Vehicle monitoring and positioning data processing method based on multi-source data fusion

Abstract

This invention relates to the field of electronic data digital processing technology, and discloses a method for vehicle monitoring and positioning data processing based on multi-source data fusion. The method includes: acquiring GNSS, IMU, and odometer data; constructing a four-dimensional feature vector containing the number of satellites, carrier-to-noise ratio statistics, and vertical acceleration vibration energy, and identifying the driving environment state in real time using a decision tree classifier; dynamically activating the corresponding fusion processing pipeline (such as tightly coupled, loosely coupled, dead reckoning, or enhanced pipeline with vibration compensation) according to the state type; wherein, under vibration interference conditions, a gyroscope bias compensation and adaptive Butterworth filtering mechanism based on a lookup table is introduced. The system includes data acquisition, environment classification, pipeline selection, and fusion processing modules. This invention improves the positioning robustness and all-scenario adaptability in bumpy scenarios, optimizes the balance between power consumption and accuracy through an on-demand activation strategy, and ensures trajectory continuity through hysteresis switching and covariance propagation.

Description

Technical Field

[0001] This invention belongs to the field of electronic data digital processing technology, specifically relating to a method for vehicle monitoring and positioning data processing based on multi-source data fusion. Background Technology

[0002] With the rapid development of intelligent transportation systems and advanced autonomous driving technologies, high-precision vehicle positioning has become a core support for achieving safe navigation, fleet coordination, and real-time monitoring. In complex urban road environments, Global Navigation Satellite System (GNSS) signals are highly susceptible to interference from tall buildings, overpass reflections, or multipath effects, resulting in severely compromised positioning continuity. To compensate for positioning gaps during GNSS signal absence, existing solutions generally employ inertial navigation systems (INS) and sensors such as wheel speedometers for dead reckoning, relying on multi-source data fusion strategies to maintain trajectory integrity. However, such methods still face many fundamental challenges in practical applications.

[0003] Vertical dynamic disturbances generated when a vehicle travels on bumpy roads significantly affect the attitude calculation accuracy of the inertial measurement unit (IMU). In particular, the dynamic drift of pitch and roll angles is not effectively modeled and compensated, directly leading to a rapid accumulation of position and heading errors during dead reckoning. Existing fusion algorithms mostly assume that the vehicle is in a stable motion state, lacking explicit perception and utilization of instantaneous vibration signals caused by road surface excitation, causing attitude estimation to quickly become inaccurate after a short GNSS interruption.

[0004] Furthermore, mainstream Kalman filtering or factor graph optimization frameworks typically use fixed noise covariance parameters, which cannot dynamically adjust the trust level of IMU data according to the current road conditions. This causes the system to rely excessively on contaminated acceleration and angular velocity observations under severe bumps, further amplifying the positioning error.

[0005] Existing technologies generally suffer from the difficulty of balancing energy efficiency and accuracy: vibration suppression or error compensation modules often operate in a continuous mode, consuming computing resources and power regardless of whether the vehicle is cruising on a highway or in a low-vibration urban congestion scenario, resulting in unnecessary power consumption. At the same time, fusion strategies lack the ability to proactively perceive environmental context (such as GNSS availability, road type, and building density), and cannot switch to a more robust processing mode in advance in typical weak signal areas such as under overpasses, tunnel entrances, or dense urban areas.

[0006] The aforementioned defects are concentrated in key road sections such as overpass weaving areas and underground ramp connection sections, often causing trajectory jumps, positioning drifts, and even service interruptions, which seriously restricts the deployment efficiency of vehicle monitoring systems in high-reliability application scenarios such as autonomous driving decision-making, emergency dispatch response, and safety early warning. Summary of the Invention

[0007] The purpose of this invention is to provide a vehicle monitoring and positioning data processing method based on multi-source data fusion, aiming to solve the problems of dynamic attitude angle drift of inertial navigation modules on bumpy roads, lack of environmental context awareness in sensor fusion strategies, and imbalance between power consumption and accuracy caused by the continuous operation of vibration suppression models in the prior art.

[0008] To achieve the above objectives, according to an aspect of the present invention, a vehicle monitoring and positioning data processing method based on multi-source data fusion is provided, comprising the following steps:

[0009] The system acquires real-time data from multiple sources, including data from the Global Navigation Satellite System (GNSS), inertial measurement unit (INS) data, and vehicle odometer data. The GNSS data includes pseudorange, carrier phase, Doppler shift, and satellite ephemeris information. The INS data includes triaxial acceleration and triaxial angular velocity. The vehicle odometer data includes wheel rotation speed pulse signals.

[0010] Based on the multi-source real-time sensor data, a driving environment feature vector is constructed, and the current driving environment state of the vehicle is classified into a preset driving environment state type in real time according to the driving environment feature vector; the preset driving environment state types include open sky state, urban canyon state, tunnel or shaded state, and vibration interference state.

[0011] Based on the classified driving environment state type, a target data processing pipeline that uniquely corresponds to the current driving environment state type is selected and activated from a preset set of processing pipelines; the set of processing pipelines consists of independent data fusion processing units pre-configured for different driving environment state types.

[0012] The activated target data processing pipeline is used to fuse and solve the multi-source real-time sensor data to generate and output the vehicle's real-time three-dimensional position, three-dimensional velocity, and three-dimensional attitude information.

[0013] Furthermore, the step of constructing the driving environment feature vector and performing real-time classification of the driving environment state specifically includes:

[0014] Set a fixed-duration sliding time window, and extract the number of satellites, the mean and variance of the signal-to-noise ratio of all visible satellites from the global navigation satellite system data within this time window;

[0015] Within the same sliding time window, power spectral density analysis is performed on the vertical axis acceleration signal in the inertial measurement unit data to extract the signal energy integral value within the preset vibration frequency band;

[0016] Construct a four-dimensional driving environment feature vector that includes the number of satellites, the mean signal-to-noise ratio, the variance of the signal-to-noise ratio, and the integral value of the vertical axis acceleration signal energy.

[0017] The driving environment feature vector is input into a preset rule-based decision tree classifier. Through multi-level comparison with preset thresholds, a unique driving environment state type code is output. Among them, the open sky state corresponds to a high number of satellites, a high mean carrier-to-noise ratio, a low carrier-to-noise ratio variance, and low vibration energy; the urban canyon state corresponds to a medium to low number of satellites, a low mean carrier-to-noise ratio, a high carrier-to-noise ratio variance, and low vibration energy; the tunnel or obstruction state corresponds to a zero number of satellites and low vibration energy; and the vibration interference state corresponds to any satellite signal condition where the integral value of the vertical axis acceleration signal energy exceeds a preset vibration trigger threshold.

[0018] Furthermore, the step of selecting and activating the target data processing pipeline specifically includes:

[0019] When the driving environment is open sky, the tightly coupled global navigation satellite system and inertial measurement unit fusion processing pipeline is activated. This pipeline uses an extended Kalman filter to directly use the pseudorange and carrier phase observations of the global navigation satellite system as measurements to correct the position, velocity and attitude errors calculated by the inertial navigation system. At this time, the vibration suppression and compensation module is inactive.

[0020] When the driving environment state type is urban canyon state, the loosely coupled global navigation satellite system and inertial measurement unit fusion processing pipeline is activated. This pipeline first independently calculates the positioning results of the global navigation satellite system and the positioning results of the inertial navigation system, and then uses the difference between their positions and velocities as a quantity to input to the extended Kalman filter for state estimation, so as to reduce the pollution of the tightly coupled solution by the multipath effect.

[0021] When the driving environment is in a tunnel or a covered state, the inertial measurement unit and the vehicle odometer dead reckoning pipeline are activated. This pipeline uses the wheel speed pulse signal from the vehicle controller area network bus to calculate the vehicle's forward speed as an external speed observation value, constrains the speed integral drift of the inertial measurement unit, and realizes pure dead reckoning.

[0022] When the driving environment state is a vibration disturbance state, the enhanced fusion processing pipeline with vibration adaptive compensation is activated.

[0023] As one embodiment of the present invention, the enhanced fusion processing pipeline with vibration adaptive compensation includes a vibration-induced attitude drift compensation module and an adaptive vibration suppression filtering module.

[0024] The specific processing flow of the vibration-induced attitude drift compensation module is as follows:

[0025] A short-time Fourier transform is performed on the vertical axis acceleration signal acquired by the inertial measurement unit to identify the principal vibration frequency and amplitude;

[0026] Based on the identified main vibration frequency and amplitude, the dynamic bias compensation vector for the pitch and roll angle gyroscopes is obtained by querying a pre-calibrated two-dimensional lookup table; the two-dimensional lookup table is established by offline calibration of the gyroscope output bias under vibration inputs of different frequencies and amplitudes on a six-degree-of-freedom vibration platform.

[0027] In the state prediction step of the extended Kalman filter, the dynamic bias compensation vector is superimposed on the static bias estimate of the gyroscope, thereby actively canceling the cross-coupling error and angular rate integral error caused by mechanical vibration in the state propagation model.

[0028] As one embodiment of the present invention, the specific processing flow of the adaptive vibration suppression filtering module is as follows:

[0029] The adaptive vibration suppression filtering module is implemented as an eighth-order Butterworth digital low-pass filter.

[0030] Based on the main vibration frequency identified by the vibration-induced attitude drift compensation module, a set of filter coefficients corresponding to the main vibration frequency is selected from the preset filter coefficient library; the filter coefficient library pre-stores multiple sets of Butterworth filter coefficients for different cutoff frequencies.

[0031] The eighth-order Butterworth digital low-pass filter is configured using the selected filter coefficients, and the raw triaxial acceleration and triaxial angular velocity data of the inertial measurement unit are filtered. The filtered data is then used for subsequent inertial navigation calculations to filter out high-frequency vibration noise.

[0032] Furthermore, the method also includes a driving environment state switching management mechanism to ensure the smoothness and stability of the processing pipeline switching;

[0033] The driving environment state switching management mechanism specifically includes:

[0034] Set state entry threshold and state exit threshold, where the state exit threshold is lower than the state entry threshold, to form a hysteresis interval for state transition;

[0035] Set a state duration counter. Only after the triggering conditions of a new driving environment state type have been met for a preset duration will the switch to the corresponding target data processing pipeline be executed.

[0036] At the moment of pipeline switching, the state covariance matrix of the extended Kalman filter in the previous pipeline is transferred to the filter of the newly activated pipeline as its initial covariance matrix to ensure the continuity of state estimation and avoid step changes in the localization results.

[0037] This invention provides a vehicle monitoring and positioning data processing system based on multi-source data fusion, comprising:

[0038] The data acquisition module is used to connect to and acquire real-time sensor data from the global navigation satellite system receiver, inertial measurement unit, and vehicle odometer;

[0039] The driving environment state classification module is used to receive multi-source real-time sensor data output by the data acquisition module, construct a driving environment feature vector, and classify the current driving environment state of the vehicle into a preset driving environment state type in real time based on the driving environment feature vector.

[0040] The processing pipeline selection module is used to select and activate the target data processing pipeline that uniquely corresponds to the current driving environment state type from a preset set of processing pipelines, based on the state type code output by the driving environment state classification module.

[0041] The data fusion processing module is used to execute the algorithm of the target data processing pipeline activated by the processing pipeline selection module, and to fuse and solve the multi-source real-time sensor data to generate and output the real-time three-dimensional position, three-dimensional velocity and three-dimensional attitude information of the vehicle.

[0042] Furthermore, the driving environment state classification module includes:

[0043] The feature extraction unit is used to process the input multi-source real-time sensor data within a preset sliding time window to extract the number of satellites, the signal-to-noise ratio statistical characteristics of the global navigation satellite system data, and the frequency domain energy characteristics of the vertical axis acceleration signal of the inertial measurement unit data.

[0044] A feature vector construction unit is used to combine multiple feature values ​​output by the feature extraction unit into a standardized driving environment feature vector;

[0045] The decision tree classification unit receives the driving environment feature vector and outputs a unique code representing the open sky state, urban canyon state, tunnel or shaded state or vibration interference state through an internally fixed set of multi-level threshold comparison rules.

[0046] Furthermore, the data fusion processing module internally includes:

[0047] A tightly coupled fusion processing unit is used to implement the data processing pipeline under the open sky condition;

[0048] A loosely coupled fusion processing unit is used to implement a data processing pipeline in the urban canyon state.

[0049] The dead reckoning processing unit implements a data processing pipeline under the tunnel or concealed conditions.

[0050] An enhanced fusion processing unit enables a data processing pipeline under the aforementioned vibration disturbance conditions;

[0051] The processing pipeline selection module directs the output data stream of the data acquisition module to the activated unit among the above four processing units by controlling the data routing switch.

[0052] In one embodiment of the present invention, the enhanced fusion processing unit further includes:

[0053] The vibration-induced attitude drift compensation module is used to perform time-frequency analysis on the vertical axis acceleration signal, identify the main vibration characteristics, and generate a dynamic gyroscope bias compensation vector based on a preset two-dimensional lookup table to correct the state prediction model of the extended Kalman filter.

[0054] The adaptive vibration suppression filtering module is a reconfigurable eighth-order Butterworth digital low-pass filter. It is used to load the corresponding coefficients from the filter coefficient library according to the identified main vibration frequency and perform pre-filtering processing on the raw data of the inertial measurement unit.

[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0056] 1. By setting vibration interference states and matching them with an enhanced fusion processing pipeline with vibration adaptive compensation, this invention can accurately identify and quantify the impact of road bumps on inertial sensors, and actively suppress vibration-induced attitude drift errors through dynamic offset compensation and adaptive filtering, significantly improving the dead reckoning accuracy and robustness of vehicles on bumpy road sections during GNSS signal interruption.

[0057] 2. This invention introduces a driving environment state classification mechanism based on multi-source sensor features, which enables the positioning system to have environmental context awareness capabilities. It can intelligently select the optimal data fusion strategy according to the actual signal and physical environment of the vehicle, avoiding the performance degradation of a single fusion strategy in changing scenarios and enhancing the system's adaptability to all scenarios.

[0058] 3. The present invention adopts an on-demand activation mechanism, which only activates the computationally intensive vibration compensation and suppression algorithm when vibration interference is detected. Under stable road conditions such as open skies or urban canyons, these modules are inactive, thereby effectively reducing the average computational load and power consumption of the system and achieving an optimized balance between positioning accuracy and energy efficiency.

[0059] 4. By designing a state switching management mechanism that includes hysteresis intervals and the inheritance and transfer of the filter state covariance matrix, this invention ensures that the data processing pipeline transitions smoothly and stably when switching between different driving environment states, avoids sudden changes or divergence in positioning results caused by mode switching, and ensures the continuity and reliability of the output trajectory. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the overall technical solution architecture of the vehicle monitoring and positioning data processing method with multi-source data fusion proposed in this invention;

[0061] Figure 2 This is a schematic diagram of the core principle framework of the adaptive data processing pipeline selection mechanism based on driving environment state classification in this invention;

[0062] Figure 3 This is a flowchart illustrating the logical process of constructing driving environment feature vectors and classifying states in this invention.

[0063] Figure 4 This is a diagram showing the internal functional modules and compensation principle framework of the enhanced fusion processing pipeline under vibration interference conditions in this invention.

[0064] Figure 5 This is a logical flowchart of the data fusion processing pipeline switching and covariance inheritance mechanism corresponding to different driving environment states in this invention.

[0065] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the terminal sensor and each data processing pipeline in this invention. Detailed Implementation

[0066] Please refer to Figures 1 to 6 This invention provides a vehicle monitoring and positioning data processing method based on multi-source data fusion. Its core lies in constructing a driving environment feature vector to classify the vehicle's environment in real time, and dynamically activating a matching data processing pipeline based on the classification results. This optimizes system energy efficiency and robustness while ensuring positioning accuracy. The following will describe in detail the specific implementation steps of this method.

[0067] The method first executes step S1: acquiring real-time data from multiple sources. In this step, the system synchronously acquires raw data streams from the Global Navigation Satellite System (GNSS) receiver, the Inertial Measurement Unit (INS), and the vehicle odometer. The GNSS data includes pseudorange, carrier phase, Doppler shift, and satellite ephemeris information; the INS data includes triaxial acceleration and triaxial angular velocity; and the vehicle odometer data consists of wheel rotation speed pulse signals, which are transmitted to the central processing unit at fixed intervals via the vehicle controller area network (VDN) bus. All sensor data are timestamped with high precision to ensure timing alignment in subsequent processing. The data sampling frequencies are set to 200 Hz for the INS, 10 Hz for the GNSS, and 50 Hz for the vehicle odometer. Each data stream undergoes nanosecond-level timing alignment via hardware synchronization circuitry before entering processing, eliminating phase deviations caused by asynchronous sampling.

[0068] Then, step S2 is executed: Based on the multi-source real-time sensor data, a driving environment feature vector is constructed, and the current driving environment state of the vehicle is classified into a preset driving environment state type in real time according to the feature vector. This step specifically includes setting a sliding time window with a length of 5 seconds, and the window slides forward in 100-millisecond steps. Within this window, the system extracts the number of visible satellites, the mean and variance of the signal-to-noise ratio of all visible satellites from the Global Navigation Satellite System data. Simultaneously, within the same time window, power spectral density analysis is performed on the vertical axis acceleration signal output by the inertial measurement unit, and the Welch method is used to calculate its signal energy integral value in the preset vibration frequency band from 2 Hz to 20 Hz.

[0069] The four parameters mentioned above—number of satellites, mean carrier-to-noise ratio, variance of carrier-to-noise ratio, and integral value of vertical axis acceleration signal energy—together constitute a four-dimensional driving environment feature vector. This feature vector is input into a pre-defined rule-based decision tree classifier. The classifier internally incorporates multi-level threshold comparison logic: if the number of satellites is greater than 12, the average carrier-to-noise ratio (CNR) is higher than 45 dB / Hertz, the CNR variance is less than 4 dB / s², and the vibration energy integral value is less than 0.05 m / s², it is classified as an open sky condition; if the number of satellites is between 4 and 12, the average CNR is less than 38 dB / Hertz, the CNR variance is greater than 8 dB / s², and the vibration energy integral value is less than 0.05 m / s², it is classified as an urban canyon condition; if the number of satellites is 0 and the vibration energy integral value is less than 0.05 m / s², it is classified as a tunnel or obstructed condition; if the vibration energy integral value exceeds 0.1 m / s², it is classified as a vibration interference condition regardless of satellite signal conditions. The classifier outputs a four-bit binary code, uniquely identifying the current driving environment condition type.

[0070] Next, step S3 is executed: Based on the classified driving environment state type, the target data processing pipeline uniquely corresponding to the current driving environment state type is selected and activated from a preset set of processing pipelines. This set of processing pipelines contains four independently configured data fusion processing units, each corresponding to one of the four driving environment state types. When the classification result is open sky, the system activates the tightly coupled fusion processing pipeline of the global navigation satellite system and the inertial measurement unit. This pipeline uses an extended Kalman filter as the core estimator, and its state vector includes fifteen dimensions: position error, velocity error, attitude error, gyroscope bias, and accelerometer bias. Measurements directly use the pseudorange and carrier phase raw observations from the global navigation satellite system. Residual calculations are performed using the nonlinear observation equations and the predicted position solved by the inertial navigation system to correct the accumulated error of the inertial navigation system. In this state, the vibration suppression and compensation module is completely inactive, and its related computing resources are released to reduce system power consumption.

[0071] When the classification result indicates an urban canyon state, the system activates a loosely coupled fusion processing pipeline between the Global Navigation Satellite System (GNSS) and the Inertial Measurement Unit (INS). This pipeline first independently runs the GNSS positioning module, outputting 3D position and 3D velocity. Simultaneously, the INS performs pure inertial inference based on INS data, also outputting 3D position and 3D velocity. The position and velocity differences between the two constitute a six-dimensional measurement, which is input to an extended Kalman filter. This design effectively isolates the direct contamination of the INS state estimation by GNSS positioning jumps caused by multipath effects, avoiding filter divergence due to erroneous pseudorange observations in tightly coupled mode. In this mode, the system still does not enable vibration compensation, as urban canyon states typically correspond to smooth road surfaces with low vibration energy.

[0072] When the classification result indicates a tunnel or obstruction, the system activates the inertial measurement unit (IMU) and vehicle odometer dead reckoning pipeline. At this time, GNSS data is invalid, and the system relies entirely on the IMU and vehicle odometer for calculations. The wheel speed pulse signal provided by the vehicle odometer is integrated and converted into forward velocity, which is then input as an external velocity observation to the extended Kalman filter. The filter's state vector includes position, velocity, attitude, gyroscope bias, and accelerometer bias. The velocity state is periodically corrected based on odometer observations, effectively suppressing linear drift caused by IMU velocity integration. This pipeline does not involve any GNSS data processing and does not enable the vibration compensation module.

[0073] When the classification result indicates a vibration disturbance state, the system activates an enhanced fusion processing pipeline with adaptive vibration compensation. This pipeline comprises two core sub-modules: a vibration-induced attitude drift compensation module and an adaptive vibration suppression filtering module. The vibration-induced attitude drift compensation module first performs a short-time Fourier transform on the vertical axis acceleration signal acquired by the inertial measurement unit, with a window length set to 0.5 seconds and an overlap rate of 50%, to identify the dominant vibration frequency and its corresponding amplitude. The dominant vibration frequency is defined as the frequency at which the power spectral density peak is located, and the amplitude is its corresponding spectral density amplitude. Based on the identified dominant vibration frequency and amplitude, the system queries a pre-calibrated two-dimensional lookup table.

[0074] The lookup table is constructed offline on a six-DOF vibration platform: the platform is subjected to sinusoidal sweep excitation with a frequency range of 2 to 20 Hz and an amplitude range of 0.1 to 2 m / s², and the output biases of the pitch and roll gyroscopes under steady-state vibration are recorded, forming a frequency-amplitude-bias mapping relationship. The lookup result is a two-dimensional dynamic bias compensation vector, corresponding to the real-time bias corrections of the pitch and roll axis gyroscopes, respectively. In the state prediction step of the extended Kalman filter, this dynamic bias compensation vector is superimposed on the static bias estimate of the gyroscope, forming a complete bias term that participates in the angular velocity integral, thereby actively canceling the cross-coupling error and angular rate integral error caused by mechanical vibration in the state propagation model.

[0075] The adaptive vibration suppression filtering module is implemented as an eighth-order Butterworth digital low-pass filter. Based on the dominant vibration frequency identified by the vibration-induced attitude drift compensation module, this module selects a set of corresponding filter coefficients from a pre-set filter coefficient library. This library pre-stores four sets of eighth-order Butterworth filter coefficients with cutoff frequencies of 5 Hz, 10 Hz, 15 Hz, and 20 Hz, each satisfying linear phase and maximum flat amplitude characteristics. The system selects filter coefficients with a cutoff frequency slightly higher than the dominant vibration frequency; for example, when the dominant vibration frequency is 8 Hz, coefficients with a cutoff frequency of 10 Hz are selected. The selected coefficients are loaded into the filter structure to pre-filter the raw triaxial acceleration and triaxial angular velocity data from the inertial measurement unit. The filtered data is then input to the inertial navigation calculation module, effectively filtering out high-frequency vibration noise and improving the signal-to-noise ratio of the raw observation data.

[0076] To ensure smoothness and stability during transitions between different driving environment states, the method also includes a driving environment state transition management mechanism. This mechanism sets state entry and exit thresholds, with a hysteresis interval between them. For example, the entry threshold for a vibration disturbance state is set to the vibration energy integral value of 0.1 m / s², while the exit threshold is set to 0.07 m / s². Furthermore, the system sets a state duration counter, executing the pipeline switching command only when the triggering conditions for a new state are continuously met for more than 2 seconds. At the moment of switching, the state covariance matrix of the extended Kalman filter in the previous pipeline is completely transferred to the filter of the newly activated pipeline as its initial covariance matrix. This covariance inheritance mechanism ensures the continuity of state estimation, avoiding abrupt changes or divergence in positioning results due to mode switching.

[0077] At the system level, the corresponding data processing system of this invention includes a data acquisition module, a driving environment state classification module, a processing pipeline selection module, and a data fusion processing module. The data acquisition module connects to the Global Navigation Satellite System receiver, the Inertial Measurement Unit, and the vehicle controller area network bus via a dedicated interface circuit, responsible for acquiring raw data and time-series alignment. The driving environment state classification module internally includes a feature extraction unit, a feature vector construction unit, and a decision tree classification unit. The feature extraction unit performs statistical and spectral analysis within a sliding window; the feature vector construction unit normalizes the extracted features to eliminate dimensional differences; and the decision tree classification unit solidifies the classification rules and outputs state codes. The processing pipeline selection module receives the state codes, controls the data routing switch, and directs the data stream to one of the four processing units. The data fusion processing module includes a tightly coupled fusion processing unit, a loosely coupled fusion processing unit, a dead reckoning processing unit, and an enhanced fusion processing unit. The enhanced fusion processing unit integrates a vibration-induced attitude drift compensation module and an adaptive vibration suppression filtering module, the latter achieving dynamic coefficient loading through a reconfigurable filter structure.

[0078] The entire system runs on an embedded real-time operating system, with each module deployed as an independent task and interacting with data via message queues. Upon system startup, sensor self-checks and initial alignment are completed first, followed by a cyclical processing flow: environmental state classification is performed every 100 milliseconds, and the corresponding processing pipeline is scheduled based on the classification results. Under vibration interference, the system's computational load increases significantly, but due to the on-demand activation mechanism, the average power consumption is reduced by more than 40% compared to the full-time vibration compensation scheme. Experiments show that in typical complex urban scenarios such as overpasses and elevated road networks, the standard deviation of positioning error during GNSS signal interruptions is reduced from 4.5 meters in the traditional scheme to 1.8 meters, significantly improving trajectory continuity and meeting the high-reliability positioning service requirements of autonomous driving and fleet dispatching.

[0079] In summary, this invention achieves synergistic optimization of positioning accuracy, system robustness, and energy efficiency by using an environment-aware adaptive pipeline switching mechanism, combined with a dynamic compensation strategy and state switching management mechanism under vibration interference conditions. This effectively solves the positioning continuity problem faced by existing technologies in complex urban scenarios.

Claims

1. A method for vehicle monitoring and positioning data processing based on multi-source data fusion, characterized in that, include: The system acquires real-time data from multiple sources, including data from the Global Navigation Satellite System (GNSS), inertial measurement unit (INS) data, and vehicle odometer data. The GNSS data includes pseudorange, carrier phase, Doppler shift, and satellite ephemeris information. The INS data includes triaxial acceleration and triaxial angular velocity. The vehicle odometer data includes wheel rotation speed pulse signals. Based on the multi-source real-time sensor data, a driving environment feature vector is constructed, and the current driving environment state of the vehicle is classified into a preset driving environment state type in real time according to the driving environment feature vector; the preset driving environment state types include open sky state, urban canyon state, tunnel or shaded state, and vibration interference state. Based on the classified driving environment state type, a target data processing pipeline that uniquely corresponds to the current driving environment state type is selected and activated from a preset set of processing pipelines; the set of processing pipelines consists of independent data fusion processing units pre-configured for different driving environment state types. The activated target data processing pipeline is used to fuse and solve the multi-source real-time sensor data to generate and output the vehicle's real-time three-dimensional position, three-dimensional velocity, and three-dimensional attitude information. Select and activate the target data processing pipeline, including: When the driving environment is open sky, the tightly coupled global navigation satellite system and inertial measurement unit fusion processing pipeline is activated. This pipeline uses an extended Kalman filter to directly use the pseudorange and carrier phase observations of the global navigation satellite system as measurements to correct the position, velocity and attitude errors calculated by the inertial navigation system. At this time, the vibration suppression and compensation module is inactive. When the driving environment state type is urban canyon state, the loosely coupled global navigation satellite system and inertial measurement unit fusion processing pipeline is activated. This pipeline first independently calculates the positioning results of the global navigation satellite system and the positioning results of the inertial navigation system, and then uses the difference between the two positions and velocities as a quantity to input to the extended Kalman filter for state estimation. When the driving environment is in a tunnel or a covered state, the inertial measurement unit and the vehicle odometer dead reckoning pipeline are activated. This pipeline uses the wheel speed pulse signal from the vehicle controller area network bus to calculate the vehicle's forward speed as an external speed observation value to constrain the speed integral drift of the inertial measurement unit. When the driving environment state is a vibration disturbance state, the enhanced fusion processing pipeline with vibration adaptive compensation is activated. The enhanced fusion processing pipeline with vibration adaptive compensation includes a vibration-induced attitude drift compensation module and an adaptive vibration suppression filtering module.

2. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 1, characterized in that, Constructing driving environment feature vectors and performing real-time classification of driving environment states, including: Set a fixed-duration sliding time window, and extract the number of satellites, the mean and variance of the signal-to-noise ratio of all visible satellites from the global navigation satellite system data within this time window; Within the same sliding time window, power spectral density analysis is performed on the vertical axis acceleration signal in the inertial measurement unit data to extract the signal energy integral value within the preset vibration frequency band; Construct a four-dimensional driving environment feature vector that includes the number of satellites, the mean signal-to-noise ratio, the variance of the signal-to-noise ratio, and the integral value of the vertical axis acceleration signal energy. The driving environment feature vector is input into a preset rule-based decision tree classifier, and a unique driving environment state type code is output by comparing it with a preset threshold at multiple levels.

3. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 2, characterized in that, The preset vibration frequency band is from 2 Hz to 20 Hz; the signal energy integral value is obtained by integrating the power spectral density of the vertical axis acceleration signal within the preset vibration frequency band.

4. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 1, characterized in that, The specific processing flow of the vibration-induced attitude drift compensation module is as follows: A short-time Fourier transform is performed on the vertical axis acceleration signal acquired by the inertial measurement unit to identify the principal vibration frequency and amplitude; Based on the identified main vibration frequency and amplitude, the dynamic bias compensation vector for the pitch and roll angle gyroscopes is obtained by querying a pre-calibrated two-dimensional lookup table; the two-dimensional lookup table is established by offline calibration of the gyroscope output bias under vibration inputs of different frequencies and amplitudes on a six-degree-of-freedom vibration platform. In the state prediction step of the extended Kalman filter, the dynamic bias compensation vector is superimposed on the static bias estimate of the gyroscope, thereby actively canceling the cross-coupling error and angular rate integral error caused by mechanical vibration in the state propagation model.

5. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 4, characterized in that, The specific processing flow of the adaptive vibration suppression filtering module is as follows: The adaptive vibration suppression filtering module is implemented as an eighth-order Butterworth digital low-pass filter. Based on the main vibration frequency identified by the vibration-induced attitude drift compensation module, a set of filter coefficients corresponding to the main vibration frequency is selected from the preset filter coefficient library; the filter coefficient library pre-stores multiple sets of Butterworth filter coefficients for different cutoff frequencies. The eighth-order Butterworth digital low-pass filter is configured using the selected filter coefficients, and the raw triaxial acceleration and triaxial angular velocity data of the inertial measurement unit are filtered. The filtered data is then used for subsequent inertial navigation calculations.

6. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 5, characterized in that, The cutoff frequencies pre-stored in the filter coefficient library include 5 Hz, 10 Hz, 15 Hz and 20 Hz; the selected filter cutoff frequency is slightly higher than the main vibration frequency.

7. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 1, characterized in that, It also includes a driving environment state switching management mechanism to ensure the smoothness and stability of pipeline switching; The driving environment state switching management mechanism specifically includes: Set state entry threshold and state exit threshold, where the state exit threshold is lower than the state entry threshold, to form a hysteresis interval for state transition; Set a state duration counter. Only after the triggering conditions of a new driving environment state type have been met for a preset duration will the switch to the corresponding target data processing pipeline be executed. At the moment of pipeline switching, the state covariance matrix of the extended Kalman filter in the previous pipeline is passed to the filter of the newly activated pipeline as its initial covariance matrix.

8. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 7, characterized in that, The preset time length is 2 seconds; the entry threshold for the vibration interference state is 0.1 m / s² square Hertz, the energy integral value of the vertical axis acceleration signal, and the exit threshold is 0.07 m / s² square Hertz.

9. The vehicle monitoring and positioning data processing method based on multi-source data fusion according to claim 1, characterized in that, The multi-source real-time sensor data all have high-precision timestamps and are synchronized at the nanosecond level via hardware synchronization circuitry.

Images