forward propagation, calculate each output state, and update the indicator Determine the current state The update ratio is reset by The previous output state With the current input Combine to generate candidate activations for the current time step Used to update the current output status;

[0040] Update indicators ,

[0041] Reset indicator ,

[0042] Candidate Activation ,

[0043] , , Respectively represent the weight matrices of update index, reset index and candidate plan, , , All represent bias terms; represents the activation function, t represents the time;

[0044] Current output status , compare the output value obtained by forward propagation with the true value and calculate the loss function; then perform back propagation, calculate the gradient of the loss function for all parameters by the derivative chain rule, and update the weights according to the calculated gradient until the maximum number of iterations is reached or the loss function converges;

[0045] The gradient update is expressed as: , , represents the first-order moment vector, represents the second-order moment vector, represents the gradient, , All represent coefficients.

[0046] The present invention also provides a computer device, comprising:

[0047] processor;

[0048] A memory for storing executable instructions;

[0049] Wherein, the processor is used to read the executable instructions from the memory and execute the executable instructions to implement the underground personnel positioning method based on step length estimation.

[0050] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the underground personnel positioning method based on step length estimation.

[0051] The beneficial effect of the present invention is that the underground personnel positioning method of the present invention locates the coordinate position of the target personnel based on step length estimation. Compared with the use of UWB positioning technology, a large number of positioning base stations can be saved, reducing the difficulty of underground deployment. The present invention relies on the sensor equipment carried by the target personnel to obtain the gait information of the target personnel, predicts the step length change of the target personnel through an improved step length estimation model, and calculates the coordinate position of the target personnel using the positioning model, which not only saves equipment costs, but also the positioning results can meet the requirements (accuracy and real-time), and is more convenient in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0053] Figure 1 It is an overall flow chart of the underground personnel positioning method based on step length estimation of the present invention.

[0054] Figure 2 It is a flow chart of the improved step length estimation model of the present invention.

[0055] Figure 3 Schematic diagram of gait detection of the present invention.

[0056] Figure 4 Schematic diagram of step number detection of the present invention.

[0057] Figure 5 It is a schematic diagram of the rotation posture change of the present invention. DETAILED DESCRIPTION

[0058] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0059] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0060] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0061] like Figures 1 to 4As shown, the underground personnel positioning method based on step length estimation of the present invention comprises: S1, using a gait detection model to obtain the historical step length, step number N and direction data of the target person S2. Use the historical step length as a training set to train an improved step length estimation model. The improved step length estimation model is used to output the estimated step length of the target person when walking. S3, the estimated step length , number of steps N and direction data Input the positioning model and output the coordinate position of the target person.

[0062] It should be explained that the "step length" mentioned in this article refers to the length of a person's step, the number of steps refers to the total number of steps a person takes from one position to another, and the direction data refers to the change in direction of a person during walking. Underground workers carry sensor equipment (including accelerometers, gyroscopes, and magnetometers) to monitor parameters such as acceleration and angular velocity generated by personnel while walking.

[0063] The difference between underground personnel positioning and ground personnel positioning and indoor personnel positioning is that:

[0064] Environmental aspects: The underground working environment is more complex, the tunnels are intricate, the light is dim, the space is changeable, and the terrain has slopes, potholes, steps, etc. When personnel walk underground, the changes in sensor data when passing through slopes, potholes, etc. are more complex, which increases the difficulty of extracting key feature data.

[0065] Data processing: The data types that need to be processed for step length estimation of underground personnel are different from those in other scenarios. Considering the complexity of underground terrain, such as the impact of terrain changes on step length, changes in personnel activity status, etc., special attention needs to be paid to the vertical motion characteristics. When people step on a slope or step in a pit, the changes in the vertical direction of the person will be more obvious than on a flat road. Moreover, changes in the vertical characteristics will also affect the current step length estimate. For example, if the length of a person's step on a flat road is 0.5m, then the step length when walking on a slope / in a pit will be smaller than 0.5m. If the positioning prediction is still based on 0.5m, the final coordinate position will not match the actual position.

[0066] Real-time performance: Downhole positioning systems usually require real-time or near real-time feedback, so the computational efficiency and response speed of the model are key. Downhole equipment is usually limited by battery capacity, so the model needs to find a balance between computational complexity and power consumption, and the calculation process needs to be optimized to meet real-time requirements.

[0067] Considering the difficulties in the above three aspects, the present invention makes the following improvements:

[0068] 1. In view of the complexity of the data, the present invention proposes a special improved step length estimation model. When performing step length estimation, the attention weight of the step length feature itself, the attention weight in the vertical direction, the weighted features of the time step, etc. are added, thereby improving the attention to key features and the calculation efficiency, making the output step length estimation value more consistent with the actual state of the person and improving the accuracy of the step length estimation value.

[0069] 2. In terms of real-time performance, the present invention has made improvements in acquiring directional data, thereby improving the real-time performance of data acquisition. Attention has been added to the improved step-length estimation model to improve computational efficiency. The training input data has been normalized to improve the computational efficiency of the model and reduce the estimation time.

[0070] The details are explained below.

[0071] The gait detection model is mainly used to detect the gait of a person when walking, and obtain the acceleration, angular velocity and other data collected by the sensor equipment to calculate the number of steps and direction data. The gait detection model can also obtain the historical step length of a person when walking to train the improved step length estimation model. The trained improved step length estimation model can estimate the current step length based on the step length characteristics.

[0072] Direction data The calculation process is as follows: Calculate the current rotation posture of the target person , is the real part, which represents the cosine value of the rotation angle, represents the three rotation axes in three-dimensional space, represents a vector; represents the imaginary unit, The angular velocity is , rotation posture , convert the angular velocity into a quaternion with a pure imaginary part ,Right now is the quaternion corresponding to the angular velocity. The update amount of the rotation attitude is , the updated rotation posture is , Indicates the sampling time (for example, 0.01 seconds), , , Respectively represent the angular velocity of the three axes output by the gyroscope, It indicates the influence of angular velocity on rotation attitude within one sampling period, that is, the change value of angular velocity within one sampling period.

[0073] like Figure 5 As shown, the direction data With rotation posture The relationship is: ,in, , , Represents a unit vector; the rotation posture after rotation is: , α represents the angle of change in the horizontal plane. In order to facilitate calculation, the rotation attitude formula Normalize , Represents the normalized rotational posture. The rotation matrix can be expressed as: ,use Expressed as the rotation angle around the x-axis, y-axis, and z-axis, , then the direction data The calculation formula is: .

[0074] It should be noted that in traditional attitude update algorithms, the common method is to perform rotation updates through Euler angles or rotation matrices. The Euler angle update requires multiple trigonometric function calculations, which has a high computational complexity. The rotation matrix update involves multiple matrix multiplications, and the amount of calculation for each update is large. In addition, the matrix operation has a large cumulative error, and frequent updates will cause the orthogonality of the matrix to fail, requiring additional orthogonalization processing. In an underground environment, computing resources are often limited, and the present invention has a large amount of orthogonalization processing for directional data. The update of the present invention has been innovated, and has advantages in computational efficiency. Its computational complexity is relatively low. The posture update can be achieved mainly through multiplication and addition, avoiding complex matrix operations or high-dimensional calculations. Compared with the rotation matrix, the update process of the present invention is more stable and can maintain the accuracy of the posture data; the present invention only involves four parameters, which is five parameters less than the rotation matrix, so that the update calculation of the direction angle can be completed more quickly under limited computing resources, thereby improving the real-time performance of positioning. In addition, in the update formula of the rotation posture, the updated part is combined with the posture of the previous moment, so that at each update, the current rotation posture can be taken into account, and the rotation posture of the next moment can be predicted by introducing angular velocity and time factors, thereby speeding up the update calculation of the direction angle. Since the formula of the present invention can more directly reflect the changing trend of the rotation posture, it is easier to converge to the accurate solution of the direction angle when performing iterative calculations. This means that with the same number of calculations, the formula of the present invention can obtain stable direction data more quickly and improve computational efficiency. Directly using the angular velocity , , The introduction of the rotation attitude update formula enables it to respond quickly to changes in the downhole environment. When the downhole equipment is subjected to sudden changes, the angular velocity vector will change rapidly, and the formula can update the rotation attitude faster to adapt to this change.

[0075] That is, the present invention introduces the , can speed up the acquisition of directional data. In an underground environment, the formula The calculation efficiency of the direction data is effectively improved, which is conducive to improving the real-time performance of positioning.

[0076] like Figure 3 and Figure 4 As shown, the gait detection model performs step detection including: a walking cycle of the target person includes: the foot is gradually lifted, the body moves forward, and the foot gradually lands; when the foot is gradually lifted, the accelerometer detects a positive acceleration; when the foot gradually lands, the accelerometer detects a negative acceleration; that is, the acceleration data collected by the accelerometer changes roughly in a sinusoidal period after filtering; the filtered acceleration data is marked with maximum and minimum values. For example, two thresholds (threshold a and threshold b) are set. When the acceleration data exceeds threshold a, it is regarded as a maximum value and the value is marked; when the acceleration data is less than threshold b, it is regarded as a minimum value and the value is marked; the average value of all extreme value points is calculated as the balance point of the current pedestrian's movement; the minimum value above the extreme value average value and the maximum value below the extreme value average value are filtered out. When calculating the number of steps, starting from the periodic changes of pedestrians' walking, the acceleration data will pass through a peak and a trough for each step of the pedestrian. Each time a group of adjacent peaks and troughs is detected, and there is only one intersection with the extreme mean between these peaks and troughs, it is regarded as a step. Initialize the pedometer to 0, traverse all the maximum points, and always look for the minimum point behind the current maximum point; if there are no other extreme points between the current maximum point and the minimum point found, then the interval is a valid step, the counter is increased by 1, and the step interval is added to the index list; otherwise, the interval is skipped. In this way, the number of steps N of the target person can be obtained.

[0077] like Figure 2 As shown in the figure, the improved stride length estimation model includes: input unit, estimation unit and output unit. The input unit extracts the stride length feature after filtering the input acceleration, angular velocity and other data (i.e., inertial navigation data), and uses the stride length feature as the input of the estimation unit. The estimation unit outputs the optimal output value and sends it to the output unit. In this process, attention weights are introduced in the input unit to capture gait information more accurately. Specifically, it includes: calculating the stride length feature The attention weight , Q, K, and T represent query value, key value, and value matrix respectively; Represents the dimension of the key; calculates the attention weight of the vertical component , the vertical component includes the vertical acceleration and angular velocity, Q V , K V , T V Respectively represent the query value, key value, and value matrix in the vertical direction; Indicates the dimension of the vertical direction key; the step length feature The attention weight and the attention weight of the vertical component Fusion , α and β represent weight coefficients; the weighted step length feature is normalized (for example, maximum-minimum normalization) and converted to a value of [0,1] As the input of the estimation unit; the estimation unit Perform iterative updates until the output state Satisfy the requirements; output unit output status Output step size estimate after denormalization , represents the weight matrix of the output unit, Indicates the bias term. Output status Represents the state information output at time t, including the step length feature.

[0078] It should be noted that α+β = 1 is set to ensure that the relative weights of the two can be reasonably compared and combined. When the vertical direction changes slightly (the measured values ​​in the vertical direction always fluctuate slightly around a mean value, and the fluctuation range does not exceed a certain threshold (10% to 20% of the mean)), α = 0.7 and β = 0.3 can be taken, which means that 70% of the model trusts the step length feature and 30% in the vertical direction component. When the vertical direction changes greatly (the measured values ​​in the vertical direction frequently change greatly within a large range, and the fluctuation range exceeds a certain large threshold (more than 30% of the mean)), the weight coefficient can be adjusted to increase the focus on the vertical component. For example, α = 0.4 and β = 0.6 can be set. This means that the model trusts the vertical direction component more (60%) in this case, while still retaining the consideration of the step length feature (40%). By reasonably selecting and adjusting these two weight coefficients, the performance of the model on a specific task can be optimized. Considering that inertial navigation data is easily disturbed by the downhole environment, extreme values ​​or abnormal values ​​are generated. Compared with other normalization methods, the present invention adopts maximum-minimum normalization processing, takes the maximum and minimum values ​​of the training set as intervals, and normalizes the step length feature data, which can eliminate the influence of extreme values. At the same time, considering the real-time positioning of underground personnel, the input of the estimation unit is always controlled in the range of [0,1], which can improve the speed of step length estimation, ensure that the influence range of step length estimation features in the model is consistent, help stabilize the model training, and improve the robustness of the model to abnormal data.

[0079] The estimation unit is based on the input value Iterate and update the output state The process includes: the estimation unit will normalize the value Forward propagation, calculate each output state, and update the indicator Determine the current state The update ratio is reset by The previous output state With the current input Combine to generate candidate activations for the current time step Used to update the current output status; update indicators , reset the indicator , candidate activation , , , Respectively represent the weight matrices of update index, reset index and candidate plan, , , Both represent bias terms, represents the activation function, t represents the time; the current output state , compare the output value obtained by forward propagation with the true value, and calculate the loss function; then perform back propagation, calculate the gradient of the loss function for all parameters by the derivative chain rule, and update the weights according to the calculated gradient until the maximum number of iterations is reached or the loss function converges; the gradient update is expressed as: , , represents the first-order moment vector, represents the second-order moment vector, represents the gradient, , All represent coefficients (e.g., Take 0.9, Take 0.999).

[0080] It should be noted that the sensor data fluctuates greatly when people walk in the lane, and there is a lot of noise. Therefore, the moment related to gait is more important than other moments. Multivariate sensor data allows more information to be included in the gait expression. The step length feature of the present invention It is not a conventional text feature or image feature, but gait information obtained after filtering and smoothing, which has time dependence and periodicity. Therefore, in order to more accurately capture the key moments of gait (such as stepping and landing), the present invention assigns a higher weight to the key moments, which not only includes time information but also related sensor data. Capturing the key moments of gait mainly relies on the dynamic analysis of sensor data. After filtering and smoothing the sensor data, interference data can be eliminated. Using the index list obtained during step detection (the list of time intervals between the start and end times of each step), the time steps of stepping and landing are obtained. Calculate the features of each time step The attention weight , the weighted feature of the time step is ; Using weighted features Step length feature Correction yields: , represents the corrected step length feature, t represents the time, k represents the total number of steps before time t, n=1, 2, ..., k; Represents the weight coefficient of the step length feature before time t; the corrected step length feature Enter the estimation unit.

[0081] It should be noted that the correction formula Incorporating short-term information (i.e. , the weighted features of the current time step) and long-term information (i.e. , the product of the weighted average of the step length features of the previous k time steps and the weight) to improve the stability and accuracy of the step length estimation, represents the weighted average of the features of the first k time steps, Indicates the proportion of attention weights in the previous time steps, and the sum of the attention weights in the current time step is 1, which can be obtained by The size of reflects the degree of dependence on the historical step characteristics. If If the input unit is small, it means that the current step is highly dependent on the historical step. In other words, the input unit is very good at extracting step length features. Finally, weighted features are added Correct the step length feature, and the corrected feature It not only includes time dependence and periodicity, but also combines the dynamic changes of the previous step length (i.e., the step length before time t), so that the key features of the step length have a greater impact on the output of the model, which is beneficial to improve the accuracy of the model output results. The improved step length estimation model of the present invention adds a variety of attention weights in the input unit. These attention weights are not set randomly, but are set in consideration of the impact of the particularity of the underground environment on the walking of personnel, in order to be able to more accurately predict the step length. In addition, adding attention weights to the input unit and normalizing the features are beneficial to simplifying the calculation process of the estimation unit, improving the efficiency of the model result output, and meeting the real-time requirements.

[0082] After the improved step length estimation model outputs the step length estimation value at the current moment, it combines the number of steps N and direction data collected at time t , we can get the coordinate position of the target person. Assume the starting coordinate position of the target person is , the starting direction is , the direction of the target person changes to ; Then every time the target person takes a step, the direction data is updated to , k represents the kth step taken by the target person; the coordinate position is updated to , ; When k=N, the position update and direction update are completed, and the final coordinate position of the target person is The total walking distance is .

[0083] It should be noted that the starting coordinate position and starting direction of the target person are known, and the corresponding direction data can be calculated through the sensor data every time the target person takes a step. , subtract the direction data of the previous step from the current direction data to get According to the direction data and step length, the coordinate position of the target person after taking this step can be calculated and updated accordingly. Every time the target person takes a step, the coordinate position can be updated accordingly until the target person reaches the specified position or stops moving.

[0084] The present invention also provides a computer device, comprising: a processor and a memory for storing executable instructions; wherein the processor is used to read the executable instructions from the memory and execute the executable instructions to implement an underground personnel positioning method based on step length estimation.

[0085] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements an underground personnel positioning method based on step length estimation.

[0086] In summary, the underground personnel positioning method of the present invention locates the coordinate position of the target person based on step length estimation. Compared with the UWB positioning technology, a large number of positioning base stations can be saved, reducing the difficulty of underground deployment. The present invention relies on the sensor equipment carried by the target person to obtain the gait information of the target person, and predicts the step length change of the target person through an improved step length estimation model. The coordinate position of the target person is calculated using the positioning model, which not only saves equipment costs, but also the positioning results can meet the requirements (accuracy and real-time), and is more convenient in practical applications.

[0087] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for underground personnel positioning based on step length estimation, characterized in that: include: S1. Use the gait detection model to obtain the historical step length, number of steps N and direction data of the target person. ; S2. Using the historical step length as a training set to train an improved step length estimation model, the improved step length estimation model is used to output the step length estimation value of the target person when walking ; S3, the step length estimation value , number of steps N and direction data Input the positioning model and output the coordinate position of the target person; The improved step length estimation model comprises: an input unit, an estimation unit and an output unit. The processing of the input unit includes: Extract the step length feature after preprocessing the input inertial navigation data , Calculate step length characteristics The attention weight , Q, K, T Represent query value, key value, and value matrix respectively; Represents the dimension of the key; Calculate the attention weight of the vertical component , the vertical component includes the vertical acceleration and angular velocity, Q V 、K V 、T V Respectively represent the query value, key value, and value matrix in the vertical direction; Indicates the dimension of the vertical direction key; The step length feature The attention weight and the attention weight of the vertical component Fusion , α and β represent weight coefficients; The weighted step length feature is normalized and converted to a value of [0,1] As input to the estimation unit; The estimation unit is based on the input value Perform iterative updates until the output state Satisfy requirements; output status Represents the state information output at time t, including the step length feature; The output unit outputs the state Output step size estimate after denormalization , represents the weight matrix of the output unit, Represents the bias term.

2. The underground personnel positioning method based on step length estimation according to claim 1, characterized in that: Gait detection model obtains direction data include: Calculate the current rotation posture of the target person , is the real part, which represents the cosine value of the rotation angle, represents the three rotation axes in three-dimensional space, represents the imaginary unit, ; Rotation posture , angular velocity ; , , Respectively represent the angular velocities of the three axes output by the gyroscope; Updated rotation attitude , Indicates sampling time; Direction data With rotation posture The relationship is: ,in, , , represents a unit vector; Normalize the rotation attitude formula , Represents the normalized rotation posture; The rotation matrix is: ， Direction data The calculation formula is: .

3. The underground personnel positioning method based on step length estimation according to claim 1, characterized in that: The gait detection model performs step detection including: A walking cycle of the target person includes: the foot is gradually lifted, the body moves forward, and the foot gradually lands; when the foot is gradually lifted, the accelerometer detects a positive acceleration; when the foot gradually lands, the accelerometer detects a negative acceleration; that is, the acceleration data collected by the accelerometer changes roughly in a sinusoidal period after filtering; the filtered acceleration data is marked with maximum and minimum values; Initialize the pedometer to 0, traverse all the maximum points, and look for the minimum point behind the current maximum point; if there is no other extreme point in the interval between the current maximum point and the minimum point found, then the interval is a valid step, the counter is increased by 1, and the interval is added to the index list; otherwise, skip the interval.

4. The underground personnel positioning method based on step length estimation according to claim 1, characterized in that: The positioning model outputs the coordinate position of the target person, including: Assume the starting coordinate position of the target person is , the starting direction is , the direction of the target person changes to ; Then every time the target person takes a step, the direction data is updated to , k represents the kth step taken by the target person; the coordinate position is updated to , ; When k=N, the position update and direction update are completed, and the final coordinate position of the target person is The total walking distance is .

5. The underground personnel positioning method based on step length estimation according to claim 1, characterized in that: The estimation unit is based on the input value Iterate and update the output state The process includes: The estimation unit will normalize the value Forward propagation, calculate each output state, and update the indicator Determine the current output status The update ratio is reset by The previous output state With the current input Combine to generate candidate activations for the current time step Used to update the current output status; Update indicators , Reset indicator , Candidate Activation , , , Respectively represent the weight matrices of update index, reset index and candidate plan, , , All represent bias terms; represents the activation function, t represents the time; Current output status , compare the output value obtained by forward propagation with the true value and calculate the loss function; then perform back propagation, calculate the gradient of the loss function for all parameters by the derivative chain rule, and update the weights according to the calculated gradient until the maximum number of iterations is reached or the loss function converges; The gradient update is expressed as: , , represents the first-order moment vector, represents the second-order moment vector, represents the gradient, , All represent coefficients.

6. A computer device, characterized in that: include: processor; A memory for storing executable instructions; The processor is used to read the executable instructions from the memory and execute the executable instructions to implement the underground personnel positioning method based on step length estimation as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the underground personnel positioning method based on step length estimation as described in any one of claims 1 to 5.

Images