Multi-sensor dynamic adaptive fusion method and system for unmanned aerial vehicle pose estimation

Abstract

The present invention discloses a multi-sensor dynamic adaptive fusion method and system for unmanned aerial vehicle posture estimation. The present invention constructs a support function in combination with a self-supporting factor, thereby reducing the influence of the precision error of the unmanned aerial vehicle sensor due to the movement of the unmanned aerial vehicle on the fusion result; the present invention constructs the self-supporting factor based on the grey theory, and then constructs the support function in combination with the self-supporting factor, so that the weight design is more reasonable, thereby improving the posture measurement accuracy of the multi-sensor when the unmanned aerial vehicle is in motion; the present invention fully considers the problem of insufficient sensor abnormal data detection standards, uses the sensor data itself for detection, has sufficient objectivity, uses different processing methods for different types of abnormal data, fully improves the utilization rate of the data, and improves the quality of the posture data when the unmanned aerial vehicle is in motion, thereby obtaining a high-precision unmanned aerial vehicle posture estimation result.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-sensor fusion, and in particular, relates to a multi-sensor dynamic adaptive fusion method and system for unmanned aerial vehicle posture estimation. Background Art

[0002] In the process of collecting UAV flight data, there is often a lot of local and fragmentary information, which may have extensive conflicts and overlaps. Therefore, the use of multi-sensor data fusion technology to effectively integrate this information can not only solve the problem of information overlap and conflict, but also improve the accuracy of system decision-making, making the final decision and judgment faster, more accurate, timely and effective. If sensor information containing interference information or much error information and high uncertainty is involved in data fusion, the result will only get worse. Therefore, multi-sensor data fusion needs to select a suitable fusion strategy based on the information characteristics contained in the fused data.

[0003] In the field of multi-sensor data fusion, traditional fusion methods face multiple challenges, including insufficient processing of abnormal data, slow fusion speed, low accuracy and reliability, etc. Therefore, a system is needed that can dynamically adapt to data changes, effectively process abnormal data, and improve data fusion speed and accuracy.

[0004] A major advantage of the support-based multi-sensor information fusion algorithm is that it does not rely on prior knowledge. This means that the algorithm does not need additional information to process the data, making the algorithm more adaptable when dealing with unknown or changing environments. However, the fusion accuracy depends largely on the design of the support function. The purpose of the support function is to measure the consistency and reliability of the data, which in turn affects the final weight distribution. Since the calculation of the weight is based only on the data itself, the quality of data preprocessing is directly related to the accuracy and reliability of the fusion result. If the preprocessing is not appropriate, even the optimal fusion algorithm may not achieve the desired results. Therefore, when applying this type of fusion algorithm, it is key to ensure high-quality data preprocessing. Summary of the invention

[0005] Traditional UAV posture estimation is basically provided by a single sensor, such as the UAV flight control with built-in IMU or the depth camera built-in IMU, but a single sensor often cannot accurately provide the UAV posture data in complex environments. In addition, most of the existing sensor fusion algorithms are filtering algorithms or neural network algorithms, which often believe that there is too much intervention and the resulting fusion results lack good objectivity. In order to obtain fusion data better and more objectively, the present invention provides a multi-sensor dynamic adaptive fusion method, system, medium and equipment for UAV posture estimation, which aims to solve the problems of low efficiency, insufficient accuracy and limited ability to process abnormal data in existing sensor fusion technology.

[0006] On the one hand, the present invention provides a multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation, comprising:

[0007] S1: Construct the self-support factor of the posture sensor based on the grey theory, and then construct the support function in combination with the self-support factor;

[0008] S2: Calculate data consistency based on the support function, establish an abnormal data detection method based on the dual indicators of data consistency and self-support factor, and then classify and process the abnormal data according to the length of the abnormal data;

[0009] S3: Design a window forgetting function, and calculate the consistency mean and reliability measure of the pose observation data to obtain the weight of the data, and combine the weights to obtain the pose sensor fusion result.

[0010] On the other hand, the present invention provides a multi-sensor dynamic adaptive fusion system for unmanned aerial vehicle posture estimation, comprising:

[0011] Support function construction module, used to calculate the self-support factor and construct the support function based on the grey relational theory;

[0012] The abnormal data processing module uses the support function to calculate data consistency, establishes an abnormal data detection method based on the dual indicators of data consistency and self-support factor, and then classifies and processes the abnormal data according to the length of the abnormal data;

[0013] The weight calculation module is used to calculate the data consistency mean in combination with the window forgetting function, and obtain the data reliability measure based on the data consistency mean, combine the two to design the weights, combine the weights to perform data fusion, and perform error assessment on the data fusion.

[0014] On the other hand, the present invention provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute the multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation as described in any one of the above.

[0015] On the other hand, the present invention provides a multi-sensor dynamic adaptive fusion device for unmanned aerial vehicle posture estimation, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the contents of any of the above methods when executing the computer program.

[0016] Compared with the prior art, the present invention has the following advantages or beneficial effects:

[0017] 1. The present invention combines the self-support factor to construct the support function, thereby making up for the defect of the traditional support algorithm that ignores the reliability of the sensor data itself, and reduces the influence of the precision error of the drone sensor due to the movement of the drone on the fusion result.

[0018] 2. The present invention constructs self-supporting factors based on grey theory, and then constructs support functions in combination with the self-supporting factors, which solves the defect that the traditional support function fails to take into account the credibility of the sensor data itself, making the weight design more reasonable, thereby improving the accuracy of multi-sensor posture measurement when the drone is in motion.

[0019] 3. The present invention fully considers the problem of insufficient detection standards for abnormal sensor data, uses the sensor data itself for detection, has sufficient objectivity, uses different processing methods for different types of abnormal data, fully improves data utilization, and improves the quality of the posture data during UAV movement, thereby obtaining a higher-precision UAV posture estimation result. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a multi-sensor dynamic adaptive fusion method for drone pose estimation according to an embodiment of the present application.

[0021] Figure 2 This is a comparison diagram of the embodiments of the present application and other methods. DETAILED DESCRIPTION

[0022] The present invention will be described in detail below with reference to the accompanying drawings and preferred examples, and the purpose and effects of the present invention will become more clear. It should be understood that the specific examples described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] like Figure 1 As shown, the embodiment of the present application provides a multi-sensor dynamic adaptive fusion method for drone pose estimation, comprising:

[0024] S1: Based on the grey theory, the self-support factor of the posture sensor is constructed, and then the support function is constructed in combination with the self-support factor.

[0025] S2: Calculate data consistency based on the support function, establish an abnormal data detection method based on the dual indicators of data consistency and self-support factor, and then classify and process the abnormal data according to the length of the abnormal data.

[0026] S3: Design a window forgetting function, and calculate the consistency mean and reliability measure of the pose observation data to obtain the weight of the data, and combine the weights to obtain the pose sensor fusion result.

[0027] Further, the S1 specifically includes:

[0028] S11: Establish a UAV posture data observation model based on the sensor measurement model.

[0029] S12: Construct self-supporting factors for the drone posture observation values ​​of each sensor.

[0030] S13: Define the difference of the posture observation data of different sensors at the same time as the measurement distance and normalize it.

[0031] S14: Construct a support function based on the self-support factor in step S12.

[0032] Furthermore, the S12 specifically includes:

[0033] S121: Based on the data observation value of the drone at a certain moment, the observation values ​​with a time interval of t are taken from both ends to construct a reference sequence.

[0034] S122: Combining the priori and a posteriori data, a self-supporting factor is constructed according to the grey correlation theory to measure the credibility of the posture observation sensor data at that moment.

[0035] Further, the S2 specifically includes:

[0036] S21: Calculating data consistency according to the support function of each sensor at each time obtained in step S14;

[0037] S22: Calculate the detection coefficient by combining data consistency and self-support factor, and calculate the length of abnormal data;

[0038] S23: Classify the abnormal data according to their length and formulate abnormal data processing strategies respectively.

[0039] In particular, the abnormal data processing is specifically as follows:

[0040] When the abnormal data is peak abnormal data, that is, when the previous and subsequent data are qualified, the elimination processing method is adopted;

[0041] When the abnormal data is short-term abnormal data, that is, the abnormal data is continuous but the time is short, the Prophet forecasting method is used;

[0042] When the abnormal data is long-term abnormal data, that is, the abnormal data is continuous and lasts for a long time, the mean of the posture data at the same moment is used for replacement. If other posture data at the same moment are short-term abnormal data, the processed predicted data is used for mean replacement.

[0043] Further, the S3 specifically includes:

[0044] S31: Design a window forgetting function and calculate the consistency mean of the pose data according to the window length;

[0045] S32: Calculate the reliability measure of the pose data according to the consistency mean;

[0046] S33: Calculate weights based on consistency mean and reliability measure, and perform error assessment.

[0047] In a preferred embodiment, the present application provides a multi-sensor dynamic adaptive fusion method for drone pose estimation, which specifically includes the following steps:

[0048] Step 1: As shown in formula (1), let z i (k) is the observation value of each drone IMU sensor i at time k, then the observation value z i (k) is the true value x(k) at time k and the measurement noise v i (k) and.

[0049] z i (k) = x(k) + v i (k),i=1,2,...,m;k=1,2,...,n (1)

[0050] Where m is the number of sensors in the system and n is the maximum time.

[0051] Step 2: The observation value z of IMU sensor i at time k i (k) is set as the reference sequence x i (k), with time k as the center, the sensor data before and after time t are taken as the comparison sequence y i , as shown in formula (2).

[0052]

[0053] Step 3: Based on the grey relational theory, i As the reference sequence, calculate the self-support factor S of sensor i at time k i (k), as shown in formula (3).

[0054] The self-support factor indicates the credibility of the IMU data at that moment and is an indicator of the accuracy of the sensor itself. i The closer (k) is to 1, the higher the credibility.

[0055]

[0056] Where r is the resolution coefficient, assuming r = 0.5, Δx(w) = |x i (k)-y i(w)|,w=1,2,...,2t,Δ max and Δ min It is the maximum and minimum difference between the comparison sequences.

[0057] Step 4: Observe the sequence z according to the IMU sensor i (k), calculate the difference between sensors i and j at time k, and normalize the data to define the support distance D ij (k), as shown in formula (4).

[0058] The closer the data of the two sensors at the same time, the smaller the distance between them, D ij The closer (k) is to 0.

[0059]

[0060] Among them, z i (k) and z j (k) is the observation value of sensor i and sensor j at time k.

[0061] Step 5: Combine the support distance and self-support factor to construct the support function S of IMU sensors i and j at time k ij (k), the calculation method is shown in formula (5). ij (k) Having the following properties:

[0062] (1)S ij (k) is about D ij (k) is a monotonically decreasing function, that is, the smaller the distance between sensor data, the greater the mutual support.

[0063] (2) When D ij When (k) = 0, S ij (k)=1, that is, when the sensor data are the same, the support is 1, which means that the sensor data are fully trusted.

[0064] (3)0 ij (k)￡1, the support value range is between 0 and 1, and the degree of trust increases from small to large.

[0065] (4)S ij (k) = S ji (k), the mutual support between the two sensors is equal.

[0066]

[0067] ​Step 6: Construct the support degree matrix ISD(k) at time k according to the self - support factor and the support degree function, as shown in Equation (6). This matrix is a symmetric matrix, the diagonal elements of which are the self - support factors at time k, the rows and columns of which represent the sensor numbers, and the elements of which represent the support degrees between two sensors at time k.

[0068]

[0069] Step 7: Calculate the data consistency metric r i (k) at time k according to the support degrees between IMU sensors, as shown in Equation (7).

[0070]

[0071] Where m is the number of system sensors and n is the maximum time.

[0072] Step 8: Calculate the detection coefficient A i (k). m is the number of sensors, and the detection coefficient A i (k) is the calculation result weighted by 1 and m - 1, as shown in Equation (8).

[0073] Set the detection threshold λ. If A i (k) ≥ λ, the data is qualified data. If A i (k) < λ, the data is abnormal data. When abnormal data is detected, count it until the data is qualified.

[0074]

[0075] Step 9: Define the count of abnormal data during abnormal data detection as the length l of abnormal data. Based on this, classify the abnormal data and use different processing methods for it.

[0076] When the abnormal data is peak abnormal data, that is, l = 1, it is directly removed;

[0077] When the abnormal data is short - term abnormal data, that is, 1 < l < a, use the Prophet method for prediction;

[0078] When the abnormal data is long - term abnormal data, that is, l ≥ a, use the mean of the qualified data at the same time to replace it. If there is sensor data as short - term abnormal data at the same time, use the predicted data after processing for mean replacement. Then the replaced data is z i ′(k), as shown in Equation (9).

[0079]

[0080] Where m dis the number of data removed by the self-detection mechanism at time k, m p To see the number of short-term abnormal data over time, z j (k) is the qualified sensor data sequence that does not need to be processed at time k, z J (k) is the sensor data sequence after replacement by the Prophet model prediction at time k.

[0081] Step 10: Design the window forgetting function F(t), as shown in formula (10).

[0082] The window forgetting function means that the data of L moments before moment k is retained, and the data before moment kL is forgotten.

[0083]

[0084] Among them, L is the forgetting depth.

[0085] Step 11: Combined with the window forgetting function, the data consistency at the moment in the window is averaged to obtain the consistency mean As shown in formula (11).

[0086]

[0087] Step 12: Combine the window forgetting function and the consistency mean to find the consistency of the IMU sensor data within the window, and define it as a reliability measure As shown in formula (12).

[0088]

[0089] Step 13: From steps 11 and 12, if the IMU pose data are closer, the support is higher, and the mean consistency within the window is The higher the value, the higher the mutual support of the data. At the same time, if the reliability metric is The lower it is, the more stable the data consistency is and the more reliable the data is. The weight q is calculated accordingly. i (k), as shown in formula (13).

[0090] Among them, if The bigger, The smaller the weight q is, the i (k)The higher the

[0091]

[0092] Step 14: Combine weights q i (k) Calculate the fusion result at time k The fusion result is shown in formula (14).

[0093]

[0094] Step 15: Combine the fusion results and define the error evaluation E(k) as shown in formula (15).

[0095]

[0096] Step 16: As shown in formula (16), the mean value of the error evaluation is used as the measurement index E of the fusion effect of the whole system. The smaller E is, the better the IMU data fusion effect is.

[0097]

[0098] The experimental verification results are as follows Figure 2 As shown in the figure, this experiment measures the attitude angles roll (°) and pitch (°) of the same UAV over a period of time using three inertial navigation sensors. Figure 2 This is roll's multi-sensor fusion results and error analysis. The upper half of the figure shows the sensor raw data and sensor fusion results. SDM-ISF is a method provided in an embodiment of the present application, which is compared with other multi-sensor fusion methods, such as the single active multiple passive sensor (SAMPS) measurement fusion algorithm (SAMPS-AA) based on angle association (AA), the improved adaptive weighting algorithm (IAWA) and the magnetic correction weighted extended Kalman filter (MEKF); the lower half of the figure shows the error comparison between the embodiment of the present application and other methods.

[0099] from Figure 2 It can be seen from the comparison results that the error of the sensor fusion method provided in the embodiment of the present application is smaller than the error of other algorithms.

[0100] The above is a detailed introduction to a multi-sensor dynamic adaptive fusion method for drone pose estimation provided by an embodiment of the present application. The description of the above embodiment is only used to help understand the method and its core idea of ​​the present application; at the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0101] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation, characterized in that: The method comprises the following steps: S1: Construct the self-support factor of the posture sensor based on the grey theory, and then construct the support function in combination with the self-support factor; S2: Calculate data consistency based on the support function, establish an abnormal data detection method based on the dual indicators of data consistency and self-support factor, and then classify and process the abnormal data according to the length of the abnormal data; S3: Design a window forgetting function, and calculate the consistency mean and reliability measure of the pose observation data to obtain the weight of the data, and combine the weights to obtain the pose sensor fusion result; Among them, S2 is specifically: S21: Calculate data consistency according to the support function of each sensor at each moment; S22: Calculate the detection coefficient by combining data consistency and self-support factor, and calculate the length of abnormal data; S23: classify the abnormal data according to their length and formulate abnormal data processing strategies respectively; The abnormal data processing is specifically as follows: When the abnormal data is peak abnormal data, that is, when the previous and subsequent data are qualified, the elimination processing method is adopted; When the abnormal data is short-term abnormal data, that is, the abnormal data is continuous but the time is short, the Prophet forecasting method is used; When the abnormal data is long-term abnormal data, that is, the abnormal data is continuous and lasts for a long time, the mean value of the pose data at the same moment is used for replacement. If other pose data at the same moment are short-term abnormal data, the processed predicted data is used for mean replacement. S3 is specifically: S31: Design a window forgetting function and calculate the consistency mean of the pose data according to the window length; S32: Calculate the reliability measure of the pose data according to the consistency mean; S33: Calculate weights based on consistency mean and reliability measure, and perform error assessment.

2. The multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation according to claim 1, characterized in that: S1 is specifically: S11: Establish a UAV posture data observation model based on the sensor measurement model; S12: construct self-supporting factors for the drone posture observation values ​​of each sensor; S13: defining the difference of the posture observation data of different sensors at the same time as the measurement distance, and normalizing it; S14: Construct a support function based on the self-support factor in step S12.

3. The multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation according to claim 2 is characterized in that: S12 is specifically: S121: According to the data observation value of the UAV at a certain time, the observation values ​​with a time interval of t are taken from both ends to construct a reference sequence; S122: Combining the priori and a posteriori data, a self-supporting factor is constructed according to the grey correlation theory to measure the credibility of the posture observation sensor data at that moment.

4. A multi-sensor dynamic adaptive fusion system for UAV pose estimation, characterized in that include: Support function construction module, used to calculate the self-support factor and construct the support function based on the grey relational theory; The abnormal data processing module uses the support function to calculate data consistency, establishes an abnormal data detection method based on the dual indicators of data consistency and self-support factor, and then classifies and processes the abnormal data according to the length of the abnormal data; The weight calculation module is used to calculate the mean of data consistency in combination with the window forgetting function, and obtain the data reliability measure based on the mean of data consistency, combine the two to design the weight, combine the weight to perform data fusion, and perform error assessment on the data fusion; The abnormal data processing module performs the following operations: Calculate data consistency based on the support function of each sensor at each moment; Calculate the detection coefficient by combining data consistency and self-support factor, and calculate the length of abnormal data; Classify abnormal data according to their length and formulate abnormal data processing strategies respectively; The abnormal data processing is specifically as follows: When the abnormal data is peak abnormal data, that is, when the previous and subsequent data are qualified, the elimination processing method is adopted; When the abnormal data is short-term abnormal data, that is, the abnormal data is continuous but the time is short, the Prophet forecasting method is used; When the abnormal data is long-term abnormal data, that is, the abnormal data is continuous and lasts for a long time, the mean value of the pose data at the same moment is used for replacement. If other pose data at the same moment are short-term abnormal data, the processed predicted data is used for mean replacement. The weight calculation module performs the following operations: Design a window forgetting function and calculate the mean consistency of pose data based on the window length; Calculate the reliability measure of the pose data based on the consistency mean; Weights are calculated based on the consistency mean and reliability measures, and error assessment is performed.

5. The multi-sensor dynamic adaptive fusion system for unmanned aerial vehicle posture estimation according to claim 4, characterized in that: The abnormal data processing module: The data consistency is calculated based on the support function, and the detection coefficient is calculated in combination with the self-support factor. The detection coefficient is used as the basis for abnormal data detection, and the length of the abnormal data is recorded as the abnormal data classification standard. The abnormal data is classified and processed according to the category of abnormal data, and the detection coefficient is solved again to check whether the data processing results are qualified.

6. A computer-readable storage medium comprising instructions, which, when executed on a computer, enables the computer to execute the multi-sensor dynamic adaptive fusion method for unmanned aerial vehicle pose estimation as described in any one of claims 1 to 3.

7. A multi-sensor dynamic adaptive fusion device for unmanned aerial vehicle posture estimation, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 3 is implemented.

Images