A zero speed detection method based on adaptive fusion of multi-source inertial information and GNSS

Abstract

The application discloses a zero-speed detection method based on adaptive fusion of multi-source inertial information and GNSS, comprising: acquiring original measurement data of an inertial measurement unit (IMU) and performing preprocessing to generate filtered inertial data; based on the filtered inertial data, calculating inertial statistical characteristics in a statistical window, and outputting a first intermediate zero-speed state according to a comparison result of the inertial statistical characteristics and a preset threshold; based on the filtered inertial data and external speed input, calculating motion characteristics, and outputting a second intermediate zero-speed state according to the motion characteristics and a preset state switching restriction strategy; acquiring and evaluating signal quality of an auxiliary information source to generate a quality evaluation parameter; according to the quality evaluation parameter, selecting one from preset fusion strategies, adaptively selecting the first intermediate zero-speed state and the second intermediate zero-speed state, and generating and outputting a final zero-speed state. The application improves the robustness in the GNSS recovery / interruption stage.

Description

Technical Field

[0001] This invention relates to the field of integrated navigation and inertial signal processing technology, and in particular to a zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS. Background Technology

[0002] Zero-velocity update (ZUPT) is a key technology for suppressing positioning error divergence in inertial navigation systems, and its effectiveness is highly dependent on the accuracy of zero-velocity detection.

[0003] Existing zero-velocity detection methods are mainly classified into the following categories: (1) Threshold detection methods based on inertial statistical features, which calculate the variance, modulus, GLRT statistics, etc. of acceleration / angular velocity and compare them with preset thresholds, but fixed thresholds are difficult to adapt to different IMU noise levels and environmental changes; (2) Adaptive threshold methods, which can dynamically adjust the threshold, are still within the framework of a single detector and have limited adaptability to complex working conditions; (3) Detection methods based on velocity information, which use GNSS or wheel speed to determine stationary conditions, but their performance deteriorates sharply when GNSS is interrupted or signal quality deteriorates; (4) Detection methods based on machine learning, which rely on the quality and coverage of training data, have insufficient generalization ability and high computational overhead, and are not conducive to real-time embedded implementation.

[0004] The above methods have the following main shortcomings: First, the criteria dimensions of a single detector are limited. Pure inertial methods are prone to misjudgment / missed judgment in vibration and slow start-stop scenarios, while velocity information methods are difficult to work when GNSS signals are weak or interrupted. Second, the generalization ability of fixed thresholds is insufficient, requiring repeated manual calibration. Third, there is a lack of adaptive mechanism for GNSS signal quality, making it impossible to dynamically adjust the decision strategy according to its effectiveness and continuous stability. Fourth, point-by-point independent decision-making lacks temporal stability constraints, and short-term disturbances can easily lead to frequent state jumps.

[0005] Therefore, a zero-velocity detection method is needed that integrates multi-source inertial characteristics and external velocity information, adaptively selects the decision source based on GNSS quality, and has the ability to perform state-time constraints and adaptive threshold updates. Summary of the Invention

[0006] To address the technical problems existing in the prior art, this invention proposes a zero-velocity detection method based on the adaptive fusion of multi-source inertial information and GNSS. This method can enhance the separability between stationary and moving states by utilizing multi-source inertial features, suppress jitter through state switching constraints and hysteresis, and adaptively select a more reliable zero-velocity decision source based on GNSS quality, supporting real-time operation and engineering implementation.

[0007] To achieve the above objectives, this invention provides a zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS, comprising: The raw measurement data from the inertial measurement unit (IMU) is acquired and preprocessed to generate filtered inertial data. Based on the filtered inertial data, several inertial statistical features, including at least the standard deviation of acceleration, the standard deviation of angular velocity, the generalized likelihood ratio GLRT statistic, and the angular velocity energy ARE statistic, are calculated within the statistical window. The first intermediate zero-velocity state is then output based on the comparison results of the inertial statistical features with a preset threshold. Based on the filtered inertial data and external velocity input, calculate at least several motion characteristics including the variance within the acceleration group, the mean difference of forward acceleration, the mean difference of angular velocity, and the velocity magnitude, and output a second intermediate zero-velocity state according to the motion characteristics and a preset state switching restriction strategy. Acquire and evaluate the signal quality of auxiliary information sources, and generate quality assessment parameters; Based on the quality assessment parameters, a preset fusion strategy is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, and the final zero-speed state is generated and output.

[0008] Preferably, generating filtered inertial data includes: Based on the installation orientation of the IMU, the raw measurement data is mapped from the IMU coordinate system to the carrier coordinate system; The mapped data for each axis are subjected to an infinite impulse response (IIR) low-pass filter to generate the filtered inertial data.

[0009] Preferably, the preprocessing further includes a data caching step, specifically including: The filtered inertial data is stored in a circular buffer queue; wherein the length of the circular buffer queue is not less than the length of the statistical window.

[0010] Preferably, the preset threshold is set using an adaptive update strategy, which includes: When it is determined that the current moment is a stationary state, the preset threshold is updated using an exponential moving average method based on the inertial statistical characteristics of the current moment, and the updated threshold is set to be no lower than the preset lower limit.

[0011] Preferably, the preset state switching restriction strategy includes: Set the state transition counter; When a candidate state is different from the current state and meets the corresponding state triggering condition, the counter increments; otherwise, it decrements or is set to zero. The state flip is allowed only when the counter value reaches the preset switching threshold, and the second intermediate zero speed state is output.

[0012] Preferably, based on the quality assessment parameters, one of the preset fusion strategies is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, including: When the auxiliary information source is valid and the historical valid data of the signal is greater than the first preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid and the historical valid data of the signal is less than or equal to the first preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the invalid historical data of the signal is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the historical invalid data of the signal is less than or equal to the second preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is less than or equal to the second preset stability threshold, the first intermediate zero-speed state is selected as the final zero-speed state.

[0013] Preferably, the first preset threshold is a short missing threshold, and the second preset threshold is a long missing threshold, wherein the duration of the short missing threshold is less than the duration of the long missing threshold.

[0014] Preferably, the quality assessment parameters include the validity indicator of the auxiliary information source, the duration of signal loss, the signal loss count, and the number of consecutive valid signals; The duration of signal loss is maintained based on data flag bits and timestamps.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: (1) The present invention sets up a first detector based on inertial statistical features and a second detector based on velocity consistency, and performs adaptive gating fusion on the outputs of the two, so that the final decision result has both the robustness of pure inertial criteria and the sensitivity of velocity-assisted criteria.

[0016] (2) This invention suppresses decision jitter caused by short-term disturbances through a dual mechanism of hysteresis counting and state switching restriction. Through a gating strategy driven by GNSS validity flag, missing duration and consecutive valid count, it adaptively selects the optimal detection source at different stages such as short-term GNSS interruption, long-term interruption and just recovery. The threshold is automatically updated through exponential moving average. This enables the algorithm to maintain stable and reliable detection capabilities under complex engineering conditions such as vehicle vibration, slow start and stop, IMU noise difference and environmental changes. It has good generalization and engineering applicability. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the detector A process according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the detector B process according to an embodiment of the present invention; Figure 4 This is a schematic diagram of GNSS gating according to an embodiment of the present invention; Figure 5 This is a test vehicle trajectory diagram according to an embodiment of the present invention; Figure 6 This is a comparison chart of zero-speed detection results in an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0020] This embodiment proposes a zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS, including: The raw measurement data from the inertial measurement unit (IMU) is acquired and preprocessed to generate filtered inertial data. Based on the filtered inertial data, several inertial statistical features, including at least the standard deviation of acceleration, the standard deviation of angular velocity, the generalized likelihood ratio GLRT statistic, and the angular velocity energy ARE statistic, are calculated within the statistical window. The first intermediate zero-velocity state is then output based on the comparison results of the inertial statistical features with a preset threshold. Based on the filtered inertial data and external velocity input, calculate at least several motion characteristics including the variance within the acceleration group, the mean difference of forward acceleration, the mean difference of angular velocity, and the velocity magnitude, and output a second intermediate zero-velocity state according to the motion characteristics and a preset state switching restriction strategy. Acquire and evaluate the signal quality of auxiliary information sources, and generate quality assessment parameters; Based on the quality assessment parameters, a preset fusion strategy is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, and the final zero-speed state is generated and output.

[0021] Specifically, this technical solution does not employ a neural network model, but instead uses a decision-making process that combines inertial statistical detection, velocity consistency detection, state machine constraints, and GNSS quality gating. Let the sampling time be k, and the original IMU six-axis measurements be: ; In the formula, , , These represent the triaxial acceleration measurements in the IMU's original coordinate system. , , These represent the three-axis angular velocity measurements in the IMU's original coordinate system.

[0022] The output of the first detector A is recorded as the first intermediate zero-speed state. The output of the second detector B is recorded as the second intermediate zero-speed state. The final zero-speed state is recorded as ,and: ; Where 1 represents a stationary state and 0 represents a moving state.

[0023] The final zero-speed state is determined by the gating function. right and The selection yields: ; in, As a marker of GNSS validity, For GNSS missing duration, This represents the number of consecutive valid GNSS events.

[0024] Further, the filtered inertial data is generated, including: Based on the installation orientation of the IMU, the raw measurement data is mapped from the IMU coordinate system to the carrier coordinate system; The mapped data for each axis are subjected to an infinite impulse response (IIR) low-pass filter to generate the filtered inertial data.

[0025] Specifically, establish a circular buffer queue. Set the length of the sliding window Group window length Statistical window length Circular buffer queue length Filter order State transition limit counting threshold GNSS short missing threshold GNSS long missing threshold and the threshold for the number of consecutive stable GNSS operations .

[0026] Circular buffer queue length satisfy: ; Initialize the low-pass filter coefficients, the standard deviation threshold of the first detector A, the GLRT statistic threshold, the ARE (Angular Rate Energy) statistic threshold, the adaptive threshold coefficients, and the variance threshold, differential threshold, velocity threshold, and state machine counter of the second detector B. Simultaneously initialize the GNSS validity flag, GNSS missing duration, GNSS missing count, GNSS continuous valid count, and zero-velocity state variable.

[0027] Acquiring IMU six-axis measurements And based on the relationship between the IMU mounting direction and the carrier coordinate system, axial mapping is performed to obtain the six-axis measurement values ​​in the carrier coordinate system: ; In the formula, This is the axial mapping matrix. These are the mapped six-axis measurements.

[0028] For each axis signal after mapping Perform separately IIR low-pass filter, filter output Represented as: ; In the formula, These are the feedback filter coefficients of the low-pass filter. These are the forward filter coefficients. i For filter order index, The signal at time ki is the six-axis signal. This is the filtered output at time ki.

[0029] After filtering the six-axis data separately, the filtered six-axis inertial data are obtained: ; In the formula, These are the filtered x, y, and z accelerometer data at time k. These are the filtered x, y, and z axis gyroscope data at time k.

[0030] Will Write to circular buffer queue This provides unified filtered inertial data for the first detector A and the second detector B.

[0031] Furthermore, the preset threshold is set using an adaptive update strategy, which includes: When it is determined that the current moment is a stationary state, the preset threshold is updated using an exponential moving average method based on the inertial statistical characteristics of the current moment, and the updated threshold is set to be no lower than the preset lower limit.

[0032] Specifically, the first detector A uses filtered six-axis inertial data. For input, in the statistics window Calculate the mean acceleration, mean angular velocity, standard deviation of acceleration, and standard deviation of angular velocity.

[0033] The mean acceleration and mean angular velocity are expressed as follows: ; In the formula, Let be the mean acceleration at time k. The data are the three-axis acceleration data at time ki. Let be the average angular velocity at time k. This represents the data from the three-axis gyroscope at time ki.

[0034] The standard deviations of acceleration and angular velocity are expressed as follows: ; In the formula, Let be the standard deviation of acceleration at time k. Let be the standard deviation of the angular velocity at time k. Let be the mean acceleration at time k. Let be the average angular velocity at time k.

[0035] The direction of gravity is estimated based on the mean acceleration, and the residual is obtained by removing gravity from the acceleration: ; In the formula, It is the acceleration due to gravity. This represents the acceleration residual after removing gravity.

[0036] We further constructed the generalized likelihood ratio GLRT statistic and the angular velocity amplitude ARE statistic.

[0037] If the standard deviation of acceleration, standard deviation of angular velocity, GLRT statistic, and ARE statistic meet the preset static threshold condition, then the first detector A outputs a static candidate; otherwise, it outputs a moving candidate.

[0038] After processing by static hysteresis counting and dynamic hysteresis counting, the first intermediate zero-velocity state is obtained. .

[0039] when When =1, the relevant thresholds are updated adaptively using an exponential sliding method: ; In the formula, The current threshold, For threshold update coefficients, This is the lower limit of the threshold. It is the inertial statistical feature value corresponding to the currently updated threshold at time k.

[0040] Furthermore, the preset state transition restriction strategy includes: Set the state transition counter; When a candidate state is different from the current state and meets the corresponding state triggering condition, the counter increments; otherwise, it decrements or is set to zero. The state flip is allowed only when the counter value reaches the preset switching threshold, and the second intermediate zero speed state is output.

[0041] Specifically, the second detector B uses filtered six-axis inertial data. and speed input The information is for input.

[0042] First in the sliding window Internal calculation of the mean forward acceleration and its mean difference: ; in, Indicates forward acceleration. This represents the mean difference of forward acceleration.

[0043] Simultaneously calculate the mean value of the Z-axis angular velocity and its mean difference: ; In the formula, Let W be the average Z-axis angular velocity within the sliding window W at time k. This is the filtered Z-axis angular velocity measurement value at time k-1. Let be the mean difference of the Z-axis angular velocity at time k. This represents the average Z-axis angular velocity within the sliding window at time kW.

[0044] In the group window Internal calculation of the variance within the acceleration group And calculate the speed input. Length of the module: ; In the formula, It can be obtained from GNSS speed, wheel speed meter speed, inertial recursive speed, or the output speed of the combined navigation filter.

[0045] Candidate triggering conditions for the state machine are constructed based on the variance within the acceleration group, the mean difference of the forward acceleration, the mean difference of the Z-axis angular velocity, and the velocity modulus.

[0046] When the current state is stationary and the motion triggering condition is met, a motion candidate state is generated; when the current state is in motion and the stationary triggering condition is met, a stationary candidate state is generated.

[0047] Let the state transition counter be When the candidate state is different from the current state and the candidate triggering condition remains true: ; otherwise: ; when At that time, the state is allowed to flip and the second intermediate zero-speed state is output. ;when At that time, maintain the state of the previous moment. .

[0048] This suppresses zero-speed state jitter caused by short-term disturbances by limiting state switching.

[0049] Further, based on the quality assessment parameters, one of the preset fusion strategies is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, including: When the auxiliary information source is valid and the historical valid data of the signal is greater than the first preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid and the historical valid data of the signal is less than or equal to the first preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the invalid historical data of the signal is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the historical invalid data of the signal is less than or equal to the second preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is less than or equal to the second preset stability threshold, the first intermediate zero-speed state is selected as the final zero-speed state.

[0050] Specifically, based on GNSS validity indicators and GNSS missing duration GNSS missing count and GNSS continuous effective count Determine the current GNSS quality and use a gating function. For the first intermediate zero-speed state Second intermediate zero speed state Perform adaptive selection, such as Figure 4 .

[0051] When GNSS is valid and historical GNSS data is validly counted. At that time, assuming the speed information is reliable, the output of the second detector B is used as the final zero-speed state: .

[0052] When GNSS is valid but historical GNSS data is valid counted. At that time, the GNSS is in a recovery or unstable phase, and the output of the first detector A is used: .

[0053] When GNSS is valid and historical GNSS data is invalid, the count is as follows. At that time, assuming the speed information is reliable, the output of the second detector B is used as the final zero-speed state: .

[0054] Counting when GNSS is valid but historical GNSS data is invalid. At that time, the GNSS is in a recovery or unstable phase, and the output of the first detector A is used: .

[0055] When GNSS is invalid or interrupted, the effective count of historical GNSS data is as follows. At that time, short-time velocity recursion is still available, using the output of the second detector B: ; When GNSS is invalid or interrupted, the effective count of historical GNSS data is as follows. At this time, the speed recursion error may accumulate, and the backoff uses the output of the first detector A: .

[0056] Through the above steps, the first detector A provides inertial statistical zero-velocity candidate results independent of GNSS velocity, and the second detector B provides zero-velocity candidate results combining velocity consistency and state machine constraints. The GNSS gating module adaptively selects between the two based on GNSS validity, missing duration, and number of consecutive stable states, thereby obtaining the final zero-velocity state. .

[0057] To more clearly illustrate the technical solution of the present invention, specific embodiments are provided below for description: A zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS, such as Figure 1 ,include: Step 1 At sampling time The system acquires raw triaxial acceleration and triaxial angular velocity measurements from the IMU. In this embodiment, the inertial data is acquired by an automotive-grade MEMS-IMU, specifically the STMicroelectronics ASM330LHHX, with a data output frequency of 100Hz. The vehicle odometer information is output from the vehicle chassis domain controller and transmitted via the CAN interface; this includes the wheel speeds of all four wheels, with a data output frequency of 50Hz. GNSS position and velocity information is output from the Quectel LG69T board, with a data output frequency of 5Hz.

[0058] To facilitate multi-source data fusion processing, the 100Hz sampling time of the IMU is used as the main time reference for zero-velocity detection. For 50Hz wheel velocity data, the wheel velocity input corresponding to the current IMU sampling time is obtained by nearest neighbor hold or linear interpolation. For 5Hz GNSS velocity and position information, the most recent valid GNSS solution result is used as the GNSS input for the current sampling time, and the GNSS data quality is judged by combining the GNSS validity flag, missing duration, and consecutive valid count.

[0059] Original six-axis measurements Represented as: ; In the formula, , , These represent the triaxial acceleration measurements in the IMU's original coordinate system. , , These represent the three-axis angular velocity measurements in the IMU's original coordinate system.

[0060] During initialization, set the length of the sliding window. Group window length Statistical window length Circular buffer queue length Filter order State transition limit counting threshold , hysteresis count threshold Motion hysteresis counting threshold GNSS short missing threshold GNSS long missing threshold and the threshold for the number of consecutive stable GNSS operations .

[0061] Wherein, the length of the circular buffer queue Not less than , and The maximum value in, that is: ; In this embodiment =20、 =100、 =40.

[0062] The IMU output frequency is 100Hz, and the filter is a 6th-order Butterworth low-pass filter, i.e., n=6. Sliding window length... Take 10 to 50 sampling points, group window length Take 20 to 100 sampling points and calculate the statistical window length. Take 50 to 200 sampling points, and set a state switching limit counting threshold. Threshold for the number of consecutive stable GNSS cycles, ranging from 3 to 10. The range is 3 to 20. The above parameters can be adjusted according to the IMU sampling frequency, vehicle motion characteristics, sensor noise level, and actual application scenario.

[0063] Initialize the feedback filter coefficients of the low-pass filter and forward filter coefficients The feedback coefficients of a 6th-order Butterworth low-pass filter are: ; The forward coefficients are: ; The aforementioned filter coefficients are used for low-pass filtering of the IMU's six-axis measurement signals. In the specific implementation, independent filter states are established for the three-axis acceleration and three-axis angular velocity to avoid crosstalk in the filtering results caused by different axial signals sharing the same internal filter state.

[0064] Initialize the standard deviation threshold, GLRT statistic threshold, ARE statistic threshold, and threshold adaptation coefficient required for the first detector A; initialize the variance threshold, difference threshold, velocity threshold, and state machine counter required for the second detector B.

[0065] Let the output of the first detector A be denoted as The output of the second detector B is denoted as The final zero-speed state is recorded as ,in: ; A state value of 1 indicates that the carrier is in a stationary state, and a state value of 0 indicates that the carrier is in motion. During initialization, it can be... , and It can be set to 0, or it can be set to 1 based on the initial static calibration results.

[0066] Simultaneously, initialize the circular buffer queue Q to store the filtered six-axis inertial data; initialize the wheel speed input variable. GNSS velocity input variable GNSS validity indicator GNSS missing duration GNSS missing count and GNSS continuous effective count .in, It is calculated from the wheel speeds of the vehicle's four wheels. The speed information is obtained from the GNSS board output. These variables are used together for subsequent speed consistency determination of the second detector B, as well as GNSS quality assessment and gating fusion.

[0067] Step Two At each sampling time k First, read the raw measurement values ​​of the six axes of the IMU. And perform axial mapping based on the correspondence between the IMU installation direction and the carrier coordinate system.

[0068] Axial mapping can be expressed as: ; In the formula, These are the six-axis measurement values ​​mapped to the carrier coordinate system. This is the axial mapping matrix. The axial mapping matrix P is determined according to the actual installation orientation of the IMU and is used to realize coordinate axis swapping, orientation reversal, or coordinate arrangement adjustment.

[0069] Mapped six-axis measurements Represented as: ; To suppress the impact of high-frequency noise, vehicle vibration, and IMU transient glitches on the zero-speed detection results, the mapped signal of each axis was... Perform separately IIR low-pass filter Its filtered output Represented as: ; In the formula, This is the input signal at the current sampling time. This is the filtered output signal. These are the forward filter coefficients. For feedback filter coefficients, This represents the filter order.

[0070] After filtering the six-axis data separately, the filtered six-axis inertial data are obtained: ; In the formula, the subscript f This indicates that the corresponding measured value has undergone low-pass filtering.

[0071] Subsequently, the filtered six-axis inertial data Write to circular buffer queue Q .

[0072] Circular Buffer Queue Q The length is When the queue is not full, data is written sequentially according to the sampling time order; when the queue is full, the queue is updated by overwriting the oldest data with new data.

[0073] Circular Buffer Queue It can be represented as: ; When the actual amount of cached data is insufficient At that time, calculations are performed using existing sampled data; once the amount of cached data meets the window length requirement, calculations are performed according to the sliding window length. Group window length and statistical window length Extract the corresponding window data to provide data support for the calculation of statistics for the first detector A and the state decision for the second detector B.

[0074] Step 3 like Figure 2 The first detector A uses the filtered six-axis inertial data obtained in step two. For input, acceleration and angular velocity are mainly used in the statistics window. Zero velocity is determined using the mean, standard deviation, gravity-de-accelerated residual, GLRT statistic, and ARE statistic. This detector does not directly rely on GNSS velocity or integrated navigation velocity; therefore, it can serve as a relatively robust candidate criterion for zero velocity when GNSS has just recovered, GNSS quality is unstable, or the reliability of velocity information decreases. The output of the first detector A is recorded as the first intermediate zero velocity state. .

[0075] The mean and standard deviation are as follows: ; ; In the formula, Let be the mean acceleration at time k. The data are the three-axis acceleration data at time ki. Let be the average angular velocity at time k. The data is from the three-axis gyroscope at time ki. Let be the standard deviation of acceleration at time k. Let be the standard deviation of the angular velocity at time k. Let be the mean acceleration at time k. Let be the average angular velocity at time k.

[0076] The gravity-free residual is: ; ; ; In the formula, Let be the mean acceleration at time k. , , These are the components of the mean acceleration vector along the x, y, and z axes, respectively. Let be the magnitude of the mean acceleration vector. A positive number is set to prevent the denominator from being zero. Let k be the filtered triaxial acceleration vector. The acceleration due to gravity is constant. For the first The triaxial acceleration residual vector after removing the gravitational component at any given moment. , , These are the components of the acceleration residual vector along the x, y, and z axes, respectively. Let be the magnitude of the acceleration residual vector. Let be the filtered three-axis angular velocity vector at time k. , , These are the axial components of the triaxial angular velocity vector, respectively. Let be the magnitude of the triaxial angular velocity vector.

[0077] The GLRT and ARE statistics are as follows: Given noise scale parameters ,structure: ; ; In the formula, Let be the generalized likelihood ratio GLRT statistic calculated within the statistical window at time k. The ARE statistic is the angular velocity energy calculated within the statistical window at time k. To calculate the window length, This is the index of the sampling points within the statistics window. Let be the triaxial acceleration residual vector after removing the gravitational component at time ki. For its modulus length, Let be the filtered three-axis angular velocity vector at time ki. For its modulus length, The scale parameter for noise in acceleration measurement. The scale parameter for noise in gyroscope angular velocity measurement.

[0078] Joint judgments and deferred actions include: Set the standard deviation threshold vector With statistical threshold Candidates for stationary / moving states are obtained based on the following logic: If the standard deviation count satisfies the static condition, and or If the candidate is stationary, then it is a candidate for motion; otherwise, it is a candidate for motion. A stationary / motion hysteresis counter is used for output. .

[0079] Threshold adaptive updates include: When the condition of being stationary is met, the threshold is updated using an exponential sliding mechanism and a lower limit is set: ; in , This is the lower limit of the threshold.

[0080] Step Four like Figure 3 The second detector B also uses the filtered six-axis inertial data obtained in step two. As input, and further introduce speed output. Zero-speed discrimination is achieved using the mean difference of acceleration, the mean difference of angular velocity, the within-group variance, and the state transition constraint count. This detector can improve the discrimination stability under scenarios of slow vehicle start-up, deceleration and stopping, and short-term disturbances by utilizing speed consistency and state machine constraints.

[0081] The output of the second detector B is recorded as the second intermediate zero-speed state. .

[0082] Moving average and difference of mean: Forward acceleration : ; Z-axis angular velocity : ; Within-group variance: ; ; Velocity modulus: ; Among them, velocity modulus It can be calculated from GNSS velocity, wheel speedometer velocity, inertial recursive velocity, or the output velocity of the combined navigation filter.

[0083] In a preferred embodiment, when the GNSS signal is valid, the velocity output after fusing the GNSS and IMU signals using a combined navigation filter is used; when the GNSS signal is briefly interrupted, the velocity obtained by the combined navigation filter based on inertial recursion is used as the velocity input to maintain the continuous discrimination capability of detector B.

[0084] State machine triggering conditions: Set the threshold set: Variance threshold: (normal), (start up), (High dynamic / parking); Differential threshold: (Acceleration Difference) (Angular velocity difference); Speed ​​threshold: ; Magnification factor: ; Define candidate trigger logic: Static → Motion Candidate Trigger: ; Movement → Stationary Candidate Trigger: ; State transition limit count: Set switching limit threshold The counter increments when the candidate state differs from the current state and the candidate trigger remains valid. ; When satisfied Flipping is only allowed at certain times: .

[0085] Step 5 GNSS missing duration: Maintain valid GNSS counts based on data flags and timestamps. Or missing count Set the first missing threshold. (Short missing threshold) and second missing threshold (Long missing threshold), and set a threshold for the number of stable GNSS occurrences. .

[0086] Gating fusion strategy: Final output By gate function Decide: ; One feasible strategy is: When GNSS is valid and historical GNSS data is validly counted. At that time, assuming the speed information is reliable, the output of the second detector B is used as the final zero-speed state: .

[0087] When GNSS is valid but historical GNSS data is valid counted. At that time, the GNSS is in a recovery or unstable phase, and the output of the first detector A is used: .

[0088] When GNSS is valid and historical GNSS data is invalid, the count is as follows. At that time, assuming the speed information is reliable, the output of the second detector B is used as the final zero-speed state: .

[0089] Counting when GNSS is valid but historical GNSS data is invalid. At that time, the GNSS is in a recovery or unstable phase, and the output of the first detector A is used: .

[0090] When GNSS is invalid or interrupted, the effective count of historical GNSS data is as follows. At that time, short-time velocity recursion is still available, using the output of the second detector B: ; When GNSS is invalid or interrupted, the effective count of historical GNSS data is as follows. At this time, the speed recursion error may accumulate, and the backoff uses the output of the first detector A: .

[0091] Through the above steps, the first detector A provides inertial statistical zero-velocity candidate results independent of GNSS velocity, and the second detector B provides zero-velocity candidate results combining velocity consistency and state machine constraints. The GNSS gating module adaptively selects between the two based on GNSS validity, missing duration, and number of consecutive stable states, thereby obtaining the final zero-velocity state. .

[0092] This strategy is reflected in the following ways: when GNSS is first restored, robust statistical detector A is used first; after GNSS stabilizes, detector B, which combines speed consistency and state machine constraints, is used; after a long GNSS interruption, it reverts to detector A, which is less dependent on speed.

[0093] The experimental inertial data acquisition used an automotive-grade MEMS-IMU manufactured by STMicroelectronics, model ASM330LHHX, with a data output frequency of 100Hz. Odometer data was acquired from the vehicle chassis domain controller and transmitted via the CAN port, collecting wheel speeds for all four wheels, with a data output frequency of 50Hz. The GNSS position and velocity calculation module used a Quectel LG69T board, with a data output frequency of 5Hz.

[0094] like Figure 5 This is a test vehicle trajectory diagram. The reference trajectory was obtained by the Novatel CPT7 integrated navigation system through post-processing forward and reverse smoothing. The reference position, speed, and attitude output frequency is 100Hz. The Novatel CPT7 GNSS module uses the OEM7 board, and the IMU module is also a MEMS-IMU, using the Honeywell-HG4930 module.

[0095] To verify the effectiveness of the zero-speed detection method in this embodiment, three different data source scenarios were selected for comparative experiments: kxw_odo_1_insert_full, huace_raw_data, and kxw_odo_1.0_insert. These datasets were collected from actual sports car experiments in Nanjing, Wuhan, and other locations.

[0096] The comparison method adopts the traditional MAG&ARE detection strategy, while the method in this embodiment adopts a fusion decision strategy of "six-axis inertial statistical features + parallel dual detectors + adaptive switching of GNSS quality gating".

[0097] To compare the performance of zero-rate detection methods, the evaluation metrics used are false negative rate, false positive rate, and accuracy. Each metric is defined as follows: ; ; ; In the formula, PMD is the false negative rate; PFD is the false positive rate; PRD is the accuracy rate; N0 represents the total number of epochs that are actually stationary; N1 represents the total number of epochs that are actually in motion; N0,1 represents the number of epochs that are actually stationary but are detected as moving; N1,0 represents the number of epochs that are actually moving but are detected as stationary; N0,0 represents the number of epochs that are actually stationary and are also detected as stationary; N1,1 represents the number of epochs that are actually moving and are also detected as moving.

[0098] Table 1 shows the zero-speed test results of vehicle test data under two different schemes in three different situations.

[0099] Table 1 From Table 1 and Figure 6 It can be seen that the zero-velocity detection method based on multi-source inertial information and GNSS adaptive fusion proposed in this embodiment reduces the false detection rate and missed detection rate to the order of 0.3% in three different data source scenarios, with an accuracy rate that stably reaches 99.25%-99.39%. Compared with the comparative methods, this technical solution can more effectively cope with complex working conditions such as vehicle vibration, slow start-stop, and GNSS interruption / recovery, and has significant engineering application value. It can provide a reliable zero-velocity criterion for subsequent ZUPT updates and improvement of integrated navigation accuracy.

[0100] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS, characterized in that, include: The raw measurement data from the inertial measurement unit (IMU) is acquired and preprocessed to generate filtered inertial data. Based on the filtered inertial data, several inertial statistical features, including at least the standard deviation of acceleration, the standard deviation of angular velocity, the generalized likelihood ratio GLRT statistic, and the angular velocity energy ARE statistic, are calculated within the statistical window. The first intermediate zero-velocity state is then output based on the comparison results of the inertial statistical features with a preset threshold. Based on the filtered inertial data and external velocity input, calculate at least several motion characteristics including the variance within the acceleration group, the mean difference of forward acceleration, the mean difference of angular velocity, and the velocity magnitude, and output a second intermediate zero-velocity state according to the motion characteristics and a preset state switching restriction strategy. Acquire and evaluate the signal quality of auxiliary information sources, and generate quality assessment parameters; Based on the quality assessment parameters, a preset fusion strategy is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, and the final zero-speed state is generated and output.

2. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 1, characterized in that, Generate filtered inertial data, including: Based on the installation orientation of the IMU, the raw measurement data is mapped from the IMU coordinate system to the carrier coordinate system; The mapped data for each axis are subjected to an infinite impulse response (IIR) low-pass filter to generate the filtered inertial data.

3. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 2, characterized in that, The preprocessing also includes a data caching step, specifically including: The filtered inertial data is stored in a circular buffer queue; wherein the length of the circular buffer queue is not less than the length of the statistical window.

4. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 1, characterized in that, The preset threshold is set using an adaptive update strategy, which includes: When it is determined that the current moment is a stationary state, the preset threshold is updated using an exponential moving average method based on the inertial statistical characteristics of the current moment, and the updated threshold is set to be no lower than the preset lower limit.

5. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 1, characterized in that, The preset state switching restriction strategy includes: Set the state transition counter; When a candidate state is different from the current state and meets the corresponding state triggering condition, the counter increments; otherwise, it decrements or is set to zero. The state flip is allowed only when the counter value reaches the preset switching threshold, and the second intermediate zero speed state is output.

6. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 1, characterized in that, Based on the quality assessment parameters, one of the preset fusion strategies is selected to adaptively select the first intermediate zero-speed state and the second intermediate zero-speed state, including: When the auxiliary information source is valid and the historical valid data of the signal is greater than the first preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid and the historical valid data of the signal is less than or equal to the first preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the invalid historical data of the signal is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is valid but the historical invalid data of the signal is less than or equal to the second preset stability threshold, the first intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is greater than the second preset stability threshold, the second intermediate zero speed state is selected as the final zero speed state. When the auxiliary information source is invalid and the signal loss duration is less than or equal to the second preset stability threshold, the first intermediate zero-speed state is selected as the final zero-speed state.

7. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 6, characterized in that, The first preset stability threshold is a short missing threshold, and the second preset stability threshold is a long missing threshold, wherein the duration of the short missing threshold is less than the duration of the long missing threshold.

8. The zero-velocity detection method based on adaptive fusion of multi-source inertial information and GNSS according to claim 6, characterized in that, The quality assessment parameters include the validity indicator of the auxiliary information source, the duration of signal loss, the count of signal loss, and the number of consecutive valid signals. The duration of signal loss is maintained based on data flag bits and timestamps.

Images