Internet of Things interactive grid control method based on in-shoe inertial measurement

Abstract

The invention relates to the technical field of data interaction, in particular to an internet of things interactive grid control method based on in-shoe inertial measurement, which comprises the following steps: setting an inertial measurement module in an intelligent shoe and establishing communication with a networking terminal running an application program; collecting a multi-axis angular velocity and a multi-axis acceleration, and carrying out low-pass filtering, zero offset estimation deduction and gravity-based coordinate calibration; adaptive sliding window discrete Fourier transform is adopted for the angular velocity to extract spectral peaks to obtain a stride frequency sequence, and window adjustment is carried out along with stride frequency changes; determining a zero-speed interval threshold value according to the stride frequency, calculating test statistics fusing angular velocity energy and acceleration stability, and comparing to output a foot static interval; suppressing attitude drift and segmenting effective trampling by using a static interval, extracting angular velocity time sequence characteristics, and comparing the angular velocity time sequence characteristics with a template library to determine a target grid; a direction instruction is generated, the center is homing, and the periphery is moving; and sending to the terminal, mapping as a direction key, and performing direction-changing gating by homing to realize continuous movement control. According to the invention, the limitation of the gyroscope when the game terminal is applied in the shoe can be solved, and the limitation of the intelligent shoe game terminal and scene can be solved.

Description

Technical Field

[0001] This invention relates to the field of data interaction technology, and in particular to a grid control method for Internet of Things (IoT) interaction based on in-shoe inertial measurement. Background Technology

[0002] In existing IoT sports interaction applications, common human-computer interaction methods include touch control, handheld controllers, and motion recognition based on external cameras / vision. There is a lack of human-computer interaction methods based on shoes, which fails to fully utilize the flexibility of the human body and does not combine entertainment and exercise anytime and anywhere. If it can drive the intelligent upgrading of games and smart shoes, the market potential and market size will be huge.

[0003] If an inertial measurement module is placed inside the shoe for interactive control, users can exercise while having fun, which can help address the physical harm caused by children's addiction to games and thus respond to the national call for fitness for all. However, it faces challenges such as large-map games requiring continuous movement but limited real-world space, existing game shoes limiting interaction to left, right, and jumping, resulting in monotonous gameplay, high error rates and decreased haptic feedback from using both feet as input, and the decreasing accuracy of algorithms in complex in-shoe scenarios, making it difficult to smoothly adapt to mainstream popular games.

[0004] Furthermore, in the prior art, the gyroscope angular velocity and acceleration signals are easily affected by noise and zero bias during integration and attitude calculation, resulting in cumulative drift. This leads to inaccurate identification of the zero velocity range (the static period of the support phase) and unstable attitude and displacement estimation, thereby affecting motion recognition. Summary of the Invention

[0005] In view of the above technical problems, the present invention provides a grid control method for Internet of Things interaction based on in-shoe inertial measurement, which aims to address the limitations of gyroscopes in in-shoe gaming terminals and the limitations of smart shoe gaming terminals and scenarios, for the novel input terminal form of in-shoe inertial measurement for human-computer interaction control.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to one aspect of the present invention, a grid control method for Internet of Things (IoT) interaction based on in-shoe inertial measurement is proposed, the method comprising: An inertial measurement module is installed inside the smart shoe and establishes data communication with a networked terminal running an application; The inertial measurement module acquires multi-axis angular velocity data and multi-axis acceleration data, and performs filtering, zero-bias estimation subtraction, and coordinate reference calibration based on the gravity direction on the angular velocity data and the acceleration data. The angular velocity data is analyzed in the frequency domain using an adaptive sliding window to obtain a step frequency sequence. The frequency domain analysis includes performing a discrete Fourier transform on the signal within the window and extracting the spectral peaks. The window length is then adaptively adjusted based on the step frequency changes of adjacent windows. A zero-velocity interval threshold is determined based on the step frequency sequence, and zero-velocity interval detection is performed. The zero-velocity interval detection includes calculating a test statistic that fuses angular velocity energy and acceleration stability within the sliding window and comparing it with the zero-velocity interval threshold to output the foot stationary interval. The foot stationary interval is used to suppress drift in attitude estimation and complete the start and end segmentation of effective stepping action; the angular velocity temporal features of the segmented effective stepping action are extracted and compared with the feature template library to determine whether the stepping belongs to the target interaction area in the preset grid interaction area. Directional commands are generated based on the target interaction area, wherein the central interaction area is configured as a return command, and the directional interaction areas set around the central interaction area are configured as multiple movement commands respectively. The directional command is sent to the network terminal and mapped to the directional key input in the application. The return command is used as the directional gating condition, so that switching from the current movement command to another movement command is allowed only after the return command is detected, and the current movement command is kept output when the return command is not detected to achieve continuous movement control.

[0008] Furthermore, the smart shoe includes a shoe body, an inertial measurement module disposed in the shoe body, a data processing unit electrically connected to the inertial measurement module, and a wireless communication unit communicatively connected to the network terminal. The inertial measurement module includes a multi-axis gyroscope and a multi-axis accelerometer arranged orthogonally to each other. The inertial measurement module is fixed to a preset part of the shoe body and establishes a sensor coordinate system consistent with the geometric direction of the shoe body. The forward axis of the sensor coordinate system is consistent with the direction of travel of the shoe body, the lateral axis is consistent with the lateral direction of the shoe body, and the vertical axis is consistent with the direction of gravity. The network terminal is a mobile terminal or a computing terminal connected to a large-screen display device. The application runs on the network terminal and receives the directional instructions to drive the game character to move. The smart shoe is a single shoe or a pair of shoes.

[0009] Furthermore, when filtering the angular velocity data and the acceleration data, the process includes: performing low-pass filtering on the angular velocity data and the acceleration data to suppress high-frequency noise and jitter caused by human body micro-vibration, wherein the low-pass filtering adopts any one or more combinations of Butterworth filter, finite impulse response filter or moving average filter. When performing the zero bias estimation subtraction, the process includes: collecting a segment of angular velocity data and acceleration data when the foot is in the foot stationary range, calculating their statistics as the zero bias of the inertial measurement module, and then subtracting the subsequent collected data in real time. The coordinate reference calibration based on the direction of gravity includes: calculating the pitch and roll attitude components corresponding to the direction of gravity from the acceleration data in the foot stationary range, and using them to calibrate the initial orientation of the attitude estimation, and eliminating the difference in the initial orientation of the shoe body by rotating the angular velocity data to a stable reference coordinate system. The effective stepping action start and end segmentation includes: detecting a sudden increase in motion based on the vector amplitude of the acceleration data or the modulus of the angular velocity data to determine the start point of the action, and determining the end point of the action after the vector amplitude or modulus falls back to a preset condition, so as to extract a complete action segment and use it for the similarity comparison.

[0010] Furthermore, the frequency domain analysis of the adaptive sliding window specifically includes: Initialize the window length and zero-padded the window boundaries to improve frequency resolution; Within each window, the axial component of the angular velocity data that is related to the gait periodicity or the magnitude of the angular velocity data is selected as the analysis signal, and the discrete Fourier transform is performed on the analysis signal to obtain the spectrum; Multiple peak frequencies and their amplitudes are identified in the spectrum. The peak frequency with the largest amplitude is selected as the human gait frequency of the current window within the preset candidate range of the fundamental frequency of human gait frequency. When there is no peak frequency that meets the confidence condition within the candidate range of the fundamental frequency, the peak frequency with the largest amplitude in the spectrum is selected and corrected in combination with its continuity constraint with the gait frequency sequence to suppress harmonic misselection. The gait frequency of adjacent windows is differentially calculated to obtain the gait frequency change. When the gait frequency change is greater than a preset positive threshold, the window length is reduced; when the gait frequency change is less than a preset negative threshold, the window length is increased. The gait frequency is then recalculated under the updated window length. This process is iterated to obtain a continuous and stable gait frequency sequence. The frequency domain analysis is an adaptive short-time frequency domain analysis, which can dynamically adjust the trade-off between time resolution and frequency resolution as the gait frequency changes.

[0011] Furthermore, the zero-velocity interval detection includes: Within each detection window, the window-averaged acceleration vector is calculated and normalized to obtain an estimate of the gravity direction. The deviation between the acceleration vector at each moment within the window and the gravitational component corresponding to the gravitational direction estimate is calculated to characterize acceleration stability, and the sum of squares of the angular velocity vector at each moment within the window is calculated to characterize angular velocity energy. The test statistic is obtained by normalizing and fusing the acceleration stability and the angular velocity energy according to the preset noise weights. Using the gait frequency sequence as input, a threshold function is established through the pre-calibrated gait frequency and the optimal threshold sample set, and the zero-velocity interval threshold is obtained in real time. The threshold function adopts polynomial fitting, spline interpolation, or piecewise function mapping, and the optimal threshold sample set covers the gait frequency changes under different motion states. When the test statistic is less than or equal to the zero-velocity interval threshold, it is determined that the current moment belongs to the foot stationary interval and is regarded as the support phase zero-velocity interval; otherwise, it is determined as a non-stationary interval. The support phase zero-velocity interval corresponds to the period when the foot is in complete contact with the support surface and the acceleration modulus exhibits gravity-dominated characteristics.

[0012] Furthermore, the attitude estimation is represented using quaternions and sensor fusion is performed through extended Kalman filtering or complementary filtering, wherein: The prediction stage of the extended Kalman filter includes calculating the quaternion increment based on the angular velocity data and updating the attitude quaternion, rotating the acceleration data to a preset navigation coordinate system based on the attitude quaternion and subtracting the gravity component to update the velocity and displacement, wherein the preset navigation coordinate system is a North-East-Earth coordinate system or a local coordinate system equivalent to the North-East-Earth coordinate system. The update phase of the extended Kalman filter constrains the velocity to zero or near zero when the foot is detected to be stationary, and introduces heading or attitude constraints to generate measurement residuals. It corrects the state vector containing position error, velocity error, attitude error and the zero bias error of the inertial measurement module, and achieves filter stabilization by adjusting the state covariance, process noise covariance and measurement noise covariance. The corrected state vector is used to update the attitude quaternion, velocity, and displacement, and to update the zero bias of the inertial measurement module to suppress cumulative drift caused by integration.

[0013] Furthermore, the angular velocity time-series features include peak features characterizing action intensity, mean features characterizing action bias, zero-crossing rate features characterizing waveform changes, energy features characterizing energy distribution, and angular change trajectory features obtained by integrating the angular velocity data. The angular velocity timing characteristics also include the main frequency component, harmonic components, and frequency band energy distribution characteristics obtained through fast Fourier transform or discrete Fourier transform; The angular velocity time-series features, together with the impact vector, stamping duration, vector amplitude change rate, and post-impact decay characteristics extracted from the acceleration data, constitute a multi-dimensional feature vector. Before entering the feature template library for comparison, the vector vector is normalized, time-aligned, detrended, or resampled. Furthermore, the multi-dimensional feature vector can be updated to the baseline within the foot stationary interval.

[0014] Furthermore, the feature template library is established by: guiding the user to perform multiple standard stepping actions for each interactive area in the preset grid interactive area during the initialization or calibration phase, collecting the corresponding multidimensional feature vectors respectively, and averaging, clustering to select representative samples, constructing statistical distributions or generating prototype sequences for multiple samples of the same interactive area to form templates for the corresponding interactive areas. The similarity comparison includes using a dynamic time warping algorithm to elastically align the real-time angular velocity temporal features with the template and calculate a distance metric, and determining the similarity between the angular velocity temporal features and the templates in the feature template library based on the distance metric; or, the similarity comparison includes: The multidimensional feature vector is input into the trained classifier to output the target interaction region. The classifier includes one of support vector machine, decision tree, random forest or neural network. When the difference between the maximum similarity and the second largest similarity does not meet the confidence condition or when the confidence of the classifier output does not meet the preset threshold condition, the rejection result is output and the state is maintained or reverted to the state of the return instruction. In the rejection state, samples can be re-collected to incrementally update the feature template library.

[0015] Furthermore, the mapping of the direction command also includes: The application displays a virtual directional key layout corresponding to the preset grid interaction area, and maps the movement commands to continuous press events in the preset movement directions of up, down, left, right and diagonal, and maps the return command to a release event or a stop event. The directional gating conditions also include suppressing unintentional swaying based on the foot stationary zone and the impact characteristics extracted from the acceleration data, and triggering the directional command only when a valid pedaling action that meets the combination of impact intensity, action duration, and stationary zone conditions is detected.

[0016] Furthermore, the method also includes: When an invalid stepping action is detected or an action that does not match the feature template library is detected, the direction command or the return command is not output; and on the network terminal, it is allowed to work in conjunction with a handheld remote control, touch interface or other peripheral inputs to respectively undertake perspective, skill or confirmation operations, while the smart shoe undertakes the movement direction operation, so that a stepping action in a confined space can generate continuous movement in the game and stop and change direction by returning to the position.

[0017] The technical solution disclosed herein has the following beneficial effects: This invention utilizes the interaction rules of a grid-based directional key and a centering gating system to transform in-shoe motion recognition results into stable semantics for continuous directional key presses / releases. This allows for continuous movement while maintaining the same direction of motion without returning to center, and stops and allows for direction changes upon returning to center. From an interaction perspective, this reduces the high error rate and haptic feedback issues caused by synchronized two-foot direction changes, and enables smooth operation of large-map games within limited spaces. Simultaneously, it avoids the problem of zero-speed intervals or interval mismatches caused by fixed thresholds when step frequency changes, thus more stably obtaining the foot's stationary interval and using it to suppress inertial integral drift, improving stability and reliability over long-term operation. Attached Figure Description

[0018] Figure 1 This is a flowchart of a grid control method for IoT interaction based on in-shoe inertial measurement, as described in an embodiment of this specification. Figure 2 This is a schematic diagram of the interface of the network terminal in the embodiments of this specification; Figure 3 This is a schematic diagram showing the installation location of the inertial measurement module in the embodiments of this specification. Detailed Implementation

[0019] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0020] Furthermore, the accompanying drawings are merely illustrative of this disclosure. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0021] This invention provides a grid control method for IoT interaction of products based on in-shoe inertial measurement. (Refer to...) Figure 1 The diagram shown is a flowchart illustrating a grid control method for IoT interaction based on in-shoe inertial measurement, according to an embodiment of the present invention. This method can be applied to devices such as personal computers, servers, tablets, and mobile phones. The method can be implemented by software and / or hardware, and specifically includes the following steps S101-S106: In step S101, an inertial measurement module is installed inside the smart shoe and establishes data communication with a network terminal running the application.

[0022] The smart shoe serves as the control hardware for the interactive system. The overall interactive system also includes mobile terminals connected to the internet or display devices such as TVs, projectors, mobile phones, and tablets with installed applications. The application receives control information from the smart shoe to drive the interactive content, thus forming a basic link between shoe-end data acquisition and command issuance, and terminal reception and execution. The design focus of this link is to place inertial measurement-related data acquisition hardware inside the shoe, allowing for the continuous and stable acquisition and uploading of inertial information generated by the user's natural movements such as stepping and swinging, providing a data foundation for subsequent action recognition and command generation.

[0023] The smart shoe includes a shoe body, an inertial measurement module installed inside the shoe body, a data processing unit electrically connected to the inertial measurement module, and a wireless communication unit connected to a network terminal. The inertial measurement module includes a multi-axis gyroscope and a multi-axis accelerometer arranged orthogonally to each other. The inertial measurement module is fixed to a preset part of the shoe body and establishes a sensor coordinate system consistent with the geometric direction of the shoe body. The forward axis of the sensor coordinate system is consistent with the direction of travel of the shoe body, the lateral axis is consistent with the lateral direction of the shoe body, and the vertical axis is consistent with the direction of gravity. The network terminal is a mobile terminal or a computing terminal connected to a large-screen display device. The application runs on the network terminal and receives directional commands to drive the game character to move. The smart shoe can be a single shoe body or a pair of shoe bodies.

[0024] The inertial measurement module can be set at any position on the shoe body, for example, such as Figure 3 As shown, it can be placed inside the shoe at the junction of the heel and the arch.

[0025] Specifically, the data processing unit is electrically connected to the inertial measurement module, used to read multi-axis angular velocity and multi-axis acceleration in real time through the sensor interface, forming a continuous inertial data stream (e.g., reading triaxial angular velocity data and organizing it synchronously with acceleration data). To ensure directional consistency and comparability in subsequent calculations, the inertial measurement module is fixed to a preset part of the shoe body (e.g., fixed to the toe area), and a sensor coordinate system consistent with the geometric orientation of the shoe body is established: the origin of the coordinate system can correspond to the position of the carrier's center of mass, the x-axis points forward along the longitudinal direction of the shoe body, the y-axis points to the right along the transverse direction of the shoe body, and the z-axis points downward along the vertical direction, thus forming an orthogonal coordinate system; through the above fixed installation and coordinate system agreement, the same action can still have a consistent axial physical meaning under different wearing states.

[0026] In terms of data communication, the wireless communication unit is used to establish a wireless link between the smart shoe and the networked terminal. In one embodiment, the inertial measurement module (or its integrated communication module) establishes a connection with the networked terminal via Bluetooth, thereby transmitting the acquired inertial data to the networked terminal for reception and subsequent processing by the data acquisition program or application running on the networked terminal. Establishing data communication here may include processes such as device pairing, connection maintenance, and data reporting: after the connection is established, the smart shoe continuously outputs multi-axis inertial data frames corresponding to the time sequence, and the networked terminal continuously receives and buffers the data frames, thereby providing input data for subsequent steps such as filtering, calibration, gait analysis, and motion segmentation and recognition.

[0027] In step S102, the inertial measurement module acquires multi-axis angular velocity data and multi-axis acceleration data, and performs filtering, zero-bias estimation subtraction, and coordinate reference calibration based on gravity direction on the angular velocity data and acceleration data.

[0028] The filtering process for angular velocity and acceleration data includes: performing low-pass filtering on the angular velocity and acceleration data to suppress high-frequency noise and jitter caused by human body micro-vibrations. The low-pass filtering employs any one or more combinations of Butterworth filters, finite impulse response filters, or moving average filters. Since the acquired raw inertial data typically contains high-frequency noise and jitter caused by human body micro-vibrations, low-pass filtering is prioritized for angular velocity and acceleration data to reduce the interference of high-frequency components on subsequent feature extraction and threshold determination. The low-pass filter can be a Butterworth filter, a finite impulse response filter, or a moving average filter, or a combination thereof, depending on the implementation requirements. This ensures that the main frequency band information of gait and pedaling is preserved while reducing signal jitter caused by muscle micro-tremors.

[0029] The process of zero-bias estimation and subtraction includes: acquiring a segment of angular velocity and acceleration data when the foot is in a stationary position, calculating its statistics as the zero bias of the inertial measurement module, and then subtracting it from subsequent data in real time. Because gyroscopes have inherent zero-point drift, directly integrating the angular velocity will cause attitude angle drift and accumulate errors over time. Therefore, a short segment of angular velocity and acceleration data is acquired before the start of the movement or during a stable foot position, and the statistics (e.g., the mean) of this segment are calculated as the zero bias. This data is then subtracted from subsequent data in real time to eliminate drift.

[0030] Gravity-oriented coordinate reference calibration includes: calculating the pitch and roll attitude components corresponding to the gravity direction from acceleration data during the foot's static phase, and using this information to calibrate the initial orientation of the attitude estimation; and eliminating differences in the initial orientation of the shoe by rotating the angular velocity data to a stable reference coordinate system. Since gravity-oriented coordinate reference calibration utilizes the characteristic that accelerometers can sense gravity in static or low-speed states, the pitch and roll angles relative to the gravity direction are obtained by calculating the accelerometer data, and this attitude information is used to calibrate the initial orientation. Based on this, the angular velocity data can be transformed into a stable reference coordinate system to eliminate the influence of different initial shoe orientations, thus ensuring that the same type of pedaling action still has comparable temporal patterns under different initial wearing postures.

[0031] The effective pedaling motion segmentation includes: detecting motion spikes based on the vector amplitude of acceleration data or the magnitude of angular velocity data to determine the start point of the motion, and determining the end point of the motion after the vector amplitude or magnitude falls back to a preset condition, thus extracting a complete motion segment for similarity comparison. Specifically, to extract a complete motion segment from the continuous inertial data stream that can be used for subsequent similarity comparison, effective pedaling motion segmentation is necessary. The segmentation strategy can utilize the vector amplitude (SVM) synthesized from the accelerometer or the magnitude of the angular velocity to detect motion spikes: when the signal amplitude exceeds a preset threshold condition, it is determined as the start point of a valid pedaling motion; when the amplitude falls back below the threshold condition, it is determined as the end point of the motion, thereby extracting a complete motion data segment for subsequent comparison. This segmentation method, combined with the aforementioned noise reduction, zero-bias subtraction, and gravity reference calibration, can reduce the disturbance of motion boundaries caused by jitter and drift, improving the integrity and consistency of the motion segment.

[0032] In step S103, the angular velocity data is analyzed in the frequency domain using an adaptive sliding window to obtain the step frequency sequence. The frequency domain analysis includes performing a discrete Fourier transform on the signal within the window and extracting the spectral peaks. Then, the window length is adaptively adjusted according to the step frequency changes of adjacent windows. The zero velocity interval threshold is determined based on the step frequency sequence, and zero velocity interval detection is performed. The zero velocity interval detection includes calculating the test statistic of the fused angular velocity energy and acceleration stability within the sliding window and comparing it with the zero velocity interval threshold to output the foot stationary interval.

[0033] The adaptive sliding window frequency domain analysis specifically includes: initializing the window length and zero-padding the window boundaries to improve frequency resolution; selecting the axial component of the angular velocity data that is periodically related to the gait or selecting the magnitude of the angular velocity data as the analysis signal in each window, and performing a discrete Fourier transform on the analysis signal to obtain the spectrum; identifying multiple peak frequencies and their amplitudes in the spectrum, selecting the peak frequency with the largest amplitude within the preset candidate range of the fundamental frequency of human gait as the human gait frequency of the current window; when there is no peak frequency within the candidate range of the fundamental frequency that meets the confidence condition, selecting the peak frequency with the largest amplitude in the spectrum and combining it with the continuity constraint of the gait frequency sequence for correction to suppress harmonic misselection; differentiating the human gait frequencies of adjacent windows to obtain the gait frequency change; when the gait frequency change is greater than a preset positive threshold, reducing the window length; when the gait frequency change is less than a preset negative threshold, increasing the window length; and recalculating the human gait frequency under the updated window length, iterating cyclically to obtain a continuous and stable gait frequency sequence; wherein, the frequency domain analysis is an adaptive short-time frequency domain analysis, which can dynamically adjust the trade-off between time resolution and frequency resolution as the gait frequency changes.

[0034] As an explanation, the gait frequency acquisition of angular velocity data employs short-time frequency domain analysis using an adaptive sliding window: first, the angular velocity sequence is divided into sliding windows over time, with zero-padding at the boundaries of each window to enhance frequency resolution; then, within each window, the axial angular velocity related to gait periodicity is selected as the analysis signal (for example, the angular velocity of the axis around which the foot rotates can well characterize periodic changes); finally, a discrete Fourier transform is performed on the discrete signal to obtain the spectrum. Discrete signal sequence The discrete Fourier transform is ; Where N is the length of the sequence within the window, and k is the spectral line number. Let be the spectral density function. For the spectrum of each window, multi-peak spectral lines are identified to obtain the set of peak frequencies and the set of amplitudes, and the amplitude of the largest peak in the full spectrum is extracted. and their corresponding frequencies To suppress the misidentification of harmonics as fundamental frequencies, the frequency corresponding to the largest peak preceding the largest peak in the peak sequence can be selected as the human gait frequency for that window. The core idea is that the fundamental frequency peak of human gait often appears before several harmonic peaks, and multiple peaks in the spectrum have harmonic relationships. Then, sliding to an adjacent window yields... Calculate the step frequency difference between adjacent windows: ; And accordingly, the window length is adaptively adjusted: when When the window width is reduced, Increase the window width and then recalculate the new window length. The process is iterated repeatedly to obtain a continuous and stable step frequency sequence; where the step frequency difference threshold is... The setting can be based on the range of variation between the average cadence and the maximum cadence (for example, when the average cadence is about 0.8Hz and the maximum running cadence is about 5Hz, you can take...). This enables adaptive short-time frequency domain analysis that dynamically adjusts the tradeoff between time resolution and frequency resolution as the step frequency changes.

[0035] Furthermore, the zero-velocity interval detection includes: calculating the window-average acceleration vector within each detection window and normalizing the window-average acceleration vector to obtain a gravity direction estimate; calculating the deviation between the acceleration vector at each moment within the window and the gravity component corresponding to the gravity direction estimate to characterize acceleration stability, and simultaneously calculating the sum of squares of the angular velocity vectors at each moment within the window to characterize angular velocity energy; normalizing and fusing acceleration stability and angular velocity energy according to a preset noise weight to obtain a test statistic; using the step frequency sequence as input, establishing a threshold function through a pre-calibrated step frequency and an optimal threshold sample set, and obtaining the zero-velocity interval threshold in real time. The threshold function uses polynomial fitting, spline interpolation, or piecewise function mapping, and the optimal threshold sample set covers step frequency changes under different motion states; when the test statistic is less than or equal to the zero-velocity interval threshold, the current moment is determined to belong to the foot stationary interval and is taken as the support phase zero-velocity interval; otherwise, it is determined to be a non-stationary interval. The support phase zero-velocity interval corresponds to the period when the foot is in complete contact with the support surface and the acceleration modulus exhibits gravity-dominated characteristics.

[0036] Based on the above step frequency sequence, the zero-velocity interval threshold is further determined, and zero-velocity interval detection is performed within the sliding detection window. The detection employs a test statistic that fuses angular velocity energy and acceleration stability: assuming triaxial acceleration output... The angular velocity triaxial output is Within a sliding window of length w, first calculate the average acceleration vector of the window. : ; Its direction is then used for gravity direction estimation; the deviation of acceleration relative to the gravity direction at each time point within the window (characterizing acceleration stability) and the sum of squares of angular velocities (characterizing angular velocity energy) are then calculated to construct the generalized likelihood ratio test statistic: ; in , These are the variances of acceleration noise and angular velocity noise, respectively. For local gravitational acceleration (e.g., 9.78) When the foot is fully in contact with the support surface, the sensor output is approximately constant, exhibiting an angular velocity close to zero and acceleration dominated by gravity (e.g., acceleration along a certain axis is close to gravitational acceleration). Therefore, this statistic tends to be smaller during the stationary phase. Finally, the statistic is compared with the zero-velocity threshold. Compare and output the resting intervals of the feet: ; in The current moment is determined to be within the zero-speed range. It is determined to be a non-stationary interval.

[0037] In step S103, the zero velocity range threshold Instead of using a fixed value, a dynamic threshold driven by a step frequency sequence is employed: by conducting experimental statistics on the collected data under different step frequency conditions, a step frequency-optimal threshold sample set can be obtained, and a threshold function can be established based on this to achieve real-time threshold calculation; the threshold function can be modeled using linear, quadratic, or cubic spline methods, where cubic fitting can be expressed as:

[0038] Where a, b, c, and d are fitting coefficients.

[0039] For example, a=12499.609; b=105547.262; c=-68955.180; d=12312.054. This allows the threshold to adaptively adjust as the gait frequency changes: lower gait frequency (slow walking) and higher gait frequency (brisk walking / running) correspond to different optimal threshold levels, thus more robustly distinguishing between stationary and non-stationary intervals at various gait speeds and reducing false positives and false negatives caused by threshold mismatch.

[0040] In step S104, the drift in attitude estimation is suppressed by using the foot stationary interval and the start and end segmentation of the effective stepping action is completed; the angular velocity temporal features of the segmented effective stepping action are extracted and compared with the feature template library to determine whether the stepping belongs to the target interaction area in the preset grid interaction area.

[0041] Attitude estimation is represented using quaternions and fused from sensors using extended Kalman filtering or complementary filtering, where: The prediction stage of the extended Kalman filter includes calculating the quaternion increment based on the angular velocity data and updating the attitude quaternion, rotating the acceleration data to a preset navigation coordinate system based on the attitude quaternion and subtracting the gravity component to update the velocity and displacement. The preset navigation coordinate system is the North-East-Earth coordinate system or a local coordinate system equivalent to the North-East-Earth coordinate system.

[0042] When using the previously obtained foot stillness interval to suppress attitude estimation drift, the attitude can be represented by quaternions and band-limited updated in the North-East-Earth (NED) coordinate system or equivalent local coordinate system: After obtaining one frame of inertial data, the attitude is recursively updated using quaternions as follows: ; in Update the matrix for quaternions. Let be the angular velocity at time k. The sampling interval and Meanwhile, position and velocity are recursively derived using inertial navigation: ; ; in For location, For speed, For acceleration, This indicates that the acceleration is rotated to the navigation coordinate system using attitude quaternions. It is the constant of gravitational acceleration.

[0043] The update phase of the extended Kalman filter constrains the velocity to zero or near zero when the foot is detected to be stationary, and introduces heading or attitude constraints to generate measurement residuals. It corrects the state vector containing position error, velocity error, attitude error and zero bias error of the inertial measurement module, and achieves filter stabilization by adjusting the state covariance, process noise covariance and measurement noise covariance.

[0044] In the extended Kalman filter update performed within the foot stationary region, the core principle is to introduce physical constraints with zero or near-zero velocity into the measurement, and combine this with heading / attitude-related constraints to form measurement residuals, thereby correcting the error state and suppressing integral accumulation drift. The error state vector can be defined as follows: ; in For positional error, For speed error, For attitude error, For gyroscope zero bias error, This is the zero bias error of the accelerometer. The measurement equation is: ; When a stationary zone is detected at the feet, the error measurement vector can be obtained by subtracting the velocity calculated by the inertial navigation system from the velocity within the stationary zone (which is approximately zero) and the heading / attitude constraints. The corresponding measurement matrix can be represented as: ; In implementation, the stability of filtering can be improved by adjusting the state estimation covariance matrix P, the process noise covariance matrix Q, and the measurement noise covariance matrix R; and after the update is completed, the corrected error state is fed back to the attitude quaternion, velocity, displacement, and zero bias term to achieve continuous suppression of drift.

[0045] As a supplement, the angular velocity time-series features include peak features characterizing motion intensity, mean features characterizing motion bias, zero-crossing rate features characterizing waveform changes, energy features characterizing energy distribution, and angular change trajectory features obtained by integrating the angular velocity data. The angular velocity time-series features also include the dominant frequency component, harmonic components, and frequency band energy distribution features obtained through Fast Fourier Transform or Discrete Fourier Transform. The angular velocity time-series features, together with the impact vector, stamping duration, vector amplitude change rate, and post-impact attenuation characteristics extracted from the acceleration data, constitute a multi-dimensional feature vector. This vector undergoes normalization, time alignment, detrending, or resampling before being compared with the feature template library, and the multi-dimensional feature vector can be updated to the baseline within the foot's stationary region.

[0046] After drift suppression, the stationary area of ​​the foot can serve as a stable anchor point for the action segment: when the signal enters motion from the stationary area and returns to the stationary area, a complete and clearly defined effective stamping action segment can be formed, facilitating subsequent feature extraction and comparison. The determination and filtering of effective stamping actions can be combined with accelerometer verification of action effectiveness. By analyzing the peak value and pattern of the impact force at the moment of stamping, intentional stamping and unintentional swaying can be distinguished, thereby reducing the false trigger rate. For each captured action segment, the angular velocity will exhibit a unique sequence pattern on the time axis. Peak features related to action intensity, mean features related to action bias, zero-crossing rate features related to waveform changes, and energy features related to energy distribution can be extracted from the angular velocity sequence. Furthermore, time integration of the angular velocity can yield the trajectory features of angular changes. Simultaneously, Fast Fourier Transform or Discrete Fourier Transform can be performed on the angular velocity sequence to obtain the main frequency component, harmonic components, and frequency band energy distribution features, enhancing the separability of stamping at different rhythms. Furthermore, angular velocity features can be fused with impact force vectors derived from acceleration, stomping duration, vector amplitude change rate, and post-impact decay characteristics into a multi-dimensional feature vector. Normalization, time alignment, detrending, or resampling processing are then performed before comparison. Simultaneously, the baseline is updated using the static interval to improve robustness across users, shoe types, and force levels.

[0047] When the feature template library is established, it includes: guiding users to perform multiple standard stepping actions for each interactive area in the preset grid interactive area during the initialization or calibration phase, collecting corresponding multi-dimensional feature vectors, and averaging and clustering multiple samples of the same interactive area to select representative samples, constructing statistical distributions or generating prototype sequences to form templates for the corresponding interactive areas.

[0048] Similarity comparison includes using a dynamic time warping algorithm to elastically align real-time angular velocity temporal features with templates and calculate a distance metric, and determining the similarity between the angular velocity temporal features and templates in the feature template library based on the distance metric; or, similarity comparison includes: inputting a multi-dimensional feature vector into a trained classifier to output the target interaction region, wherein the classifier includes one of support vector machines, decision trees, random forests, or neural networks; when the difference between the maximum similarity and the second largest similarity does not meet the confidence condition or when the confidence of the classifier output does not meet the preset threshold condition, a rejection result is output and the state is maintained or reverted to the return instruction state, and in the rejection state, samples can be re-collected to incrementally update the feature template library.

[0049] The feature template library can be established during the initialization or calibration phase: multiple standard stomping actions are recorded for each interactive area of ​​the nine-square grid to form a representative template for each area; during real-time recognition, the angular velocity sequence of the current action is matched with each template in the template library, and the similarity is calculated through dynamic time warping (DTW) or correlation analysis. Finally, the interactive area corresponding to the template with the highest similarity is determined as the stomping location. In another implementation, the multi-dimensional feature vector can be input into the classifier trained to output the target interactive area, and a rejection strategy can be set when the confidence level is insufficient: when the difference between the maximum and second-largest similarity is insufficient or the classifier confidence level is insufficient, rejection is output and the system is maintained or reverted to the middle state. At the same time, samples can be re-collected to incrementally update the template library to maintain stability and reliability in real complex environments.

[0050] In step S105, directional instructions are generated based on the target interaction area, wherein the central interaction area is configured as a return instruction, and the directional interaction areas set around the central interaction area are configured as multiple movement instructions.

[0051] The generation of directional commands is based on the predefined nine-grid interaction area, such as... Figure 2 As shown, the interactive interface is structured as a 3x3 grid with directional keys and a return key, making the smart shoe the command terminal for directional control. The side directional keys are modified into a 3x3 grid to serve as the display and interaction carrier. The core of the directional commands is to map the recognition result of the target interaction area to the directional or return key semantics of the corresponding cells in the 3x3 grid. The center cell is configured as the return / return key, and the other eight cells are configured as eight directional keys, thus using a unified 3x3 grid rule to cover common movement direction control needs.

[0052] The specific generation rules can establish a one-to-one correspondence based on the position numbers of the nine-square grid: the center square is the return-to-center key, and the eight areas surrounding the center square correspond to the eight directions: top left, top, top right, left, right, bottom left, bottom, and bottom right. When the recognition result falls on the center square, a return-to-center command is generated; when the recognition result falls on any direction square, a movement command corresponding to that direction is generated. To ensure consistent interaction, the direction command can be structured as a direction type field + a status field. The direction type field indicates a direction among the eight directions or a return-to-center, and the status field is used to characterize the current movement state or return-to-center state, enabling the application to implement the interaction logic of stopping after returning to center and allowing another change of direction.

[0053] In step S106, the direction command is sent to the network terminal and mapped to the direction key input in the application. The return command is used as the direction change gating condition, so that the switch from the current movement command to another movement command is allowed only after the return command is detected, and the current movement command is kept output when the return command is not detected to achieve continuous movement control.

[0054] The mapping of directional commands also includes: displaying a virtual directional key layout corresponding to the preset grid interaction area in the application, and mapping the movement commands to continuous press events in the preset movement directions of up, down, left, right and diagonal, respectively, and mapping the return command to a release event or a stop event; the directional gating conditions also include suppressing unintentional swaying based on the foot stationary interval and the impact characteristics extracted from acceleration data, and triggering directional commands only when a valid pedaling action that meets the combination of impact intensity, action duration and stationary interval conditions is detected.

[0055] In this system, directional commands are transmitted wirelessly to the networked terminal, where the application handles input mapping and state maintenance. The application displays the directional keys in a 3x3 grid, mapping the directional commands output by the smart shoe to continuous press semantics in the corresponding direction, and mapping the return-to-center command to release / stop semantics. This allows for continuous output of the same directional input within the program after a single step. In the 3x3 grid interaction, the center square is the center key, and the other squares correspond to the eight directions. When the recognition result falls on a directional square, the application triggers and maintains the press event of that directional key; when the recognition result falls on the center key, the application triggers the release event of the directional key and enters a stop state. In large map scenarios, one can freely take a step in any direction, and without returning to square 5, the movement will continue in that direction. Returning to the center and stopping allows for reversal, achieving continuous movement control and a controllable stop / reversal experience.

[0056] To implement the return-to-center command as a gating condition for direction changes, the application can maintain a current movement direction state: when the current state is a certain direction, the current movement direction cannot be switched to another direction unless a return-to-center command corresponding to the center button is received; during the period when no return-to-center command is detected, the press event corresponding to the current movement direction is continuously output, thereby avoiding direction jitter and frequent direction changes. Only when a return-to-center command corresponding to the center button is detected is the gating released and the next direction command allowed to take effect, forming a closed-loop control logic of "direction - return to center - then direction", consistent with the interaction rule of "always returning to the center of grid 5".

[0057] Alternatively, if you miss a step back to center, you can step on the ground twice to return to center.

[0058] In one embodiment, the method further includes: When an invalid stepping action is detected or an action that does not match the feature template library is detected, no directional command or a return command is output; and on networked terminals, it is allowed to work in conjunction with handheld remote control, touch interface or other peripheral inputs to respectively handle perspective, skill or confirmation operations, while the smart shoe handles the movement direction operation, so that a single step in a confined space can generate continuous movement in the game and stop and change direction by returning to the original position.

[0059] To reduce false triggers and maintain interaction stability, when an invalid stepping action is detected, directional commands can be withheld or a return command can be issued to keep the application in a stopped state. The identification of invalid stepping actions can utilize accelerometer data to distinguish between "intentional stepping" and "unintentional shaking (such as walking or jumping)." Invalid actions are filtered by analyzing the impact force (peak acceleration) at the moment of stepping and the pattern, thereby reducing the false trigger rate. Furthermore, in advanced implementations, in addition to angular velocity sequence patterns, features derived from accelerometer data, such as impact force vectors and stepping duration, can be fused to form multi-dimensional feature vectors, improving accuracy and robustness. Simultaneously, when an action does not match the feature template library, it can be considered a recognition failure. Since the recognition stage compares real-time angular velocity data with the templates in each cell and calculates similarity, then the cell corresponding to the template with the highest similarity is used as the judgment result. Therefore, when the overall similarity is low or a clear optimal matching relationship is not formed, directional commands can be withheld or a return command can be issued to avoid forcibly mapping uncertain actions as directional key input.

[0060] Furthermore, on networked terminals, it allows for collaborative input with handheld remote controls, touch interfaces, or other peripherals, enabling different inputs to perform different functions: the smart shoes primarily handle movement direction, while handheld remote controls and other peripherals can be used for viewpoint, skill, or confirmation operations. This avoids overloading too many complex operations onto the in-shoe movements, keeping the algorithm and interaction simple and enhancing the haptic experience. Users only need to take one step in a certain direction to move continuously within the game. When it is necessary to stop or change direction, the user can stop by returning to the center and then change direction again. At the same time, other operations can be completed in conjunction with the handheld remote control, thus enabling smooth control and a combination of entertainment and exercise in large-map games within a limited space.

[0061] As can be seen from the above embodiments, this invention, through the interaction rules of the grid directional keys and centering gating, transforms the in-shoe motion recognition results into stable directional key press / release semantics. This allows for continuous movement while maintaining the same direction of motion without returning to center, and stops and allows for direction changes after returning to center. From an interaction perspective, this reduces the high error rate and haptic issues caused by simultaneous two-foot direction changes, and enables smooth operation of large-map games within a limited space. Simultaneously, it avoids the problem of zero-speed intervals or interval mismatches caused by fixed thresholds when step frequency changes, thus more stably obtaining the foot's stationary interval and using it to suppress inertial integral drift, improving stability and reliability under long-term operation.

[0062] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0063] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A grid control method for IoT interaction based on in-shoe inertial measurement, characterized in that, The method includes: An inertial measurement module is installed inside the smart shoe and establishes data communication with a networked terminal running an application; The inertial measurement module acquires multi-axis angular velocity data and multi-axis acceleration data, and performs filtering, zero-bias estimation subtraction, and coordinate reference calibration based on the gravity direction on the angular velocity data and the acceleration data. The angular velocity data is analyzed in the frequency domain using an adaptive sliding window to obtain a step frequency sequence. The frequency domain analysis includes performing a discrete Fourier transform on the signal within the window and extracting the spectral peaks. The window length is then adaptively adjusted based on the step frequency changes of adjacent windows. A zero-velocity interval threshold is determined based on the step frequency sequence, and zero-velocity interval detection is performed. The zero-velocity interval detection includes calculating a test statistic that fuses angular velocity energy and acceleration stability within the sliding window and comparing it with the zero-velocity interval threshold to output the foot stationary interval. The foot stationary interval is used to suppress drift in attitude estimation and complete the start and end segmentation of effective stepping action; the angular velocity temporal features of the segmented effective stepping action are extracted and compared with the feature template library to determine whether the stepping belongs to the target interaction area in the preset grid interaction area. Directional commands are generated based on the target interaction area, wherein the central interaction area is configured as a return command, and the directional interaction areas set around the central interaction area are configured as multiple movement commands respectively. The directional command is sent to the network terminal and mapped to the directional key input in the application. The return command is used as the directional gating condition, so that switching from the current movement command to another movement command is allowed only after the return command is detected, and the current movement command is kept output when the return command is not detected to achieve continuous movement control.

2. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The smart shoe includes a shoe body, an inertial measurement module disposed in the shoe body, a data processing unit electrically connected to the inertial measurement module, and a wireless communication unit communicatively connected to the network terminal. The inertial measurement module includes a multi-axis gyroscope and a multi-axis accelerometer arranged orthogonally to each other. The inertial measurement module is fixed to a preset part of the shoe body and establishes a sensor coordinate system consistent with the geometric direction of the shoe body. The forward axis of the sensor coordinate system is consistent with the direction of travel of the shoe body, the lateral axis is consistent with the lateral direction of the shoe body, and the vertical axis is consistent with the direction of gravity. The network terminal is a mobile terminal or a computing terminal connected to a large-screen display device. The application runs on the network terminal and receives the directional commands to drive the game character to move. The smart shoe is a single shoe or a pair of shoes.

3. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, When filtering the angular velocity data and the acceleration data, the process includes: performing low-pass filtering on the angular velocity data and the acceleration data to suppress high-frequency noise and jitter caused by human body micro-vibration, wherein the low-pass filtering adopts any one or more combinations of Butterworth filter, finite impulse response filter or moving average filter; When performing the zero bias estimation subtraction, the process includes: collecting a segment of angular velocity data and acceleration data when the foot is in the foot stationary range, calculating their statistics as the zero bias of the inertial measurement module, and then subtracting the subsequent collected data in real time. The coordinate reference calibration based on the direction of gravity includes: calculating the pitch and roll attitude components corresponding to the direction of gravity from the acceleration data in the foot stationary range, and using them to calibrate the initial orientation of the attitude estimation, and eliminating the difference in the initial orientation of the shoe body by rotating the angular velocity data to a stable reference coordinate system. The effective stepping action start and end segmentation includes: detecting a sudden increase in motion based on the vector amplitude of the acceleration data or the modulus of the angular velocity data to determine the start point of the action, and determining the end point of the action after the vector amplitude or modulus falls back to a preset condition, so as to extract a complete action segment and use it for the similarity comparison.

4. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The frequency domain analysis of the adaptive sliding window specifically includes: Initialize the window length and zero-padded the window boundaries to improve frequency resolution; Within each window, the axial component of the angular velocity data that is related to the gait periodicity or the magnitude of the angular velocity data is selected as the analysis signal, and the discrete Fourier transform is performed on the analysis signal to obtain the spectrum; Multiple peak frequencies and their amplitudes are identified in the spectrum. The peak frequency with the largest amplitude is selected as the human gait frequency of the current window within the preset candidate range of the fundamental frequency of human gait frequency. When there is no peak frequency that meets the confidence condition within the candidate range of the fundamental frequency, the peak frequency with the largest amplitude in the spectrum is selected and corrected in combination with its continuity constraint with the gait frequency sequence to suppress harmonic misselection. The gait frequency of adjacent windows is differentially calculated to obtain the gait frequency change. When the gait frequency change is greater than a preset positive threshold, the window length is reduced; when the gait frequency change is less than a preset negative threshold, the window length is increased. The gait frequency is then recalculated under the updated window length. This process is iterated to obtain a continuous and stable gait frequency sequence. The frequency domain analysis is an adaptive short-time frequency domain analysis, which can dynamically adjust the trade-off between time resolution and frequency resolution as the gait frequency changes.

5. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The zero-velocity interval detection includes: Within each detection window, the window-averaged acceleration vector is calculated and normalized to obtain an estimate of the gravity direction. The deviation between the acceleration vector at each moment within the window and the gravitational component corresponding to the gravitational direction estimate is calculated to characterize acceleration stability, and the sum of squares of the angular velocity vector at each moment within the window is calculated to characterize angular velocity energy. The test statistic is obtained by normalizing and fusing the acceleration stability and the angular velocity energy according to the preset noise weights. Using the gait frequency sequence as input, a threshold function is established through the pre-calibrated gait frequency and the optimal threshold sample set, and the zero-velocity interval threshold is obtained in real time. The threshold function adopts polynomial fitting, spline interpolation, or piecewise function mapping, and the optimal threshold sample set covers the gait frequency changes under different motion states. When the test statistic is less than or equal to the zero-velocity interval threshold, it is determined that the current moment belongs to the foot stationary interval and is regarded as the support phase zero-velocity interval; otherwise, it is determined as a non-stationary interval. The support phase zero-velocity interval corresponds to the period when the foot is in complete contact with the support surface and the acceleration modulus exhibits gravity-dominated characteristics.

6. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The attitude estimation is represented by quaternions and sensor fusion is performed using extended Kalman filtering or complementary filtering, wherein: The prediction stage of the extended Kalman filter includes calculating the quaternion increment based on the angular velocity data and updating the attitude quaternion, rotating the acceleration data to a preset navigation coordinate system based on the attitude quaternion and subtracting the gravity component to update the velocity and displacement, wherein the preset navigation coordinate system is a North-East-Earth coordinate system or a local coordinate system equivalent to the North-East-Earth coordinate system; The update phase of the extended Kalman filter constrains the velocity to zero or near zero when the foot is detected to be stationary, and introduces heading or attitude constraints to generate measurement residuals. It corrects the state vector containing position error, velocity error, attitude error and the zero bias error of the inertial measurement module, and achieves filter stabilization by adjusting the state covariance, process noise covariance and measurement noise covariance. The corrected state vector is used to update the attitude quaternion, velocity, and displacement, and to update the zero bias of the inertial measurement module to suppress cumulative drift caused by integration.

7. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The angular velocity time-series features include peak features characterizing motion intensity, mean features characterizing motion bias, zero-crossing rate features characterizing waveform changes, energy features characterizing energy distribution, and angular change trajectory features obtained by integrating the angular velocity data. The angular velocity timing characteristics also include the main frequency component, harmonic components, and frequency band energy distribution characteristics obtained through fast Fourier transform or discrete Fourier transform; The angular velocity time-series features, together with the impact vector, stamping duration, vector amplitude change rate, and post-impact decay characteristics extracted from the acceleration data, constitute a multi-dimensional feature vector. Before entering the feature template library for comparison, the vector vector is normalized, time-aligned, detrended, or resampled. Furthermore, the multi-dimensional feature vector can be updated to the baseline within the foot's stationary interval.

8. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 7, characterized in that, When the feature template library is established, it includes: guiding the user to perform multiple standard stepping actions for each interactive area in the preset grid interactive area during the initialization or calibration phase, collecting the corresponding multidimensional feature vectors respectively, and averaging, clustering to select representative samples, constructing statistical distribution or generating prototype sequences for multiple samples of the same interactive area to form templates for the corresponding interactive areas. The similarity comparison includes using a dynamic time warping algorithm to elastically align the real-time angular velocity temporal features with the template and calculate a distance metric, and determining the similarity between the angular velocity temporal features and the templates in the feature template library based on the distance metric; or, the similarity comparison includes: The multidimensional feature vector is input into the trained classifier to output the target interaction region. The classifier includes one of support vector machine, decision tree, random forest or neural network. When the difference between the maximum similarity and the second largest similarity does not meet the confidence condition or when the confidence of the classifier output does not meet the preset threshold condition, the rejection result is output and the state is maintained or reverted to the state of the return instruction. In the rejection state, samples can be re-collected to incrementally update the feature template library.

9. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The mapping of the direction command also includes: The application displays a virtual directional key layout corresponding to the preset grid interaction area, and maps the movement commands to continuous press events in the preset movement directions of up, down, left, right and diagonal, and maps the return command to a release event or a stop event. The directional gating conditions also include suppressing unintentional swaying based on the foot stationary zone and the impact characteristics extracted from the acceleration data, and triggering the directional command only when a valid pedaling action that meets the combination of impact intensity, action duration, and stationary zone conditions is detected.

10. The grid control method for IoT interaction based on in-shoe inertial measurement according to claim 1, characterized in that, The method further includes: When an invalid stepping action is detected or an action that does not match the feature template library is detected, the direction command or the return command is not output; and on the network terminal, it is allowed to work in conjunction with a handheld remote control, touch interface or other peripheral inputs to respectively undertake perspective, skill or confirmation operations, while the smart shoe undertakes the movement direction operation, so that a stepping action in a confined space can generate continuous movement in the game and stop and change direction by returning to the position.

Images