Disclosure of Invention
The invention provides a fusion positioning method and a fusion positioning system for an automatic driving vehicle, which are used for realizing the reliability of a positioning result.
In order to solve the above technical problems, the present invention provides a fusion positioning method for an autopilot vehicle, including:
converting the format of the original data through a pre-established multi-source data standardization model to generate a first data set;
screening heterogeneous data in the first data set, and if the deviation value of one data exceeds a preset deviation threshold value, eliminating the corresponding record to obtain a second data set;
performing weighted calculation on the direction information of different data in the second data set to generate a first direction estimated value;
Performing reliability evaluation on the first direction estimated value, and triggering a data correction flow to obtain a second direction estimated value if the reliability is lower than a preset evaluation threshold;
carrying out smoothing processing and prediction on the direction information in the second direction estimated value to generate a third direction estimated value;
acquiring real-time feedback data of an external environment, and generating a fourth direction estimated value if the deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold value;
Detecting short-term fluctuation of the direction information in the fourth direction estimated value, and if abnormal fluctuation is detected, carrying out smoothing correction by combining historical data to obtain a fifth direction estimated value;
the positioning coordinate data obtained in real time carries out joint optimization on the direction information and the position information in the fifth direction estimated value to generate a fusion positioning result;
and periodically updating parameters and weight distribution rules of the multi-source data standardization model according to the fusion positioning result to generate new model parameter configuration for multi-source data processing circulation of the next round.
Preferably, the screening the heterogeneous data in the first data set, if the deviation value of one of the data exceeds a preset deviation threshold, removing the corresponding record to obtain a second data set, including:
comparing heterogeneous data in the first data set item by adopting a data verification tool, calculating a deviation value of each data and a preset deviation threshold value, and deleting relevant records of the data if the deviation value exceeds the threshold value range to obtain a primarily screened data set;
Cleaning the primarily screened data set through a data cleaning tool to generate a cleaned data set;
adopting a data integration tool to reorganize the cleaned data set according to a unified structure, and if the field is missing, filling the data set through a pre-established field complement rule to determine the reorganized data set;
And summarizing the recombined data set and formatting the data set into a standard structure to generate the second data set. Preferably, the weighting calculation is performed on the direction information of the different data in the second data set, and generating a first direction estimated value includes:
acquiring direction information and real-time data in different data sources from the second data set, and giving initial weight to the direction information through a pre-established weight distribution rule to obtain a preliminarily weighted direction information set;
Judging the signal intensity change in the real-time data, if the signal intensity is higher than a preset change threshold range, increasing the weight proportion of the corresponding data source, otherwise, reducing the weight proportion, and obtaining a weight combination after dynamic adjustment;
according to the dynamically adjusted weight combination, carrying out weighted average calculation on the preliminarily weighted direction information set to obtain a fused direction information value, and obtaining a first direction estimated value;
And the method further comprises the step of storing the first direction estimated value, the data source and the weight distribution record in a correlated manner through a data storage tool, and generating a structured direction estimated data record.
Preferably, the performing reliability evaluation on the first direction estimated value, if the reliability is lower than a preset evaluation threshold, triggering a data correction flow to obtain a second direction estimated value, including:
Acquiring signal interference degree and sensor data record from dynamic environment according to the first direction estimated value to obtain a tidied data set;
According to a pre-established evaluation standard, if the reliability value in the sorted data set is lower than a preset reliability threshold value, determining a corrected data range;
Re-acquiring sensor data from a dynamic environment according to the corrected data range, and combining the signal interference degree to obtain an adjusted direction reference value;
And carrying out association storage on the adjusted direction reference value and the credibility value to obtain the second direction estimated value.
Preferably, the smoothing and predicting the direction information in the second direction estimated value, generating a third direction estimated value includes:
acquiring vehicle speed change and road curvature information from a dynamic environment according to the second direction estimated value to obtain a sorted parameter set;
According to the sorted parameter set, comprehensively evaluating the influence of the vehicle speed and the road curvature by combining with a pre-established dynamic environment perception rule, and determining fused direction reference data;
Continuously adjusting the direction information of the fused direction reference data, and if the adjusted data fluctuation exceeds a preset fluctuation threshold value, re-acquiring real-time parameters in a dynamic environment to obtain the adjusted direction information;
correcting deviation of the prediction result and the actual environment sensing data according to the adjusted direction information, storing the corrected direction estimated value, and generating the third direction estimated value.
Preferably, the acquiring feedback data of the real-time external environment, if a deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold, generating a fourth direction estimated value includes:
Acquiring satellite signal intensity and road characteristic point data from external environment feedback to obtain a tidied environment data set;
performing deviation analysis on the third direction estimated value and the environment data set, and if the deviation exceeds a preset dynamic threshold value, starting a local data updating process to determine a corrected direction data range;
acquiring the latest external environment feedback information, and adjusting the corrected direction data range to obtain a preliminary corrected direction reference value;
And finally comparing the initially corrected direction reference value with real-time data to generate the fourth direction estimated value.
Preferably, the detecting short-term fluctuation of the direction information in the fourth direction estimated value, if abnormal fluctuation is detected, performing smoothing correction in combination with the historical data to obtain a fifth direction estimated value, includes:
Acquiring a history record of the fourth direction estimated value, and carrying out segmentation processing on the distribution condition of the history record in a plurality of time windows by adopting a time sequence decomposition tool to obtain a separated data set of short-term fluctuation and long-term change;
Detecting the short-term fluctuation by using a data comparison tool, and marking the abnormal data as a range to be adjusted if the short-term fluctuation exceeds a preset fluctuation threshold;
Extracting a corresponding direction estimation record from a historical data storage unit according to the range to be adjusted, and carrying out weighting treatment on the range to be adjusted by adopting a data smoothing tool to obtain a primarily corrected direction data set;
And according to the primarily corrected direction data set, a data verification tool is used for carrying out comparison and adjustment in combination with the long-term change, and the fifth direction estimated value is determined.
Preferably, the performing joint optimization on the direction information and the position information in the fifth direction estimated value according to the positioning coordinate data acquired in real time to generate a fusion positioning result includes:
Performing preliminary registration on the fifth direction estimated value and the positioning coordinates by using a space geometric mapping tool, and performing calibration processing by using a weighted average tool to determine a preliminarily integrated space information set;
According to the preliminarily integrated space information set, if the deviation between the fifth direction estimated value and the position of the positioning coordinate is detected to exceed a preset deviation threshold value, dynamically adjusting by a deviation correcting tool to obtain an adjusted data set;
And finally integrating the direction estimated value in the adjusted data set and the positioning coordinate, and carrying out smoothing treatment by combining a history record to obtain a fusion positioning result.
Preferably, the updating the parameters and weight allocation rules of the multi-source data standardization model according to the fusion positioning result periodically generates new model parameter configuration for the next round of multi-source data processing cycle, including:
Acquiring historical data and real-time feedback data of the fusion positioning result, and performing synchronous processing by adopting a time alignment tool to obtain a data set under a unified time reference;
classifying and summarizing the historical data and the real-time feedback data through a data integration tool according to the data set under the unified time reference;
the classified features are subjected to weight distribution adjustment by adopting a weighted average tool, and weight combinations suitable for current circulation processing are determined;
According to the weight combination, if the detected weight distribution deviates from a preset weight threshold, the weight is finely adjusted through a dynamic optimization tool, and an adjusted weight set meeting the requirement of cyclic processing is judged by combining the characteristics of multi-source data;
and updating parameters of the multisource data standardized model configuration by adopting an information fusion tool through the adjusted weight set, and acquiring configuration parameters applicable to the next round of multisource data processing by combining training results.
In a second aspect, the present invention provides a fusion positioning system for an autonomous vehicle comprising:
the first acquisition module is used for carrying out format conversion on the original data through a pre-established multi-source data standardization model to generate a first data set;
The second acquisition module is used for screening heterogeneous data in the first data set, and if the deviation value of one data exceeds a preset deviation threshold value, the corresponding record is removed to obtain a second data set;
the third acquisition module is used for carrying out weighted calculation on the direction information of different data in the second data set to generate a first direction estimated value;
a fourth obtaining module, configured to perform reliability evaluation on the first direction estimated value, and if the reliability is lower than a preset evaluation threshold, trigger a data correction procedure to obtain a second direction estimated value;
A fifth obtaining module, configured to perform smoothing and prediction on the direction information in the second direction estimated value, and generate a third direction estimated value;
A sixth obtaining module, configured to obtain feedback data of the real-time external environment, and if a deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold, generate a fourth direction estimated value;
A seventh obtaining module, configured to detect short-term fluctuations of the direction information in the fourth direction estimated value, and if abnormal fluctuations are detected, perform smoothing correction in combination with the historical data to obtain a fifth direction estimated value;
an eighth obtaining module, configured to obtain positioning coordinate data in real time, perform joint optimization on direction information and position information in the fifth direction estimated value, and generate a fusion positioning result;
And the configuration module is used for periodically updating the parameters and weight distribution rules of the multi-source data standardization model according to the fusion positioning result to generate new model parameter configuration for the multi-source data processing cycle of the next round.
Compared with the prior art, the invention discloses a fusion positioning method of an automatic driving vehicle, which comprises the steps of carrying out format conversion and consistency verification on original data of different sources through a pre-established multi-source data standardization model, generating a first direction estimated value by adopting a weighted average fusion algorithm, and carrying out credibility evaluation and correction by combining a Bayesian probability model. Then, the invention utilizes Kalman filtering algorithm to carry out smoothing treatment and prediction on the direction information, carries out local update according to real-time external environment feedback data, and detects abnormal fluctuation and corrects through multi-time window trend analysis. Finally, the optimized direction information and the position information are combined and optimized to generate a final fusion positioning result, and model parameters are updated regularly through a machine learning method, so that high-precision, high-reliability and high-adaptability positioning under a complex scene are realized.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a first embodiment of the present invention provides a flow chart of a fusion positioning method for an automatic driving vehicle, comprising the following steps:
s11, converting the format of the original data through a pre-established multi-source data standardization model to generate a first data set;
s12, screening heterogeneous data in the first data set, and if the deviation value of one data exceeds a preset deviation threshold, eliminating the corresponding record to obtain a second data set;
S13, carrying out weighted calculation on the direction information of different data in the second data set to generate a first direction estimated value;
s14, carrying out reliability evaluation on the first direction estimated value by combining a Bayesian probability model, and triggering a data correction flow if the reliability is lower than a preset evaluation threshold value to obtain a second direction estimated value;
S15, carrying out smoothing processing and prediction on the direction information in the second direction estimated value by using a Kalman filtering algorithm to generate a third direction estimated value;
S16, acquiring real-time feedback data of an external environment, and if the deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold value, generating a fourth direction estimated value;
s17, detecting short-term fluctuation of the direction information in the fourth direction estimated value, and if abnormal fluctuation is detected, carrying out smoothing correction by combining historical data to obtain a fifth direction estimated value;
S18, positioning coordinate data obtained in real time are subjected to joint optimization on direction information and position information in the fifth direction estimated value, and a fusion positioning result is generated;
and S19, periodically updating parameters and weight distribution rules of the multi-source data standardization model according to the fusion positioning result, and generating new model parameter configuration for multi-source data processing circulation of the next round.
In step S11, format conversion is performed on the original data by a pre-established multi-source data standardization model, and a first data set is generated, including:
Raw data are acquired from a plurality of data sources, and initial data sets are obtained by extracting satellite signals, sensor data and motion trail calculation data through a pre-established acquisition tool. And processing the original data of different sources by adopting a format conversion tool aiming at the initial data set, and if the data fields are inconsistent, adjusting through a field mapping table to determine a unified data structure. And according to the unified data structure, the processed data are encoded by using a standardized encoding rule, and data conforming to a standard format are generated through a preset encoding template, so that a standardized data set is obtained. And summarizing the standardized data set through a data integration tool to generate a first data set, and storing the first data set into a designated database for a subsequent processing link.
For example, in processing satellite signals, sensor data, and motion trajectory estimation data, embodiments of this flow may be understood through a specific scenario. It is assumed that in an intelligent traffic monitoring system, vehicle operation data needs to be obtained from multiple sources. The satellite signals may come from GPS equipment to provide real-time position of vehicle, the sensor data come from vehicle-mounted equipment to record speed and acceleration, and the motion track calculation data can be used for predicting possible path of vehicle by means of algorithm. These data sources are in different formats, and the field names and units may be different, so that the original data needs to be extracted by a pre-established collection tool to form an initial data set.
For format conversion of an initial data set, a format conversion tool can be used for unifying data of different sources, and specifically, a pre-established multi-source data standardization model defines unified processing rules for three main original data sources, namely satellite signals, sensor data and motion trail calculation data. The model contains a specific format conversion template and consistency check logic. In the conversion process, the model firstly identifies the source type (such as GPS signals, IMU data, wheel speed meter data, visual positioning data and the like) of input data, then applies a preset analysis rule for the type, deconstructs the original format (such as NMEA sentences, CAN bus messages, custom binary streams, image coordinates and the like) and extracts key information fields. Then, the model uses the built-in field mapping table to forcedly map the fields (such as 'time stamp' and 'time_record', longitude and latitude degree minute and decimal system, speed m/s and km/h) with the same meaning but inconsistent naming or unit in different source data onto the predefined standard field names and unified unit.
For example, the time field of GPS data may be "time" and the sensor data "time_record", both of which are unified to "time" by the field mapping table. If the GPS position data is in units of degrees and the sensor data is in units of meters, the units are adjusted through the mapping table, so that the consistency of the data fields is ensured. The unified data structure lays a foundation for subsequent processing, avoids data conflict caused by format difference, and improves the accuracy of data processing.
The construction of the model begins with the depth analysis of multi-source heterogeneous data in an automatic driving scene, and covers physical characteristics and logic structures of data sources such as satellite positioning (such as GPS/Beidou NMEA protocol), inertial sensors (IMU angular velocity/acceleration original values), wheel speed meter pulse signals, visual positioning coordinates, high-precision map feature points and the like. Based on the method, a three-layer standardized architecture is designed, wherein a bottom layer analysis layer is used for customizing binary/text analysis rules (such as resolving GPGGA fields of NMEA sentences into longitude, latitude, altitude and satellite numbers, analyzing messages of CAN bus ID 0x0A0 into yaw rate) for each data source, a middle mapping layer is used for establishing a dynamic field mapping table and unifying semantic conflicts (such as unifying the 'direction angle' fields in different systems to be named as 'head', the units are forcedly converted into 0-360 degrees, and unifying 'time stamps' to be the nanosecond UTC time in an ISO 8601 format), a top layer check layer is used for implanting physical logic constraints (such as a vehicle speed range of 0-200km/h and a road curvature radius threshold of 500 meters), and performing self-adaptive filling (for missing GPS elevation data, adopting Kalman filtering to fusion IMU Z-axis acceleration complementation). The model verifies the post-curing parameters through historical multi-source data training, and finally forms a multi-source data standardized model comprising data pattern recognition, outlier filtering, unit conversion and a spatial coordinate system (such as a WGS 84-local ENU coordinate system).
In one possible implementation, the application of standardized coding rules may further optimize the data quality. Assuming that the processed data is required to meet a specific industry standard, a preset coding template can be used to unify the time format into a 'YYYY-MM-DDHH:MM-SS', and the position data is coded into a theodolite format, such as 'latitude:39.9042, longitude:116.4074'. By the coding, the data set conforming to the standard format is generated, so that seamless transmission of the data among different systems is ensured, and errors caused by inconsistent formats are reduced.
For example, the use of a data integration tool may aggregate a standardized set of data into a first set of data. Assuming that the GPS locations, speeds, and predicted trajectories of all vehicles are integrated into a table, each record contains fields for vehicle ID, time, location, and speed, etc., a first data set is ultimately generated and stored in a designated database. The integration mode is convenient for subsequent analysis, such as traffic flow prediction or congestion early warning, and the data utilization efficiency is remarkably improved.
It should be noted that, the process of storing in the designated database also needs to pay attention to data security and access speed. The first data set can be stored through the distributed database, so that efficient query of large-scale data is ensured, and meanwhile, authority control is set, and data privacy is protected. The method not only improves the reliability of data processing, but also provides solid support for subsequent links such as real-time monitoring or historical data analysis. Through the flow, from data acquisition to standardization to integration, each link is tightly connected, so that the consistency and usability of the data are ensured, and finally, a high-quality data base is provided for decision support of an intelligent traffic system, and higher service value and operation efficiency are brought.
In step S12, heterogeneous data in the first data set is screened, if the deviation value of one of the data exceeds a preset deviation threshold, the corresponding record is removed, and a second data set is obtained, including:
comparing heterogeneous data in the first data set item by adopting a data verification tool, calculating a deviation value of each data and a preset deviation threshold value, and deleting relevant records of the data if the deviation value exceeds the threshold value range to obtain a primarily screened data set;
Cleaning the primarily screened data set through a data cleaning tool to generate a cleaned data set;
adopting a data integration tool to reorganize the cleaned data set according to a unified structure, and if the field is missing, filling the data set through a pre-established field complement rule to determine the reorganized data set;
and summarizing the recombined data set and formatting the data set into a standard structure to generate the second data set.
For example, in the context of an intelligent traffic monitoring system, specific embodiments of each technical topic may be further discussed from multiple angles with respect to the processing flow of the first data set, and a detailed analysis may be performed in conjunction with a business context.
For example, for the use of data verification tools, the core is item-by-item alignment and bias value calculation for heterogeneous data. Assuming that the first data set contains position data from the GPS device and speed data from the onboard sensors, the verification tool will compare the records of each vehicle item by item, check if the position data is within a reasonable range, such as if the longitude and latitude are outside of city boundaries, and if the speed data meets normal ranges, such as 0 to 120 km/h. If the display speed of a record is 200 km/h and is obviously beyond the range of 80 to 150 km/h of the preset threshold value, calculating the deviation value of the record and marking the record as the item to be removed. The method can effectively identify the abnormal data and provides a more reliable basis for subsequent processing.
For example, in the application of a data cleansing tool, deleting marked records is a critical step for a primarily screened data set. It is assumed that one of the records of a vehicle is marked for speed anomalies and is removed by the cleaning tool while other normal records, such as records with both location and time fields meeting a threshold, are maintained. The cleaned data set is more simplified, and the subsequent analysis is ensured not to be interfered by abnormal values.
For example, with respect to the reorganization process of the data consolidation tool, unified structure and field completion are important. Assuming that part of the records in the cleaned data set lack acceleration fields, the integration tool can calculate the complement missing values from the speed change trend according to a pre-established rule. For example, the speed of a vehicle in two seconds is changed from 40 km/h to 50 km/h, and the acceleration can be estimated to be a certain value and the field can be filled. The completion mode enables the recombined data set to be more complete, and unified analysis is facilitated.
For example, in the use of data storage tools, formatting the reorganized data set into a standard structure and generating a second data set is the final link. Assuming that the data is formatted as a table containing fields for vehicle ID, time, location, speed, etc., the time of each record is unified into a standard format, such as 2023-10-0108:00:00, and the location data is stored in latitude and longitude form. The standard structure is convenient for subsequent system reading and processing, and ensures efficient circulation of data among different modules. Through the refinement treatment of the links, each step is tightly spread around the requirement of intelligent traffic monitoring from verification to cleaning and integration to storage, so that the accuracy and consistency of data are ensured, and a solid foundation is laid for subsequent business analysis.
In step S13, performing a weighted calculation on the direction information of the different data in the second data set to generate a first direction estimated value, including:
acquiring direction information and real-time data in different data sources from the second data set, and giving initial weight to the direction information through a pre-established weight distribution rule to obtain a preliminarily weighted direction information set;
Judging the signal intensity change in the real-time data, if the signal intensity is higher than a preset change threshold range, increasing the weight proportion of the corresponding data source, otherwise, reducing the weight proportion, and obtaining a weight combination after dynamic adjustment;
according to the dynamically adjusted weight combination, carrying out weighted average calculation on the preliminarily weighted direction information set to obtain a fused direction information value, and obtaining a first direction estimated value;
And the method further comprises the step of storing the first direction estimated value, the data source and the weight distribution record in a correlated manner through a data storage tool, and generating a structured direction estimated data record.
For example, in the context of an intelligent traffic monitoring system, specific embodiments may be discussed from multiple angles for direction information processing of the second data set, with refined analysis in conjunction with business context. For processes that acquire direction information and real-time data from different data sources.
It is understood that the data sources may include road side cameras, car navigation devices, and positioning modules of mobile terminals. Assuming that the camera provides vehicle direction angle data, the car navigation device provides direction estimation based on the historical track, and the mobile terminal provides real-time direction information through the gyroscope. These sources of data may be accompanied by historical error rate values and signal strength recordings, for example, the camera may have a light affected error rate of 5% and the vehicle may have an error rate of up to 15% in a weak tunnel.
For example, for a link to which an initial weight is given by a weight distribution rule established in advance, it is assumed that the weight distribution rule is based on a comprehensive evaluation of a historical error rate and signal strength. The camera may be given an initial weight of 0.5 due to low error rate, the in-vehicle device is 0.3, and the mobile terminal is 0.2. This initial weight reflects the difference in credibility of the data sources, forming a preliminary weighted set of directional information.
It should be noted that, the design of the weight allocation rule needs to consider the specificity of the service scenario, for example, in a dense urban area, the camera data may be more reliable, and in suburban areas, the vehicle-mounted devices may be relied on.
For example, in dynamically adjusting weights using data processing tools, it is critical to incorporate signal strength changes in real-time data. Assuming that the signal intensity of the camera at a certain moment drops below a preset threshold value due to weather, the weight of the camera may drop from 0.5 to 0.3, and the signal intensity of the vehicle-mounted equipment is higher than the threshold value, and the weight is increased from 0.3 to 0.4. The dynamic adjustment can be better adapted to the real-time environment change, and the accuracy of the direction information is ensured.
For example, for the link of weighted average calculation of the data fusion tool, assuming that the direction information of three data sources is respectively 10 degrees, 15 degrees and 20 degrees in north and east, the weights after dynamic adjustment are 0.3, 0.4 and 0.3, and the value of the direction information after fusion can be obtained by weighted average to be about 14.5 degrees and used as the first direction estimated value. The fusion mode can integrate advantages of multi-source data, and stability of direction estimation is improved.
For example, in the step of storing data using a data storage tool association, it is assumed that the first direction estimate value of 14.5 degrees is stored as a structured record, including fields such as a time stamp, a device identification, etc., along with the data source, the weight allocation record. Such structured storage facilitates subsequent traceability and analysis.
It should be noted that, when judging whether the preset credibility standard is met, comparing the historical data, and if the deviation between the estimated value and the actual track is less than 2 degrees, considering that the standard is met. The judging mechanism is helpful for screening out high-quality direction estimation data, and provides support for subsequent traffic flow analysis or path planning.
For example, for the implementation of the links described above, the business context is always stretched around intelligent traffic monitoring. And from data acquisition to dynamic weight adjustment, fusion and storage, each step closely meets the requirement of vehicle direction estimation, and the practicability and reliability of the data are ensured. The refinement processing can effectively improve the accuracy of the direction information and provide more valuable data support for traffic management.
In step S14, the reliability evaluation is performed on the first direction estimated value, and if the reliability is lower than a preset evaluation threshold, a data correction procedure is triggered to obtain a second direction estimated value, including:
Acquiring signal interference degree and sensor data record from dynamic environment according to the first direction estimated value to obtain a tidied data set;
According to a pre-established evaluation standard, if the reliability value in the sorted data set is lower than a preset reliability threshold value, determining a corrected data range;
Re-acquiring sensor data from a dynamic environment according to the corrected data range, and combining the signal interference degree to obtain an adjusted direction reference value;
And carrying out association storage on the adjusted direction reference value and the credibility value to obtain the second direction estimated value.
For example, in the context of intelligent traffic monitoring, for subsequent processing of the first direction estimate, a refinement analysis may be performed from the perspective of signal interference and data differences. The degree of signal interference may be due to environmental factors such as weather changes or electromagnetic interference, while the sensor data records include directional information collected by different devices. It is assumed that in urban road monitoring, sensor data originate from a roadside camera and vehicle-mounted equipment, the degree of signal interference causes blurring of the camera data due to heavy rain and weather, and the vehicle-mounted equipment generates data fluctuation due to signal shielding. Through the data processing tool, the data can be cleaned and classified, obvious abnormal values are removed, a data set after arrangement is formed, and a basis is provided for subsequent evaluation.
For example, for links that employ probabilistic computation tools to evaluate confidence values.
It will be appreciated that the pre-established evaluation criteria may be based on a comprehensive determination of historical data bias and real-time signal quality. Assuming that the sorted data set shows that the reliability value of the estimated value of a certain direction is 0.6 and the preset threshold value is 0.8, the data set is lower than the standard, and the subsequent processing flow is triggered.
It should be noted that, this evaluation method can find potential problems in the data in time, and ensure reliability of the direction information.
For example, in the process of determining a data range to be corrected and reacquiring data, it is assumed that sensor data of the vehicle-mounted device is reacquired from a dynamic environment for camera data having a large disturbance by a data correction tool, and meanwhile, direction estimation is adjusted in combination with a signal disturbance degree, such as signal attenuation caused by heavy rain. Assuming that the original estimated value is 14.5 degrees north-east, the adjusted direction reference value may be 13.8 degrees north-east. This approach can effectively cope with data bias caused by environmental changes.
For example, the step of storing the adjusted direction reference value and the confidence value is directed to using a data storage tool association.
It will be appreciated that the stored content may include a time stamp, a device source, and an adjustment record. Assuming that the adjusted direction reference value is 13.8 degrees, the reliability value is increased to 0.85, and the direction reference value meets the requirement compared with the preset evaluation standard of 0.8, and finally the direction reference result is confirmed. This storage facilitates subsequent trace back and data analysis.
For example, in the implementation of the links described above, the business context is always stretched around intelligent traffic monitoring. From the recognition of signal interference to data arrangement, probability evaluation, data correction and final storage, each step closely meets the requirement of vehicle direction estimation, and the practicability of the data is ensured. The refinement processing can provide more reliable direction reference for traffic management, particularly improves the adaptability of data processing in complex environments, and lays a foundation for subsequent path planning or flow analysis.
In step S15, the smoothing and predicting the direction information in the second direction estimated value, to generate a third direction estimated value, including:
acquiring vehicle speed change and road curvature information from a dynamic environment according to the second direction estimated value to obtain a sorted parameter set;
According to the sorted parameter set, comprehensively evaluating the influence of the vehicle speed and the road curvature by combining with a pre-established dynamic environment perception rule, and determining fused direction reference data;
Continuously adjusting the direction information of the fused direction reference data, and if the adjusted data fluctuation exceeds a preset fluctuation threshold value, re-acquiring real-time parameters in a dynamic environment to obtain the adjusted direction information;
correcting deviation of the prediction result and the actual environment sensing data according to the adjusted direction information, storing the corrected direction estimated value, and generating the third direction estimated value.
For example, in the context of intelligent traffic monitoring, for subsequent processing of the second direction estimate, a refinement analysis may be performed in conjunction with specific business requirements, starting with vehicle speed changes and road curvature information in a dynamic environment. For the use of data acquisition tools.
It will be appreciated that vehicle speed variations may be affected by real-time traffic conditions, while road curvature information is related to road design and topography. In urban road monitoring, vehicle speed data are acquired through a road side sensor and vehicle-mounted equipment, abnormal values caused by equipment faults, such as data with speed suddenly changed to 0 or exceeding a reasonable range, are primarily filtered, a finished parameter set is formed, and a reliable basis is provided for subsequent fusion.
For example, for application of a data fusion tool, the impact of vehicle speed and road curvature can be comprehensively evaluated in combination with pre-established dynamic environment awareness rules. Assuming that on a curved road the speed of the vehicle decreases from 60 km/h to 40 km/h while the road curvature shows a medium curvature, the fusion rules may prioritize the impact of the speed decrease on the directional reference data, resulting in a more realistic directional reference value. This approach can be effective to accommodate changes in dynamic environments.
For example, in the use of the smoothing tool, the continuity adjustment is performed with respect to the fused direction reference data, so that the abrupt change of the direction information can be avoided. Assuming that the adjusted data fluctuation range is preset to be plus or minus 2 degrees, and the actual fluctuation reaches 3 degrees, the threshold value range is exceeded, and real-time parameters in the dynamic environment, such as vehicle speed and road curvature data, need to be acquired again, and direction information is readjusted. This approach helps to maintain the stability of the direction information.
For example, for applications of the information calibration tool, offset correction of the prediction result from the actual environment-aware data is particularly important. Assuming that the predicted direction is 13.8 degrees north-east and the actual perceived data is 14.2 degrees north-east, the calibration tool will be fine-tuned according to the deviation to obtain a direction estimated value closer to the actual direction. Such a calibration can promote the accuracy of the directional references.
For example, the corrected direction estimated value is saved by the data storage tool, and whether the direction estimated value meets the stability requirement or not is judged, so that a basis can be provided for subsequent analysis. Assuming that the corrected direction value is 14.0 degrees north-east, the stability evaluation shows that the fluctuation range is within a preset threshold, and the final direction reference basis is confirmed. This storage and judgment mechanism facilitates data trace back and long-term monitoring.
For example, from an overall business context, the above links are deployed closely around intelligent traffic monitoring, whether data collection, fusion, smoothing, or calibration and storage, to provide reliable support for vehicle direction estimation. In particular, in a complex urban road environment, the multi-loop cooperative processing can adapt to dynamic changes, ensure the continuity and stability of direction information, and provide important reference values for traffic management and path planning.
In step S16, feedback data of the real-time external environment is obtained, and if the deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold, a fourth direction estimated value is generated, including:
Acquiring satellite signal intensity and road characteristic point data from external environment feedback to obtain a tidied environment data set;
performing deviation analysis on the third direction estimated value and the environment data set, and if the deviation exceeds a preset dynamic threshold value, starting a local data updating process to determine a corrected direction data range;
acquiring the latest external environment feedback information, and adjusting the corrected direction data range to obtain a preliminary corrected direction reference value;
And finally comparing the initially corrected direction reference value with real-time data to generate the fourth direction estimated value.
For example, in the field of intelligent traffic monitoring, satellite signal intensity and road feature point data acquired in external environment feedback can be primarily processed through a data sorting tool. The data sorting tool has the main functions of cleaning and classifying the collected original information and ensuring the reliability of the data of subsequent analysis. In the urban road environment, the satellite signal intensity is influenced by high-rise shielding, partial data are missing or abnormal, the data sorting tool can preferentially reject the data with the signal intensity lower than a certain threshold value, such as a signal value lower than 30%, and meanwhile coordinate normalization processing is carried out on the road characteristic point data to form a sorted environment data set, so that a foundation is provided for subsequent comparison.
For example, for the application of the data comparison tool, in the deviation analysis of the third direction estimation value and the environment data set, whether the correction is needed or not can be judged through a preset dynamic threshold value. Assuming that the third direction estimated value is 15.0 degrees north-east, and the reference direction in the environment data set is 16.5 degrees north-east, the deviation is 1.5 degrees, if the preset dynamic threshold value is 1.0 degrees, the range is exceeded, and the local data updating flow is required to be started. The comparison method can find potential problems in direction estimation in time and ensure the accuracy of data.
For example, in the local data update process, it is particularly critical to obtain the latest external environment feedback information. Combining the signal intensity fluctuation with the road feature matching data, assuming that the signal intensity is reduced to 40% due to temporary construction interference on an urban arterial road, and the road feature points are displayed as straight roads, the data fusion tool can obtain a preliminary corrected direction reference value by more depending on the road feature matching data according to the weight corrected in the direction of reducing the signal intensity fluctuation, such as being adjusted to be north-east-deviation 16.2 degrees. The method can dynamically adapt to environmental changes and improve the reliability of direction reference.
For example, for the use of information verification tools, it is particularly important to perform final comparison of the initially corrected directional reference values in combination with the environmental adaptation adjustment rules. Assuming that the preliminary correction value is 16.2 degrees in north and the real-time data is 16.0 degrees in north after being updated, the deviation is 0.2 degrees, and the requirement of the preset dynamic threshold value of 0.5 degrees is met, the fourth direction estimated value is 16.0 degrees in north. The checking mechanism can further refine the direction data and ensure that the final result is fit with the actual environment.
For example, from the perspective of the implementation of data fusion tools, the core is the comprehensive evaluation of multi-source data. Under the urban road scene, the satellite signal intensity is assumed to be recovered to 80%, the road characteristic point data is also highly matched with the history record, the fusion tool can synthesize the influence of the satellite signal intensity and the history record, the weight of the signal intensity is properly improved, and finally the direction reference value is adjusted. The multi-dimensional fusion can effectively cope with uncertainty of a single data source, and stability of direction estimation is improved.
For example, in the whole business context, the links are tightly spread around intelligent traffic monitoring, and a complete closed loop processing flow is formed from data arrangement to comparison, updating and verification. In particular in a complex urban environment, the multi-level data processing mode can dynamically adapt to external environment changes, provides reliable support for vehicle direction estimation, and further provides important references for traffic management and path planning.
In step S17, short-term fluctuations of the direction information in the fourth direction estimated value are detected, and if abnormal fluctuations are detected, smoothing correction is performed in combination with the history data to obtain a fifth direction estimated value, including:
Acquiring a history record of the fourth direction estimated value, and carrying out segmentation processing on the distribution condition of the history record in a plurality of time windows by adopting a time sequence decomposition tool to obtain a separated data set of short-term fluctuation and long-term change;
Detecting the short-term fluctuation by using a data comparison tool, and marking the abnormal data as a range to be adjusted if the short-term fluctuation exceeds a preset fluctuation threshold;
Extracting a corresponding direction estimation record from a historical data storage unit according to the range to be adjusted, and carrying out weighting treatment on the range to be adjusted by adopting a data smoothing tool to obtain a primarily corrected direction data set;
And according to the primarily corrected direction data set, a data verification tool is used for carrying out comparison and adjustment in combination with the long-term change, and the fifth direction estimated value is determined.
For example, in the field of intelligent traffic monitoring, the process flow of the history analysis of the fourth direction estimation value can be studied from various angles. First, for the link of acquiring the history from the storage unit, it is assumed that the system stores fourth direction estimated value data generated every minute in the past 24 hours in the urban road scene. These data record the direction change of the vehicle over a particular road segment, such as a fluctuation between 15.0 degrees and 16.5 degrees north-east. When the time series decomposition tool is used for carrying out segmentation processing, the data can be divided into two parts of short-term fluctuation and long-term change. Short term fluctuations may reflect temporary disturbances, such as short jumps in direction values caused by signal occlusion, while long term changes may indicate gradual changes in road layout or environmental conditions.
For example, for the subject of short-term fluctuation detection, it is assumed that the preset threshold range is plus or minus 0.5 degrees when short-term fluctuation is analyzed using the data comparison tool. If the direction value suddenly jumps to 17.0 degrees in north of the east within a certain time window and exceeds the threshold range, the direction value is marked as a range to be adjusted. The detection mode can quickly identify abnormal data and provide basis for subsequent adjustment.
The purpose of the abnormal data is to avoid misleading of the short-term interference to the estimation of the overall direction, and to ensure the continuity and reliability of the data.
For example, when weighting the adjustment range, it is assumed that the record of 5 minutes before and after the extraction of the abnormal data from the history data storage unit, the direction value is found to be stable at about 16.0 degrees north-east before the abnormal point. When the data smoothing tool is adopted, lower weight can be given to the abnormal points, higher weight is given to the peripheral stable data, and finally a primarily corrected direction data set is obtained, for example, the north is adjusted to be 16.1 degrees in the east. The smoothing process can effectively reduce the interference of abnormal data and keep the stability of direction data.
For example, for the final checksum adjustment link, when the data verification tool is used to combine with the long-term variation trend to perform comparison, it is assumed that the long-term variation trend display direction value gradually deviates to the north by 16.2 degrees in the past hours, the preliminary correction value is 16.1 degrees to the north, and the deviation is within the preset fluctuation range of 0.3 degrees, and then the fifth direction estimated value can be determined to be 16.1 degrees to the north. The verification mode combining the long-term trend can ensure that the final result reflects the rationality of short-term correction and accords with the integral change rule.
For example, from the overall process, the links form a complete data processing chain in urban traffic monitoring. Whether short term fluctuations are detected or long term changes are referenced, it is intended to promote the stability of the direction estimation. Particularly in a complex urban environment, signal interference and road characteristic change are frequent, the multi-level processing mode can dynamically adapt to various conditions, powerful support is provided for accuracy of vehicle direction data, and reliable basis is provided for traffic management and path optimization.
In step S18, positioning coordinate data acquired in real time, and the direction information and the position information in the fifth direction estimated value are jointly optimized to generate a fused positioning result, which includes:
Performing preliminary registration on the fifth direction estimated value and the positioning coordinates by using a space geometric mapping tool, and performing calibration processing by using a weighted average tool to determine a preliminarily integrated space information set;
According to the preliminarily integrated space information set, if the deviation between the fifth direction estimated value and the position of the positioning coordinate is detected to exceed a preset deviation threshold value, dynamically adjusting by a deviation correcting tool to obtain an adjusted data set;
And finally integrating the direction estimated value in the adjusted data set and the positioning coordinate, and carrying out smoothing treatment by combining a history record to obtain a fusion positioning result.
For example, in the field of urban traffic monitoring, the integration of vehicle direction and position data can be performed in a number of ways, and specific embodiments thereof can be studied in depth. For the link of acquiring the fifth direction estimated value and the positioning coordinate data acquired in real time from the storage unit, it is assumed that the system stores the direction estimated value updated every minute in the past hour and the corresponding position coordinate data. The direction estimate may be 16.1 degrees north-east, and the position coordinate data records the specific latitude and longitude position of the vehicle at a road segment. When the data alignment tool is adopted for synchronous processing, the system can correspond the data alignment tool and the data alignment tool according to the time stamp, so that each group of data is ensured to be completely consistent in the time dimension, and a direction position data set with consistent time is formed.
For example, for the topic of preliminary registration with a spatial geometric mapping tool, assume that in an urban road scenario, the system spatially preliminary matches the direction estimate with the position coordinates through geometric mapping. If the direction indicates that the vehicle is moving 16.1 degrees north and the position coordinates indicate that the vehicle is on a straight line path on a road segment, the system will attempt to align the two. However, due to sensor errors or environmental disturbances, there may be some deviation, such as a deviation of 0.2 degrees from the actual path. At this time, the calibration process is performed by using a weighted average tool, and the system may assign different weights according to the reliability of the data, for example, assign a higher weight to the positioning coordinates, assign a lower weight to the direction estimation value, and finally determine the preliminarily integrated spatial information set.
For example, in the step of detecting whether the deviation exceeds the preset threshold range and performing the dynamic adjustment, it is assumed that the preset threshold is 0.3 degrees, and the deviation between the direction and the position is found to be 0.4 degrees after the preliminary registration, and the deviation exceeds the threshold range. The system can dynamically adjust through the deviation correcting tool, possibly perform interpolation processing by combining the data of the peripheral time points, and judge the adjusted data set which is more in line with the actual scene, for example, reduce the deviation to 0.1 degree. This approach can effectively address the signal interference or data loss problems common in urban environments.
For example, for the final integration and smoothing combined with the history, assume that the adjusted data set is displayed in a direction of 16.0 degrees north-east, and the position coordinates point to a specific point on a road segment. The system can adopt an information fusion tool to deeply integrate the two, and simultaneously extract the history of the past 10 minutes, and find that the direction value fluctuates between 15.9 degrees and 16.1 degrees in the north of the east. Based on the method, the system performs smoothing treatment on the final result to obtain a fusion positioning result of 16.0 degrees north-east. The processing mode can reduce data jump, ensure continuity of results, and has important significance on stability of vehicle tracks in urban traffic monitoring.
For example, from the overall process, the links form a complete data integration chain in urban road scenarios. Whether time alignment, preliminary registration, or offset calibration and final fusion, it is intended to promote the matching of the direction to the position data. The multi-level processing mode can dynamically adapt to complex conditions in urban environments and provide reliable support for subsequent traffic management and path planning.
In step S19, the parameters and weight allocation rules of the multi-source data standardization model are updated periodically according to the fusion positioning result, and a new model parameter configuration is generated for the multi-source data processing cycle of the next round, including:
Acquiring historical data and real-time feedback data of the fusion positioning result, and performing synchronous processing by adopting a time alignment tool to obtain a data set under a unified time reference;
classifying and summarizing the historical data and the real-time feedback data through a data integration tool according to the data set under the unified time reference;
the classified features are subjected to weight distribution adjustment by adopting a weighted average tool, and weight combinations suitable for current circulation processing are determined;
According to the weight combination, if the detected weight distribution deviates from a preset weight threshold, the weight is finely adjusted through a dynamic optimization tool, and an adjusted weight set meeting the requirement of cyclic processing is judged by combining the characteristics of multi-source data;
and updating parameters of the multisource data standardized model configuration by adopting an information fusion tool through the adjusted weight set, and acquiring configuration parameters applicable to the next round of multisource data processing by combining training results.
For example, in the field of urban traffic monitoring, specific embodiments may be discussed from multiple angles with respect to the processing of fused positioning data, particularly in the integration of historical data with real-time feedback data. For the link of acquiring the fusion positioning historical data and the real-time feedback data from the storage unit, the system is assumed to store the historical data updated every 5 minutes in the past 24 hours, wherein the historical data comprises the direction and the position information of the vehicle on the urban road, and the real-time feedback data is the vehicle state information acquired by the sensor at the current moment. Because there may be time differences between the two types of data, such as a 14:00 historic data record time and a 14:02 real-time data record time, the system may employ a time alignment tool to synchronize the two to a uniform time reference, such as aligning to 14:00 in minutes, to form a uniform data set.
For example, for the subject of classifying and summarizing historical data and real-time feedback data by using a data integration tool, it can be understood in principle that the classifying and summarizing are mainly based on the source and the characteristics of the data, such as classifying the historical data into long-term trend references, and classifying the real-time data into real-time state reflection. Assuming that in an urban road scenario, the historical data reflects that the average speed of the past vehicle on a certain road segment is 40 km/h, while the real-time data shows that the current speed is 35 km/h, the system classifies according to these characteristics and provides basis for subsequent weight distribution.
For example, in a link where weight distribution is adjusted using a weighted average tool, it is assumed that the system distributes weights according to freshness of data, real-time data is given a weight of 0.7 due to its immediacy, and history data is given a weight of 0.3 due to its stability. This weight combination is suitable for the current loop processing, and can balance timeliness and reliability of data. If the weight distribution is detected to deviate from the preset threshold range, for example, the weight of the real-time data exceeds 0.8, the system can be finely adjusted through a dynamic optimization tool, and the weight of the system is properly reduced to 0.75 by combining the characteristics of the multi-source data, for example, the fluctuation of the real-time data is larger, so that an adjusted weight set is formed.
For example, for the topic of updating model configuration parameters by adjusting the set of weights, assume that the system utilizes an information fusion tool to apply the adjusted weights 0.75 and 0.25 to the fusion algorithm, and update the parameters in conjunction with training results, such as model performance data for the past week, to accommodate the next round of multi-source data processing. The method can ensure that the model is continuously optimized in urban traffic monitoring and adapts to dynamically-changed road environments. The final configuration parameters provide references more fitting the actual demands for subsequent processing, and the adaptability of data processing is improved.
In summary, the invention discloses a fusion positioning method of an automatic driving vehicle, which carries out format conversion and consistency verification on satellite signals, sensor data and motion track calculation data through a pre-established multi-source data standardization model, adopts a weighted average fusion algorithm to generate a preliminary direction estimated value, and combines a Bayesian probability model to carry out reliability evaluation and correction. Then, the invention utilizes Kalman filtering algorithm to carry out smoothing treatment and prediction on the direction information, carries out local update according to real-time external environment feedback data, and detects abnormal fluctuation and corrects through multi-time window trend analysis. Finally, the optimized direction information and the position information are combined and optimized to generate a final fusion positioning result, and model parameters are updated regularly through a machine learning method, so that high-precision, high-reliability and high-adaptability positioning under a complex scene are realized.
Referring to fig. 2, a second embodiment of the present invention provides a structure diagram of a fusion positioning system for an autonomous vehicle, comprising:
a first obtaining module 201, configured to perform format conversion on original data through a pre-established multi-source data standardization model, and generate a first data set;
The second obtaining module 202 is configured to screen heterogeneous data in the first data set, and if a deviation value of one of the heterogeneous data exceeds a preset deviation threshold, reject a corresponding record to obtain a second data set;
A third obtaining module 203, configured to perform weighted calculation on direction information of different data in the second data set, and generate a first direction estimated value;
A fourth obtaining module 204, configured to perform reliability evaluation on the first direction estimated value, and if the reliability is lower than a preset evaluation threshold, trigger a data correction procedure to obtain a second direction estimated value;
A fifth obtaining module 205, configured to perform smoothing and prediction on the direction information in the second direction estimated value, and generate a third direction estimated value;
A sixth obtaining module 206, configured to obtain feedback data of the real-time external environment, and if a deviation between the feedback data and the third direction estimated value exceeds a preset dynamic threshold, generate a fourth direction estimated value;
a seventh obtaining module 207, configured to detect short-term fluctuations of the direction information in the fourth direction estimated value, and if abnormal fluctuations are detected, perform smoothing correction in combination with the historical data to obtain a fifth direction estimated value;
an eighth obtaining module 208, configured to obtain positioning coordinate data in real time, perform joint optimization on direction information and position information in the fifth direction estimated value, and generate a fused positioning result;
And a configuration module 209, configured to periodically update parameters and weight allocation rules of the multi-source data standardization model according to the fusion positioning result, and generate a new model parameter configuration for a multi-source data processing cycle of a next round.
It should be noted that, the fusion positioning system for an automatic driving vehicle provided by the embodiment of the present invention is used for executing all the flow steps of the fusion positioning method for an automatic driving vehicle in the foregoing embodiment, and the working principles and beneficial effects of the two correspond one to one, so that the description is omitted.
It should be noted that the above-described apparatus embodiments are merely illustrative, and the units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the embodiment of the device provided by the invention, the connection relation between the modules represents that the modules have communication connection, and can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
The foregoing embodiments have been provided for the purpose of illustrating the general principles of the present invention, and are not to be construed as limiting the scope of the invention. It should be noted that any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without departing from the spirit and principles of the present invention are intended to be included in the scope of the present invention.