Unmanned aerial vehicle precision positioning method and system based on multi-source data fusion

Abstract

The application relates to the technical field of unmanned aerial vehicle positioning, in particular to an unmanned aerial vehicle accurate positioning method and system based on multi-source data fusion, which comprises the following steps: acquiring multiple-dimension measurement data collected by different sensors in the flight process of an unmanned aerial vehicle; constructing an error marker sequence of each dimension according to the fluctuation characteristics of the measurement data of each dimension; determining the historical credibility of each dimension according to the distribution characteristics of the error markers in the error marker sequence; determining the abnormal correlation between any two dimensions according to the error markers of each dimension within a preset historical period, and determining the final error corresponding to each sampling moment; performing weighted filtering processing on the measurement data of all dimensions within a preset filtering window according to the final error corresponding to each sampling moment, fusing the filtered measurement data of all dimensions, and outputting the positioning result of the unmanned aerial vehicle. The application improves the positioning accuracy of the unmanned aerial vehicle by identifying the cooperative anomaly of multi-dimension data.

Description

Technical Field

[0001] This invention relates to the field of UAV positioning technology, specifically to a method and system for precise UAV positioning based on multi-source data fusion. Background Technology

[0002] The widespread application of drones in agricultural surveying, disaster monitoring, and logistics delivery places extremely high demands on their flight positioning accuracy. To achieve precise positioning, existing technologies typically employ multi-source sensor data fusion, integrating measurement data from heterogeneous sensors such as the Global Positioning System (GPS), Inertial Measurement Unit (IMU), and lidar (e.g., Light Detection and Ranging, LiDAR) to improve the reliability and accuracy of the positioning system.

[0003] However, in complex real-world environments, multiple sensors may simultaneously malfunction due to common external interference or faults. Existing methods struggle to effectively identify coordinated anomalies, resulting in insufficient ability to identify and compensate for anomalous data, which in turn affects the overall positioning accuracy and robustness of UAVs under varying conditions. Summary of the Invention

[0004] To address the technical problem of poor positioning accuracy caused by insufficient identification of anomalies in multi-dimensional data collaboration, the present invention aims to provide a precise positioning method and system for unmanned aerial vehicles (UAVs) based on multi-source data fusion. The specific technical solution adopted is as follows:

[0005] Firstly, a method for precise positioning of unmanned aerial vehicles (UAVs) based on multi-source data fusion is provided. This method includes: acquiring measurement data from multiple dimensions collected by different sensors during the UAV's flight; for each dimension, determining the initial error of each dimension at each sampling time based on the fluctuation characteristics of the measurement data within a preset historical period before each sampling time, and constructing an error label sequence for each dimension based on the initial error, the error label sequence being used to characterize the stability of the measurement data for each dimension at each sampling time; determining the historical reliability of each dimension based on the distribution characteristics of the error labels in the error label sequence of each dimension, the historical reliability being used to characterize the long-term reliability of the measurement data for the corresponding dimension; for each At each sampling time, based on the error markers in the error marker sequence of each dimension within a preset historical time period, the abnormal correlation between any two dimensions is determined. Then, based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions, the final error corresponding to each sampling time is determined. The abnormal correlation characterizes the probability that measurement data of different dimensions will exhibit anomalies at the same sampling time due to a common cause. The number of abnormal dimensions is the number of error markers used to characterize the instability of measurement data at each sampling time. Based on the final error corresponding to each sampling time, the measurement data of all dimensions within a preset filtering window are weighted and filtered. The filtered measurement data of all dimensions are then fused to output the UAV's positioning result.

[0006] In one possible design, for each of the multiple dimensions, the initial error of each dimension at each sampling time is determined based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period prior to each sampling time. This includes: acquiring the measurement data sequence of each dimension within the preset historical period; identifying the maximum data points in the measurement data sequence; determining the extreme value change rate based on the numerical difference and time interval between adjacent maximum data points in the time series; and determining the initial error of each dimension at each sampling time based on the number density of maximum data points and the maximum value of the extreme value change rate within the preset historical period.

[0007] In one possible design, an error label sequence for each dimension is constructed based on the initial error, including: for each dimension, comparing the initial error of each dimension at each sampling time with a preset error value; if the initial error at the target sampling time is greater than or equal to the preset error value, determining the error label corresponding to the target sampling time as the first error label, which is used to characterize the instability of the measurement data of each dimension at the target sampling time; the target sampling time is any sampling time during the flight of the UAV; if the initial error at the target sampling time is less than the preset error value, determining the error label corresponding to the target sampling time as the second error label, which is used to characterize the stability of the measurement data of each dimension at the target sampling time; and constructing an error label sequence for each dimension based on the error label corresponding to each sampling time.

[0008] In one possible design, the historical reliability of each dimension is determined based on the distribution characteristics of error markers in the error marker sequence of each dimension, including: for each dimension, determining the stable marker segment in the error marker sequence of each dimension, the stable marker segment consisting of continuous error markers used to characterize the stability of the measurement data; determining the maximum value among the average values ​​of the initial errors of all stable marker segments in each dimension as the maximum average value; and determining the historical reliability of each dimension based on the longest segment length, the total number of stable marker segments, and the maximum average value among all stable marker segments in each dimension.

[0009] In one possible design, for each sampling time, based on the error markers in the error marker sequence of each dimension within a preset historical period, the abnormal correlation between any two dimensions is determined, including: for each sampling time, based on the error markers in the error marker sequence of each dimension within a preset historical period, determining the abnormal time set for each dimension, the abnormal time set consisting of sampling times corresponding to error markers representing instability of measurement data within the preset historical period; for any two dimensions, determining the intersection of the abnormal time sets of the two dimensions as the target time set; using a preset sequence similarity measurement algorithm, determining the degree of difference between the measurement data sequences composed of measurement data at each sampling time of the two dimensions within the target time set; and determining the abnormal correlation between the two dimensions based on the number of times included in the target time set and the degree of difference.

[0010] In one possible design, the final error corresponding to each sampling moment is determined based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions. This includes: for each sampling moment, determining all dimensions whose error labels represent the instability of the measurement data, forming an abnormal dimension set for that sampling moment; determining the correlation factor for each sampling moment based on the abnormal correlation between any two dimensions in the abnormal dimension set and the historical reliability of each dimension in the abnormal dimension set; constructing an abnormal dimension number sequence based on the number of abnormal dimensions at each sampling moment within a preset historical period; determining the trend of the number of abnormal dimensions at each sampling moment based on the abnormal dimension number sequence; and determining the final error corresponding to each sampling moment based on the correlation factor, the number of abnormal dimensions, and the trend.

[0011] In one possible design, the measurement data of all dimensions within a preset filtering window are weighted and filtered based on the final error corresponding to each sampling time. This includes: determining the time weight corresponding to each sampling time based on the final error corresponding to each sampling time, where the time weight is negatively correlated with the final error; and for each dimension, calculating a weighted average of the measurement data at each sampling time within the preset filtering window, based on the time weight corresponding to each sampling time within the preset filtering window ending at the current sampling time, to obtain the filtering result of each dimension at the current sampling time.

[0012] In one possible design, the UAV's localization result is output by fusing and filtering measurement data from all dimensions. This includes: combining the filtered results of all dimensions at the current sampling time into an observation vector at the current sampling time; determining the noise covariance of the observation vector based on the time weight corresponding to the current sampling time; predicting the state at the current sampling time based on the UAV's motion model and the state estimate from the previous sampling time; fusing the observation vector, noise covariance, state prediction value, and prediction covariance using a preset state estimation algorithm to obtain a state estimate at the current sampling time, which includes the UAV's position information; and extracting the position information from the state estimate as the UAV's localization result at the current sampling time.

[0013] In one possible design, multiple dimensions of measurement data are acquired from different sensors during the flight of the UAV. This includes: different sensors carried by the UAV, including a GPS sensor, an inertial measurement unit, and a lidar, with each sensor connected to the UAV's flight control system; the synchronous trigger signal output by the UAV's flight control system controls each sensor to simultaneously start data acquisition; and the raw data acquired by each sensor is converted into standardized data in a unified format to obtain multiple dimensions of measurement data.

[0014] Secondly, a UAV precise positioning system based on multi-source data fusion is provided, comprising: a data acquisition unit for acquiring measurement data of multiple dimensions collected by different sensors during UAV flight; an error analysis unit for determining the initial error of each dimension at each sampling time based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period prior to each sampling time, and constructing an error label sequence for each dimension based on the initial error, the error label sequence being used to characterize the stability of the measurement data of each dimension at each sampling time; and a reliability determination unit for determining the historical reliability of each dimension based on the distribution characteristics of the error labels in the error label sequence of each dimension, the historical reliability being used to characterize the long-term reliability of the measurement data of the corresponding dimension. The error determination unit, for each sampling time, determines the anomaly correlation between any two dimensions based on the error markers in the error marker sequence of each dimension within a preset historical time period. It then determines the final error corresponding to each sampling time based on the historical reliability of each dimension, the anomaly correlation between any two dimensions, and the number of anomaly dimensions. The anomaly correlation characterizes the probability that measurement data from different dimensions will exhibit anomalies at the same sampling time due to a common cause. The number of anomaly dimensions represents the number of error markers used to characterize measurement data instability at each sampling time. The fusion positioning unit, based on the final error corresponding to each sampling time, performs weighted filtering on the measurement data of all dimensions within a preset filtering window, fuses the filtered measurement data of all dimensions, and outputs the UAV's positioning result.

[0015] The present invention has the following beneficial effects:

[0016] The UAV precise positioning method based on multi-source data fusion provided in this invention comprehensively quantifies the integrated error at each sampling moment by analyzing the fluctuation characteristics, long-term reliability, and anomaly correlation of data from various dimensions. Then, based on the error, the filtering weights are dynamically adjusted and multi-source data is fused, effectively solving the problem of insufficient consideration of data stability and anomaly correlation in traditional multi-source data fusion positioning. This method ensures full utilization of multi-source data while mitigating the impact of environmental interference and sensor errors on positioning through precise error quantification and dynamic filtering strategies. This allows the positioning results to accurately adapt to the complex state changes during UAV flight, meeting the stringent positioning accuracy requirements of applications such as precision agriculture. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a UAV precise positioning system based on multi-source data fusion, provided in one embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating a method for precise positioning of unmanned aerial vehicles (UAVs) based on multi-source data fusion, as provided in one embodiment of the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a UAV precise positioning method and system based on multi-source data fusion proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] In embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] The following description, in conjunction with the accompanying drawings, details a specific scheme for a UAV precise positioning method and system based on multi-source data fusion provided by the present invention.

[0025] Please see Figure 1 This illustrates a schematic diagram of a UAV precise positioning system based on multi-source data fusion provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the UAV precise positioning system 10 based on multi-source data fusion includes a data acquisition unit 11, an error analysis unit 12, a credibility determination unit 13, an error determination unit 14, and a fusion positioning unit 15.

[0026] The data acquisition unit 11 is used to collect and synchronize raw data from different sensors on the UAV.

[0027] Optionally, the data acquisition unit 11 controls the simultaneous start-up of sensors such as GPS, IMU, and LiDAR through the synchronous trigger signal output by the UAV flight control system, ensuring strict alignment of timestamps for multi-dimensional data. Subsequently, the data acquisition unit 11 converts the raw data from each sensor into standardized data in a unified format, generating a synchronous multi-dimensional measurement data stream for subsequent processing.

[0028] Error analysis unit 12 is used to determine the initial error of each dimension at each sampling time and to construct an error label sequence.

[0029] The error analysis unit 12 includes an initial error calculation module and an error label sequence construction module.

[0030] The initial error calculation module is used to extract the measurement data of each dimension within a preset historical period before each sampling time for each dimension, forming a corresponding measurement data sequence; and by analyzing the distribution characteristics of the maximum value data points in the measurement data sequence (such as the ratio of the number of maximum values ​​to the duration of the preset historical period) and the changes in the maximum values ​​(such as the correlation between the numerical differences of adjacent maximum values ​​on the time series and the time interval), the rate of change of the extreme values ​​is determined. Then, by combining the number density of the maximum value data points and the maximum value of the rate of change of the extreme values, the initial error of each dimension at each sampling time is quantified.

[0031] The error label sequence construction module is used to compare the initial error at each sampling time with the preset error value. If the initial error is greater than or equal to the preset error value, it is marked as the first error label (indicating that the measurement data is unstable, such as represented by 1). If it is less than the preset error value, it is marked as the second error label (indicating that the measurement data is stable, such as represented by 0). All labels are arranged in chronological order of sampling time to form the error label sequence for each dimension.

[0032] It should be noted that the preset historical time period corresponding to each sampling moment is a time period formed by a preset number of consecutive sampling moments ending at that sampling moment. The preset number can be, for example, 50, 60, 70, etc., and the embodiments of the present invention do not specifically limit this.

[0033] The credibility determination unit 13 is used to determine the historical credibility of each dimension by analyzing the distribution characteristics of error markers in the error marker sequence of each dimension.

[0034] The credibility determination unit 13 includes a stable marker segment identification module and a credibility calculation module.

[0035] The stable marker segment identification module is used to identify stable marker segments consisting of consecutive second error markers for each dimension of the error marker sequence, and to determine the length of each stable marker segment and the sampling time it covers.

[0036] The reliability calculation module calculates the average initial error across all sampling times within each stable marker segment and selects the maximum average from all averages. Further, by combining the longest stable marker segment length, the total number of stable marker segments, and the maximum average, the historical reliability for each dimension is obtained. This value directly reflects the long-term reliability of the measurement data for the corresponding dimension.

[0037] The error determination unit 14 is used to dynamically calculate the final error at each sampling time by comprehensively utilizing the error label sequence and historical reliability.

[0038] The error determination unit 14 includes an anomaly correlation analysis module and a final error calculation module.

[0039] The anomaly correlation analysis module is used to extract sampling times marked as the first error mark within a preset historical time period based on the error mark sequence of each dimension, forming an anomaly time set for each dimension. Further, for any two dimensions, the intersection of their anomaly time sets is taken as the target time set. A preset sequence similarity measurement algorithm is used to calculate the degree of difference between the measurement data sequences of the two dimensions within the target time set. Combining the number of times in the target time set with the degree of difference, the anomaly correlation between the two dimensions is determined (characterizing the probability that the anomaly at the same sampling time is caused by a common trigger).

[0040] The final error calculation module first filters out the dimensions that characterize the instability of the measurement data based on the error markers of each dimension at each sampling time, forming an abnormal dimension set. Then, it combines the abnormal correlation between any two dimensions in the abnormal dimension set with the historical reliability of each dimension, and determines the correlation factor for that sampling time through preset integration rules. Further, it constructs an abnormal dimension quantity sequence based on the number of abnormal dimensions at each sampling time within a preset historical period, analyzes this sequence to obtain the changing trend of the number of abnormal dimensions at each sampling time, and then combines the correlation factor, the number of abnormal dimensions, and the changing trend to quantify the final error at each sampling time.

[0041] The fusion positioning unit 15 is used to guide data fusion by using the final error determined by the error determination unit 14 to achieve accurate positioning.

[0042] The fusion positioning unit 15 includes a weighted filtering processing module and a fusion positioning module.

[0043] The weighted filtering module uses the final error at each sampling time as the basis for calculation to determine the time weight that is negatively correlated with the final error (the smaller the final error, the larger the time weight). Further, for each dimension, a preset filtering window is defined with the current sampling time as the end time. The time weight of each sampling time within the window is used to perform a weighted average calculation on the corresponding measurement data to obtain the filtering result for each dimension at the current sampling time.

[0044] The fusion positioning module first combines the filtering results of all dimensions into an observation vector at the current sampling time, and determines the noise covariance of the observation vector according to the time weight. Then, it makes a prediction based on the UAV's motion model and the state estimate at the previous sampling time to obtain the predicted state value and prediction covariance at the current sampling time. Finally, the observation vector, noise covariance, predicted state value, and prediction covariance are input into a preset state estimation algorithm for data fusion to obtain a state estimate containing the UAV's position information. The position information is then extracted from the state estimate as the output of the UAV's accurate positioning result at the current sampling time.

[0045] Please see Figure 2 The diagram illustrates a flowchart of a UAV precise positioning method based on multi-source data fusion provided by an embodiment of the present invention, including steps S201-S207.

[0046] S201. Acquire measurement data from multiple dimensions collected by different sensors during the flight of the UAV.

[0047] In some embodiments, the UAV is equipped with multiple heterogeneous positioning sensors, including at least a GPS sensor, an IMU sensor, and a lidar, all of which are connected to the UAV's flight control system. During flight, the flight control system generates and outputs a high-precision synchronization trigger signal. This synchronization trigger signal is simultaneously sent to the GPS sensor, IMU sensor, and lidar, controlling them to synchronously initiate data acquisition for each frame, obtaining multi-dimensional measurement data, including GPS measurement data, IMU measurement data, and lidar point cloud data, thereby eliminating time deviations introduced by differences in sensor activation timing at the source.

[0048] Because the raw data collected by various sensors differ in format—for example, GPS sensors output data in the National Marine Electronics Association (NMEA) protocol format, IMU sensors output binary data, and lidar outputs point cloud data—the raw data needs to be standardized to facilitate subsequent fusion processing. This can be done by converting it to a unified decimal floating-point standardized data format. The standardization process follows the formula:

[0049] ;

[0050] In the formula, Standardized data corresponding to the original data in a certain dimension. This refers to the raw data values ​​for this dimension, i.e., the unprocessed data directly collected by the sensor. This is the preset maximum value of the raw data for this dimension, that is, the maximum possible value of the data for this dimension that is preset according to the sensor's measurement range. This is the preset minimum value for the raw data in this dimension, that is, the minimum possible value of the data in this dimension that is preset according to the measurement range of the sensor. Through the standardization process of the above formula, the numerical range of the raw data in different dimensions can be uniformly mapped to the interval of 0 to 1, eliminating the interference of data magnitude differences on subsequent data processing.

[0051] It should be noted that, due to as well as The measurement range of the sensor is preset, therefore the raw data collected by the sensor is... It must be within this range, thus the raw data collected by the sensor can be standardized using the above formula.

[0052] In some embodiments, each sampled and standardized dimension of data is accompanied by a corresponding timestamp. The timestamp is generated by a high-precision clock, and its value corresponds one-to-one with the sampling time, used to identify the acquisition time of the data in that dimension. This ensures that when extracting measurement data sequences within a preset historical time period, all data within the same time period can be accurately filtered out. Ultimately, at each sampling time, standardized GPS positioning data for the GPS sensor, standardized inertial measurement data for the IMU sensor, and standardized laser ranging data for the LiDAR sensor can be acquired. These three data together constitute the multi-dimensional measurement data at that sampling time. The measurement data from all sampling times are arranged in chronological order of their timestamps, forming a multi-dimensional measurement data sequence.

[0053] S202. For each of the multiple dimensions, determine the initial error of each dimension at each sampling time based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period before each sampling time.

[0054] As one possible implementation, in terms of the third dimension Taking one dimension as an example, at each sampling time The preset historical time period includes the current sampling time. and the current sampling time Previous The time period of consecutive sampling moments, i.e., the preset historical time period is [ ], This is a pre-set fixed value that can be adjusted according to real-time positioning requirements. For example, it can be set to 60, and this time period is used to extract the current sampling time. Previous historical measurement data was used to analyze the fluctuation characteristics of the measurement data.

[0055] Among them, for the current sampling time and the For each dimension, the following steps are performed to determine its initial error. .

[0056] The first step is to extract the first historical time period based on the above-mentioned preset historical time period. All measurement data for each dimension within that time period are sorted according to the chronological order of sampling time to form a measurement data sequence. .

[0057] The second step is to identify all the maximum data points in the measurement data sequence. For any given data point... If its value is greater than its previous sampling time With the next sampling time If a data value is found to be a maximum value, then it is determined to be a maximum value data point.

[0058] The third step is to determine the rate of change of extreme values ​​based on the numerical differences and time intervals between adjacent maximum data points in the time series. Assume that extreme values ​​are identified within a preset historical time period. There are 1000 maximum data points, and their corresponding time indices are: For any two adjacent maxima, their values ​​are respectively and The time interval between the two is This allows us to obtain the rate of change of the extreme value. .

[0059] In the formula, For the measurement data sequence of the first The rate of change of each extreme value corresponds to a maximum value. For the measurement data sequence of the first The value of the maximum value, For the measurement data sequence of the first The value of the maximum value, For the measurement data sequence of the first The maximum value and the first The time interval between sampling times corresponding to each maximum value.

[0060] The fourth step is to analyze all the calculated rates of change of extreme values. , … Find the maximum value in the middle. This value represents the most dramatic instantaneous fluctuation of this dimension within the current analysis period.

[0061] The fifth step is to determine the number of maximum data points within a preset historical time period. and the number of times included in the preset historical period. The number density of maximum data points within the current preset historical time period is calculated. This value reflects the frequency of data fluctuations.

[0062] Step 6: Taking into account the frequency of data fluctuations and the most dramatic instantaneous fluctuations within a preset historical period, the sixth step is to calculate the first step using a normalization function. Each dimension at the sampling time initial error .

[0063] In some embodiments, the initial error is calculated using the following formula:

[0064] ;

[0065] In the formula, For the first Each dimension at the sampling time The initial error, To preset the number of maximum data points within a historical time period, To preset the number of moments included in the historical period, For the first The maximum rate of change of extreme values ​​for each dimension within the preset historical period. This indicates that normalization processing is being performed, for example, maximum-minimum normalization can be used. Using the historical maximum and minimum values ​​as benchmarks, if the currently calculated value is greater than the historical maximum, then the initial error is... The value is set to 1. If the currently calculated value is less than the historical minimum value, then the initial error is... The value is 0.

[0066] Among them, the Each dimension at the sampling time Within the corresponding preset historical time period, if the number density of maximum value data points The larger the value, the greater the rate of change of the extreme value. The larger the value, the higher the dimension at the sampling time. The more frequent and severe the fluctuations, the more likely the drone's positioning accuracy will be reduced, resulting in larger errors.

[0067] Furthermore, based on the above steps, the initial error of each dimension at each sampling time can be obtained.

[0068] Understandably, in this embodiment of the invention, by extracting the maximum value data points and calculating their rate of change and density, it is possible to keenly capture the violent fluctuations and frequent jumps representing instability in the sensor measurement data. Quantifying the maximum value of the extreme value change rate by combining it with the density allows the calculated initial error to simultaneously reflect the severity and frequency of data fluctuations, providing a precise and objective numerical basis for accurately judging the stability of each dimension at each moment.

[0069] S203. Construct an error label sequence for each dimension based on the initial error.

[0070] The error label sequence is used to characterize the stability of the measurement data for each dimension at each sampling time.

[0071] In some embodiments, a preset error value is first set. This preset error value is a critical quantization value used to distinguish between stable and unstable measurement data. Its value can be pre-calibrated based on the performance parameters of the sensors carried by the UAV, the accuracy requirements of the positioning task, and the characteristics of the actual application scenario. For example, it can be 0.8. This embodiment of the invention does not specifically limit this value.

[0072] As one possible implementation, for each dimension, the initial error of each dimension at each sampling time is compared with a preset error value.

[0073] If the initial error at the target sampling time is greater than or equal to a preset error value, the error marker corresponding to the target sampling time is determined as the first error marker. The first error marker is used to characterize the instability of the measurement data of each dimension at the target sampling time. The target sampling time is any sampling time during the flight of the UAV.

[0074] If the initial error at the target sampling time is less than the preset error value, the error marker corresponding to the target sampling time is determined as the second error marker, wherein the second error marker is used to characterize the stability of the measurement data of each dimension at the target sampling time;

[0075] Furthermore, based on the error markers corresponding to each sampling time, an error marker sequence for each dimension is constructed.

[0076] In some embodiments, the first error marker can be represented by 1, and the second error marker can be represented by 0, thereby obtaining an error marker sequence for each dimension from the start time to the current sampling time. In the error marker sequence, consecutive 0s indicate relatively stable data with minimal variation, meaning the positioning information at the corresponding sampling time is relatively accurate, while consecutive 1s indicate unstable data with significant fluctuations, requiring further analysis or adjustment.

[0077] S204. Determine the historical reliability of each dimension based on the distribution characteristics of error markers in the error marker sequence of each dimension.

[0078] Historical reliability is used to characterize the long-term reliability of measurement data for the corresponding dimension.

[0079] As one possible approach, the historical credibility of each dimension can be determined by following these steps.

[0080] The first step is to determine the stable marker segment in the error marker sequence corresponding to the dimension, which consists of consecutive error markers (i.e. consecutive second error markers with a value of 0) used to characterize the stability of the measurement data. Each stable marker segment reflects the state in which the sensor maintains stable data over a continuous period of time in that dimension.

[0081] The second step is to extract the distribution features of the stable labeled segments. These distribution features include:

[0082] The longest segment length among all stable marker segments. This distribution characteristic is the length of the stable marker segment containing the most consecutive 0 markers among all stable marker segments, that is, the number of 0 markers contained in the stable marker segment. This value characterizes the sensor's optimal ability to maintain stable operation for a long time.

[0083] The total number of stable marker segments; this distribution characteristic is the number of all stable marker segments. This value reflects the frequency of interruptions in the steady state of the measurement data; a higher number indicates that the steady state of the measurement data in this dimension is more easily broken.

[0084] The maximum value among the average initial errors of all stable labeled segments is determined by first calculating the average initial error across all sampling times within each stable labeled segment, which reflects the average error level of the data within that stable labeled segment. Then, the maximum value is found among the average initial errors of all stable labeled segments as the maximum average value, which characterizes the upper limit of potential error that may exist in this dimension of the data even within stable periods.

[0085] The third step is to determine the historical reliability of each dimension based on the longest segment length, the total number of stable segments, and the maximum average value among all stable segments in each dimension.

[0086] In some embodiments, the formula for calculating the historical credibility of each dimension is as follows:

[0087] ;

[0088] In the formula, For the first Historical credibility in several dimensions For the first The longest segment length among all stable labeled segments (0 sequence segments) in the error labeled sequence of each dimension. For the first The number of all stable labeled segments (0 sequence segments) in the error labeled sequence of each dimension. For the first The maximum value among the average initial errors of all stable labeled segments (0 sequence segments) in the error labeled sequence of each dimension. For the natural constant An exponential function with base 0.

[0089] in, As a product factor, it directly contributes to historical credibility; the longer the longest segment, the higher the historical credibility value. An error decay factor is formed, in which the more stable marker segments there are (the more easily the stable state of the measurement data in this dimension is broken), or the higher the maximum average value (the higher the average error level of the stable marker segments), the smaller the error decay factor is. The error decay factor plays a moderating role in historical reliability. The more easily the stable state of the measurement data is broken or the higher the average error level of the stable marker segments, the lower the final historical reliability assessment will be.

[0090] Furthermore, based on the above steps, the historical reliability of each dimension can be obtained. The higher the historical reliability value, the better the long-term stability of the sensor in that dimension, the fewer times it has been disturbed, and the higher the data quality during stable periods. Therefore, it should be given higher weight and trust in subsequent fusion decisions.

[0091] Understandably, in this embodiment of the invention, by identifying continuous stable state segments and combining the length of the longest stable segment, the total number of stable segments, and the average error level within each stable segment, a comprehensive long-term performance evaluation of a sensor can be achieved from historical data. This not only considers whether the sensor can maintain stability for a long period but also the frequency of disruptions to its stability and the data quality during stable periods. The resulting historical reliability is a composite index integrating persistence, robustness, and accuracy, providing important historical context and weighted references for judging the severity of current abnormal data, and distinguishing between high-performance sensors that occasionally malfunction and sensors that are unreliable over the long term.

[0092] S205. For each sampling time, based on the error markers in the error marker sequence of each dimension within the preset historical time period, determine the abnormal correlation between any two dimensions.

[0093] Among them, anomaly correlation is used to characterize the probability that measurement data of different dimensions will show anomalies due to common causes at the same sampling time.

[0094] As one possible implementation, for each sampling time The abnormal correlation between any two dimensions can be determined by following these steps.

[0095] The first step is to determine the set of anomalous moments for each dimension based on the error markers in the error marker sequence within a preset historical time period. Specifically, for each dimension, all sampling moments marked with the first error marker within the preset historical time period are extracted to form the set of anomalous moments for that dimension.

[0096] The second step is to consider any two dimensions (in terms of dimensions). and dimensions For example, the intersection of the sets of anomalous moments in two dimensions is determined as the target set of moments, and the number of moments contained in this target set is determined, denoted as . This target time set contains data within a preset historical time period, with dimensions... and dimensions All sampling moments that are simultaneously deemed to have unstable measurement data.

[0097] The third step is to extract outlier data sequences. This involves extracting outlier data sequences from various dimensions. and dimensions From the measurement data, extract the measurement data values ​​corresponding to each sampling time in the target time set, and arrange them in time sequence to obtain the dimension. Abnormal data sequences and dimensions Abnormal data sequences .

[0098] Step 4: Determine the dimensions and dimensions The degree of difference between anomalous data sequences.

[0099] Optionally, a pre-defined sequence similarity measurement algorithm (such as Dynamic Time Warping (DTW)) can be used. This algorithm finds the optimal nonlinear alignment path between two sequences and calculates their cumulative distance to obtain a representation of two anomalous data sequences. and Distance values ​​of morphological differences That is, dimension and dimensions The degree of difference between anomalous data sequences is the measure of the difference between them. The smaller the value, the more similar the fluctuation patterns of the two sequences are; conversely, the greater the difference in patterns, the more similar the fluctuation patterns are.

[0100] It should be noted that the preset sequence similarity measurement algorithm used in the embodiments of the present invention can be, in addition to the DTW algorithm, the correlation coefficient calculation method (such as the Pearson coefficient, the closer its value is to 1, the stronger the positive linear correlation, and the consistent fluctuation trend of the two abnormal data sequences), the mutual information calculation method, or the distance calculation method based on sequence feature extraction (such as Euclidean distance, Manhattan distance), etc. The embodiments of the present invention do not make specific limitations on this.

[0101] The fifth step is to determine the abnormal correlation between the two dimensions based on the number of times included in the target time set and the degree of difference.

[0102] In some embodiments, the formula for determining the abnormal correlation between two dimensions is as follows:

[0103] ;

[0104] In the formula, For dimension and dimensions Abnormal correlation, The number of times contained in the target time set. For dimension and dimensions The degree of difference between anomalous data sequences It is a very small positive number, for example, it can take the value 0.001, to prevent the denominator from being zero.

[0105] Among them, when two dimensions (dimensions) and dimensions The more times an anomaly occurs simultaneously ( The larger the value, the more similar the fluctuation patterns of their measurement data at these times ( The smaller the value, the higher the calculated abnormal correlation. The larger the value, the more likely the anomalies in the two dimensions are caused by the same common cause (such as strong electromagnetic interference or violent vibration), indicating a high degree of synergy rather than independent occurrence.

[0106] It should be noted that, in one possible scenario, there are no anomalies causing synchronization between the two dimensions. A value of 0 indicates a weak correlation between the two dimensions, meaning there's no need to further calculate the difference between them; the abnormal correlation between the two dimensions can be directly considered. Setting it to 0 improves computational efficiency.

[0107] Furthermore, based on the above steps, we can obtain the abnormal correlation between any two dimensions at any sampling time.

[0108] Understandably, in this embodiment of the invention, synchronization is confirmed by finding the intersection of abnormal moments, and then the similarity is verified by comparing the data sequence patterns at the abnormal moments. This dual verification mechanism greatly improves the accuracy of correlation judgment. Accurate quantification of abnormal correlation is crucial for subsequent judgment of whether systemic coordinated anomalies exist, enabling a more sensitive response and stronger compensation for coordinated anomalies caused by common triggers (such as strong electromagnetic interference) that pose greater harm.

[0109] S206. Based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions, determine the final error corresponding to each sampling time.

[0110] The number of outlier dimensions refers to the number of error markers used to characterize the instability of the measurement data at each sampling time.

[0111] As one possible implementation, for each sampling time, the final error corresponding to each sampling time can be determined according to the following steps.

[0112] The first step is to determine, based on the error markers of each dimension, all dimensions that represent the instability of the measurement data and form the set of abnormal dimensions for that sampling time.

[0113] For each sampling time Iterate through the error markers of all dimensions and filter out those at the sampling time. Error markers characterize the dimensions where the measurement data is unstable (error marker is the first error marker "1"). The selected dimensions are then used as the dimensions for that sampling time. The set of abnormal dimensions is determined, and the number of dimensions contained in this set of abnormal dimensions is denoted as . .

[0114] The second step is to determine the correlation factor at each sampling time based on the abnormal correlation between any two dimensions in the abnormal dimension set and the historical reliability of each dimension in the abnormal dimension set.

[0115] In some embodiments, the sampling time is determined. The formula for calculating the correlation factor is as follows:

[0116] ;

[0117] In the formula, Sampling time The correlation factors, The number of dimensions contained in the abnormal dimension set. For abnormal dimensions with abnormal dimensions The abnormal correlation between them For abnormal dimensions Historical credibility and anomaly dimensions The historical average credibility, For the normalization function, the maximum-minimum normalization method can be used, where the maximum and minimum values ​​are preset empirical extreme values ​​derived from a large amount of historical calculation data. If the calculation result exceeds the [0,1] interval, it is restricted to the [0,1] range by a truncation function (i.e., if the result is less than 0, it is taken as 0, and if it is greater than 1, it is taken as 1) to eliminate the influence of outliers on the evaluation index.

[0118] Among them, correlation factors The larger the value, the faster the sampling time. Not only are there multiple anomalous dimensions, but these anomalous dimensions also exhibit strong synergy, and the historical reliability of the sensors involved is high, suggesting that there may be a common trigger that caused the anomalousness in multiple dimensions.

[0119] It should be noted that if the sampling time The number of abnormal dimensions included in the abnormal dimension set Less than 2, that is, the sampling time If there are no outlier dimensions or only one outlier dimension, the above formula for determining the correlation factor is no longer used; instead, the sampling time is directly used. Correlation factors The value is 0.

[0120] The third step is to construct an anomaly dimension count sequence based on the number of anomaly dimensions at each sampling time within a preset historical time period. Then, based on this anomaly dimension count sequence, the changing trend of the number of anomaly dimensions at each sampling time is determined.

[0121] Specifically, based on the number of outlier dimensions at each sampling moment within a preset historical time period, the outlier dimension count sequence is obtained by sorting the samples chronologically. A linear fitting method is then used to calculate the trend representing the rate of change in the number of outlier dimensions at each sampling moment; this trend is represented by the slope corresponding to each sampling moment. A larger absolute value of this trend indicates a faster increase or decrease in the number of outlier dimensions.

[0122] The fourth step is to determine the final error corresponding to each sampling time based on the correlation factors, the number of outlier dimensions, and the trend of change.

[0123] In some embodiments, the sampling time is determined. The formula for calculating the corresponding final error is as follows:

[0124] ;

[0125] In the formula, Sampling time The corresponding final error, Sampling time The correlation factors, This is a preset adjustment coefficient, ranging from 0.1 to 0.5, used to control the sampling time. The impact of the changing trend in the number of outlier dimensions on the error should be considered to avoid overestimation. Sampling time The number of abnormal dimensions, Sampling time The changing trend of the number of outlier dimensions (this value is normalized before being included in the above formula calculation by dividing the value by the historical maximum slope). for The absolute value is used to quantify the strength of the trend and avoid interference from the fusion of positive and negative signs. Ensure that even in (When there is no trend of change) Number of outlier dimensions It can still participate in calculations normally. Sampling time The weights of the correlation factors, Sampling time abnormal size The weight, Its value can be set based on experience to 0.6, 0.4, or 0.7, 0.3, etc. For the normalization function, the maximum-minimum normalization method can be used, where the maximum and minimum values ​​are preset empirical extreme values ​​derived from a large amount of historical calculation data. If the calculation result exceeds the [0,1] interval, it is restricted to the [0,1] range by a truncation function (i.e., if the result is less than 0, it is taken as 0, and if it is greater than 1, it is taken as 1) to eliminate the influence of outliers on the evaluation index.

[0126] Among them, correlation factors Quantified sampling time The strength of synergy among multiple anomalous dimensions and the historical reliability of these sensors, The higher the value, the greater its impact on the final error. The greater the direct contribution, the higher the weight. This determines the level of importance attached to anomalies in synergy. This directly reflects how many sensors are malfunctioning simultaneously. The magnitude of changes in the number of outlier dimensions was quantified; whether it's an increase or a decrease, rapid changes indicate state instability. The larger the value, the longer the sampling time. The larger the anomaly size, the greater its impact on the final error. The greater the direct contribution, the higher the weight. This determines the level of importance attached to the abnormal scale.

[0127] Furthermore, based on the above steps, the final error corresponding to each sampling time can be obtained.

[0128] Understandably, in this embodiment of the invention, the anomaly dimension set is first screened. Then, a correlation factor reflecting the strength of coordinated anomalies is calculated based on the correlation and historical reliability of the dimensions within the set. Simultaneously, the dynamic trend of the number of anomalies is analyzed. Finally, the correlation factor is combined with the dynamically changing anomaly scale. The resulting final error simultaneously encompasses the strength of anomaly coordination, the historical reliability of the involved sensors, the anomaly scale, and its dynamic evolution, providing a comprehensive, accurate, and forward-looking measure of overall unreliability. This measure serves as the weighting basis for subsequent filtering and fusion, enabling the data fusion process to intelligently and adaptively respond to anomalies of different natures and risk levels.

[0129] S207. Based on the final error corresponding to each sampling time, perform weighted filtering on the measurement data of all dimensions within the preset filtering window, fuse the filtered measurement data of all dimensions, and output the UAV positioning result.

[0130] As one possible implementation, the time weight corresponding to each sampling time is first determined based on the final error corresponding to each sampling time. This time weight is negatively correlated with the final error.

[0131] In some embodiments, the reciprocal of the sum of the final error and a very small positive number can be used as the time weight for the corresponding sampling time, which can be expressed as: ,in, Sampling time The corresponding final error, It should be a very small positive number, for example, 0.001, to prevent the denominator from being zero. This way, in the final error... Smaller sampling time Its corresponding time weight The larger the value, the faster the sampling time. The sensor data is generally more reliable.

[0132] Furthermore, a preset containing A preset filtering window for each sampling time, the preset filtering window being based on the current sampling time. The preset filter window covers a time range of [time range], which is the end time. ].

[0133] Furthermore, for each dimension According to the preset filtering window [ Each sampling time within ] The corresponding time weights, and their measurement data Perform a weighted average calculation to obtain this dimension. At the current sampling time The filtering result is calculated using the following formula:

[0134] ;

[0135] In the formula, For dimension At the current sampling time The filtering results The number of sampling times included in the preset filtering window. To filter within the preset window [ ]No. The time weight corresponding to each sampling moment The value of is always greater than 0, therefore the denominator Always greater than 0, For dimension In the preset filter window [ ]No. Measurement data at each sampling time.

[0136] The above calculation formula is equivalent to a dynamic low-pass filter. Measurement data at sampling moments with small errors and high weights are enhanced, while measurement data at sampling moments with large errors and low weights are suppressed, thereby smoothing out the noise and abnormal jumps of each sensor in the time dimension.

[0137] Based on the above steps, we can obtain the value of each dimension at the current sampling time. The filtering results.

[0138] Secondly, after obtaining the current sampling time for each dimension... After filtering the results, the measurement data from all dimensions after filtering are fused together to output the UAV's positioning results.

[0139] In some embodiments, taking the use of a Kalman filter for fusion localization as an example, firstly, all dimensions are sampled at the current sampling time. The filtered results are combined into a vector, which is used as the current sampling time. observation vector .

[0140] Observation vector The expression is: ,in, For all In this context, each dimension at the current sampling time Measurement data, The matrix transpose operation yields the final observation vector. for 3D column vectors, representing the current sampling time. Measurement information from each sensor.

[0141] Further based on the current sampling time Corresponding time weights Determine the noise covariance of the observation vector.

[0142] In some embodiments, the formula for calculating the noise covariance based on time weighting is as follows:

[0143] ;

[0144] In the formula, The current sampling time observation vector The noise covariance matrix, This is an empirical coefficient used to adjust the overall noise level, and it can typically be calibrated between 0.1 and 1.0 depending on the sensor's accuracy. It is the identity matrix. The current sampling time Time weighting.

[0145] Among them, time weight The smaller the value, the more likely it is to be the current sampling time. final error The larger the value, the higher the corresponding observation noise covariance. The larger the value, the lower the confidence level the Kalman filter assigns to that observation during fusion.

[0146] Furthermore, based on the posterior state estimate of the UAV at the previous sampling time, including position, velocity, attitude, etc., numerical integration is performed using the UAV's motion model (such as a uniform velocity model or a six-degree-of-freedom dynamic model) to predict the state prediction value and prediction covariance at the current sampling time.

[0147] Furthermore, the Kalman filter calculates the Kalman gain based on the observation vector, the noise covariance of the observation vector, the predicted state value, and the prediction covariance. The Kalman gain automatically determines whether to assign greater confidence to the prediction or the observation based on the relative magnitude of the prediction covariance and the measurement noise covariance. Subsequently, the gain is used to fuse the prediction and the observation to obtain the fused posterior state estimate.

[0148] Finally, the position state components, such as three-dimensional coordinates, are directly extracted from the posterior state estimate at the current sampling time. These coordinates represent the precise positioning result of the UAV at the current sampling time.

[0149] In some embodiments, if velocity or attitude information is required, it can also be extracted from the corresponding component of the posterior state estimate at the current sampling time.

[0150] In some embodiments, to further improve accuracy and suppress long-term drift, a factor graph containing state prediction constraints, observation constraints, and possible prior information (such as constraints provided by a digital twin model) can be constructed at multiple consecutive time points. Alternatively, a sliding window optimization can be used for local batch optimization, jointly optimizing the state sequence within the window to obtain a smoother and more consistent trajectory estimate. The positional information in the optimized state sequence is the final output with higher accuracy.

[0151] It should be noted that after obtaining the filtering results for each dimension at the current sampling time, fusion localization can also be performed by using filtering algorithms such as particle filter (PF), extended Kalman filter (EKF), and information filter. This embodiment of the invention does not specifically limit this.

[0152] Understandably, in the UAV precise positioning method based on multi-source data fusion provided in this invention embodiment, by analyzing the fluctuation characteristics, long-term reliability, and anomaly correlation of data from various dimensions, the comprehensive error at each sampling moment is fully quantified. Then, based on the error, the filtering weights are dynamically adjusted and multi-source data is fused, effectively solving the problem of insufficient consideration of data stability and anomaly correlation in traditional multi-source data fusion positioning. This method ensures full utilization of multi-source data and, through precise error quantification and dynamic filtering strategies, weakens the impact of environmental interference, sensor errors, and other factors on positioning. This allows the positioning results to accurately adapt to the complex state changes during UAV flight, meeting the stringent positioning accuracy requirements of application scenarios such as precision agriculture.

[0153] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0154] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for precise positioning of unmanned aerial vehicles (UAVs) based on multi-source data fusion, characterized in that, The method includes: Acquire multi-dimensional measurement data collected by different sensors during the flight of the drone; For each of the multiple dimensions, based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period before each sampling time, the initial error of each dimension at each sampling time is determined, and an error label sequence for each dimension is constructed based on the initial error. The error label sequence is used to characterize the stability of the measurement data of each dimension at each sampling time. Based on the distribution characteristics of error markers in the error marker sequence of each dimension, the historical reliability of each dimension is determined, and the historical reliability is used to characterize the long-term reliability of the measurement data of the corresponding dimension. For each sampling time, based on the error markers in the error marker sequence of each dimension within the preset historical time period, the abnormal correlation between any two dimensions is determined. Based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions, the final error corresponding to each sampling time is determined. The abnormal correlation is used to characterize the possibility that measurement data of different dimensions will be abnormal due to a common cause at the same sampling time. The number of abnormal dimensions is the number of error markers used to characterize the instability of measurement data at each sampling time. Based on the final error corresponding to each sampling time, the measurement data of all dimensions within the preset filtering window are weighted and filtered, and the measurement data of all dimensions after filtering are fused to output the positioning result of the UAV. Based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions, the final error corresponding to each sampling time is determined, including: For each sampling time, based on the error markers of each dimension, all dimensions that characterize the instability of the measurement data are determined, forming the set of abnormal dimensions for that sampling time; Based on the abnormal correlation between any two dimensions in the abnormal dimension set and the historical reliability of each dimension in the abnormal dimension set, the correlation factor at each sampling time is determined. Based on the number of abnormal dimensions at each sampling time within the preset historical period, construct an abnormal dimension count sequence; Based on the sequence of abnormal dimension counts, determine the trend of change in the number of abnormal dimensions at each sampling time. Based on the correlation factor, the number of abnormal dimensions, and the trend of change, the final error corresponding to each sampling time is determined.

2. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, For each of the multiple dimensions, based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period prior to each sampling time, the initial error of each dimension at each sampling time is determined, including: Obtain the measurement data sequence for each dimension during the preset historical time period; Identify the maximum data points in the measurement data sequence; The rate of change of extreme values ​​is determined based on the numerical differences and time intervals between adjacent maximum data points in the time series. The initial error of each dimension at each sampling time is determined based on the number density of the maximum value data points and the maximum value of the extreme value change rate within the preset historical time period.

3. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, Constructing an error label sequence for each dimension based on the initial error includes: For each dimension, the initial error of each dimension at each sampling time is compared with a preset error value; If the initial error at the target sampling time is greater than or equal to the preset error value, the error marker corresponding to the target sampling time is determined as the first error marker. The first error marker is used to characterize the instability of the measurement data of each dimension at the target sampling time. The target sampling time is any sampling time during the flight of the UAV. If the initial error at the target sampling time is less than the preset error value, the error marker corresponding to the target sampling time is determined as the second error marker. The second error marker is used to characterize the stability of the measurement data of each dimension at the target sampling time. Based on the error markers corresponding to each sampling time, construct the error marker sequence for each dimension.

4. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, Based on the distribution characteristics of error markers in the error marker sequence of each dimension, the historical reliability of each dimension is determined, including: For each dimension, a stable marker segment is determined in the error marker sequence for each dimension, the stable marker segment consisting of continuous error markers used to characterize the stability of the measurement data; The maximum value among the average values ​​of the initial errors of all stable labeled segments in each dimension is determined as the maximum average value; The historical reliability of each dimension is determined based on the longest segment length among all stable segments in each dimension, the total number of stable segments, and the maximum average value.

5. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, For each sampling time, based on the error markers in the error marker sequence of each dimension within the preset historical time period, determine the abnormal correlation between any two dimensions, including: For each sampling time, based on the error markers in the error marker sequence of each dimension within the preset historical time period, an abnormal time set for each dimension is determined. The abnormal time set consists of the sampling times corresponding to the error markers that characterize the instability of the measurement data within the preset historical time period. For any two dimensions, the intersection of the sets of abnormal moments in the two dimensions is determined as the set of target moments; The degree of difference between the measurement data sequences composed of the measurement data at each sampling time within the target time set of the two dimensions is determined by a preset sequence similarity measurement algorithm. Based on the number of times included in the target time set and the degree of difference, the abnormal correlation between the two dimensions is determined.

6. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, Based on the final error corresponding to each sampling time, weighted filtering is performed on the measurement data of all dimensions within the preset filtering window, including: Based on the final error corresponding to each sampling time, a time weight corresponding to each sampling time is determined, and the time weight is negatively correlated with the final error; For each dimension, a weighted average is calculated on the measurement data at each sampling time within the preset filtering window, based on the time weight corresponding to each sampling time within the preset filtering window with the current sampling time as the end time, to obtain the filtering result of each dimension at the current sampling time.

7. The UAV precise positioning method based on multi-source data fusion according to claim 6, characterized in that, The UAV's positioning result is output by fusing and filtering the measurement data from all dimensions, including: Combine the filtering results of all dimensions at the current sampling time into the observation vector at the current sampling time; The noise covariance of the observation vector is determined based on the time weight corresponding to the current sampling time. Based on the motion model of the UAV and the state estimate at the previous sampling time, a prediction is made to obtain the state prediction value and prediction covariance at the current sampling time. Based on the observation vector, the noise covariance, the state prediction value, and the prediction covariance, data fusion is performed using a preset state estimation algorithm to obtain the state estimate at the current sampling time. The state estimate includes the position information of the UAV. The location information is extracted from the state estimate as the positioning result of the UAV at the current sampling time.

8. The UAV precise positioning method based on multi-source data fusion according to claim 1, characterized in that, Acquire multi-dimensional measurement data collected by different sensors during the drone's flight, including: The drone is equipped with various sensors, including a global positioning system sensor, an inertial measurement unit, and a lidar, and each sensor is connected to the drone's flight control system. The flight control system of the UAV outputs a synchronous trigger signal to control each sensor to start data acquisition simultaneously; The raw data collected by each sensor is converted into standardized data in a unified format to obtain measurement data in multiple dimensions.

9. A precise positioning system for unmanned aerial vehicles (UAVs) based on multi-source data fusion, characterized in that, include: The data acquisition unit is used to acquire measurement data in multiple dimensions collected by different sensors during the flight of the UAV; An error analysis unit is used to determine the initial error of each dimension at each sampling time based on the fluctuation characteristics of the measurement data of each dimension within a preset historical period before each sampling time for each dimension among multiple dimensions, and to construct an error label sequence for each dimension based on the initial error. The error label sequence is used to characterize the stability of the measurement data of each dimension at each sampling time. A credibility determination unit is used to determine the historical credibility of each dimension based on the distribution characteristics of error markers in the error marker sequence of each dimension. The historical credibility is used to characterize the long-term reliability of the measurement data of the corresponding dimension. An error determination unit is used to determine the abnormal correlation between any two dimensions for each sampling time based on the error markers in the error marker sequence of each dimension within the preset historical time period, and to determine the final error corresponding to each sampling time based on the historical reliability of each dimension, the abnormal correlation between any two dimensions, and the number of abnormal dimensions. The abnormal correlation is used to characterize the possibility that measurement data of different dimensions will be abnormal at the same sampling time due to a common cause. The number of abnormal dimensions is the number of error markers used to characterize the instability of measurement data at each sampling time. The fusion positioning unit is used to perform weighted filtering on the measurement data of all dimensions within a preset filtering window based on the final error corresponding to each sampling time, fuse the measurement data of all dimensions after fusion filtering, and output the positioning result of the UAV. The error determination unit is specifically used to determine, for each sampling time, all dimensions that characterize the instability of the measurement data based on the error markers of each dimension, thereby forming an abnormal dimension set for the sampling time. Based on the abnormal correlation between any two dimensions in the abnormal dimension set and the historical reliability of each dimension in the abnormal dimension set, the correlation factor at each sampling time is determined. Based on the number of abnormal dimensions at each sampling time within the preset historical period, construct an abnormal dimension count sequence; Based on the sequence of abnormal dimension counts, determine the trend of change in the number of abnormal dimensions at each sampling time. Based on the correlation factor, the number of abnormal dimensions, and the trend of change, the final error corresponding to each sampling time is determined.

Images