A Fusion Localization Method and System for Autonomous Vehicles

Abstract

This invention relates to the technical field of modern transportation and intelligent navigation, and discloses a fusion positioning method and system for autonomous vehicles. It utilizes a pre-established multi-source data standardization model to perform format conversion and consistency verification on satellite signals, sensor data, and motion trajectory estimation data. A weighted average fusion algorithm is used to generate a preliminary direction estimate, which is then evaluated and corrected using a Bayesian probability model. Subsequently, the invention employs a Kalman filter algorithm to smooth and predict the direction information, and performs local updates based on real-time external environmental feedback data. Multi-time-window trend analysis is used to detect and correct abnormal fluctuations. Finally, the optimized direction and position information are jointly optimized to generate the final fusion positioning result. Machine learning methods are used to periodically update the model parameters, achieving high-precision, high-reliability, and highly adaptive positioning in complex scenarios.

Description

Technical Field

[0001] This invention relates to the technical field of modern transportation and intelligent navigation, and in particular to a fusion positioning method and system for autonomous vehicles. Background Technology

[0002] Currently, in the field of modern transportation and intelligent navigation, precise positioning technology is the cornerstone of ensuring safe travel and improving efficiency; its importance is self-evident. Especially in scenarios such as autonomous driving and drone navigation, the accuracy and real-time performance of positioning are directly related to the reliability of the system and the safety of users. However, current mainstream positioning methods often have significant shortcomings. Single positioning methods, such as the Global Positioning System (GPS), are easily affected by signal interference or obstruction in complex environments, leading to a decrease in accuracy. Furthermore, the fusion of multi-source data lacks a unified standard, making it difficult to achieve stable output.

[0003] In existing technologies, the core challenges in this field lie primarily in effectively integrating multi-source data while ensuring high information reliability. Firstly, data from different sources, such as satellite signals, vehicle sensor information, and motion trajectory estimation data, exhibit significant differences in format and accuracy, making data fusion exceptionally complex. A further major challenge arises from the lack of unified standards for processing and encoding this data, particularly insufficient standardization of directional information, leading to difficulties in accurately determining directional changes in dynamic environments. Furthermore, due to an imperfect reliability assessment mechanism for directional information, the system often fails to update and correct erroneous data in a timely manner when facing complex scenarios, ultimately affecting the reliability of positioning results.

[0004] Therefore, how to construct a unified data processing framework to standardize and encode multi-source information, evaluate the reliability of directional data in real time, and ensure timely updates of positioning information in dynamic environments has become a key problem that this research urgently needs to solve. Summary of the Invention

[0005] This invention provides a fusion positioning method and system for autonomous vehicles to achieve reliable positioning results.

[0006] In a first aspect, to address the aforementioned technical problems, the present invention provides a fusion positioning method for autonomous vehicles, comprising:

[0007] The original data is converted into a format using a pre-established multi-source data standardization model to generate the first dataset.

[0008] Filter the heterogeneous data in the first dataset. If the deviation value of one of the data exceeds the preset deviation threshold, remove the corresponding record to obtain the second dataset.

[0009] The directional information of different data in the second dataset is weighted and calculated to generate a first directional estimate.

[0010] The credibility of the first direction estimate is evaluated. If the credibility is lower than a preset evaluation threshold, a data correction process is triggered to obtain the second direction estimate.

[0011] The direction information in the second direction estimate is smoothed and predicted to generate a third direction estimate.

[0012] Obtain real-time feedback data of the external environment; if the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, generate a fourth-direction estimate.

[0013] The short-term fluctuations in the direction information in the fourth direction estimate are detected. If abnormal fluctuations are detected, they are smoothed and corrected by combining historical data to obtain the fifth direction estimate.

[0014] The real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result;

[0015] Based on the fusion positioning results, the parameters and weight allocation rules of the multi-source data standardization model are updated periodically to generate new model parameter configurations for the next round of multi-source data processing.

[0016] Preferably, the step of filtering heterogeneous data in the first dataset, wherein if the deviation value of one data point exceeds a preset deviation threshold, the corresponding record is removed to obtain a second dataset, includes:

[0017] A data verification tool is used to compare the heterogeneous data in the first dataset item by item, calculate the deviation value of each data and the preset deviation threshold. If the deviation value exceeds the threshold range, the relevant records of the data are deleted to obtain a preliminary filtered data set.

[0018] The initially screened dataset is cleaned using data cleaning tools to generate a cleaned dataset.

[0019] Data integration tools are used to reorganize the cleaned data set according to a unified structure. If a field is missing, it is filled in according to a pre-established field completion rule to determine the reorganized data set.

[0020] The recombined datasets are merged and formatted into a standard structure to generate the second dataset. Preferably, the step of weighting the directional information of different data in the second dataset to generate a first directional estimate includes:

[0021] Obtain directional information and real-time data from different data sources from the second dataset, and assign initial weights to the directional information according to a pre-established weight allocation rule to obtain a pre-weighted set of directional information.

[0022] The signal strength change in the real-time data is determined. If the signal strength is higher than the preset change threshold range, the weight ratio of the corresponding data source is increased, and vice versa, the weight ratio is decreased to obtain the dynamically adjusted weight combination.

[0023] Based on the dynamically adjusted weight combination, a weighted average calculation is performed on the initially weighted directional information set to obtain the fused directional information value and a first directional estimate is obtained.

[0024] The process then includes: using a data storage tool to associate and save the first direction estimate with the data source and weight allocation records, thereby generating a structured direction estimate data record.

[0025] Preferably, the step of evaluating the credibility of the first direction estimate, and triggering a data correction process to obtain a second direction estimate if the credibility is lower than a preset evaluation threshold, includes:

[0026] Based on the first direction estimate, the signal interference level and sensor data records are obtained from the dynamic environment to obtain a sorted data set.

[0027] According to the pre-established evaluation criteria, if the confidence value in the sorted data set is lower than the preset confidence threshold, then the range of data to be corrected is determined.

[0028] Based on the corrected data range, sensor data is reacquired from the dynamic environment, and combined with the degree of signal interference, the adjusted direction reference value is obtained;

[0029] The adjusted direction reference value is associated with and saved with the confidence value to obtain the second direction estimate.

[0030] Preferably, the step of smoothing and predicting the direction information in the second direction estimate to generate a third direction estimate includes:

[0031] Based on the second direction estimate, vehicle speed change and road curvature information are obtained from the dynamic environment to obtain a sorted parameter set;

[0032] Based on the sorted parameter set and combined with the pre-established dynamic environment perception rules, the influence of vehicle speed and road curvature is comprehensively evaluated to determine the fused direction reference data.

[0033] The directional information of the fused directional reference data is continuously adjusted. If the fluctuation of the adjusted data exceeds the preset fluctuation threshold, the real-time parameters in the dynamic environment are reacquired to obtain the adjusted directional information.

[0034] Based on the adjusted direction information, the deviation between the prediction result and the actual environmental perception data is corrected, the corrected direction estimate is saved, and the third direction estimate is generated.

[0035] Preferably, the step of acquiring real-time feedback data of the external environment, and generating a fourth-direction estimate if the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, includes:

[0036] Satellite signal strength and road feature point data are obtained from external environmental feedback to obtain a set of organized environmental data;

[0037] A deviation analysis is performed between the third-party directional estimate and the environmental data set. If the deviation exceeds a preset dynamic threshold, a local data update process is initiated to determine the direction and range of the correction data.

[0038] Obtain the latest external environment feedback information, adjust the range of the corrected direction data, and obtain a preliminary corrected direction reference value;

[0039] The preliminary corrected direction reference value is compared with the real-time data to generate the fourth direction estimate.

[0040] Preferably, the step of detecting short-term fluctuations in the direction information in the fourth direction estimate, and if abnormal fluctuations are detected, performing smoothing corrections by combining historical data to obtain the fifth direction estimate, includes:

[0041] The historical records of the fourth direction estimate are obtained, and the distribution of the historical records in multiple time windows is segmented using a time series decomposition tool to obtain a separate dataset of short-term fluctuations and long-term changes.

[0042] The short-term fluctuations are detected using a data comparison tool. If the short-term fluctuations exceed a preset fluctuation threshold, the abnormal data is marked as being within the range to be adjusted.

[0043] Based on the range to be adjusted, the corresponding direction estimation records are extracted from the historical data storage unit, and the range to be adjusted is weighted using a data smoothing tool to obtain a preliminary corrected direction data set.

[0044] Based on the preliminary corrected direction data set, the fifth direction estimate is determined by comparing and adjusting it with the long-term changes using data verification tools.

[0045] Preferably, the real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result, including:

[0046] The estimated value of the fifth direction and the positioning coordinates are initially registered using a spatial geometric mapping tool, and then calibrated using a weighted average tool to determine the initially integrated set of spatial information.

[0047] Based on the preliminarily integrated set of spatial information, if the deviation between the estimated value of the fifth direction and the position of the positioning coordinates exceeds a preset deviation threshold, then a deviation correction tool is used to dynamically adjust the data and obtain the adjusted data set.

[0048] The direction estimates and positioning coordinates in the adjusted data set are finally integrated, and smoothed by combining historical records to obtain the fused positioning result.

[0049] Preferably, the step of periodically updating the parameters and weight allocation rules of the multi-source data standardization model based on the fused positioning results to generate new model parameter configurations for the next round of multi-source data processing includes:

[0050] Historical data and real-time feedback data of the fused positioning results are obtained, and time alignment tools are used for synchronization processing to obtain a data set under a unified time reference.

[0051] Based on the data set under the unified time benchmark, the historical data and the real-time feedback data are classified and summarized using data integration tools.

[0052] For the classified features, a weighted average tool is used to adjust the weight allocation and determine the weight combination suitable for the current loop processing;

[0053] According to the weight combination, if the detected weight allocation deviates from the preset weight threshold, the weights are fine-tuned by a dynamic optimization tool. Combining the characteristics of multi-source data, the adjusted weight set that meets the requirements of cyclic processing is determined.

[0054] Using the adjusted weight set, the parameters of the multi-source data standardization model configuration are updated using an information fusion tool. Combined with the training results, configuration parameters suitable for the next round of multi-source data processing are obtained.

[0055] Secondly, the present invention provides a fusion positioning system for autonomous vehicles, comprising:

[0056] The first acquisition module is used to convert the format of the raw data through a pre-established multi-source data standardization model to generate the first dataset.

[0057] The second acquisition module is used to filter heterogeneous data in the first dataset. If the deviation value of one of the data exceeds a preset deviation threshold, the corresponding record is removed to obtain the second dataset.

[0058] The third acquisition module is used to perform weighted calculations on the directional information of different data in the second dataset to generate a first directional estimate.

[0059] The fourth acquisition module is used to evaluate the credibility of the first direction estimate. If the credibility is lower than a preset evaluation threshold, a data correction process is triggered to obtain the second direction estimate.

[0060] The fifth acquisition module is used to smooth and predict the direction information in the second direction estimate to generate a third direction estimate;

[0061] The sixth acquisition module is used to acquire real-time feedback data of the external environment. If the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, a fourth-direction estimate is generated.

[0062] The seventh acquisition module is used to detect short-term fluctuations in the direction information in the fourth direction estimate. If abnormal fluctuations are detected, the module combines historical data to perform smoothing corrections and obtain the fifth direction estimate.

[0063] The eighth acquisition module is used to acquire positioning coordinate data in real time, and to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result.

[0064] The configuration module is used to periodically update the parameters and weight allocation rules of the multi-source data standardization model based on the fusion positioning results, and generate new model parameter configurations for the next round of multi-source data processing.

[0065] Compared to existing technologies, this invention discloses a fusion localization method for autonomous vehicles. It utilizes a pre-established multi-source data standardization model to convert and verify the format of raw data from different sources. A weighted average fusion algorithm is employed to generate a first direction estimate, which is then evaluated and corrected using a Bayesian probability model. Subsequently, the invention uses a Kalman filter algorithm to smooth and predict the direction information, and performs local updates based on real-time external environmental feedback data. Multi-time-window trend analysis is used to detect and correct abnormal fluctuations. Finally, the optimized direction and position information are jointly optimized to generate the final fusion localization result. Machine learning methods are used to periodically update the model parameters, achieving high-precision, high-reliability, and highly adaptive localization in complex scenarios. Attached Figure Description

[0066] Figure 1This is a flowchart of a fusion positioning method for an autonomous vehicle provided by an embodiment of the present invention;

[0067] Figure 2 This is a schematic diagram of a fusion positioning system for an autonomous vehicle provided in an embodiment of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Reference Figure 1 The first embodiment of the present invention provides a flowchart of a fusion localization method for autonomous vehicles, including the following steps:

[0070] S11, the original data is converted into a format using a pre-established multi-source data standardization model to generate the first dataset;

[0071] S12, filter the heterogeneous data in the first dataset. If the deviation value of one of the data exceeds the preset deviation threshold, remove the corresponding record to obtain the second dataset.

[0072] S13, perform weighted calculation on the directional information of different data in the second dataset to generate a first directional estimate;

[0073] S14, Combine the Bayesian probability model to evaluate the credibility of the first direction estimate. If the credibility is lower than the preset evaluation threshold, trigger the data correction process to obtain the second direction estimate.

[0074] S15, using the Kalman filter algorithm, smooth and predict the direction information in the second direction estimate to generate a third direction estimate;

[0075] S16, obtain real-time feedback data of the external environment; if the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, generate a fourth-direction estimate.

[0076] S17, detect short-term fluctuations in the direction information in the fourth direction estimate; if abnormal fluctuations are detected, combine historical data for smoothing correction to obtain the fifth direction estimate.

[0077] S18, The real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result;

[0078] S19, based on the fusion positioning results, periodically update the parameters and weight allocation rules of the multi-source data standardization model to generate new model parameter configurations for the next round of multi-source data processing cycle.

[0079] In step S11, the original data is format-converted using a pre-established multi-source data standardization model to generate the first dataset, including:

[0080] Raw data is acquired from multiple sources, including satellite signals, sensor data, and motion trajectory estimation data. These are extracted using pre-established acquisition tools to obtain an initial dataset. For this initial dataset, a format conversion tool processes the raw data from different sources. If data fields are inconsistent, adjustments are made using a field mapping table to determine a unified data structure. Based on this unified data structure, the processed data is encoded using standardized encoding rules. Data conforming to a standard format is generated using a preset encoding template, resulting in a standardized dataset. This standardized dataset is then aggregated using a data integration tool to generate a first dataset, which is stored in a designated database for subsequent processing.

[0081] For example, when processing satellite signals, sensor data, and motion trajectory estimation data, the implementation of this process can be understood through specific scenarios. Suppose a smart traffic monitoring system needs to acquire vehicle operation data from multiple sources. Satellite signals may come from GPS devices, providing the vehicle's real-time location; sensor data comes from onboard equipment, recording speed and acceleration; and motion trajectory estimation data uses algorithms to predict the vehicle's possible path. These data sources have different formats, and the field names and units may differ, therefore, it is necessary to extract the raw data using pre-established acquisition tools to form an initial dataset.

[0082] For format conversion of the initial dataset, a format conversion tool can be used to unify data from different sources. Specifically, a pre-established multi-source data standardization model defines unified processing rules for three main types of raw data sources: satellite signals, sensor data, and motion trajectory estimation data. This model includes specific format conversion templates and consistency verification logic. During the conversion process, the model first identifies the source type of the input data (such as GPS signals, IMU data, wheel speedometer data, visual positioning data, etc.), and then applies the pre-defined parsing rules for that type to deconstruct its original format (such as NMEA statements, CAN bus messages, custom binary streams, image coordinates, etc.) and extract key information fields. Next, the model uses a built-in field mapping table to force fields with the same meaning but inconsistent naming or units (e.g., "timestamp" and "time_record", latitude and longitude degrees, minutes, seconds and decimal degrees, speed m / s and km / h) from different sources to predefined standard field names and unified units.

[0083] For example, the time field in GPS data might be "timestamp," while sensor data might be "time_record." A field mapping table unifies both to "time." If GPS location data is in degrees while sensor data is in meters, the mapping table adjusts the units to ensure data field consistency. This unified data structure lays the foundation for subsequent processing, avoiding data conflicts caused by format differences and improving the accuracy of data processing.

[0084] It should be noted that the process of establishing the multi-source data standardization model begins with an in-depth analysis of multi-source heterogeneous data in autonomous driving scenarios, covering the physical characteristics and logical structure of data sources such as satellite positioning (e.g., GPS / BeiDou's NMEA protocol), inertial sensors (raw values ​​of angular velocity / acceleration from IMU), wheel speedometer pulse signals, visual positioning coordinates, and high-precision map feature points. Based on this, a three-layer standardized architecture is designed: the bottom parsing layer customizes binary / text parsing rules for each data source (e.g., decomposing the GPGGA field of the NMEA statement into latitude, longitude, altitude, and number of satellites; parsing the CAN bus ID0x0A0 message into yaw rate); the middle mapping layer establishes a dynamic field mapping table to unify semantic conflicts (e.g., unifying the "heading" field in different systems as "heading" and forcibly converting the unit to 0-360 degrees; unifying the "timestamp" to nanosecond-level UTC time in ISO 8601 format); the top verification layer incorporates physical logic constraints (e.g., vehicle speed range 0-200km / h, road curvature radius threshold 500 meters) and performs adaptive filling (for missing GPS elevation data, Kalman filtering is used to fuse the Z-axis acceleration of the IMU for estimation and completion). After training and verification with historical multi-source data, the model's parameters are solidified, ultimately forming a multi-source data standardization model that includes data pattern recognition, outlier filtering, unit conversion, and spatial coordinate system unification (e.g., WGS84 to local ENU coordinate system).

[0085] In one possible implementation, the application of standardized encoding rules can further optimize data quality. Assuming the processed data needs to conform to specific industry standards, a pre-defined encoding template can be used to unify the time format as "YYYY-MM-DDHH:MM:SS" and the location data as latitude and longitude, such as "latitude:39.9042,longitude:116.4074". This encoding generates a dataset that conforms to the standard format, ensuring seamless data transfer between different systems and reducing errors caused by format inconsistencies.

[0086] For example, data integration tools can be used to aggregate standardized datasets into a first dataset. Suppose that the GPS location, speed, and predicted trajectory of all vehicles are integrated into a single table, with each record containing fields such as vehicle ID, time, location, and speed. This ultimately generates the first dataset and stores it in a designated database. This integration method facilitates subsequent analysis, such as traffic flow prediction or congestion warning, significantly improving data utilization efficiency.

[0087] It's important to note that data security and access speed must be considered during the storage process in the designated database. A distributed database can be used to store the initial dataset, ensuring efficient querying of large-scale data while implementing access controls to protect data privacy. This approach not only improves the reliability of data processing but also provides solid support for subsequent stages such as real-time monitoring or historical data analysis. Through this process, from data collection to standardization and integration, each step is closely linked, ensuring data consistency and availability. Ultimately, this provides a high-quality data foundation for decision support in intelligent transportation systems, leading to higher business value and operational efficiency.

[0088] In step S12, heterogeneous data in the first dataset is filtered. If the deviation value of any data exceeds a preset deviation threshold, the corresponding record is removed to obtain the second dataset, which includes:

[0089] A data verification tool is used to compare the heterogeneous data in the first dataset item by item, calculate the deviation value of each data and the preset deviation threshold. If the deviation value exceeds the threshold range, the relevant records of the data are deleted to obtain a preliminary filtered data set.

[0090] The initially screened dataset is cleaned using data cleaning tools to generate a cleaned dataset.

[0091] Data integration tools are used to reorganize the cleaned data set according to a unified structure. If a field is missing, it is filled in according to a pre-established field completion rule to determine the reorganized data set.

[0092] The recombined datasets are merged and formatted into a standard structure to generate the second dataset.

[0093] For example, in the context of intelligent traffic monitoring systems, the processing flow of the first dataset can be explored from multiple perspectives, with detailed analysis combined with the business background.

[0094] For example, the core of using data validation tools lies in the item-by-item comparison and deviation calculation of heterogeneous data. Assuming the first dataset contains location data from GPS devices and speed data from vehicle sensors, the validation tool will compare each vehicle's record item by item, checking whether the location data is within a reasonable range (e.g., whether the latitude and longitude exceed city boundaries) and whether the speed data conforms to a normal range, such as 0 to 120 kilometers per hour. If a record shows a speed of 200 kilometers per hour, significantly exceeding the preset threshold range of 80 to 150 kilometers per hour, its deviation value is calculated and it is marked as an item to be removed. This method effectively identifies abnormal data, providing a more reliable foundation for subsequent processing.

[0095] For example, in the application of data cleaning tools, deleting marked records is a crucial step in initially screening the dataset. Suppose that among multiple records for a vehicle, one is marked due to abnormal speed. The cleaning tool removes this record while retaining other normal records, such as those whose location and time fields meet the thresholds. The cleaned dataset is thus more concise, ensuring that subsequent analysis is not affected by outliers.

[0096] For example, in the data integration tool's restructuring process, standardized structure and field completion are key aspects. Suppose that some records in the cleaned dataset lack an acceleration field. The integration tool will calculate and fill in the missing value based on pre-established rules and the speed change trend. For instance, if a car's speed changes from 40 km / h to 50 km / h within two seconds, the acceleration can be calculated and filled into the field. This completion method makes the restructured dataset more complete and facilitates unified analysis.

[0097] For example, in the use of data storage tools, formatting the reorganized dataset into a standard structure and generating a second dataset is the final step. Assume the data is formatted as a table containing fields such as vehicle ID, time, location, and speed. The time for each record is standardized, such as 2023-10-01 08:00:00, and location data is stored in latitude and longitude. This standardized structure facilitates subsequent system reading and processing, ensuring efficient data flow between different modules. Through the detailed processing described above—from verification and cleaning to integration and storage—each step closely revolves around the needs of intelligent traffic monitoring, ensuring data accuracy and consistency and laying a solid foundation for subsequent business analysis.

[0098] In step S13, the directional information of different data in the second dataset is weighted and calculated to generate a first directional estimate, including:

[0099] Obtain directional information and real-time data from different data sources from the second dataset, and assign initial weights to the directional information according to a pre-established weight allocation rule to obtain a pre-weighted set of directional information.

[0100] The signal strength change in the real-time data is determined. If the signal strength is higher than the preset change threshold range, the weight ratio of the corresponding data source is increased, and vice versa, the weight ratio is decreased to obtain the dynamically adjusted weight combination.

[0101] Based on the dynamically adjusted weight combination, a weighted average calculation is performed on the initially weighted directional information set to obtain the fused directional information value and a first directional estimate is obtained.

[0102] The process then includes: using a data storage tool to associate and save the first direction estimate with the data source and weight allocation records, thereby generating a structured direction estimate data record.

[0103] For example, in the context of intelligent traffic monitoring systems, the processing of directional information from the second dataset can be explored from multiple perspectives, with detailed analysis tailored to the business background. This includes the process of acquiring directional information and real-time data from different data sources.

[0104] Understandably, data sources may include roadside cameras, in-vehicle navigation devices, and positioning modules on mobile devices. Assuming cameras provide vehicle orientation angle data, in-vehicle navigation devices provide direction estimation based on historical trajectories, and mobile devices provide real-time orientation information via gyroscopes, this data will be accompanied by historical error rate values ​​and signal strength records. For example, the error rate of a camera might be 5% due to lighting conditions, while the error rate of an in-vehicle device might be as high as 15% in a tunnel with weak signal.

[0105] For example, regarding the step of assigning initial weights through pre-established weighting rules, assume that the weighting rules are based on a comprehensive evaluation of historical error rates and signal strength. A camera, due to its low error rate, might be assigned an initial weight of 0.5, an in-vehicle device 0.3, and a mobile terminal 0.2. This initial weight reflects the differences in the reliability of the data sources, forming a pre-weighted set of directional information.

[0106] It should be noted that the design of weight allocation rules needs to take into account the specific characteristics of the business scenario. For example, in densely populated urban areas, camera data may be more reliable, while in suburban areas, vehicle-mounted equipment may be relied upon.

[0107] For example, when dynamically adjusting weights using data processing tools, incorporating real-time signal strength changes is crucial. Suppose that at a certain moment, the camera's signal strength drops below a preset threshold due to weather conditions, its weight might decrease from 0.5 to 0.3, while the onboard device's signal strength exceeds the threshold, increasing its weight from 0.3 to 0.4. This dynamic adjustment better adapts to real-time environmental changes, ensuring accurate directional information.

[0108] For example, in the weighted average calculation step of the data fusion tool, assuming the directional information from the three data sources is 10 degrees, 15 degrees, and 20 degrees east of north, respectively, and the dynamically adjusted weights are 0.3, 0.4, and 0.3, the weighted average yields a fused directional information value of approximately 14.5 degrees, which is used as the first directional estimate. This fusion method can combine the advantages of multiple data sources and improve the stability of the directional estimation.

[0109] For example, in the step of using data storage tools to associate and save data, assume that the estimated value of the first direction, 14.5 degrees, is stored as a structured record along with the data source and weight allocation records, including fields such as timestamps and device identifiers. This structured storage facilitates subsequent traceability and analysis.

[0110] It should be noted that when determining whether a value meets the preset confidence standard, historical data can be used for comparison. If the estimated value deviates from the actual trajectory by less than 2 degrees, it is considered to meet the standard. This judgment mechanism helps to filter out high-quality direction estimation data, providing support for subsequent traffic flow analysis or route planning.

[0111] For example, the implementation of each of the above stages is consistently grounded in intelligent traffic monitoring. From data acquisition to dynamic weight adjustment, and then to fusion and storage, each step is closely aligned with the needs of vehicle direction estimation, ensuring the usability and reliability of the data. This refined processing effectively improves the accuracy of directional information, providing more valuable data support for traffic management.

[0112] In step S14, the first direction estimate is evaluated for confidence. If the confidence level is lower than a preset evaluation threshold, a data correction process is triggered to obtain a second direction estimate, including:

[0113] Based on the first direction estimate, the signal interference level and sensor data records are obtained from the dynamic environment to obtain a sorted data set.

[0114] According to the pre-established evaluation criteria, if the confidence value in the sorted data set is lower than the preset confidence threshold, then the range of data to be corrected is determined.

[0115] Based on the corrected data range, sensor data is reacquired from the dynamic environment, and combined with the degree of signal interference, the adjusted direction reference value is obtained;

[0116] The adjusted direction reference value is associated with and saved with the confidence value to obtain the second direction estimate.

[0117] For example, in the context of intelligent traffic monitoring, the subsequent processing of the first direction estimate can be analyzed in detail from the perspectives of signal interference and data discrepancies. Signal interference may originate from environmental factors such as weather changes or electromagnetic interference, while sensor data records include directional information collected by different devices. Suppose that in urban road monitoring, sensor data comes from roadside cameras and vehicle-mounted equipment. Signal interference might cause blurry camera data due to heavy rain, while vehicle-mounted equipment might experience data fluctuations due to signal obstruction. Data processing tools can clean and categorize this data, removing obvious outliers and forming a processed dataset, providing a foundation for subsequent evaluation.

[0118] For example, in the process of using probability calculation tools to assess credibility values.

[0119] Understandably, pre-established evaluation criteria may be based on a comprehensive judgment of historical data bias and real-time signal quality. For example, if the processed dataset shows a confidence level of 0.6 for an estimated value in a certain direction, while the preset threshold is 0.8, this is below the standard and triggers subsequent processing.

[0120] It should be noted that this evaluation method can promptly identify potential problems in the data, ensuring the reliability of directional information.

[0121] For example, in determining the data range requiring correction and reacquiring data, assuming a data correction tool is used to reacquire sensor data from the vehicle's equipment in a dynamic environment for camera data that is significantly affected by interference, and the direction estimate is adjusted by considering the degree of signal interference, such as signal attenuation caused by heavy rain. For instance, the original estimate might be 14.5 degrees east of north, but after adjustment it might become 13.8 degrees east of north, which can be used as the adjusted direction reference value. This method effectively addresses data deviations caused by environmental changes.

[0122] For example, the step of using data storage tools to associate and save the adjusted direction reference value and confidence value.

[0123] Understandably, the stored content may include timestamps, device origin, and adjustment records. Assuming the adjusted direction reference value is 13.8 degrees, the confidence value increases to 0.85, meeting the requirements of the preset evaluation standard of 0.8, and is ultimately confirmed as the direction reference result. This storage method facilitates subsequent traceability and data analysis.

[0124] For example, throughout the implementation of each of the above stages, the business context consistently revolves around intelligent traffic monitoring. From signal interference identification to data processing, probability assessment, data correction, and final storage, each step closely aligns with the needs of vehicle direction estimation, ensuring the practicality of the data. This refined processing provides more reliable directional references for traffic management, particularly enhancing the adaptability of data processing in complex environments, and laying the foundation for subsequent route planning or traffic flow analysis.

[0125] In step S15, the direction information in the second direction estimate is smoothed and predicted to generate a third direction estimate, including:

[0126] Based on the second direction estimate, vehicle speed change and road curvature information are obtained from the dynamic environment to obtain a sorted parameter set;

[0127] Based on the sorted parameter set and combined with the pre-established dynamic environment perception rules, the influence of vehicle speed and road curvature is comprehensively evaluated to determine the fused direction reference data.

[0128] The directional information of the fused directional reference data is continuously adjusted. If the fluctuation of the adjusted data exceeds the preset fluctuation threshold, the real-time parameters in the dynamic environment are reacquired to obtain the adjusted directional information.

[0129] Based on the adjusted direction information, the deviation between the prediction result and the actual environmental perception data is corrected, the corrected direction estimate is saved, and the third direction estimate is generated.

[0130] For example, in intelligent traffic monitoring scenarios, the subsequent processing of the second-direction estimate can start with vehicle speed changes and road curvature information in the dynamic environment, and then be further analyzed in conjunction with specific business needs. This also relates to the use of data acquisition tools.

[0131] It is understandable that vehicle speed changes may be affected by real-time traffic conditions, while road curvature information is related to road design and terrain. Assuming that in urban road monitoring, vehicle speed data is acquired through roadside sensors and onboard equipment, outliers caused by equipment malfunctions, such as sudden speed drops to 0 or exceeding reasonable ranges, are initially filtered out, forming a processed parameter set that provides a reliable foundation for subsequent fusion.

[0132] For example, when applying data fusion tools, combined with pre-established dynamic environment perception rules, the impact of vehicle speed and road curvature can be comprehensively evaluated. Suppose on a curved road, a vehicle's speed decreases from 60 km / h to 40 km / h, while the road curvature is displayed as moderate. The fusion rule might prioritize the impact of the speed reduction on the direction reference data, resulting in a more realistic direction reference value. This approach effectively adapts to changes in the dynamic environment.

[0133] For example, in the use of smoothing tools, continuous adjustments to the fused direction reference data can prevent abrupt changes in direction information. Suppose the adjusted data fluctuation range is preset to ±2 degrees, but the actual fluctuation reaches 3 degrees, which exceeds the threshold range. In this case, real-time parameters from the dynamic environment, such as vehicle speed and road curvature data, need to be reacquired, and the direction information readjusted. This processing method helps maintain the stability of direction information.

[0134] For example, in the application of information calibration tools, the correction of the deviation between the predicted results and the actual environmental perception data is particularly important. Suppose the predicted direction is 13.8 degrees east of north, while the actual perceived data is 14.2 degrees east of north, the calibration tool will fine-tune according to the deviation to obtain a direction estimate closer to the actual value. This calibration can improve the accuracy of the direction reference.

[0135] For example, storing the corrected direction estimate using data storage tools and determining whether it meets stability requirements can provide a basis for subsequent analysis. Assuming the corrected direction value is 14.0 degrees east of north, and the stability assessment shows the fluctuation range is within a preset threshold, it is confirmed as the final direction reference. This storage and judgment mechanism facilitates data traceability and long-term monitoring.

[0136] For example, from an overall business perspective, all the above-mentioned aspects are closely related to intelligent traffic monitoring. Whether it's data collection, fusion, smoothing, calibration, or storage, the aim is to provide reliable support for vehicle direction estimation. Especially in complex urban road environments, this multi-stage collaborative processing can adapt to dynamic changes, ensuring the continuity and stability of direction information, and providing important reference value for traffic management and route planning.

[0137] In step S16, real-time feedback data of the external environment is acquired. If the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, a fourth-direction estimate is generated, including:

[0138] Satellite signal strength and road feature point data are obtained from external environmental feedback to obtain a set of organized environmental data;

[0139] A deviation analysis is performed between the third-party directional estimate and the environmental data set. If the deviation exceeds a preset dynamic threshold, a local data update process is initiated to determine the direction and range of the correction data.

[0140] Obtain the latest external environment feedback information, adjust the range of the corrected direction data, and obtain a preliminary corrected direction reference value;

[0141] The preliminary corrected direction reference value is compared with the real-time data to generate the fourth direction estimate.

[0142] For example, in the field of intelligent traffic monitoring, satellite signal strength and road feature point data obtained from external environmental feedback can be initially processed using data processing tools. The main function of these tools is to clean and classify the raw information collected, ensuring the reliability of the data for subsequent analysis. Suppose that in an urban road environment, satellite signal strength is affected by tall buildings, resulting in missing or abnormal data. The data processing tool will prioritize removing data with signal strength below a certain threshold, such as below 30% of the signal value. Simultaneously, it will perform coordinate normalization on the road feature point data, forming a processed environmental data set that provides a foundation for subsequent comparisons.

[0143] For example, in the application of data comparison tools, in the deviation analysis between the third-party orientation estimate and the environmental dataset, a preset dynamic threshold can be used to determine whether correction is needed. Suppose the third-party orientation estimate is 15.0 degrees east of north, while the reference direction in the environmental dataset is 16.5 degrees east of north, the deviation is 1.5 degrees. If the preset dynamic threshold is 1.0 degree, it exceeds the range, and a local data update process needs to be initiated. This comparison method can promptly identify potential problems in the orientation estimation, ensuring data accuracy.

[0144] For example, in a local data update process, obtaining the latest external environmental feedback is crucial. Combining signal strength fluctuations and road feature matching data, suppose on a main urban road, the signal strength drops to 40% due to temporary construction interference, while the road feature points show a straight road. The data fusion tool will reduce the weight of direction correction based on signal strength fluctuations, relying more on road feature matching data to derive a preliminary corrected direction reference value, such as adjusting it to 16.2 degrees north of east. This approach can dynamically adapt to environmental changes, improving the reliability of the direction reference.

[0145] For example, when using information verification tools, it is crucial to perform a final comparison of the initially corrected direction reference value with environmental adaptation adjustment rules. Suppose the initial corrected value is 16.2 degrees east of north, while the real-time data update shows 16.0 degrees east of north, a deviation of 0.2 degrees, which meets the preset dynamic threshold of 0.5 degrees. Then, the generated fourth direction estimate is 16.0 degrees east of north. This verification mechanism can further refine the direction data, ensuring that the final result closely matches the actual environment.

[0146] For example, in the specific implementation of data fusion tools, the core lies in the comprehensive evaluation of multi-source data. In an urban road scenario, assuming satellite signal strength recovers to 80% and road feature point data closely matches historical records, the fusion tool will integrate the influence of both, appropriately increasing the weight of signal strength, and ultimately adjusting the direction reference value. This multi-dimensional fusion can effectively address the uncertainty of a single data source and improve the stability of direction estimation.

[0147] For example, within the overall business context, the aforementioned stages are closely integrated with intelligent traffic monitoring, forming a complete closed-loop processing flow from data processing to comparison, updating, and verification. Especially in complex urban environments, this multi-layered data processing approach can dynamically adapt to changes in the external environment, providing reliable support for vehicle direction estimation, and thus offering important references for traffic management and route planning.

[0148] In step S17, short-term fluctuations in the direction information of the fourth direction estimate are detected. If abnormal fluctuations are detected, smoothing corrections are performed using historical data to obtain the fifth direction estimate, including:

[0149] The historical records of the fourth direction estimate are obtained, and the distribution of the historical records in multiple time windows is segmented using a time series decomposition tool to obtain a separate dataset of short-term fluctuations and long-term changes.

[0150] The short-term fluctuations are detected using a data comparison tool. If the short-term fluctuations exceed a preset fluctuation threshold, the abnormal data is marked as being within the range to be adjusted.

[0151] Based on the range to be adjusted, the corresponding direction estimation records are extracted from the historical data storage unit, and the range to be adjusted is weighted using a data smoothing tool to obtain a preliminary corrected direction data set.

[0152] Based on the preliminary corrected direction data set, the fifth direction estimate is determined by comparing and adjusting it with the long-term changes using data verification tools.

[0153] For example, in the field of intelligent traffic monitoring, the historical data analysis of fourth-direction estimates can be explored from multiple perspectives. First, regarding the step of retrieving historical data from storage units, let's assume, in an urban road scenario, the system stores fourth-direction estimates generated every minute over the past 24 hours. This data records the directional changes of vehicles on specific road segments, such as fluctuations between 15.0 and 16.5 degrees north of east. When segmenting the data using time-series decomposition tools, it can be divided into short-term fluctuations and long-term changes. Short-term fluctuations may reflect temporary disturbances, such as brief jumps in direction values ​​caused by signal obstruction, while long-term changes may indicate gradual changes in road layout or environmental conditions.

[0154] For example, regarding the topic of short-term fluctuation detection, suppose that when using a data comparison tool to analyze short-term fluctuations, the preset threshold range is ±0.5 degrees. If the direction value suddenly jumps to 17.0 degrees north of east within a certain time window, exceeding the threshold range, it is marked as a range to be adjusted. This detection method can quickly identify abnormal data and provide a basis for subsequent adjustments.

[0155] It should be noted that the purpose of marking outlier data is to avoid short-term disturbances from misleading the overall direction estimation and to ensure the continuity and reliability of the data.

[0156] For example, when weighting the range to be adjusted, suppose we extract the 5-minute records before and after the outlier from historical data storage units and find that the direction value stabilizes at around 16.0 degrees north of east before the outlier. Using a data smoothing tool, we can assign a lower weight to the outlier and a higher weight to the surrounding stable data, ultimately obtaining a preliminarily corrected set of direction data, such as adjusting it back to 16.1 degrees north of east. This smoothing process effectively reduces the interference of outlier data and maintains the stability of the direction data.

[0157] For example, in the final verification and adjustment phase, when using data verification tools to compare with long-term trends, assuming the long-term trend shows that the direction value has gradually shifted 16.2 degrees east of north over the past few hours, while the initial correction value is 16.1 degrees east of north, the deviation is within the preset fluctuation range of 0.3 degrees. Therefore, the estimated value for the fifth direction can be determined to be 16.1 degrees east of north. This verification method, which combines long-term trends, ensures that the final result reflects both the rationality of the short-term correction and conforms to the overall trend.

[0158] For example, from an overall process perspective, the above-mentioned steps form a complete data processing chain in urban traffic monitoring. Whether it's detecting short-term fluctuations or referencing long-term changes, the aim is to improve the stability of direction estimation. Especially in complex urban environments where signal interference and road characteristics change frequently, this multi-layered processing approach can dynamically adapt to various situations, providing strong support for the accuracy of vehicle direction data, and thus providing a reliable basis for traffic management and route optimization.

[0159] In step S18, the real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result, including:

[0160] The estimated value of the fifth direction and the positioning coordinates are initially registered using a spatial geometric mapping tool, and then calibrated using a weighted average tool to determine the initially integrated set of spatial information.

[0161] Based on the preliminarily integrated set of spatial information, if the deviation between the estimated value of the fifth direction and the position of the positioning coordinates exceeds a preset deviation threshold, then a deviation correction tool is used to dynamically adjust the data and obtain the adjusted data set.

[0162] The direction estimates and positioning coordinates in the adjusted data set are finally integrated, and smoothed by combining historical records to obtain the fused positioning result.

[0163] For example, in the field of urban traffic monitoring, the integration and processing of vehicle direction and location data can be explored in depth from multiple perspectives. Regarding the step of retrieving the fifth direction estimate and real-time acquired positioning coordinate data from the storage unit, assume the system stores the direction estimate and corresponding position coordinate data updated every minute over the past hour. The direction estimate might be 16.1 degrees east of north, while the position coordinate data records the vehicle's specific latitude and longitude position on a certain road segment. When using data alignment tools for synchronization, the system will match the two based on timestamps, ensuring that each set of data is completely consistent in the time dimension, forming a time-consistent direction and location dataset.

[0164] For example, regarding the topic of initial registration using spatial geometric mapping tools, assuming a city road scenario, the system uses geometric mapping to perform an initial spatial match between the direction estimate and the location coordinates. If the direction indicates the vehicle is moving 16.1 degrees north of east, while the location coordinates show the vehicle is on a straight path along a certain road segment, the system will attempt to align the two. However, due to sensor errors or environmental interference, there may be some deviation, such as the direction deviating from the actual path by 0.2 degrees. In this case, a weighted averaging tool is used for calibration. The system assigns different weights based on data reliability, such as giving higher weight to the location coordinates and lower weight to the direction estimate, ultimately determining the initially integrated set of spatial information.

[0165] For example, in the process of detecting whether the deviation exceeds a preset threshold and making dynamic adjustments, suppose the preset threshold is 0.3 degrees, and after initial registration, the deviation in direction and position is found to be 0.4 degrees, exceeding the threshold range. The system will make dynamic adjustments using a deviation correction tool, possibly by interpolating data from surrounding time points to determine an adjusted data set that better reflects the actual scenario, such as reducing the deviation to 0.1 degrees. This approach can effectively address common signal interference or data loss issues in urban environments.

[0166] For example, regarding the topic of final integration and smoothing based on historical data, suppose the adjusted data set displays a direction of 16.0 degrees east of north, with location coordinates pointing to a specific point on a road segment. The system will use information fusion tools to deeply integrate the two data sets, while simultaneously extracting historical data from the past 10 minutes. It will find that the direction value fluctuates between 15.9 and 16.1 degrees east of north. Based on this, the system will smooth the final result, obtaining a fused location result of 16.0 degrees east of north. This processing method reduces data jumps and ensures the continuity of results, which is particularly important for the stability of vehicle trajectories in urban traffic monitoring.

[0167] For example, from an overall process perspective, the aforementioned steps form a complete data integration chain in an urban road scenario. Whether it's time alignment, initial registration, deviation calibration, or final fusion, all aim to improve the matching degree between directional and location data. This multi-layered processing approach can dynamically adapt to the complexities of the urban environment, providing reliable support for subsequent traffic management and route planning.

[0168] In step S19, the parameters and weight allocation rules of the multi-source data standardization model are periodically updated based on the fusion positioning results to generate new model parameter configurations for the next round of multi-source data processing, including:

[0169] Historical data and real-time feedback data of the fused positioning results are obtained, and time alignment tools are used for synchronization processing to obtain a data set under a unified time reference.

[0170] Based on the data set under the unified time benchmark, the historical data and the real-time feedback data are classified and summarized using data integration tools.

[0171] For the classified features, a weighted average tool is used to adjust the weight allocation and determine the weight combination suitable for the current loop processing;

[0172] According to the weight combination, if the detected weight allocation deviates from the preset weight threshold, the weights are fine-tuned by a dynamic optimization tool. Combining the characteristics of multi-source data, the adjusted weight set that meets the requirements of cyclic processing is determined.

[0173] Using the adjusted weight set, the parameters of the multi-source data standardization model configuration are updated using an information fusion tool. Combined with the training results, configuration parameters suitable for the next round of multi-source data processing are obtained.

[0174] For example, in the field of urban traffic monitoring, the processing of fused positioning data can be explored from multiple perspectives, particularly in the integration of historical data and real-time feedback data. Regarding the step of retrieving fused positioning historical data and real-time feedback data from the storage unit, assuming the system stores historical data updated every 5 minutes over the past 24 hours, including vehicle direction and location information on urban roads, while the real-time feedback data is the vehicle status information collected by sensors at the current moment. Since the two types of data may have time differences—for example, historical data recorded at 14:00 while real-time data is recorded at 14:02—the system will use a time alignment tool to synchronize both to a unified time base, such as aligning to 14:00 in minutes, forming a unified data set.

[0175] For example, regarding the topic of using data integration tools to classify and summarize historical and real-time feedback data, we can first understand the principle: classification and summarization mainly involve grouping data based on its source and characteristics. For instance, historical data can be categorized as a long-term trend reference, while real-time data can be categorized as an immediate state reflection. Suppose that in an urban road scenario, historical data reflects the average speed of vehicles on a certain road segment in the past as 40 km / h, while real-time data shows the current speed as 35 km / h. The system will classify the data based on these characteristics and provide a basis for subsequent weight allocation.

[0176] For example, in the step of adjusting weight allocation using a weighted average tool, suppose the system assigns weights based on the freshness of the data. Real-time data is assigned a weight of 0.7 due to its immediacy, while historical data is assigned a weight of 0.3 due to its stability. This weight combination is suitable for the current cycle processing and can balance the timeliness and reliability of the data. If the weight allocation is detected to deviate from the preset threshold range, such as the weight of real-time data exceeding 0.8, the system will make fine adjustments using a dynamic optimization tool. Taking into account the characteristics of multi-source data, such as the greater volatility of real-time data, the weight will be appropriately reduced to 0.75, forming an adjusted weight set.

[0177] For example, regarding the topic of updating model configuration parameters through an adjusted weight set, suppose the system uses an information fusion tool to apply the adjusted weights of 0.75 and 0.25 to the fusion algorithm. Combined with training results, such as model performance data from the past week, the parameters are updated to adapt to the next round of multi-source data processing. This approach ensures continuous model optimization in urban traffic monitoring, adapting to dynamically changing road environments. The final configuration parameters will provide a more practical reference for subsequent processing, improving the adaptability of data processing.

[0178] In summary, this invention discloses a fusion positioning method for autonomous vehicles. It utilizes a pre-established multi-source data standardization model to perform format conversion and consistency verification on satellite signals, sensor data, and motion trajectory estimation data. A weighted average fusion algorithm is employed to generate preliminary direction estimates, which are then evaluated and corrected using a Bayesian probability model. Subsequently, the invention uses a Kalman filter algorithm to smooth and predict the direction information, and performs local updates based on real-time external environmental feedback data. Multi-time-window trend analysis is used to detect and correct abnormal fluctuations. Finally, the optimized direction and position information are jointly optimized to generate the final fusion positioning result. Machine learning methods are used to periodically update model parameters, achieving high-precision, high-reliability, and highly adaptive positioning in complex scenarios.

[0179] Reference Figure 2 The second embodiment of the present invention provides a structural diagram of a fusion positioning system for an autonomous vehicle, including:

[0180] The first acquisition module 201 is used to convert the format of the original data through a pre-established multi-source data standardization model to generate the first dataset.

[0181] The second acquisition module 202 is used to filter heterogeneous data in the first dataset. If the deviation value of one of the data exceeds a preset deviation threshold, the corresponding record is removed to obtain the second dataset.

[0182] The third acquisition module 203 is used to perform weighted calculations on the directional information of different data in the second dataset to generate a first directional estimate.

[0183] The fourth acquisition module 204 is used to evaluate the credibility of the first direction estimate. If the credibility is lower than a preset evaluation threshold, a data correction process is triggered to obtain the second direction estimate.

[0184] The fifth acquisition module 205 is used to smooth and predict the direction information in the second direction estimate to generate a third direction estimate;

[0185] The sixth acquisition module 206 is used to acquire real-time feedback data of the external environment. If the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, a fourth-direction estimate is generated.

[0186] The seventh acquisition module 207 is used to detect short-term fluctuations in the direction information in the fourth direction estimate. If abnormal fluctuations are detected, the module combines historical data to perform smoothing correction and obtain the fifth direction estimate.

[0187] The eighth acquisition module 208 is used to acquire positioning coordinate data in real time, and to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result.

[0188] The configuration module 209 is used to periodically update the parameters and weight allocation rules of the multi-source data standardization model according to the fusion positioning results, and generate new model parameter configurations for the next round of multi-source data processing cycle.

[0189] It should be noted that the fusion positioning system for autonomous vehicles provided in this embodiment of the invention is used to execute all the process steps of the fusion positioning method for autonomous vehicles in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0190] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0191] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A fusion localization method for autonomous vehicles, characterized in that, include: The original data is converted into a format using a pre-established multi-source data standardization model to generate the first dataset. Filter the heterogeneous data in the first dataset. If the deviation value of one of the data exceeds the preset deviation threshold, remove the corresponding record to obtain the second dataset. The directional information of different data in the second dataset is weighted and calculated to generate a first directional estimate. The credibility of the first direction estimate is evaluated. If the credibility is lower than a preset evaluation threshold, a data correction process is triggered to obtain the second direction estimate. The direction information in the second direction estimate is smoothed and predicted to generate a third direction estimate. Obtain real-time feedback data of the external environment; if the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, generate a fourth-direction estimate. The short-term fluctuations in the direction information in the fourth direction estimate are detected. If abnormal fluctuations are detected, they are smoothed and corrected by combining historical data to obtain the fifth direction estimate. The real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result; Based on the fusion positioning results, the parameters and weight allocation rules of the multi-source data standardization model are updated periodically to generate new model parameter configurations for the next round of multi-source data processing. The real-time acquired positioning coordinate data is used to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result, including: The estimated value of the fifth direction and the positioning coordinates are initially registered using a spatial geometric mapping tool, and then calibrated using a weighted average tool to determine the initially integrated set of spatial information. Based on the preliminarily integrated set of spatial information, if the deviation between the estimated value of the fifth direction and the position of the positioning coordinates exceeds a preset deviation threshold, then a deviation correction tool is used to dynamically adjust the data and obtain the adjusted data set. The direction estimates and positioning coordinates in the adjusted data set are finally integrated, and smoothed by combining historical records to obtain the fused positioning result. The dynamic adjustment through deviation correction tools includes: interpolating data from surrounding time points to determine the adjusted data set that better reflects the actual scenario, thereby reducing the deviation.

2. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The process of filtering heterogeneous data in the first dataset involves removing the corresponding record if the deviation value of any data exceeds a preset deviation threshold, thus obtaining a second dataset, which includes: A data verification tool is used to compare the heterogeneous data in the first dataset item by item, calculate the deviation value of each data and the preset deviation threshold. If the deviation value exceeds the threshold range, the relevant records of the data are deleted to obtain a preliminary filtered data set. The initially screened dataset is cleaned using data cleaning tools to generate a cleaned dataset. Data integration tools are used to reorganize the cleaned data set according to a unified structure. If a field is missing, it is filled in according to a pre-established field completion rule to determine the reorganized data set. The recombined datasets are merged and formatted into a standard structure to generate the second dataset.

3. The fusion positioning method for autonomous vehicles according to claim 1 or 2, characterized in that, The step of weighting the directional information of different data in the second dataset to generate a first directional estimate includes: Obtain directional information and real-time data from different data sources from the second dataset, and assign initial weights to the directional information according to a pre-established weight allocation rule to obtain a pre-weighted set of directional information. The signal strength change in the real-time data is determined. If the signal strength is higher than the preset change threshold range, the weight ratio of the corresponding data source is increased, and vice versa, the weight ratio is decreased to obtain the dynamically adjusted weight combination. Based on the dynamically adjusted weight combination, a weighted average is calculated on the initially weighted set of direction information to obtain the fused direction information value and thus obtain the first direction estimate.

4. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The process of evaluating the credibility of the first direction estimate, and if the credibility is lower than a preset evaluation threshold, triggers a data correction process to obtain a second direction estimate, including: Based on the first direction estimate, the signal interference level and sensor data records are obtained from the dynamic environment to obtain a sorted data set. According to the pre-established evaluation criteria, if the confidence value in the sorted data set is lower than the preset confidence threshold, then the range of data to be corrected is determined. Based on the corrected data range, sensor data is reacquired from the dynamic environment, and combined with the degree of signal interference, the adjusted direction reference value is obtained; The adjusted direction reference value is associated with and saved with the confidence value to obtain the second direction estimate.

5. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The step of smoothing and predicting the direction information in the second direction estimate to generate a third direction estimate includes: Based on the second direction estimate, vehicle speed change and road curvature information are obtained from the dynamic environment to obtain a sorted parameter set; Based on the sorted parameter set and combined with the pre-established dynamic environment perception rules, the influence of vehicle speed and road curvature is comprehensively evaluated to determine the fused direction reference data. The directional information of the fused directional reference data is continuously adjusted. If the fluctuation of the adjusted data exceeds the preset fluctuation threshold, the real-time parameters in the dynamic environment are reacquired to obtain the adjusted directional information. Based on the adjusted direction information, the deviation between the prediction result and the actual environmental perception data is corrected, the corrected direction estimate is saved, and the third direction estimate is generated.

6. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The step of acquiring real-time feedback data from the external environment, and generating a fourth-direction estimate if the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, includes: Satellite signal strength and road feature point data are obtained from external environmental feedback to obtain a set of organized environmental data; A deviation analysis is performed between the third-party directional estimate and the environmental data set. If the deviation exceeds a preset dynamic threshold, a local data update process is initiated to determine the direction and range of the correction data. Obtain the latest external environment feedback information, adjust the range of the corrected direction data, and obtain a preliminary corrected direction reference value; The preliminary corrected direction reference value is compared with the real-time data to generate the fourth direction estimate.

7. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The method involves detecting short-term fluctuations in the direction information in the fourth direction estimate. If abnormal fluctuations are detected, a smoothing correction is performed using historical data to obtain the fifth direction estimate, including: The historical records of the fourth direction estimate are obtained, and the distribution of the historical records in multiple time windows is segmented using a time series decomposition tool to obtain a separate dataset of short-term fluctuations and long-term changes. The short-term fluctuations are detected using a data comparison tool. If the short-term fluctuations exceed a preset fluctuation threshold, the abnormal data is marked as being within the range to be adjusted. Based on the range to be adjusted, the corresponding direction estimation records are extracted from the historical data storage unit, and the range to be adjusted is weighted using a data smoothing tool to obtain a preliminary corrected direction data set. Based on the preliminary corrected direction data set, the fifth direction estimate is determined by comparing and adjusting it with the long-term changes using data verification tools.

8. The fusion positioning method for autonomous vehicles according to claim 1, characterized in that, The step of periodically updating the parameters and weight allocation rules of the multi-source data standardization model based on the fused positioning results to generate new model parameter configurations for the next round of multi-source data processing includes: Historical data and real-time feedback data of the fused positioning results are obtained, and time alignment tools are used for synchronization processing to obtain a data set under a unified time reference. Based on the data set under the unified time benchmark, the historical data and the real-time feedback data are classified and summarized using data integration tools. For the classified features, a weighted average tool is used to adjust the weight allocation and determine the weight combination suitable for the current loop processing; According to the weight combination, if the detected weight allocation deviates from the preset weight threshold, the weights are fine-tuned by a dynamic optimization tool. Combining the characteristics of multi-source data, the adjusted weight set that meets the requirements of cyclic processing is determined. Using the adjusted weight set, the parameters of the multi-source data standardization model configuration are updated using an information fusion tool. Combined with the training results, configuration parameters suitable for the next round of multi-source data processing are obtained.

9. A fusion positioning system for an autonomous vehicle, used to implement the fusion positioning method for an autonomous vehicle as described in any one of claims 1-8, characterized in that, include: The first acquisition module is used to convert the format of the raw data through a pre-established multi-source data standardization model to generate the first dataset. The second acquisition module is used to filter heterogeneous data in the first dataset. If the deviation value of one of the data exceeds a preset deviation threshold, the corresponding record is removed to obtain the second dataset. The third acquisition module is used to perform weighted calculations on the directional information of different data in the second dataset to generate a first directional estimate. The fourth acquisition module is used to evaluate the credibility of the first direction estimate. If the credibility is lower than a preset evaluation threshold, a data correction process is triggered to obtain the second direction estimate. The fifth acquisition module is used to smooth and predict the direction information in the second direction estimate to generate a third direction estimate; The sixth acquisition module is used to acquire real-time feedback data of the external environment. If the deviation between the feedback data and the third-direction estimate exceeds a preset dynamic threshold, a fourth-direction estimate is generated. The seventh acquisition module is used to detect short-term fluctuations in the direction information in the fourth direction estimate. If abnormal fluctuations are detected, the module combines historical data to perform smoothing corrections and obtain the fifth direction estimate. The eighth acquisition module is used to acquire positioning coordinate data in real time, and to jointly optimize the direction information and position information in the fifth direction estimate to generate a fused positioning result. The configuration module is used to periodically update the parameters and weight allocation rules of the multi-source data standardization model based on the fusion positioning results, and generate new model parameter configurations for the next round of multi-source data processing.

Images