Pedestrian inertial navigation positioning method and system based on gait features and environmental semantics

Abstract

The invention provides a pedestrian inertial navigation positioning method and system based on gait features and environmental semantics, and belongs to the technical field of pedestrian positioning. According to the method, three-axis acceleration and angular velocity data of a foot IMU are collected, the motion state of a pedestrian is judged in combination with a height difference threshold value, and the number of steps is accurately recognized through an LSTM step number recognition model so as to update the height value. Meanwhile, speed drift in the horizontal direction is corrected by means of a zero-speed updating ZUPT algorithm, and accurate calculation of horizontal displacement is achieved. By adopting the pedestrian inertial navigation positioning method and system based on the gait characteristics and the environmental semantics, the problems of precision, continuity and stability in pedestrian three-dimensional positioning can be effectively solved, and the positioning precision of a pedestrian foot inertial navigation system in the vertical direction and the horizontal direction is improved; and meanwhile, the method has good adaptability and energy consumption optimization capability.

Description

Technical Field

[0001] This invention relates to the field of pedestrian positioning technology, and in particular to a pedestrian inertial navigation and positioning method and system based on gait features and environmental semantics. Background Technology

[0002] In both indoor and outdoor environments, pedestrian positioning and navigation technologies have significant applications in smart buildings, emergency rescue, and logistics warehousing. However, existing pedestrian positioning technologies have many limitations, making it difficult to meet the requirements for high precision, high continuity, and high stability.

[0003] Because Global Navigation Satellite System (GNSS) signals are severely blocked in certain indoor and outdoor environments, failing to provide effective positioning services, autonomous positioning technology based on inertial measurement units (IMUs) has become the mainstream solution. Foot-mounted IMUs can estimate a pedestrian's three-dimensional position using strapdown inertial navigation algorithms. However, due to the inherent limitations of low-cost micro-electro-mechanical systems (MEMS) inertial measurement units, their positioning accuracy, especially the vertical (height) estimation accuracy, is easily affected by error accumulation, making it difficult to meet practical needs.

[0004] Existing optimization methods for pedestrian 3D localization have the following limitations: Error accumulation problem: The accelerometer and gyroscope measurement data of the IMU contain errors such as zero-bias drift and noise. After integration, the height error accumulates rapidly over time. Even if the Zero-Rate Update (ZUPT) algorithm is used to constrain the horizontal error, the vertical drift is still difficult to suppress, especially in scenarios with drastic height changes such as going up and down stairs, where the error can reach several meters in a short period of time.

[0005] Poor scene adaptability: Traditional height update methods rely on barometer assistance, but barometer values ​​are easily affected by environmental factors such as temperature and airflow, and the accuracy drops sharply in enclosed spaces or air-conditioned environments; some methods assume that the stair slope is fixed or the step height is known, but the step parameters in actual buildings are different (such as the step heights in office areas and shopping malls are different), resulting in insufficient generalization ability of the algorithm.

[0006] Low accuracy in step count recognition: When going up or down stairs, pedestrians may step over 1-3 steps. Existing technologies can only distinguish between "going up / going down stairs" and cannot accurately identify the number of steps. For example, the threshold method based on the difference in height between adjacent steps can only determine the trend of height change and has difficulty distinguishing between "stepping over 1 step" and "stepping over 2 steps", resulting in a step-like error in height updates.

[0007] Insufficient dynamic response: Pedestrian gait varies from person to person (e.g., different stride frequency and stride length), and walking patterns may switch suddenly (e.g., walking on flat ground → going up stairs → stopping on a platform). Existing algorithms have weak real-time recognition capabilities for dynamic scenes and are prone to misjudgment (e.g., misjudging stopping on a platform as going up stairs).

[0008] Insufficient 3D positioning accuracy: Existing methods have insufficient positioning accuracy in both the horizontal and vertical directions. Especially in long-distance travel and complex environments, positioning errors will gradually accumulate, leading to a decrease in the continuity and stability of positioning results.

[0009] Therefore, for complex indoor and outdoor scenarios, especially the problem of accurate 3D positioning of height when going up and down stairs and walking long distances, there is an urgent need for a technical solution that can adaptively identify the walking environment, accurately determine the number of steps, and dynamically correct the height error based on this, so as to improve the reliability and accuracy of low-cost IMU in pedestrian 3D positioning. Summary of the Invention

[0010] The purpose of this invention is to provide a pedestrian inertial navigation and positioning method and system based on gait features and environmental semantics, which can effectively solve the problems of accuracy, continuity and stability in pedestrian three-dimensional positioning, improve the positioning accuracy of pedestrian foot inertial navigation system in the vertical and horizontal directions, and at the same time have good adaptability and energy consumption optimization capabilities.

[0011] To achieve the above objectives, this invention provides a pedestrian inertial navigation and localization method based on gait features and environmental semantics, comprising the following steps: Step S1: Collect triaxial acceleration and angular velocity data from the foot IMU; Step S2: Determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles; Step S3: Extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; Step S4: Update the three-dimensional coordinate positioning based on the motion state.

[0012] Preferably, in step S2, the pedestrian's motion state is determined based on the height difference threshold between adjacent gait cycles, specifically as follows: ; in, Indicates the first The height value calculated by the inertial navigation system. Indicates the first The height value calculated by the inertial navigation system. Indicates walking on flat ground. Indicates going up the stairs. Indicates going down the stairs. This indicates the height determination threshold.

[0013] Preferably, in step S3, the IMU data features include: Vertical data characteristics: sample mean, sample standard deviation, sample kurtosis, sample skewness, and sample interquartile range of Z-axis acceleration in the three-axis acceleration; Horizontal data characteristics: Sample mean and standard deviation of X-axis acceleration; sample mean and standard deviation of Y-axis acceleration; sample mean, standard deviation, and kurtosis of Y-axis angular velocity.

[0014] Preferably, the pre-trained step number recognition model is an LSTM step number recognition model.

[0015] Preferably, the LSTM step number recognition model satisfies: The network structure is a 6-layer LSTM with 80 units per layer; The output layer uses the Softmax function to generate confidence scores for 1 / 2 / 3 steps; Output the corresponding step number only when the highest confidence level is >0.85; otherwise, it is considered an invalid gait.

[0016] Preferably, the training process of the LSTM step number recognition model includes: Apply random rotation to the input data and set a scaling factor; The model is trained using the Adam optimizer and cross-entropy loss function; Gradient explosion is suppressed by gradient clipping and learning rate decay.

[0017] Preferably, in step S4, the three-dimensional coordinate positioning update includes the horizontal coordinate update: The zero-velocity update ZUPT algorithm is triggered by the zero-velocity state of the foot support phase in the gait cycle to correct horizontal velocity drift. The X-axis and Y-axis acceleration data within the gait swing interval are integrated: first, the horizontal velocity is obtained through integration; then, the horizontal displacement is obtained through integration. The horizontal direction of motion is determined by combining the y-axis angular velocity with the estimated heading angle, thus obtaining the single-step horizontal displacement. Update the horizontal coordinates: , ,in , The first Step, First The horizontal coordinate of the X-axis, , The first Step, First The horizontal Y-axis coordinate of the step. This represents the single-step X-axis displacement component. This represents the single-step Y-axis displacement component.

[0018] Preferably, in step S4, the three-dimensional coordinate positioning update includes vertical height update: when At that moment, the height of the pedestrian's feet is: ; when or At that moment, the height of the pedestrian's feet is: ; in, Indicates the first The estimated height of the step, Indicates the first The estimated height of the step, Represents a symbolic function. Indicates the height of one step. Indicates the number of steps crossed.

[0019] This invention also provides a pedestrian inertial navigation and positioning system based on gait features and environmental semantics, comprising: The data acquisition module is used to acquire triaxial acceleration and angular velocity data from the foot IMU; The motion state determination module is used to determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles; The displacement calculation module is used to extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; The 3D coordinate positioning update module is used to update the 3D coordinate positioning based on the motion state.

[0020] Preferably, the three-dimensional coordinate positioning and updating module includes a horizontal coordinate updating unit and a vertical height updating unit; The vertical height update unit is used to update the height value of the pedestrian's feet based on the number of steps crossed and the known step heights; The horizontal coordinate update unit is used to trigger the zero-velocity update ZUPT algorithm by utilizing the zero-velocity state of the foot support phase in the gait cycle to correct the horizontal velocity drift. It also determines the horizontal movement direction by integrating the X-axis and Y-axis acceleration data within the gait swing interval and combining the heading angle estimated by the Y-axis angular velocity, thereby obtaining the single-step horizontal displacement and updating the horizontal coordinates.

[0021] Therefore, the present invention employs the above-mentioned pedestrian inertial navigation and positioning method and system based on gait features and environmental semantics, and the beneficial technical effects are as follows: (1) This invention improves the accuracy of three-dimensional positioning simultaneously through a collaborative correction mechanism in the vertical and horizontal directions. In the vertical direction, by accurately identifying the state of going up and down stairs and the number of steps, environmental semantics are modeled, and height constraints based on prior knowledge are realized, fundamentally solving the integral drift problem of height estimation. In the horizontal direction, the Zero-Voltage Update (ZUPT) algorithm is used to effectively correct velocity drift, and combined with heading angle estimation, the accuracy of horizontal displacement calculation is ensured.

[0022] (2) The present invention can intelligently distinguish various movement states such as walking on flat ground, going up stairs, and going down stairs, and can accurately identify complex gaits that cross 1 to 3 steps. The algorithm shows good adaptability to the gait differences, gait frequency changes and dynamic switching of walking modes (such as the conversion between flat ground and stairs) of different pedestrians, so that it can maintain stable and reliable positioning performance in a variety of complex indoor and outdoor scenarios.

[0023] (3) The correction strategies of this invention (such as height constraints and zero-speed updates) are mainly triggered at specific stages of the gait cycle (such as the end of a step or the instant of zero speed of the foot), avoiding complex full-process integration calculations. This event-driven processing method reduces the computational load of the algorithm, improves the battery life of wearable devices, and makes them more suitable for long-term continuous positioning tasks. Attached Figure Description

[0024] Figure 1 This is a flowchart of the pedestrian inertial navigation and localization method based on gait features and environmental semantics according to the present invention; Figure 2 The results of step data identification include, Figure 2 (a) in the image represents the results of identifying the steps. Figure 2 (b) in the figure represents the results of identifying the step below. Figure 3 The graph shows the changes in motion numbers of subject 6 during walking. Figure 4 The graph shows the changes in movement numbers during the walking process of experimenter 7; Figure 5 A comparison of experimenters before and after the constraint algorithm was applied to the six height-constrained subjects; Figure 6 A comparison of experimenters 7 before and after the constraint algorithm was applied; Figure 7 This is a graph showing the change in accuracy of the height correction algorithm over time. Figure 8 Comparison of Z-axis height before and after height correction algorithm; Figure 9 ZUPT+ZARU pedestrian foot trajectory diagram; Figure 10 This is a pedestrian foot trajectory diagram for ZUPT+ZARU+SHCA. Figure 11 This is a graph showing the variation of relative positioning error. Detailed Implementation

[0025] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0027] Example 1 I. Experimental equipment and environment.

[0028] Sensor equipment: The WT9011DCL-bt50 ultra-low cost IMU manufactured by Witt Intelligent Technology Co., Ltd. is used. Its size is 23.5mm×32.5mm. It supports USB wired connection or Bluetooth connection and has a battery life of about 2 hours. It is used to collect triaxial acceleration and angular velocity data of pedestrian feet.

[0029] True value measurement equipment: The Southern Ruide RTS-882R4X total station was used as the measurement tool for the true values ​​of height and position. Its accuracy indicators are shown in Table 1. Table 1. Accuracy Indicators of Total Station

[0030] Experimental Scenario: A staircase and corridor area of ​​an experimental building was selected, including typical indoor scenarios such as flat ground, ascending stairs, descending stairs, and platforms between floors. The route from the first floor to the second floor passes through two platforms, with 11 stairs between each pair of flat ground sections. The specific walking route is as follows: starting from the first floor, walking on flat ground, ascending stairs, platform 1, ascending stairs, platform 2, ascending stairs, reaching the second floor, turning on the second floor, descending stairs, platform 2, descending stairs, platform 1, descending stairs, reaching the first floor and briefly walking on flat ground. This walking route includes switching between walking on flat ground and ascending / descending stairs.

[0031] II. Algorithm Implementation Steps.

[0032] like Figure 1As shown, the pedestrian inertial navigation and localization method based on gait features and environmental semantics includes the following steps: Step S1: Collect triaxial acceleration and angular velocity data from the foot IMU; Step S2: Determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles, specifically including: ; in, Indicates the first The height value calculated by the inertial navigation system. Indicates the first The height value calculated by the inertial navigation system. Indicates walking on flat ground. Indicates going up the stairs. Indicates going down the stairs. This indicates the height determination threshold.

[0033] Step S3: Extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; IMU data characteristics include: Vertical data characteristics: sample mean, sample standard deviation, sample kurtosis, sample skewness, and sample interquartile range of Z-axis acceleration in the three-axis acceleration; Horizontal data characteristics: Sample mean and standard deviation of X-axis acceleration; sample mean and standard deviation of Y-axis acceleration; sample mean, standard deviation, and kurtosis of Y-axis angular velocity.

[0034] The pre-trained step number recognition model is an LSTM step number recognition model.

[0035] The LSTM step number recognition model satisfies: It consists of a 6-layer LSTM with 80 units per layer. A fully connected layer follows the LSTM to reduce the network's output dimension to 2D. The model output defaults to two steps, with the output constrained by the Softmax function, corresponding to confidence levels for 1, 2, and 3 steps respectively. False positives are filtered out by setting a confidence threshold (above 0.85).

[0036] The training process of the LSTM step number recognition model includes: Apply random rotation to the input data and set a scaling factor; Using the Adam optimizer and cross-entropy loss function Training the model; ; in, Represents the total number of samples. Indicates the first The true label of each sample Indicates the first The predicted probability of each sample; Gradient explosion is suppressed by gradient clipping and learning rate decay.

[0037] Output labels: 1-3 for the first 3 steps of the stairs going up, -1--3 for the first 3 steps of the stairs going down, and 0 for flat ground.

[0038] Step S4: Based on the motion state, realize the three-dimensional coordinate positioning update, including horizontal coordinate update and vertical height update.

[0039] (1) Horizontal coordinate update includes: The zero-velocity update ZUPT algorithm is triggered by the zero-velocity state of the foot support phase in the gait cycle to correct horizontal velocity drift. The X-axis and Y-axis acceleration data within the gait swing interval are integrated: first, the horizontal velocity is obtained through integration; then, the horizontal displacement is obtained through integration. The horizontal direction of motion is determined by combining the y-axis angular velocity with the estimated heading angle, thus obtaining the single-step horizontal displacement. Update the horizontal coordinates: , ,in , The first Step, First The horizontal coordinate of the X-axis, , The first Step, First The horizontal Y-axis coordinate of the step. This represents the single-step X-axis displacement component. This represents the single-step Y-axis displacement component.

[0040] (2) Vertical height update includes: when At that moment, the height of the pedestrian's feet is: ; when or At that moment, the height of the pedestrian's feet is: ; in, Indicates the first The estimated height of the step, Indicates the first The estimated height of the step, Represents a symbolic function. Indicates the height of one step. Indicates the number of steps crossed.

[0041] III. Results.

[0042] Using a dataset of 5 participants (as shown in Table 2, with over 100 samples of each step type per participant), the training and test sets were divided in an 8:2 ratio. The Adam optimizer and cross-entropy loss function were used to train the dataset for 300 epochs, and false positive results were filtered out using a confidence threshold (0.85).

[0043] Table 2 Statistics of Training Dataset

[0044] The average training accuracy of the 5 participants was as follows: Figure 2 As shown, the accuracy rates for recognizing the 1st, 2nd, and 3rd steps of the stairs going up are 99.5%, 100%, and 98.1%, respectively; the accuracy rates for recognizing the 1st, 2nd, and 3rd steps of the stairs going down are 99.6%, 100%, and 99.4%, respectively. The overall recognition accuracy is low because going up / down 3 steps is similar to going down 2 steps, making it easy to identify as 2 steps during confidence filtering. However, in real-world applications, where pedestrians crossing 3 steps is extremely rare, the overall recognition accuracy is sufficient to meet practical needs.

[0045] Experimenters 6 and 7 followed the walking route described above, but with different numbers of steps traversed when going up and down stairs, to verify the model's generalization ability. The motion type number of the output for each effective step in both experiments is as follows: Figure 3 , Figure 4 As shown, this represents the number of steps a pedestrian traverses in each stride. The experimental results show that the total number of steps traversed in both experiments is consistent with the actual number of steps (33 steps each for going up and down the stairs), accurately reflecting the actual walking situation.

[0046] like Figure 5 , Figure 6 As shown, the curves illustrating the change in height over time for a pedestrian moving in a staircase scenario are illustrated. The red curve represents the height change curve before height constraint, demonstrating that while zero-velocity detection constrains the height drift when the foot is in a supported state, there is still a significant deviation from the actual estimated height within the swing range. The blue curve represents the height change curve after height constraint. Whenever a pedestrian makes a stepping motion, it is accurately identified by the calculation method of this invention. When the stepping motion ends, the height is estimated based on the number of steps taken, improving the accuracy and reliability of height estimation. This can serve as an effective solution for enhancing pedestrian foot inertial navigation positioning indoors or in situations with stairs.

[0047] Table 3 shows the comparison of height error before and after algorithm correction and the improvement in accuracy. Figure 7 As shown, the algorithm proposed in this invention can effectively correct height drift caused by sensor drift, and the accuracy of the correction increases with time, making it particularly suitable for long-distance height constraints.

[0048] Table 3. Statistical table of height error before and after constraint algorithm correction for two experimenters.

[0049] For height drift, which has the greatest impact on positioning accuracy, the height constraint algorithm based on the number of steps proposed in this invention accurately identifies walking on flat ground, climbing stairs, and descending stairs, and identifies the number of steps crossed by the pedestrian for effective height estimation. The curve of height estimation results changing with the number of effective steps is shown below. Figure 8 The red and blue curves represent the changes in height with the number of effective steps before and after correction using the SHCA (Step-Based Height Constraint Algorithm) algorithm. Because the number of steps frequently traversed during the walk was specific to the experimenter's usual walking habits, the curves... Figure 8 As can be seen, the numerous platforms encountered largely conform to the actual building design. It can distinguish the brief flat walks (i.e., platforms between floors) when going up and down stairs and can accurately identify the number of steps encountered in each step.

[0050] Figure 9 , Figure 10 The figures represent pedestrian trajectories using the ZUPT (Zero-velocity Update) + ZARU (Zero-acceleration Reset Update) algorithm and the ZUPT + ZARU + SHCA algorithm, respectively. In the figures, red dots represent the 3D coordinates of the pedestrian position at the end of each step (the starting node of the support interval), adjacent points are connected by blue dots, green triangles represent feature points (assumed to be known points) surveyed by the total station for accuracy assessment, green rectangles and circles represent the start and end points of the walking route, respectively, and black arrows indicate the walking direction.

[0051] Using only the zero-rate update algorithm still results in significant drift in height constraints. For example, when pedestrians walk in long corridors of 1L and 4L, almost every step a pedestrian takes causes height drift. There are even cases where the starting height of the step interval is similar when a pedestrian is going down a staircase. The SHCA algorithm can eliminate the drift caused by steps. By recognizing the walking scene and the number of steps going up / down the stairs, it is consistent with the original intention of the algorithm design. For example, the number of stairs on each floor matches the actual number, and the height estimation of the corner platforms between floors is consistent.

[0052] To evaluate the positioning performance of the proposed algorithm in indoor 3D positioning, 10 specific location points were marked during route planning. These points were traversed during the walking process to perform quantitative analysis of the algorithm. The 3D coordinates of these 10 representative feature points were measured using a total station as reference values ​​for evaluating the positioning accuracy of the algorithm, and these were used as the true coordinates. The evaluation indicators were 3D position error and relative positioning error, calculated using the following formulas: ; ; in, , These represent 3D error and relative positioning error, respectively, indicating the three-dimensional distance and positioning error per unit distance between the estimated position and the total station measured position. and Representing estimates and actual values ​​respectively. Coordinate values and Representing estimates and actual values ​​respectively. Coordinate values and Representing estimates and actual values ​​respectively. Coordinate values and They represent the first Step and the first Step Coordinate values and They represent the first Step and the first Step Coordinate values. Among them... Indicates the first The distance walked. Specific results are shown in Table 4.

[0053] Figure 11The graph illustrates the trend of relative positioning error over time. The blue line graph reflects the relative positioning error of the positioning system at different time points, expressed as a percentage (%). The error is relatively large in the initial positioning stage, but gradually stabilizes and decreases as the walking distance increases. This demonstrates that the algorithm of this invention can adapt well to long-term positioning tasks and provide high-precision positioning results.

[0054] The magnitude of the error in the entire trajectory is commonly represented by the root mean square error (RMSE). The average of the squared errors at each location, taken as the square root, reflects the overall level of the positioning error. The calculation formula is as follows: ; In this embodiment, This represents the root mean square error. Indicates the location point index. =11, including the coordinate difference of 10 feature points and the starting point. As shown in Table 4, the root mean square error of the indoor pedestrian three-dimensional positioning enhancement scheme proposed in this invention is 1.67m.

[0055] Table 4. Statistical table of positioning error of the algorithm proposed in this invention

[0056] Example 2 Pedestrian inertial navigation and positioning systems based on gait features and environmental semantics include: The data acquisition module is used to acquire triaxial acceleration and angular velocity data from the foot IMU; The motion state determination module is used to determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles; The displacement calculation module is used to extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; The three-dimensional coordinate positioning and update module is used to update the three-dimensional coordinate positioning based on the motion state, including a horizontal coordinate update unit and a vertical height update unit.

[0057] The vertical height update unit is used to update the height value of the pedestrian's feet based on the number of steps crossed and the known step heights; The horizontal coordinate update unit is used to trigger the zero-velocity update ZUPT algorithm by utilizing the zero-velocity state of the foot support phase in the gait cycle to correct the horizontal velocity drift. It also determines the horizontal movement direction by integrating the X-axis and Y-axis acceleration data within the gait swing interval and combining the heading angle estimated by the Y-axis angular velocity, thereby obtaining the single-step horizontal displacement and updating the horizontal coordinates.

[0058] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0059] Therefore, the present invention adopts the above-mentioned pedestrian inertial navigation and positioning method and system based on gait features and environmental semantics, which can effectively solve the problems of accuracy, continuity and stability in pedestrian three-dimensional positioning, improve the positioning accuracy of pedestrian foot inertial navigation system in vertical and horizontal directions, and at the same time have good adaptability and energy consumption optimization capabilities.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A pedestrian inertial navigation and localization method based on gait features and environmental semantics, characterized in that, Includes the following steps: Step S1: Collect triaxial acceleration and angular velocity data from the foot IMU; Step S2: Determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles; Step S3: Extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; Step S4: Update the three-dimensional coordinate positioning based on the motion state.

2. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 1, characterized in that, In step S2, the pedestrian's motion state is determined based on the height difference threshold between adjacent gait cycles, specifically as follows: ； in, Indicates the first The height value calculated by the inertial navigation system. Indicates the first The height value calculated by the inertial navigation system. Indicates walking on flat ground. Indicates going up the stairs. Indicates going down the stairs. This indicates the height determination threshold.

3. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 1, characterized in that, In step S3, the IMU data features include: Vertical data characteristics: sample mean, sample standard deviation, sample kurtosis, sample skewness, and sample interquartile range of Z-axis acceleration in the three-axis acceleration; Horizontal data characteristics: Sample mean and standard deviation of X-axis acceleration; sample mean and standard deviation of Y-axis acceleration; sample mean, standard deviation, and kurtosis of Y-axis angular velocity.

4. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 1, characterized in that, The pre-trained step number recognition model is an LSTM step number recognition model.

5. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 4, characterized in that, The LSTM step number recognition model satisfies: The network structure is a 6-layer LSTM with 80 units per layer; The output layer uses the Softmax function to generate confidence scores for 1 / 2 / 3 steps; Output the corresponding step number only when the highest confidence level is >0.85; otherwise, it is considered an invalid gait.

6. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 4, characterized in that, The training process of the LSTM step number recognition model includes: Apply random rotation to the input data and set a scaling factor; The model is trained using the Adam optimizer and cross-entropy loss function; Gradient explosion is suppressed by gradient clipping and learning rate decay.

7. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 1, characterized in that, In step S4, the 3D coordinate positioning update includes the horizontal coordinate update: The zero-velocity update ZUPT algorithm is triggered by the zero-velocity state of the foot support phase in the gait cycle to correct horizontal velocity drift. The X-axis and Y-axis acceleration data within the gait swing interval are integrated: first, the horizontal velocity is obtained through integration; then, the horizontal displacement is obtained through integration. The horizontal direction of motion is determined by combining the y-axis angular velocity with the estimated heading angle, thus obtaining the single-step horizontal displacement. Update the horizontal coordinates: , ,in , The first Step, First The horizontal coordinate of the X-axis, , The first Step, First The horizontal Y-axis coordinate of the step. This represents the single-step X-axis displacement component. This represents the single-step Y-axis displacement component.

8. The pedestrian inertial navigation and localization method based on gait features and environmental semantics according to claim 2, characterized in that, In step S4, the 3D coordinate positioning update includes the vertical height update: when At that moment, the height of the pedestrian's feet is: ； when or At that moment, the height of the pedestrian's feet is: ； in, Indicates the first The estimated height of the step, Indicates the first The estimated height of the step, Represents a symbolic function. Indicates the height of one step. Indicates the number of steps crossed.

9. A pedestrian inertial navigation and positioning system based on gait features and environmental semantics, characterized in that, include: The data acquisition module is used to acquire triaxial acceleration and angular velocity data from the foot IMU; The motion state determination module is used to determine the pedestrian's motion state based on the height difference threshold between adjacent gait cycles; The displacement calculation module is used to extract IMU data features within the current gait swing interval and perform the following processing: Horizontal displacement calculation: Under any motion state, the horizontal displacement is calculated by continuously utilizing the horizontal data features in the IMU data features; Vertical height calculation: Input the IMU data features into the pre-trained step number recognition model, and output the number of steps crossed in the current step; calculate the height change based on the number of steps crossed, and update the vertical height; The 3D coordinate positioning update module is used to update the 3D coordinate positioning based on the motion state.

10. The pedestrian inertial navigation and positioning system based on gait features and environmental semantics according to claim 9, characterized in that, The 3D coordinate positioning and update module includes a horizontal coordinate update unit and a vertical height update unit; The vertical height update unit is used to update the height value of the pedestrian's feet based on the number of steps crossed and the known step heights; The horizontal coordinate update unit is used to trigger the zero-velocity update ZUPT algorithm by utilizing the zero-velocity state of the foot support phase in the gait cycle to correct the horizontal velocity drift. It also determines the horizontal movement direction by integrating the X-axis and Y-axis acceleration data within the gait swing interval and combining the heading angle estimated by the Y-axis angular velocity, thereby obtaining the single-step horizontal displacement and updating the horizontal coordinates.

Images