Low-altitude unmanned aerial vehicle dynamic visual positioning method and device under ultra-wideband

Abstract

The application discloses a low-altitude unmanned aerial vehicle dynamic visual positioning method and device under ultra-wideband, relates to the technical field of unmanned aerial vehicle positioning and navigation, and comprises the following steps: arranging UWB anchor points in a work area, acquiring a low-altitude unmanned aerial vehicle calibration work track, and configuring an anchor point group through signal fitting; when the unmanned aerial vehicle works along the track, activating the anchor point group according to a front authentication track point; receiving a UWB signal and performing abnormal identification; if a threshold value is triggered, activating a visual acquisition device and acquiring IMU data; extracting visual key feature points, initializing a dynamic fusion network, correcting a pose based on multi-source data linkage, and generating a linkage pose correction result. The application solves the technical problem that, in the non-electric variable control adjustment of the low-altitude unmanned aerial vehicle pose in the prior art, positioning deviation and failure are prone to occurring due to complex environment, and it is difficult to meet the precise operation demand, achieves the precise positioning of the low-altitude unmanned aerial vehicle pose, realizes the precise positioning in the complex environment and avoids failure, and meets the technical effect of meeting the precise operation demand.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of unmanned aerial vehicle positioning and navigation, in particular to a low-altitude unmanned aerial vehicle dynamic visual positioning method and device under ultra-wide band. BACKGROUND

[0002] In low-altitude unmanned aerial vehicle operation, accurate control and adjustment of non-electric variables such as pose are crucial to flight safety and task execution effect. Existing technologies mostly rely on single or conventional fusion positioning methods. In complex environments, it is difficult to ensure positioning accuracy and stability in the face of signal interference and scene dynamic changes. When the traditional scheme is used for low-altitude unmanned aerial vehicle positioning, positioning deviation and failure are prone to occur due to complex environments such as signal shielding and variable motion states, which cannot meet the requirements of accurate operation. SUMMARY

[0003] The application provides a low-altitude unmanned aerial vehicle dynamic visual positioning method and device under ultra-wide band, which is used to solve the technical problem that existing technologies are prone to positioning deviation and failure in the control and adjustment of non-electric variables of low-altitude unmanned aerial vehicle pose due to complex environments, and cannot meet the requirements of accurate operation.

[0004] In a first aspect, the application provides a low-altitude unmanned aerial vehicle dynamic visual positioning method under ultra-wide band, which comprises the following steps: arranging UWB anchor points in an operation area, and obtaining a calibration operation trajectory of a low-altitude unmanned aerial vehicle; performing signal fitting according to the calibration operation trajectory and the distribution of the UWB anchor points, and configuring an anchor point group mapped with the calibration operation trajectory; when the low-altitude unmanned aerial vehicle executes an operation task along the calibration operation trajectory, performing anchor point group activation configuration according to a pre-authorization trajectory point of the low-altitude unmanned aerial vehicle; receiving UWB signals emitted by the low-altitude unmanned aerial vehicle by using the activated anchor point group, and performing abnormal identification of the UWB signals in the anchor point group; if the abnormal identification result triggers a preset abnormal threshold, activating a visual acquisition device of the low-altitude unmanned aerial vehicle to capture environmental features, and obtaining a motion data set of an IMU sensor; after extracting key feature points from the visual acquisition result, initializing a dynamic fusion network by using the pre-authorization trajectory point, performing linkage pose correction based on the key feature points, the motion data set and the UWB signals, and generating a linkage pose correction result.

[0005] In a second aspect of the present application, a low-altitude unmanned aerial vehicle dynamic visual positioning device under ultra-wideband is provided, which comprises: an anchor point group configuration module, configured to arrange UWB anchor points in a work area, obtain a calibration work trajectory of a low-altitude unmanned aerial vehicle, perform signal fitting according to the calibration work trajectory and UWB anchor point distribution, and configure an anchor point group mapped with the calibration work trajectory; an anchor point group activation module, configured to perform anchor point group activation configuration according to a pre-authorization trajectory point of the low-altitude unmanned aerial vehicle when the low-altitude unmanned aerial vehicle performs a work task along the calibration work trajectory; an anchor point group anomaly identification module, configured to receive UWB signals transmitted by the low-altitude unmanned aerial vehicle by using the activated anchor point group, and perform anomaly identification of the UWB signals in the anchor point group; a motion data set acquisition module, configured to activate a visual acquisition device of the low-altitude unmanned aerial vehicle to capture environmental features and acquire a motion data set of an IMU sensor if the anomaly identification result triggers a preset anomaly threshold; and a linkage pose correction result acquisition module, configured to initialize a dynamic fusion network by using the pre-authorization trajectory point after extracting key feature points from the visual acquisition result, perform linkage pose correction based on the key feature points, the motion data set, and the UWB signals, and generate a linkage pose correction result.

[0006] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0007] In the present application, UWB anchor points are arranged in a low-altitude work scene, a dynamic fusion network is used, trajectory fitting, feature extraction, multi-source data linkage pose correction, and other processes are performed, and anchor point group dynamic management and anomaly compensation mechanisms are combined, so that the precise control and adjustment of the pose of the unmanned aerial vehicle, which is a non-electric variable, are realized, the pose positioning of the low-altitude unmanned aerial vehicle in a complex environment is more accurate and reliable, the demand for the control and adjustment of non-electric variables in low-altitude work is met, the precise positioning of the pose of the low-altitude unmanned aerial vehicle is achieved, precise positioning in a complex environment is realized and failure is avoided, and the technical effect of meeting the demand for precise work is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1 is a flowchart of a low-altitude unmanned aerial vehicle dynamic visual positioning method under ultra-wideband provided by the embodiments of the present application.

[0010] Figure 2 is a structural schematic diagram of a low-altitude unmanned aerial vehicle dynamic visual positioning device under ultra-wideband provided by the embodiments of the present application.

[0011] The reference signs are explained as follows: anchor group configuration module 1, anchor group activation module 2, anchor group anomaly identification module 3, motion data set acquisition module 4, and linkage pose correction result acquisition module 5. DETAILED DESCRIPTION

[0012] The application provides a low-altitude unmanned aerial vehicle dynamic visual positioning method and device under an ultra-wideband, and is used for solving the technical problem that in the non-electric variable control adjustment of the low-altitude unmanned aerial vehicle pose, positioning deviation and failure are prone to occur due to complex environment, and it is difficult to meet the precise operation requirement.

[0013] The technical solutions in the embodiments of the application will be clearly and completely described in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0014] It should be noted that the terms "first", "second", and the like in the specification and the above drawings of the application are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to the process, method, product or device.

[0015] Embodiment one, as shown in the figure, a low-altitude unmanned aerial vehicle dynamic visual positioning method under an ultra-wideband, wherein the method comprises: Figure 1

[0016] Step A100: deploying UWB anchors in a work area, and acquiring a calibration work trajectory of a low-altitude unmanned aerial vehicle, performing signal fitting according to the calibration work trajectory and UWB anchor distribution, and configuring anchor groups mapped with the calibration work trajectory.

[0017] In the embodiments of the application, the UWB anchors are fixed reference points deployed in the work area, and are used for UWB signal interaction with the low-altitude unmanned aerial vehicle. The calibration work trajectory is a low-altitude unmanned aerial vehicle work trajectory preset by those skilled in the art.

[0018] ​Specifically, when deploying UWB anchors in the work area, those skilled in the art need to determine the number and position of anchors according to the geometric shape and spatial scale of the area. For example, 5 anchors are deployed at the corners and center of a rectangular work area with a length of 50 meters and a width of 30 meters to form a distributed positioning network. After the anchors are deployed, the low-altitude unmanned aerial vehicle is controlled to fly along the preset calibration work trajectory. Assuming that the trajectory contains 100 equidistant trajectory points, the unmanned aerial vehicle flies at a speed of 2 meters / second and synchronously records the UWB signal time of arrival (ToA), signal strength (RSSI) and other data of each trajectory point. The average value is taken for 3 repeated flights to reduce random errors.

[0019] Based on the calibration work trajectory and the distribution of UWB anchors, the least squares method is used to fit the signal data. First, the UWB anchor signal time difference (TDoA) data corresponding to each trajectory point of the unmanned aerial vehicle flying along the calibration work trajectory is obtained. The spatial coordinates (x, y, z) of each trajectory point are taken as the independent variable, and the TDoA values corresponding to each anchor are taken as the dependent variable. By minimizing the sum of squares of errors between the actual measured TDoA values and the theoretical TDoA values calculated based on the coordinates, the coefficients of the signal propagation model are solved, thereby establishing the mapping relationship between the TDoA of each anchor signal and the spatial distance of the trajectory point, and forming a mathematical model describing the propagation law of the UWB signal in the work area. Provide accurate signal fitting basis for subsequent anchor group configuration. Exemplarily, for the trajectory point P(x, y, z), the function relationship between the signal propagation time and the coordinates of anchors A, B and C is obtained by fitting: =0.3x+0.2y+0.5z+0.1, =0.2x+0.4y+0.3z+0.2, =0.4x+0.1y+0.4z+0.15, the root mean square error (RMSE) is less than 0.05 seconds.

[0020] According to the signal fitting result, the anchor group is configured according to the trajectory. The calibration trajectory is divided into 20 continuous segments, each segment contains 5 trajectory points, the signal quality score of each anchor is calculated for each segment of the trajectory, the ToA accuracy, RSSI value and multipath effect parameters are comprehensively considered, and the top 3 anchors are selected to form the mapping anchor group. For example, the anchor group corresponding to the trajectory segment S1 is {A, B, C}, and the signal quality scores are 92, 88 and 85 respectively, all of which are higher than the threshold value 80, ensuring the positioning accuracy.

[0021] Through dynamic signal fitting and anchor group configuration, the positioning error can be controlled within 0.2 meters under the same shielding scene, which provides accurate basic anchor mapping relationship for subsequent dynamic activation and abnormal compensation of anchor groups when the unmanned aerial vehicle works along the trajectory, and effectively improves the robustness of the positioning device in complex environments.

[0022] Step A200: When the low-altitude unmanned aerial vehicle performs a task along the calibration operation trajectory, the anchor group activation configuration is performed according to the pre-authentication trajectory point of the low-altitude unmanned aerial vehicle.

[0023] In the embodiment of the application, the pre-authentication trajectory point is used to identify trajectory drift, establish a pre-drift amount, perform time sequence flight fitting, match the calibration operation trajectory, and activate the mapping anchor group when the low-altitude unmanned aerial vehicle performs a task along the calibration operation trajectory.

[0024] Optionally, when the low-altitude unmanned aerial vehicle performs a task along the calibration operation trajectory, the trajectory drift is identified according to the pre-authentication trajectory point of the low-altitude unmanned aerial vehicle to establish a pre-drift amount with a direction identifier, a tolerance zone map containing a drift threshold, available anchor mapping, and position shielding is established by combining the operation area and the UWB anchor distribution, the pre-drift amount is subjected to regional tolerance analysis, if it passes, the time sequence flight fitting is performed based on the pre-authentication trajectory point, and the anchor group is activated after the fitting result is matched with the calibration operation trajectory. The specific steps are described in detail in A210-A230.

[0025] Step A300: The activated anchor group receives the UWB signal transmitted by the low-altitude unmanned aerial vehicle, and performs abnormal identification of the UWB signal in the anchor group.

[0026] In the embodiment of the application, the UWB signal is an ultra-wideband wireless signal, which transmits data by using a nanosecond-level non-sine wave narrow pulse, measures the propagation time of the pulse signal between the transmitting end and the receiving end, and determines the distance between the two points to realize the positioning function.

[0027] In an embodiment of the application, when the activated anchor group receives the UWB signal transmitted by the low-altitude unmanned aerial vehicle, first, the signal parameter is collected in real time by the signal receiving module of each anchor point. The module has the ability to collect UWB signal parameters such as signal strength RSSI and signal arrival time ToA in real time, which can support the subsequent abnormal identification and pose correction process. The collected UWB signal parameters include signal strength (RSSI), signal arrival time (ToA), and time difference (TDoA). Taking a typical operation scenario as an example, each anchor point in the anchor group receives signals at a frequency of 10 Hz, compares the collected RSSI data with a preset threshold (such as -70 dBm), and calculates the error between ToA and the theoretical propagation time, such as setting the threshold to 0.1 seconds. If the RSSI of a certain anchor point is lower than -70 dBm for three consecutive times and the ToA error exceeds 0.1 seconds, it is determined that the signal is abnormal.

[0028] In the anomaly identification process, a sliding window algorithm is used to analyze the signals in the anchor group in batches. For example, a 5-second sliding window is set, and the proportion of abnormal signals in the window is counted. When the proportion of abnormal signals exceeds 30%, the preset abnormal threshold is triggered. Through Kalman filtering, the signal parameters are denoised, the measurement error of RSSI is reduced, the ToA error is optimized and reduced, and the accuracy of anomaly identification is improved.

[0029] Through multi-parameter joint analysis (RSSI, ToA, TDoA) and sliding window statistics, the false positive rate of anomaly identification is reduced, and stable identification accuracy can be maintained in complex environments with multipath interference. It provides reliable signal anomaly trigger conditions for subsequent visual acquisition and pose correction, effectively improving the environmental adaptability and robustness of low-altitude unmanned aerial vehicle positioning devices.

[0030] Step A400: If the anomaly identification result triggers the preset abnormal threshold, activate the visual acquisition device of the low-altitude unmanned aerial vehicle to capture environmental features and obtain the motion data set of the IMU sensor.

[0031] In the embodiments of the present application, the IMU sensor is generally composed of accelerometers, gyroscopes and other sensors. The accelerometer can measure the acceleration of the object in three axial directions, i.e. X, Y and Z axes, and the gyroscope can measure the angular velocity of the object around these three axes. By integrating these parameters, the attitude, position and motion trajectory of the object can be obtained.

[0032] Specifically, when the UWB signal anomaly identification result triggers the preset abnormal threshold, such as when the RSSI of the anchor group is lower than -70dBm for 5 consecutive times and the ToA error exceeds 0.1 seconds, the visual acquisition device carried by the low-altitude unmanned aerial vehicle is automatically activated, such as a RGB camera with a resolution of 1280x720, to capture environmental feature images at a frequency of 30 frames / second. At the same time, the motion data set of the IMU sensor with a sampling rate of 100Hz is obtained, including three-axis acceleration obtained by an accelerometer with a range of ±5g and three-axis angular velocity obtained by a gyroscope with a range of ±2000° / s.

[0033] During the visual acquisition process, the original image is first preprocessed to remove distortion and enhance contrast, ensuring that the environmental features are clear and distinguishable. IMU data is denoised through Kalman filtering algorithm, reducing the acceleration noise density to 0.005g / √Hz and optimizing the angular velocity random walk to 0.05° / √h, providing high-quality motion data for subsequent pose correction.

[0034] In addition, when the UWB signal anomaly identification result does not trigger the preset anomaly threshold, the visual acquisition device and the IMU data acquisition process are not activated, but return to the initial monitoring state of receiving the UWB signal and performing anomaly identification, and continuously monitoring the UWB signal in the anchor group in real time to ensure timely capture of subsequent possible signal anomaly conditions.

[0035] Through the synchronous acquisition and preprocessing of vision and IMU data, in the UWB signal failure scenario, the robustness problem of single UWB positioning in a complex environment is effectively solved, multi-source heterogeneous data support is provided for the dynamic fusion network, and adaptive switching and precision guarantee of low-altitude unmanned aerial vehicle pose correction are realized.

[0036] Step A500: After extracting key feature points from the visual acquisition result, the dynamic fusion network is initialized using the pre-authentication trajectory point, the linkage pose correction based on the key feature points, the motion data set and the UWB signal is performed, and the linkage pose correction result is generated.

[0037] In the embodiment of the application, the visual acquisition result is the environmental feature image captured by the visual acquisition device of the low-altitude unmanned aerial vehicle, and the original and preprocessed image data after image preprocessing are used for subsequent feature extraction. The key feature point is a representative image feature point established by constructing a multi-scale extraction resolution for multi-scale feature extraction on the preprocessed visual acquisition image, and selecting spatially uniformly distributed features according to the extraction result. The dynamic fusion network is a network initialized by configuring the joint authentication trust degree of the UWB signal using the trajectory deviation of the pre-authentication trajectory point, which is used to receive the backtracking image, the key feature point, the motion data set and the UWB signal.

[0038] Specifically, first, the key feature points are extracted from the visual acquisition result, including obtaining the original image and performing preprocessing such as distortion removal and enhancement, constructing a multi-scale extraction resolution for multi-scale feature extraction, and then selecting spatially uniformly distributed features according to the extraction result to establish key feature points. The specific steps are described in detail in A510-A520.

[0039] Next, the dynamic fusion network is initialized using the pre-authentication trajectory point, including obtaining its trajectory deviation, configuring the joint authentication trust degree of the UWB signal based on the trajectory deviation, and then initializing and managing the dynamic fusion network based on the trust degree. The specific steps are described in detail in A541-A543.

[0040] Finally, the pre-authentication trajectory point is used to initialize the dynamic fusion network and perform joint pose correction, including performing low-frequency sliding window image backtracking positioning to determine whether the nearest neighbor sliding window image meets the preset interval threshold, if it meets, calling the preset period backtracking image synchronous input dynamic fusion network containing the image, activating the first image processing layer to perform key point matching with the nearest neighbor sliding window image to generate the first pose compensation, and simultaneously activating the IMU processing layer and the UWB processing layer to establish the second and third pose compensations based on the motion data set and the UWB signal respectively, and finally outputting the joint pose correction result according to the three outputs, and the specific steps are described in detail in A531-A537.

[0041] Further, the method provided in the embodiment of the application comprises the following steps A200:

[0042] A210: trajectory drift identification is performed using the pre-authentication trajectory point to establish a pre-drift amount, wherein the pre-drift amount is provided with a drift direction identifier.

[0043] A220: a tolerance zone map is established according to the work area and the UWB anchor point distribution, wherein each area in the tolerance zone map is provided with a drift threshold, a coverage available anchor point mapping, and a position anchor point occlusion.

[0044] A230: regional tolerance analysis of the pre-drift amount is performed using the tolerance zone map, if the regional tolerance analysis result is a pass result, then time sequence flight fitting is performed according to the pre-authentication trajectory point, and after the time sequence flight fitting result is matched with the calibration work trajectory, the mapped anchor point group is activated.

[0045] In the embodiment of the application, the trajectory drift is the spatial coordinate deviation of the pre-authentication trajectory point and the corresponding point of the calibration trajectory when the low-altitude unmanned aerial vehicle flies along the calibration work trajectory.

[0046] Specifically, when the low-altitude unmanned aerial vehicle works along the calibration trajectory, first, trajectory drift identification is performed based on the pre-authentication trajectory point. The pre-drift amount is established by calculating the spatial coordinate deviation, such as the difference of three-dimensional coordinates (x, y, z), between the current trajectory point and the corresponding point of the calibration trajectory. For example, if the calibration trajectory point coordinate is (10, 5, 3) and the current pre-authentication trajectory point coordinate is (10.3, 5, 3.1), the horizontal drift amount is 0.3 meters and the vertical drift amount is 0.1 meters, and the drift direction identifier is determined to be 15° east in the north direction. This step provides a quantitative basis for dynamic adjustment of the anchor point group through real-time deviation calculation.

[0047] Then, when constructing the tolerance zone map according to the work area layout and the UWB anchor point distribution, first, the work area is divided into a plurality of regular areas according to a 5*5 meter specification by using a grid division method, for example, a 20*16 grid matrix is formed in a rectangular area of 100*80 meters, and each grid corresponds to a unique coordinate identifier, such as (row, column). Then, for each grid area, a drift threshold is preset in combination with the positioning accuracy requirement of the unmanned aerial vehicle, for example, the maximum drift allowed in the horizontal direction is 0.5 meters, and the maximum drift allowed in the vertical direction is 0.3 meters. The threshold is determined by analyzing the deviation range of the 95% confidence interval in the historical trajectory data.

[0048] Next, based on the physical layout positions and signal coverage ranges of the UWB anchor points, an anchor point mapping available for covering each grid area is associated: by actually measuring the signal strength (RSSI) and time of arrival (ToA) of each anchor point in different grids, the top three anchor points with the highest signal quality scores (comprehensive RSSI≥-70dBm and ToA error≤0.05 seconds) are selected to form an anchor point group, for example, the grid area (2, 3) is associated with the anchor point group {A2, A3, A5} because the signal scores of the anchor points A2, A3, and A5 are 90, 88, and 85 respectively. At the same time, based on the anchor point signal failure records in the past 72 hours, RSSI<-70dBm and ToA error>0.1 second are defined as failure, the shielding probability of the anchor points in each grid area is calculated, for example, the anchor point A1 in the area (1, 1) has appeared 18 times in the historical data, and the total monitoring time is 72 times, so the shielding probability is calculated as 18 / 72=25%, and finally a three-dimensional tolerance zone map including the drift threshold, the anchor point group mapping, and the shielding probability is formed, which provides accurate spatial data support for subsequent regional tolerance analysis.

[0049] Finally, when using the tolerance zone map to perform regional tolerance analysis on the pre-shift amount, if the horizontal drift threshold of a certain area is 0.5 meters and the current drift amount is 0.3 meters, which is less than the threshold, it is determined to be passed. After passing, a cubic spline interpolation algorithm is used to fit the time sequence flight of the pre-authentication trajectory points, and the coordinates of the next 5 trajectory points are predicted. Then, a dynamic time warping (DTW) algorithm is used to match the predicted trajectory points with the calibration trajectory, and when the matching error is less than 0.2 meters, the mapped anchor point group is activated.

[0050] By trajectory drift dynamic identification, accurate modeling of the tolerance zone map, and time sequence flight fitting and matching, the correct activation rate of the anchor point group is improved, and the positioning error is reduced under the same shielding scenario, effectively solving the problems of anchor point group activation lag and misplacement in complex environments, and realizing dynamic and accurate configuration of anchor point group resources in the positioning process of low-altitude unmanned aerial vehicles.

[0051] Further, step A230 in the method provided by the embodiments of the present application comprises:

[0052] A231: If the area tolerance analysis result is a failure result, a backtracking window is established, trajectory backtracking is performed using the backtracking window, and a trajectory backtracking result is established.

[0053] A232: Time sequence flight prediction is performed using the trajectory backtracking result, and a time sequence flight prediction result is established.

[0054] A233: Redistribution based on position anchor occlusion and position distance is performed in the coverage available anchor mapping using the time sequence flight prediction result, and the anchor group is reactivated according to the redistribution result.

[0055] Optionally, when the area tolerance analysis result fails, a backtracking window is first established to capture trajectory history information. According to the flight speed of 2m / s of the unmanned aerial vehicle, the backtracking window can be set to include trajectory points in the past 5 seconds, i.e. 10 meters, for example, the backtracking window of the current time t covers 5 trajectory points from t-5s to t, and one point is recorded every 1 second. By extracting the three-dimensional coordinates (x, y, z) of these historical trajectory points and the corresponding UWB signal quality parameters such as RSSI and ToA, a sliding average filtering algorithm is used to denoise the trajectory, and a trajectory backtracking result is established, so that the coordinate error of the trajectory points is reduced.

[0056] Then, when performing time sequence flight prediction based on the trajectory backtracking result, a cubic spline interpolation algorithm is used to fit the backtracked trajectory, and the specific process is as follows: first, 5 trajectory points in the past 5 seconds are extracted from the backtracking window, each point containing a timestamp and three-dimensional coordinates , , , for example, the backtracked trajectory point set is {(t0, 10, 5, 3), (t1, 12, 5.2, 3.1), (t2, 14, 5.5, 3.2), (t3, 16, 5.8, 3.3), (t4, 18, 6.0, 3.4)}.

[0057] Then, a cubic polynomial is constructed between each two adjacent trajectory points:

[0058] where is the constant term of the piecewise polynomial at the node , corresponding to the coordinate value of the trajectory point at ; is the first-order coefficient, which determines the polynomial at The linear change rate near the node is related to the first derivative (velocity) of the trajectory at the node; is the quadratic term coefficient, which affects the curvature of the polynomial and is related to the second derivative (acceleration) of the trajectory at the node; is the cubic term coefficient, which is used to ensure the continuity of the second derivative of adjacent polynomials at the connection point, ensuring smooth transition of the trajectory. By solving a system of linear equations to ensure the continuity of the function value, the first derivative and the second derivative of adjacent polynomials at the connection point, for example, at t1, , , After completing the construction of the interpolation model, the predicted coordinates are calculated by substituting the future 5 time points, t+1 to t+5 seconds, i.e. relative time 1s to 5s, into each segmented polynomial to obtain the time series flight prediction result, which can provide high-precision trajectory prediction data for anchor group reassignment.

[0059] Finally, the time series flight prediction result is used for reassignment in the coverage available anchor mapping of the tolerance zone map. The anchor reassignment rule is set: preferentially selecting anchor points with an occlusion probability < 20% and a distance < 5 meters from the predicted trajectory point. For example, among the available anchor points corresponding to the region of the predicted trajectory point (16, 5.8, 3.3) at time t3, anchor point A has an occlusion probability of 15% and a distance of 4.2 meters, anchor point B has an occlusion probability of 25% and a distance of 3.8 meters, and anchor point C has an occlusion probability of 18% and a distance of 4.5 meters. Finally, anchor points A and C are selected to form a new anchor group. During the reassignment process, the anchor signal quality score is calculated to ensure that the score of the newly activated anchor group is > 80 points.

[0060] By capturing historical trajectories through the backtracking window, optimizing future paths through time series prediction, and dynamically reassigning anchor groups based on occlusion probability and distance, the correct reassignment rate of anchor groups is improved, the positioning error is reduced under the same occlusion scenario, and the problem of inability to adaptively adjust after anchor group failure in traditional methods is effectively solved, achieving dynamic fault tolerance and intelligent reconfiguration of anchor resources for low-altitude unmanned aerial vehicle positioning devices in complex environments.

[0061] Further, step A220 in the method provided by the embodiments of the present application includes:

[0062] A221: performing dynamic occlusion identification based on historical records on the work area to establish a failure frequency.

[0063] A222: configuring a dynamic update frequency for the tolerance zone map according to the failure frequency, and performing dynamic update management of the tolerance zone map using the dynamic update frequency.

[0064] In the embodiments of the present application, the failure frequency is statistical data established after dynamic occlusion identification based on historical records of the work area, and is used to reflect the signal failure of the UWB anchor points in the area.

[0065] Specifically, when establishing the tolerance zone map, first, dynamic occlusion identification based on historical records is carried out in the work area. By collecting historical data such as signal strength (RSSI) and signal arrival time (ToA) of the UWB anchor points in the past 72 hours, the failure times of each anchor point in different areas are counted. For example, in a 200 square meter work area, it is divided into 10x10 meter grid areas, and for the anchor points in each grid, when RSSI is lower than -70dBm and ToA error exceeds 0.1 seconds, it is determined as a failure, and thus the anchor point failure frequency of each area is established. For example, anchor point A in a certain area has failed 12 times in the past 24 hours, and its failure frequency is calculated as 0.5 times / hour, while anchor point B has a failure frequency of 0.1 times / hour in another area. Through statistical analysis of historical data, a quantitative basis is provided for dynamic updating of the tolerance zone map.

[0066] Then, the dynamic updating frequency of the tolerance zone map is configured according to the failure frequency. The mapping relationship between failure frequency and updating frequency is set as shown in Table 1. For example, the tolerance zone map updating frequency of the corresponding grid of the area where anchor point A is located is configured as 5 minutes according to the failure frequency of 0.5 times / hour, while the area where anchor point B is located is configured to be updated once every 15 minutes according to the failure frequency of 0.1 times / hour. By monitoring the change of failure frequency in real time, the updating frequency is dynamically adjusted to realize dynamic updating management of the tolerance zone map, so that the map data continuously matches the occlusion state of the work area.

[0067] By establishing the failure frequency based on dynamic occlusion identification based on historical records, and dynamically configuring the map updating frequency according to the failure frequency, the real-time performance of the tolerance zone map is improved, and in the same work environment, the anchor point group activation error rate is reduced, effectively solving the problem that the map data is out of touch with the actual occlusion situation under the traditional fixed updating mechanism, and realizing the adaptive matching of the tolerance zone map to the dynamic changes of the work environment.

[0068] Table 1: Mapping relationship table of failure and updating frequency

[0069]

[0070] Further, the step A500 in the method provided by the embodiments of the present application comprises:

[0071] A510: Obtain the original image of the visual acquisition result, and perform image preprocessing, which includes image distortion correction and image enhancement.

[0072] A520: construct a multi-scale extraction resolution, perform multi-scale feature extraction on the pre-processed image using the multi-scale extraction resolution, perform spatially uniformly distributed feature selection according to the multi-scale feature extraction result, and establish key feature points.

[0073] In the embodiments of the present application, the multi-scale extraction resolution refers to constructing a multi-layer image representation with different resolutions when performing feature extraction on the pre-processed image, so as to capture feature information of different scales in the image.

[0074] Specifically, when extracting key feature points from the visual acquisition result, first, the original image output by the visual acquisition device is acquired, such as an RGB image with a resolution of 1280x720, and image preprocessing is performed. The original image is de-distorted by Zhang Zhengyou calibration method, and the lens distortion coefficients k1 and k2 are respectively corrected to 0.05 and -0.03 to eliminate the image distortion caused by the physical characteristics of the lens; then histogram equalization and contrast limited adaptive histogram equalization (CLAHE) are used for image enhancement to improve the image gray entropy and significantly improve the recognition of environmental features.

[0075] Subsequently, a multi-scale extraction resolution is constructed, and a Gaussian pyramid model is usually used to generate 8 scale layers with a scale factor of 1.2. The pre-processed image is subjected to multi-scale feature extraction from 0.5 times the original resolution to 2 times the original resolution. Taking the ORB algorithm as an example, features are extracted on each scale layer using a FAST corner detector and a BRIEF descriptor. A single image can extract about 2000 feature points, of which the feature points extracted at the 1.0 times scale layer account for about 45%, the 0.5 times and 2 times scale layers account for 25% and 30% respectively, forming an ORB feature point set covering different spatial scales.

[0076] Then, according to the multi-scale feature extraction result, spatially uniformly distributed feature selection is performed, the image is divided into 16x16 grid regions, the feature points in each grid are sorted according to the response intensity, the top 3 strongest feature points are retained, and the redundant features gathered in the local area are removed. After this processing, the spatial distribution variance of the feature points in the image is reduced to 0.3, ensuring that there are at least 3 feature points in every 100x100 pixel area, and a uniformly distributed key feature point set is constructed.

[0077] By eliminating distortion and enhancing contrast through image preprocessing, combining multi-scale feature extraction and spatially uniformly distributed selection, the repeated detection rate of feature points is improved, effectively solving the problems of insufficient number and uneven distribution of feature points in traditional methods, providing high-quality visual feature input for subsequent linkage pose correction, and significantly improving the precision and environmental adaptability of low-altitude unmanned aerial vehicle dynamic positioning.

[0078] Further, the step A500 in the method provided by the embodiments of the present application comprises:

[0079] A531: performing low-frequency sliding window image backtracking to locate a nearest neighbor sliding window image.

[0080] A532: determining whether the nearest neighbor sliding window image meets a preset interval threshold.

[0081] A533: if the nearest neighbor sliding window image meets the preset interval threshold, calling a backtracking image in a preset period, synchronously inputting the backtracking image into the dynamic fusion network, and the backtracking image including the nearest neighbor sliding window image.

[0082] A534: activating a first image processing layer to perform key point matching using the key feature points and the nearest neighbor sliding window image, and generating a first pose compensation using a key point matching result and an inter-frame sliding window matching result in the backtracking image.

[0083] A535: activating an IMU processing layer to perform motion trajectory fitting based on a motion data set to establish a second pose compensation.

[0084] A536: activating a UWB processing layer to perform position fitting based on a UWB signal to establish a third pose compensation.

[0085] A537: outputting a linkage pose correction result according to the first pose compensation, the second pose compensation, and the third pose compensation.

[0086] In the embodiments of the application, the nearest neighbor sliding window image is the sliding window image closest to the current time and is located by performing low-frequency sliding window image backtracking, and is used for subsequent key point matching and pose compensation calculation. The IMU processing layer is a functional module for processing a motion data set collected by an IMU sensor, and a second pose compensation is established by motion trajectory fitting based on the motion data set. The UWB processing layer is a functional module for position fitting based on a UWB signal, and a third pose compensation is established by processing the UWB signal.

[0087] Specifically, first, low-frequency sliding window image backtracking is performed, for example, a sliding window interval of 0.5 seconds is set, and the sliding window image closest to the current time is located by searching the image sequence. Taking the collection frequency of 30 frames / second of the unmanned aerial vehicle 30 as an example, if the current time is t, the nearest neighbor sliding window image is usually the image at the time of t-0.5 seconds. This step reduces the calculation amount by using the sliding window mechanism, and reduces the calculation load compared with the traditional full-frame processing.

[0088] When it is judged that the nearest neighbor sliding window image meets the preset interval threshold (such as time interval ≤ 0.5 seconds and image feature change rate < 30%), the backtracking image in the preset period (such as the past 5 seconds) is called and input into the dynamic fusion network synchronously with the nearest neighbor sliding window image. For example, when the unmanned aerial vehicle flies at a speed of 2 meters / second, the backtracking image in 5 seconds covers a track range of about 10 meters, ensuring that the input image sequence can reflect the recent motion characteristics and provide spatiotemporal continuous visual data for pose compensation.

[0089] After the first image processing layer is activated, ORB feature points are extracted from the preprocessed image, about 1000 per frame, and key point matching is performed on these key feature points and the nearest neighbor sliding window image. The Hamming distance matching algorithm is used to calculate the similarity of the feature point descriptors, and the Hamming distance threshold is set to 50 to screen out about 50 effective matching point pairs. Then, combined with the inter-frame sliding window matching results in the backtracking image, the matching error between adjacent frames is checked, and the matching error is required to be less than 2 pixels, so as to eliminate the mismatched points. Based on the above screened matching results, the first pose compensation is generated by calculating the position transformation relationship of the feature points between different frames, including translation compensation (such as 0.1 meters) and rotation compensation (such as 1°), thereby providing visual level compensation data support for subsequent linkage pose correction.

[0090] Meanwhile, after the IMU processing layer is activated, Kalman filtering is performed on the three-axis acceleration (range ± 5g) and three-axis angular velocity (range ± 2000° / s) data collected by the IMU sensor at a sampling rate of 100Hz. First, a state space model is constructed, taking the position, velocity and attitude angle of the unmanned aerial vehicle as state variables, and the acceleration and angular velocity as observation variables, wherein the noise density of the accelerometer is 0.01g / √Hz. During the filtering process, the motion state at the next time is estimated through the state prediction equation, and the actual collected acceleration and angular velocity data are compared with the predicted value to calculate the noise covariance matrix to optimize the state estimation. After Kalman filtering, the second pose compensation is generated, and the motion trajectory prediction error is reduced to 0.1 meters / second, effectively suppressing high-frequency noise interference, providing high-precision motion trajectory constraints for pose correction, and realizing real-time accurate fitting of the motion state of the unmanned aerial vehicle.

[0091] After the UWB processing layer is activated, the layer first receives the UWB signal transmitted by the low-altitude unmanned aerial vehicle and returned by the activated anchor group, calculates the time of arrival (ToA) or time difference (TDoA) of the signal by using a multilateration algorithm, and solves the real-time position of the unmanned aerial vehicle in combination with the coordinates (known three-dimensional position) of the laid UWB anchor points. The UWB signal itself has a positioning accuracy of 0.1 meters, and the processing layer will denoise and filter the original signal, such as using median filtering to eliminate burst noise, and adjusting the weight coefficient of position fitting according to the joint authentication trust degree configured by the trajectory deviation of the pre-authentication trajectory point. When the trust degree is 90%, the position fitting error is controlled within 0.15 meters through a weighted fusion algorithm, for example, in the calculation of the final position, the weight of the UWB signal accounts for 0.3, which is complementary to the visual weight 0.4 and the IMU weight 0.3 data, thereby establishing the third pose compensation and providing high-precision position constraints for the linkage pose correction.

[0092] Finally, according to the three types of pose compensation results, the linkage pose correction result is output through weighted fusion, with a visual weight of 0.4, an IMU weight of 0.3, and a UWB weight of 0.3.

[0093] Through dynamic fusion of multiple sources of data, the problem of insufficient robustness of a single sensor is effectively solved, and high-precision pose correction and environmental adaptability of the low-altitude unmanned aerial vehicle in a dynamic scene are realized.

[0094] Further, step A532 in the method provided in the embodiments of the application includes:

[0095] A532-1: If the nearest neighbor sliding window image cannot meet the preset interval threshold, activate the visual acquisition device to perform continuous image frame acquisition and establish a continuous frame set.

[0096] A532-2: The second image processing layer uses the key feature points and the continuous frame set to perform key point matching and establish the fourth pose compensation.

[0097] A532-3: Output the linkage pose correction result according to the second pose compensation, the third pose compensation, and the fourth pose compensation.

[0098] In one embodiment, when the nearest neighbor sliding window image does not meet the preset interval threshold, such as when the time interval exceeds 0.5 seconds or the feature change rate exceeds 50%, the visual acquisition device is automatically activated to perform continuous image frame acquisition at a frequency of 30 frames per second, and a continuous frame set containing 10-15 frames is established. For example, when the image interval becomes large due to high-speed flight of the unmanned aerial vehicle, the continuously acquired images can cover a motion trajectory of 0.3-0.5 seconds, providing dense visual data support for subsequent processing.

[0099] After the second image processing layer is activated, a FLANN (Fast Library for Approximate Nearest Neighbors) algorithm is used to perform feature matching on the key feature points (about 1000 ORB feature points per frame) and the continuous frame set. FLANN first constructs an index structure of a high-dimensional feature space, such as a KD tree or a hierarchical clustering tree, and quickly finds candidate matching points for each feature point through an approximate nearest neighbor search strategy. For the BRIEF binary descriptor of the ORB feature, the similarity between the feature points is calculated using the Hamming distance, and when the distance is less than 50, the feature points are determined to be similar features. After FLANN screening, about 800 pairs of valid matching points per frame can be retained, and this process greatly reduces the candidate set of false matches through fast indexing and distance threshold filtering, laying a foundation for subsequent accurate matching.

[0100] Then, on the basis of the FLANN preliminary matching, a RANSAC (Random Sample Consensus) algorithm is used to remove false matching points. This algorithm finds the optimal transformation model through iteration: first, randomly select 4 pairs of matching points as the minimum sample set, calculate the pose transformation matrix between adjacent frames, such as the homography matrix or the essential matrix, and then test all matching points with the matrix, and determine the points that meet the preset error threshold (such as a projection error < 2 pixels) as inliers. After multiple iterations, the model with the most inliers is selected as the final transformation matrix, and at this time the false matching rate can be reduced from the initial 20% to less than 5%. Finally, the accurate pose transformation is recalculated based on all inliers, and the fourth pose compensation is established, such as a translation compensation of 0.2 meters and a rotation compensation of 2°, effectively capturing the attitude changes of the unmanned aerial vehicle during rapid movement.

[0101] According to the second pose compensation output by the IMU processing layer, the motion trajectory is fitted based on 100Hz sampling, with an error of 0.1 meters / second; the third pose compensation of the UWB processing layer has a positioning accuracy of 0.15 meters, and the fourth pose compensation is again fused through weighted fusion, with a visual weight of 0.5, an IMU weight of 0.3, and a UWB weight of 0.2, to output the linkage pose correction result.

[0102] Through continuous frame acquisition and multi-frame feature matching, high-precision pose correction can still be maintained when the visual information is sparse, effectively solving the problem of insufficient robustness of traditional methods in dynamic scenes. Through dense visual data supplementation and multi-source data fusion, stable positioning of low-altitude unmanned aerial vehicles in complex motion states is achieved.

[0103] Further, the step A500 in the method provided by the embodiments of the present application comprises:

[0104] A541: Obtain a trajectory deviation of the pre-authentication trajectory point.

[0105] A542: Configure a joint authentication trust degree of the UWB signal according to the trajectory deviation.

[0106] A543: initializing management of the dynamic fusion network based on the joint authentication trust degree.

[0107] In the embodiments of the present application, the joint authentication trust degree is a weight parameter configured according to the trajectory deviation of the pre-authentication trajectory point to represent the reliability of the UWB signal.

[0108] Optionally, first, the three-dimensional coordinate difference (such as x, y, z axis deviation) between the pre-authentication trajectory point and the corresponding point of the calibration operation trajectory is calculated to obtain the trajectory deviation. Taking a certain trajectory as an example, if the calibration trajectory point coordinate is (10, 5, 3) and the pre-authentication trajectory point is (10.3, 5, 3.1), the horizontal deviation is 0.3 meters and the vertical deviation is 0.1 meters, forming a trajectory deviation vector. This step provides a quantitative basis for subsequent trust degree configuration through real-time deviation calculation.

[0109] Then, the joint authentication trust degree of the UWB signal is configured according to the trajectory deviation. The mapping relationship between the deviation and the trust degree is set as shown in Table 2. For example, if the trajectory deviation at a certain time is 0.3 meters, the joint authentication trust degree of the UWB signal is 90%, which is used to represent the reliability weight of the UWB signal. By dynamically adjusting the trust degree, the weight of the UWB signal is reduced when the trajectory deviation is large, avoiding the influence of false signals on positioning accuracy.

[0110] The initialization management of the dynamic fusion network is based on the joint authentication trust degree. The trust degree is used as the weight parameter of the network input layer. For example, when initializing, if the trust degree is 90%, the weight coefficient of the UWB signal processing branch is set to 0.9, and the weights of the visual and IMU branches are set to 0.5 and 0.6 respectively, and the dynamic fusion of multi-source data is realized through weight distribution.

[0111] By dynamically configuring the UWB signal trust degree based on the trajectory deviation, and initializing the fusion network, the problem of insufficient robustness caused by fixed signal trust degree in traditional methods is effectively solved, the adaptive initialization of the dynamic fusion network to the UAV trajectory deviation is realized, and the environmental adaptability of the positioning device is improved.

[0112] Table 2: Mapping relationship table of deviation and trust degree

[0113]

[0114] Further, the method provided in the embodiments of the present application comprises the following steps:

[0115] A551: matching the warning level according to the joint position correction result, and establishing a warning level matching result.

[0116] A552: issuing a warning signal using the warning level matching result, and performing warning feedback.

[0117] In one embodiment, after generating the linkage pose correction result, first, the pre-warning level is matched according to the pose deviation amount in the correction result. Three pre-warning threshold values are set: when the horizontal deviation is ≤0.5 meters and the vertical deviation is ≤0.3 meters, it is a safe level (no pre-warning); when the horizontal deviation is 0.5-1 meter or the vertical deviation is 0.3-0.5 meters, it is a yellow pre-warning; and when the horizontal deviation is >1 meter or the vertical deviation is >0.5 meters, it is a red pre-warning. For example, if the correction result shows that the horizontal deviation of the current position of the UAV from the calibration trajectory is 0.8 meters and the vertical deviation is 0.4 meters, it is matched to the yellow pre-warning level. This grading standard is obtained by statistical analysis of historical accident data, and those skilled in the art can adjust the corresponding numerical values according to the actual situation.

[0118] Then, the pre-warning signal is reported and feedback is performed using the pre-warning level matching result. When the yellow pre-warning is triggered, an audible and visual alarm is triggered, such as a beeper that sounds every 2 seconds and a yellow flashing LED light on the body, and a speed reduction command is sent to the flight control system to reduce the flight speed from 2 meters / second to 1 meter / second. When the red pre-warning is triggered, in addition to the loud and bright alarm, the beeper continuously sounds and the red LED flashes, the automatic return program is activated, and the UAV is controlled to return to the starting point along the nearest path.

[0119] Through the multi-level threshold matching and grading feedback strategy, the pre-warning accuracy is improved, and the closed-loop management from error detection to risk disposal is realized in combination with the linkage pose correction result, which effectively reduces the collision risk in low-altitude UAV operation and improves the safety and reliability of the device.

[0120] In summary, the low-altitude UAV dynamic visual positioning method under the super wide band provided by the embodiments of the present application has the following technical effects:

[0121] The present application collects environment images in the calibration trajectory region and the tolerance zone map coverage region of the operation area, obtains key feature point data through multi-scale feature extraction and spatial uniform distribution selection, calculates the spatial distribution and matching relationship of the feature points, and performs pose correction in combination with the drift threshold and anchor point mapping of the tolerance zone map, thereby accurately realizing the dynamic visual positioning of the low-altitude UAV under the super wide band, making the low-altitude UAV pose correction result more accurate and reliable, achieving accurate positioning of the low-altitude UAV pose, realizing accurate positioning in complex environments and avoiding failure, and meeting the technical effect of meeting the demand for accurate operation.

[0122] Embodiment two, as shown in Figure 2 Based on the same inventive concept as the aforementioned embodiment one, the present application embodiment provides a low-altitude UAV dynamic visual positioning device under a super wide band, which comprises:

[0123] Anchors group configuration module 1, used for deploying UWB anchors in a work area, and obtaining a calibration work trajectory of a low-altitude unmanned aerial vehicle, and performing signal fitting according to the calibration work trajectory and the distribution of UWB anchors, and configuring an anchor group mapped with the calibration work trajectory.

[0124] Anchors group activation module 2, used for activating the anchor group according to a pre-authentication trajectory point of the low-altitude unmanned aerial vehicle when the low-altitude unmanned aerial vehicle performs a work task along the calibration work trajectory.

[0125] Anchors group anomaly identification module 3, used for receiving a UWB signal emitted by the low-altitude unmanned aerial vehicle by using the activated anchor group, and performing anomaly identification of the UWB signal in the anchor group.

[0126] Motion data set acquisition module 4, used for activating a visual acquisition device of the low-altitude unmanned aerial vehicle to capture environmental features and acquiring a motion data set of an IMU sensor if the anomaly identification result triggers a preset anomaly threshold.

[0127] Linkage pose correction result acquisition module 5, used for initializing a dynamic fusion network by using the pre-authentication trajectory point after extracting key feature points from the visual acquisition result, performing linkage pose correction based on the key feature points, the motion data set and the UWB signal, and generating a linkage pose correction result.

[0128] Further, the anchors group activation module 2 is used to perform the following steps:

[0129] Trajectory drift identification is performed by using the pre-authentication trajectory point, and a pre-drift amount is established, wherein the pre-drift amount is provided with a drift direction identifier; a tolerance zone map is established according to the work area and the distribution of UWB anchors, each region in the tolerance zone map is provided with a drift threshold, a coverage available anchor mapping and a position anchor occlusion; region tolerance analysis of the pre-drift amount is performed by using the tolerance zone map, if the region tolerance analysis result is a pass result, then time sequence flight fitting is performed according to the pre-authentication trajectory point, and after the time sequence flight fitting result is matched with the calibration work trajectory, the mapped anchor group is activated.

[0130] Further, the anchors group activation module 2 is used to perform the following steps:

[0131] If the area tolerance analysis result is a failure result, a backtracking window is established, trajectory backtracking is performed using the backtracking window, and a trajectory backtracking result is established; time sequence flight prediction is performed using the trajectory backtracking result, and a time sequence flight prediction result is established; redistribution based on position anchor shielding and position distance is performed in the covering available anchor mapping using the time sequence flight prediction result, and an anchor group is reactivated according to the redistribution result.

[0132] Further, the anchor group activation module 2 is configured to perform the following steps:

[0133] The job area is subjected to dynamic shielding identification based on historical records, and a failure frequency is established; the dynamic update frequency of the tolerance zone map is configured according to the failure frequency, and dynamic update management of the tolerance zone map is performed using the dynamic update frequency.

[0134] Further, the linkage pose correction result acquisition module 5 is configured to perform the following steps:

[0135] An original image of the visual acquisition result is obtained, image preprocessing is performed, the image preprocessing includes image de-warping and image enhancement; a multi-scale extraction resolution is constructed, multi-scale feature extraction of the preprocessed image is performed using the multi-scale extraction resolution, spatially uniformly distributed feature selection is performed according to the multi-scale feature extraction result, and key feature points are established.

[0136] Further, the linkage pose correction result acquisition module 5 is configured to perform the following steps:

[0137] Low-frequency sliding window image backtracking is performed, and the nearest neighbor sliding window image is located;

[0138] It is judged whether the nearest neighbor sliding window image meets a preset interval threshold; if the nearest neighbor sliding window image meets the preset interval threshold, a backtracking image in a preset period is called, the backtracking image is input into the dynamic fusion network, the backtracking image includes the nearest neighbor sliding window image; a first image processing layer is activated to perform key point matching using the key feature points and the nearest neighbor sliding window image, a first pose compensation is generated using the key point matching result and an inter-frame sliding window matching result in the backtracking image; an IMU processing layer is activated to perform motion trajectory fitting based on a motion data set, a second pose compensation is established; a UWB processing layer is activated to perform position fitting based on a UWB signal, a third pose compensation is established; and a linkage pose correction result is output according to the first pose compensation, the second pose compensation, and the third pose compensation.

[0139] Further, the linkage pose correction result acquisition module 5 is configured to perform the following steps:

[0140] If the nearest neighbor sliding window image cannot satisfy the preset interval threshold, a visual acquisition device is activated to perform continuous image frame acquisition to establish a continuous frame set; a second image processing layer is activated to perform key point matching using the key feature points and the continuous frame set to establish a fourth pose compensation; and a linkage pose correction result is output according to the second pose compensation, the third pose compensation and the fourth pose compensation.

[0141] Further, the linkage pose correction result acquisition module 5 is configured to perform the following steps:

[0142] Obtain a trajectory deviation of the pre-authentication trajectory point; configure a joint authentication trust degree of the UWB signal according to the trajectory deviation; and perform initialization management of the dynamic fusion network based on the joint authentication trust degree.

[0143] Further, the linkage pose correction result acquisition module 5 is configured to perform the following steps:

[0144] Perform warning level matching of a pose warning according to the linkage pose correction result to establish a warning level matching result; output a warning signal using the warning level matching result and perform warning feedback.

[0145] The low-altitude unmanned aerial vehicle dynamic visual positioning device under the ultra-wide band provided by the embodiment of the application can perform the low-altitude unmanned aerial vehicle dynamic visual positioning method under the ultra-wide band provided by any embodiment of the application, has the function modules and beneficial effects corresponding to the execution method.

[0146] Although various references are made to certain modules in the device according to the embodiments of the application in the present application, however, any number of different modules can be used and run on the user terminal and / or server, and each unit and module included is only divided according to the functional logic, but is not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual differentiation, and do not limit the protection scope of the application.

[0147] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. A dynamic visual positioning method for low-altitude unmanned aerial vehicles (UAVs) under ultra-wideband conditions, characterized in that, The method includes: UWB anchor points are deployed in the operation area, and the calibration operation trajectory of the low-altitude UAV is obtained. The signal is fitted according to the calibration operation trajectory and the distribution of UWB anchor points, and the anchor point group mapped to the calibration operation trajectory is configured. When the low-altitude UAV performs its mission along the calibrated operational trajectory, the anchor point group is activated and configured based on the pre-authentication trajectory points of the low-altitude UAV. Utilize the activated anchor point group to receive UWB signals transmitted by low-altitude UAVs and perform anomaly identification of UWB signals within the anchor point group; If the abnormal identification result triggers the preset abnormal threshold, the visual acquisition device of the low-altitude UAV is activated to capture environmental features and acquire the motion dataset of the IMU sensor. After extracting key feature points from the visual acquisition results, the dynamic fusion network is initialized using the pre-authenticated trajectory points, and the linkage pose correction based on key feature points, motion dataset, and UWB signal is performed to generate linkage pose correction results. The step of activating the anchor point group configuration based on the pre-authentication trajectory points of the low-altitude UAV includes: The trajectory drift is identified using the pre-authentication trajectory points, and a pre-drift amount is established, wherein the pre-drift amount is set with a drift direction indicator; A tolerance zone map is established based on the work area and the distribution of UWB anchor points. Each area in the tolerance zone map is set with a drift threshold, coverage of available anchor points, and location anchor point occlusion. The tolerance zone map is used to perform a regional tolerance analysis of the pre-drift amount. If the regional tolerance analysis result is a pass result, then a time-series flight fitting is performed based on the pre-certified trajectory points. After matching the calibration operation trajectory with the time-series flight fitting result, the mapped anchor point group is activated.

2. The dynamic visual positioning method for low-altitude unmanned aerial vehicles under ultra-wideband conditions as described in claim 1, characterized in that, The regional tolerance analysis of the preceding drift amount using the tolerance zone map includes: If the regional tolerance analysis result is a failure, a backtracking window is established, and the trajectory is backtracked using the backtracking window to establish the trajectory backtracking result; Using the trajectory backtracking results, time-series flight prediction is performed to establish time-series flight prediction results; The time-series flight prediction results are used to perform a redistribution of anchor points based on location anchor occlusion and location distance in the available anchor point map, and the anchor point group is reactivated according to the redistribution results.

3. The dynamic visual positioning method for low-altitude UAVs under ultra-wideband conditions as described in claim 1, characterized in that, The step of establishing a tolerance zone map based on the operating area and the distribution of UWB anchor points includes: The work area is dynamically occluded based on historical records to establish a failure frequency. Configure the dynamic update frequency of the tolerance zone map according to the failure frequency, and use the dynamic update frequency to manage the dynamic update of the tolerance zone map.

4. The dynamic visual positioning method for low-altitude UAVs under ultra-wideband conditions as described in claim 1, characterized in that, The extraction of key feature points from the visual acquisition results includes: The original image of the visual acquisition result is acquired, and image preprocessing is performed, including image distortion correction and image enhancement. A multi-scale extraction resolution is constructed, and multi-scale feature extraction of the preprocessed image is performed using the multi-scale extraction resolution. Based on the multi-scale feature extraction results, spatially uniformly distributed features are selected to establish key feature points.

5. The dynamic visual positioning method for low-altitude unmanned aerial vehicles under ultra-wideband conditions as described in claim 4, characterized in that, The process of initializing the dynamic fusion network using pre-authenticated trajectory points and performing coordinated pose correction based on key feature points, motion datasets, and UWB signals includes: Perform low-frequency sliding window image backtracking to locate the nearest neighbor sliding window image; Determine whether the nearest neighbor sliding window image meets the preset interval threshold; If the nearest neighbor sliding window image meets the preset interval threshold, then the backtracking image within the preset period is called and the backtracking image is synchronously input into the dynamic fusion network. The backtracking image includes the nearest neighbor sliding window image. The first image processing layer is activated to perform key point matching using the key feature points and the nearest neighbor sliding window image. The first pose compensation is generated using the key point matching result and the inter-frame sliding window matching result in the backtracking image. Activate the IMU processing layer to fit motion trajectories based on the motion dataset and establish a second pose compensation. Activate the UWB processing layer to perform position fitting based on UWB signals and establish a third pose compensation. The linked pose correction result is output based on the first pose compensation, the second pose compensation, and the third pose compensation.

6. The dynamic visual positioning method for low-altitude UAVs under ultra-wideband conditions as described in claim 5, characterized in that, The step of determining whether the nearest neighbor sliding window image meets the preset interval threshold includes: If the nearest neighbor sliding window image cannot meet the preset interval threshold, the visual acquisition device is activated to perform continuous image frame acquisition and establish a continuous frame set. The second image processing layer is activated to perform key point matching using the key feature points and the set of consecutive frames, and to establish a fourth pose compensation. The output of the linkage pose correction result is based on the second pose compensation, the third pose compensation, and the fourth pose compensation.

7. The dynamic visual positioning method for low-altitude unmanned aerial vehicles under ultra-wideband conditions as described in claim 1, characterized in that, The initialization of the dynamic fusion network using pre-authenticated trajectory points includes: Obtain the trajectory deviation of the pre-authentication trajectory points; Configure the joint authentication trust level of the UWB signal based on the trajectory deviation; The initialization management of the dynamic fusion network is based on the aforementioned joint authentication trust level.

8. The dynamic visual positioning method for low-altitude unmanned aerial vehicles under ultra-wideband conditions as described in claim 1, characterized in that, The generated linkage pose correction result includes: Based on the linkage pose correction results, the pose warning level is matched, and a warning level matching result is established. The warning signal is reported based on the warning level matching result, and warning feedback is executed.

9. A dynamic visual positioning device for low-altitude unmanned aerial vehicles under ultra-wideband conditions, characterized in that, The apparatus for implementing the ultra-wideband low-altitude UAV dynamic visual positioning method according to any one of claims 1-8, the apparatus comprising: Anchor point group configuration module is used to deploy UWB anchor points in the operation area, acquire the calibration operation trajectory of the low-altitude UAV, perform signal fitting based on the calibration operation trajectory and the distribution of UWB anchor points, and configure anchor point groups mapped to the calibration operation trajectory. An anchor point group activation module is used to configure the anchor point group activation based on the pre-authentication trajectory points of the low-altitude UAV when the low-altitude UAV performs the operation task along the calibrated operation trajectory. The anchor point group anomaly identification module is used to receive UWB signals transmitted by low-altitude UAVs using the activated anchor point group and to perform anomaly identification of UWB signals within the anchor point group. The motion dataset acquisition module is used to activate the visual acquisition device of the low-altitude UAV to capture environmental features and acquire the motion dataset of the IMU sensor if the anomaly identification result triggers a preset anomaly threshold. The linkage pose correction result acquisition module is used to extract key feature points from the visual acquisition results, initialize the dynamic fusion network with the pre-authenticated trajectory points, perform linkage pose correction based on key feature points, motion dataset, and UWB signal, and generate linkage pose correction results.

Images