Method for improving target positioning accuracy based on UAV video data

Abstract

The present invention discloses a method for improving target positioning accuracy based on drone video data. The method is aimed at processing drone aerial video data around a stationary target and aims to solve the technical problem that the existing pod calibration method is not feasible under uncontrolled conditions. The method comprises the following steps: (1) confirming the trajectory of the target and the drone and obtaining positioning parameters; (2) performing single-image positioning on the target and confirming that the plane coordinate distribution of the positioning output is approximately circular; (3) inputting the positioning result into an error fitting algorithm to obtain the center coordinate of the circle as a virtual control point; (4) constructing an error equation based on the virtual control point and adjusting the installation offset angle and the target elevation; (5) using the center coordinate of the circle as the plane coordinate of the target and the target elevation as the elevation coordinate of the target to output the three-dimensional coordinate of the target. The positioning accuracy improvement method of the present invention is significantly better than the mean filtering method and can simultaneously improve the positioning accuracy in the plane and elevation directions.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) target positioning, and in particular to a method for improving target positioning accuracy based on UAV video data. Background Art

[0002] Electro-optical pods typically consist of cameras, thermal imagers, and laser rangefinders. They can monitor target position changes in real time and record target images and related parameter information. When tracking targets, electro-optical pods typically employ optical tracking, flight control, and data transmission technologies to precisely locate and track the target. UAV target positioning methods based on laser ranging and optical sensors are commonly used in the military, enabling accurate positioning and tracking of targets on land and at sea. Laser ranging, a key step in target positioning, calculates the target's distance by emitting a laser beam and receiving the reflected signal. The laser ranging process primarily involves emitting a laser beam, receiving the reflected signal, performing signal processing, and calculating distance and position. While this method offers high accuracy, it is significantly affected by weather. Optical sensors primarily handle target detection, tracking, and recording.

[0003] The target positioning method, based on the fusion of multiple sensors such as laser rangefinders and cameras, involves the following steps: First, the laser rangefinder on the flight platform acquires target distance information using methods such as triangulation and time difference measurement. Next, the visible light camera or thermal imager built into the electro-optical pod captures target images, and target detection and tracking is performed based on image features. After acquiring the target distance and position information, the laser rangefinder and electro-optical pod data are integrated and processed using a multi-sensor fusion algorithm to obtain more accurate target distance and position information. Finally, this processed target positioning information is transmitted to the flight control system, enabling precise positioning and tracking of the target.

[0004] This drone target positioning method offers advantages such as high precision and real-time performance, making it suitable for target reconnaissance and monitoring in complex environments. It has been widely used in military reconnaissance, monitoring, and strike operations. However, it is subject to significant limitations in weather and environmental factors, requiring appropriate selection and optimization for different situations.

[0005] Coordinate transformation-based drone target positioning is a technology commonly used in industrial and military fields. It can capture a target using an onboard camera and calculate the target's precise position in the world coordinate system through a series of coordinate transformations. Target positioning based on a single image is a common method. The specific process is as follows: First, the target is captured using a camera mounted on the drone, and the target's coordinates in the image plane coordinate system are obtained. These coordinates can be obtained by calculating information such as the camera's intrinsic parameters and the image's pixel coordinates. Next, the coordinates in the image plane coordinate system are transformed into the camera coordinate system, taking into account the effects of factors such as the camera's focal length and distortion on the image coordinates. The camera model is then used for coordinate transformation. Finally, the coordinates in the camera coordinate system are transformed into the pod coordinate system. This conversion needs to take into account factors such as the relative position and attitude between the pod and the carrier aircraft, and uses methods such as Euler angles or quaternions for coordinate conversion. The coordinates in the pod coordinate system are converted to the carrier aircraft's body coordinate system, and the relative position and attitude between the carrier aircraft's body coordinate system and the pod coordinate system need to be considered. Finally, the coordinates in the carrier aircraft's body coordinate system are converted to the carrier aircraft's geographic coordinate system, and factors such as the position and attitude of the carrier aircraft on Earth need to be considered. The coordinate conversion is performed using information such as longitude, latitude, and altitude. Through this series of coordinate conversions, the target's coordinates in the image plane coordinate system can be converted to the carrier aircraft's geographic coordinate system, thereby accurately positioning the target. This positioning process has high requirements for camera internal and external parameters and the carrier aircraft's attitude.

[0006] UAV pod calibration involves determining the precise position and attitude of the drone pod relative to the platform through a series of experiments and calculations, enabling high-precision positioning and tracking of targets. Pod calibration is often a key step in drone target positioning. Its purpose is to improve the accuracy and stability of high-precision positioning. Pod calibration typically involves data collection, mathematical model development, parameter estimation, and calibration verification, and requires optimization and refinement for specific application scenarios.

[0007] Online calibration for high-precision UAV target positioning involves processing and calculating a series of observation data during actual flight to update the position and attitude information of the UAV pod relative to the platform in real time, thereby achieving high-precision positioning and tracking of the target. Compared to traditional offline calibration methods, online calibration offers advantages such as real-time, speed, and convenience. The online calibration process primarily involves the following steps: Data Collection: First, real-time pod data is collected, including information such as the attitude and position of the pod's onboard equipment. This data can be acquired in real time using sensors such as an inertial measurement unit (IMU), a GNSS receiver, and a camera. Real-time Calculation: Based on this collected real-time data and a mathematical model, the precise position and attitude information between the pod and the platform can be calculated in real time. Calibration Verification: After the real-time calculation is complete, the online calibration results need to be calibrated and verified. This process can be verified by measuring the actual position using sensors such as onboard LiDAR.

[0008] The advantage of online calibration is that it can update the position and attitude information of the equipment in the pod in real time during actual flight, allowing the drone to more accurately locate and track targets. However, online calibration also has some disadvantages. First, because real-time calculations consume a large amount of computing resources and energy, efficient computing methods and hardware are required. Second, due to factors such as environmental complexity and noise interference, online calibration results may contain errors, requiring recalibration and verification.

[0009] Both offline and online pod calibration methods require the prior deployment of control points within the flight area, which is labor-intensive and resource-intensive. This makes them unsuitable for specialized scenarios like emergency missions and incapable of meeting rapid response requirements. Therefore, other, faster and more convenient calibration methods are needed. Mean filtering is a commonly used signal processing method that can improve positioning accuracy to a certain extent. However, mean filtering cannot eliminate positioning errors in the elevation direction. Therefore, in practical applications, more advanced positioning algorithms and technologies are needed to improve the accuracy and effectiveness of pod calibration. Summary of the Invention

[0010] The technical problem to be solved by the present invention is to provide a method for improving target positioning accuracy based on drone video data, so as to solve the technical problems that the existing pod calibration method relies on ground control points, consumes manpower and material resources, is not suitable for emergency tasks, cannot respond quickly, is not fast and convenient enough, and has low accuracy and effectiveness.

[0011] To address the above technical problems, the inventors, after extensive research, have proposed a high-precision three-dimensional coordinate positioning method for processing drone aerial video data orbiting a stationary target. This method is applicable when the drone is filming along a nearly circular trajectory around the target. This method comprehensively utilizes the drone's flight control data, pod information, and laser ranging data to achieve high-precision positioning through two components: a planar accuracy improvement method based on error fitting, and a method for jointly resolving elevation and installation offset angles using virtual control points. Specifically, the planar accuracy improvement method based on error fitting post-processes and optimizes the results of single-image positioning based on laser ranging to improve planar accuracy, and uses the improved positioning results as virtual plane control points. The method for jointly resolving elevation and installation offset angles based on virtual control points utilizes virtual plane control points, an imaging model, and laser ranging information to jointly resolve the installation offset angle between the drone's pod's onboard equipment and the pod platform, as well as the target elevation, to obtain the target's high-precision three-dimensional coordinates. The primary advantage of this method lies in its comprehensive utilization of multiple sources of information in drone aerial video data, ensuring high-precision positioning while reducing the workload of data collection and processing. At the same time, the method based on virtual control points can also effectively solve the positioning error problem in the elevation direction.

[0012] The present invention adopts the following technical solutions:

[0013] Design a method to improve target positioning accuracy based on UAV video data, including the following steps:

[0014] (1) Confirming that the target is stationary, the UAV navigates around the target in a nearly circular trajectory to obtain the target's positioning parameters, i.e., the parameters on which positioning depends;

[0015] (2) performing single-image positioning of the target based on laser ranging to confirm that the plane coordinate distribution of the positioning output is approximately circular;

[0016] (3) Inputting the result of the single-image positioning into the error fitting algorithm, fitting to obtain the coordinates of the circle center as a virtual control point;

[0017] (4) constructing an error equation based on the virtual control points, and adjusting and solving the installation offset angle and the target elevation;

[0018] (5) The center coordinates of the circle in step (3) are used as the plane coordinates of the target, and the target elevation in step (4) is used as the elevation coordinates of the target, and the three-dimensional coordinates of the target are output.

[0019] Preferably, in step (1), the positioning parameters include flight control data, pod information, laser ranging data and installation matrix.

[0020] Preferably, in step (2), the input of the single-image positioning is the coordinate position of the target on the image, the laser ranging result and the positioning parameters, and the output is the coordinates of the target in the world coordinate system. After a series of coordinate transformations, the longitude L, latitude B and elevation H of the target are finally obtained.

[0021] Preferably, in step (3), the error fitting algorithm is the equation of the fitting circle:

[0022] x 2 +y 2 +ax+by+c=0

[0023] Among them, a, b, c are constant terms; a set of coordinates containing K positioning results (x i ,y i ), i = 1, 2, 3, ..., K, and solve using linear least squares; after finding a, b, and c, the center of the circle is:

[0024] Preferably, in step (4), the error equation is V=AΔ-L; wherein V, A, L, and Δ are the residual, coefficient matrix, constant term, and unknown correction number of the error equation, respectively; and according to the error equation, the installation offset angle and the target elevation are solved by the least squares method.

[0025] Preferably, in step (5), the single image positioning is performed using the installation offset angle to update the three-dimensional coordinates of the target.

[0026] Preferably, the single image positioning algorithm is: the coordinates (x, y, z) of the local rectangular coordinate system of the target are obtained by the following conversion: T :

[0027]

[0028] Among them, (u, v) T is the image pixel coordinate of the target, R1 is the rotation matrix composed of the image plane size and the pixel size, which converts the image point coordinates into physical coordinates; R2 is the camera focal length, as well as the rotation matrix composed of the target distance and scale parameters, which converts the position of the target from the two-dimensional physical coordinate system to the three-dimensional camera coordinate system; R3 is the rotation matrix composed of the camera angle, t1 represents the coordinates of the origin of the pod platform coordinate system in the UAV body coordinate system, R3 and t1 together convert the position of the target from the camera coordinate system to the UAV body coordinate system; S1 represents the rotation matrix composed of the yaw angle ψ, pitch angle θ, and roll angle φ of the UAV, t2 represents the coordinates of the UAV in the world coordinate system, S1 and t2 together convert the position of the target from the UAV body coordinate system to the world coordinate system; Rx is the installation error of the pod, which is represented by the rotation matrix composed of the installation offset angle.

[0029] Preferably, the installation error R of the pod x It is represented by a rotation matrix consisting of two installation offset angles (ω, k) or three installation offset angles (l, ω, k); Denote (u, v), R, R3, t1, S1, t2, where R represents the parameter matrix consisting of the camera focal length, image plane size, and resolution; z represents the target elevation; D represents the length of the laser ranging; when the installation offset angle is two, the error equation is constructed using the following formula:

[0030]

[0031] f3(ω,k,z,S1,R3,D,t2)=0

[0032] When the installation offset angle is three, the error equation is constructed by the following formula:

[0033]

[0034] f3(l,ω,k,z,S1,R3,D,t2)=0.

[0035] Preferably, f 1，2 The derivation is as follows: Based on the single image positioning algorithm, the three-dimensional coordinates (x, y, z) of the target are T Convert to image plane coordinates (u, v)

[0036]

[0037]

[0038] f3 is derived as follows: According to the direction of the laser direction in the world coordinate system

[0039]

[0040] Combining the elevation component z2 of the laser direction vector in the world coordinate system, the length D of the laser ranging, and the elevation coordinate t2(3) of the UAV in the world coordinate system, the target elevation z=Dz2+t2(3) is obtained, and the equation is converted to f3.

[0041] Preferably, the installation offset angle obtained by solution is fed back to the single-image positioning algorithm for positioning solution, and the plane coordinates of the target are obtained by solution, and then output to the error fitting algorithm and the construction of the error equation. During the flight and positioning process, the solution is performed alternately and iteratively.

[0042] Compared with the prior art, the beneficial technical effects of the present invention are:

[0043] 1. The present invention is based on an error fitting method for improving the target plane positioning accuracy based on UAV video data. For an uncalibrated pod platform, the target plane positioning accuracy can be significantly improved; a joint solution method of the pod installation offset angle and the target elevation based on the virtual plane control point can perform self-calibration on the UAV pod online and solve the installation offset angle; a UAV target positioning system error elimination method based on error fitting can eliminate the systematic error in the target positioning result, and combined with the random error method, it can further improve the target positioning accuracy.

[0044] 2. The positioning accuracy improvement method implemented by the present invention is significantly superior to the mean filtering method in terms of accuracy, and can significantly improve the positioning accuracy in the plane and elevation directions at the same time.

[0045] 3. The present invention can achieve the stabilization of the plane positioning error in a short time, and will not have the fluctuation of positioning accuracy over time shown by the mean filtering method. It has a significant accuracy advantage over a longer period of time.

[0046] 4. The positioning accuracy improvement method implemented by the present invention does not require offline pod calibration and can meet some emergency mission requirements.

[0047] 5. The present invention does not need to rely on ground control points, which can save manpower and material resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the overall framework of a method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration in embodiments 1, 2, and 3 of the present invention.

[0049] Figure 2 This is the change of positioning errors in three directions over time / track in the visualization of target positioning results when the installation error is not considered in embodiments 1, 2, and 3 of the present invention.

[0050] Figure 3 This is the distribution of positioning errors in two-dimensional space in the visualization of target positioning results when the installation error is not considered in embodiments 1, 2, and 3 of the present invention.

[0051] Figure 4 This is the distribution of target positioning error in two-dimensional space when the installation error is not considered in Examples 1, 2, and 3 of the present invention. The circle in the middle represents the relative track after scale reduction corresponding to this section of observation data, and the color of each scattered point represents the relative time of observation.

[0052] Figure 5Schematic diagram of the comparison results between the positioning accuracy of the second embodiment of the present invention and the mean filtering method.

[0053] Figure 6 This is a schematic diagram of the overall framework of a method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration in Example 4 of the present invention. DETAILED DESCRIPTION

[0054] The specific implementation modes of the present invention are described below with reference to the accompanying drawings and examples. However, the following examples are only used to illustrate the present invention in detail and are not intended to limit the scope of the present invention in any way.

[0055] The procedures involved or relied upon in the following embodiments are all conventional or simple procedures in the art, and those skilled in the art can make conventional selections or adaptive adjustments based on specific application scenarios. Unless otherwise specified, the unit modules, components, structures, mechanisms, or sensors involved in the following embodiments are all conventional commercially available products.

[0056] Example 1: A method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration, see Figure 1-5 Uncontrolled means that this method does not require ground control points.

[0057] The overall framework of the first embodiment is as follows Figure 1 As shown in Figure 1, a high-precision 3D coordinate positioning method is proposed for processing UAV aerial video data orbiting a stationary target. This method is applicable when the UAV is filming along a nearly circular trajectory around the target. This method comprehensively utilizes the UAV's flight control data, pod information, and laser ranging data to achieve high-precision positioning through two components: a planar accuracy improvement method based on error fitting, and a joint solution of elevation and installation offset angles using virtual control points. Specifically, the planar accuracy improvement method based on error fitting post-processes and optimizes the results of single-image positioning to improve planar accuracy, and uses the improved positioning results as virtual planar control points. The joint solution of elevation and installation offset angles based on virtual control points utilizes virtual planar control points, an imaging model, and laser ranging information to jointly solve the installation offset angle between the UAV's pod's onboard equipment and the pod platform, as well as the target elevation, to obtain the target's high-precision 3D coordinates. The main advantage of this method lies in its comprehensive utilization of multiple sources of information in the UAV's aerial video data, ensuring high-precision positioning while reducing the workload of data collection and processing. At the same time, the method based on virtual control points can also effectively solve the positioning error problem in the elevation direction.

[0058] The specific implementation steps are as follows:

[0059] (1) Confirm that the target and aircraft's trajectory meet the algorithm's requirements

[0060] The task is to process the aerial video data of a UAV orbiting a stationary target. The UAV orbits the target in a nearly circular trajectory.

[0061] (2) Single-image positioning method based on laser ranging

[0062] Using the target's observation data and corresponding positioning parameters, the target is positioned using a single image based on laser ranging. The specific steps are as follows:

[0063] The input for single-image positioning based on laser ranging is the target's position in the video image captured by the pod-mounted camera. The output is the target's coordinates in the world coordinate system. The basic principle is a series of coordinate transformations: image plane coordinate system - camera coordinate system - pod coordinate system - carrier body coordinate system - carrier geographic coordinate system, etc. The carrier geographic coordinate system can be further transformed to a geodetic rectangular and geodetic coordinate system by selecting a reference point (with known longitude, latitude, and elevation), ultimately obtaining the target's longitude L, latitude B, and elevation H. Laser ranging information is used to provide the target's elevation difference relative to the pod. Combined with the angle information, this information is used to restore the scale from the image plane coordinate system to the camera coordinate system.

[0064] In addition to the laser ranging information, the coordinate transformation mainly relies on the flight control data, the pod angle, and the installation matrix. Among them, the installation matrix is the position and posture of the pod platform relative to the carrier body. Ideally, the pod platform coordinate system (O b -x b y b z b )'s origin O b is the origin of the camera coordinate system O c , x b The axis is parallel to the longitudinal axis of the body and points to the head of the body, y b The axis is parallel to the horizontal axis of the fuselage and points to the right wing. b The axis is parallel to the vertical axis of the fuselage and points downward. However, in reality, the position of the pod platform coordinate system relative to the carrier aircraft coordinate system does not strictly satisfy the above relationship. Instead, there is a certain installation error, namely the installation offset angle, which is reflected as a systematic error in the target positioning result.

[0065] (3) Plane accuracy improvement method based on error fitting

[0066] After accumulating the positioning results for a period of time and confirming the distribution pattern of the positioning results, that is, the plane coordinate distribution of the positioning results is approximately circular, the positioning results are input into the error fitting algorithm and the coordinates of the circle center obtained by fitting are output.

[0067] The specific method of error fitting is to fit the equation of the circle:

[0068] x2 +y 2 +ax+by+c=0

[0069] Where a, b, and c are constants. Given a set of coordinates (x i ,y i ), i = 1, 2, 3, ..., K. The above equation is a linear equation involving a, b, and c, which can be solved using linear least squares. After finding a, b, and c, the center of the circle is:

[0070] Visualize the single-image positioning result of the stationary target based on laser ranging corresponding to a circular track Figure 2 、 3 and 4. Figure 2 The visualization method is to mark the target positioning coordinates (coordinate vertical axis) of the multiple groups in three directions corresponding to the track in sequence (coordinate horizontal axis) on the coordinate plane; Figure 3 and Figure 4 The visualization method is to make scatter plots of the multiple target positioning results corresponding to the track according to the plane and elevation directions, where Figure 4 The zoomed-out track is marked for reference.

[0071] Analysis reveals a clear distribution pattern for target positioning errors: within a circular trajectory, the target positioning error remains constant, but its direction changes with the drone's position. Based on this "error circle" phenomenon, an error fitting method is used to determine the target's accurate, optimized planar position based on the center of the fitted circle.

[0072] (IV) Joint solution method of elevation and installation offset angle based on virtual control points

[0073] In subsequent observations, the center position coordinates output from step (3) are used as the accurate plane position of the target. Using the newly collected observation data and positioning parameters of at least two times, the following formulas (1) and (2) are used to construct the error equation and solve the installation offset angle and the target elevation. The solved offset angle is used to perform single-image positioning based on laser ranging to update the three-dimensional coordinates of the target. The error fitting method can obtain high-precision plane coordinates of the target. This plane coordinate can be initially used as a virtual control point to optimize the elevation of the target. The core idea is to adjust and solve the installation offset angle and elevation. The models required are the single-image target positioning equation and the laser ranging equation. The detailed steps are as follows:

[0074] The single image positioning algorithm based on laser ranging is to know the image pixel coordinates (u, v) of the target TIn the case of , the coordinates (x, y, z) of the target's local rectangular coordinate system (such as the north-east coordinate system) can be obtained by the following conversion: T :

[0075]

[0076] Among them, (u, v) T is the image pixel coordinate of the target, R1 is the rotation matrix composed of the image plane size and the pixel size, which converts the image point coordinates into physical coordinates; R2 is the rotation matrix composed of the camera focal length, the target distance, and the scale parameter, which converts the target position from the two-dimensional physical coordinate system to the three-dimensional camera coordinate system; R3 is the rotation matrix composed of the camera angle, R x is the installation error of the pod, t1 represents the coordinate of the origin of the pod platform coordinate system in the carrier body coordinate system, R3 and t1 together transform the target position from the camera coordinate system to the carrier body coordinate system; S1 represents the rotation matrix composed of the carrier's yaw angle ψ, pitch angle θ, and roll angle φ, t2 represents the coordinate of the carrier in the world coordinate system, S1 and t2 together transform the target position from the carrier coordinate system to the world coordinate system. Ideally, the installation error of the pod R x is the unit matrix. In practice, this error can be represented by a rotation matrix consisting of three installation offset angles (l, ω, k).

[0077] The three-dimensional coordinates of the target can be converted to image plane coordinates (u, v) by reversing the above steps:

[0078]

[0079]

[0080] Among them, R represents the parameter matrix composed of camera focal length, image plane size and resolution. According to the above two formulas, two observation equations can be listed:

[0081]

[0082] in, Represents positioning parameters, that is, various parameters on which positioning depends, including (u, v), R, R3, t1, S1, and t2.

[0083] According to the laser ranging, the third equation of the current observation can be listed. Specifically, the direction of the camera optical axis, that is, the direction of the laser direction in the world coordinate system is,

[0084]

[0085] Combined with the length D of the laser rangefinder, the height difference Dz2 of the target relative to the carrier can be obtained. Combined with the coordinate t2(3) of the carrier in the elevation direction in the world coordinate system, the elevation of the target can be obtained, z = Dz2 + t2(3). Rewriting the form, we can get:

[0086] f3(l,ω,k,z,S1,R3,D,t2)=0 (2)

[0087] Formulas (1) and (2) together form three nonlinear equations about (l, ω, k, z). With the elevation z of the target point and the installation offset angle (l, ω, k) as unknowns, the positioning model is linearized to obtain the error equation:

[0088] V=AΔ-L

[0089] Where V, A, L, and Δ are the residual, coefficient matrix, constant term, and correction number for unknowns of the error equation, respectively.

[0090] After two observations, the error equation is constructed according to the above formulas (1) and (2), and the installation offset angle l, ω, k and the target elevation z can be solved by the least squares method.

[0091] Example 2: A method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration. The difference from Example 1 is that in subsequent observations, Example 1 uses the installation offset angle calculated in step (4) to perform single-image positioning based on laser ranging to update the target's three-dimensional coordinates. Example 2, on the other hand, directly outputs the target's three-dimensional coordinates based on actual conditions (for example, the observation and positioning task of the current target positioning has ended, and the carrier aircraft is no longer flying around the current target). That is, Example 2 directly uses the plane coordinates output in step (3) as the final plane coordinates of the target, and uses the elevation calculated in step (4) as the final elevation coordinates of the target, while saving the installation offset angle calculated in step (4) for use in subsequent positioning of other targets by the carrier aircraft.

[0092] Using the measured data, the comparison results of the positioning accuracy of the second embodiment and the common method (mean filtering) are as follows: Figure 5 Compared with the mean filtering method, the positioning accuracy of the second embodiment in three directions is improved.

[0093] Example 3: A method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration. The difference from Example 1 is that in practice, not all three installation offset angles will affect the target positioning result, and the offset angles that do not affect the positioning result can be ignored. If the effect of the offset angle l on the positioning result can be ignored, then formulas (1) and (2) are correspondingly

[0094]

[0095] f3(ω,k,z,S1,R3,D,t2)=0

[0096] At this point, the error equation can be simplified accordingly, and only two offset angles and elevation are solved.

[0097] Example 4: A method for improving target positioning accuracy based on drone video data and uncontrolled self-calibration, see Figure 6 ( Figure 6 and Figure 1 The difference is Figure 6 ). High-precision positioning is achieved through two parts: a plane accuracy improvement method based on error fitting and a joint solution method of elevation and installation offset angle based on virtual control points. The difference from Example 1 is that this method can be further optimized and improved for specific scenarios in actual applications to meet the needs of high-precision positioning. The installation offset angle obtained by the solution is fed back to the single-image positioning algorithm based on laser ranging for positioning solution, and the plane coordinates of the target point obtained by the solution are output to the construction of the error fitting algorithm and the error equation. During the flight and positioning process, the solution is performed alternately and iteratively. That is, according to the requirements of the application scenario for the target positioning accuracy, this embodiment feeds back the installation offset angle output by the second part to the first part for positioning solution, and the plane coordinates of the target point obtained by the solution are output to the second part. That is, during the flight and positioning process, the two parts are alternately and iteratively performed to gradually improve the accuracy of the target.

[0098] The present invention has been described in detail above with reference to the accompanying drawings and embodiments; however, those skilled in the art will appreciate that, without departing from the spirit of the present invention, the specific parameters in the above embodiments may be changed, or related components and structures may be replaced by equivalents, thereby forming multiple specific embodiments, which are all within the common variation scope of the present invention and will not be described in detail one by one.

Claims

1. A method for improving target positioning accuracy based on drone video data, characterized in that: The following steps are involved: (1) Confirming that the target is stationary, the UAV navigates around the target in a substantially circular trajectory to obtain the target's positioning parameters, i.e., the parameters on which positioning depends; the positioning parameters include flight control data, pod information, laser ranging data, and installation matrix; (2) performing single-image positioning based on laser ranging on the target, and confirming that the plane coordinate distribution of the positioning output is approximately circular; the input of the single-image positioning is the coordinate position of the target on the image, the laser ranging result, and the positioning parameters, and the output is the coordinate of the target in the world coordinate system. After a series of coordinate transformations, the longitude L, latitude B, and elevation H of the target are finally obtained; (3) The result of the single-image positioning is input into the error fitting algorithm to obtain the coordinates of the circle center as the virtual control point; the error fitting algorithm is the equation of the fitted circle: x 2 +y 2 +ax+by+c=0 Among them, a, b, c are constant terms; a set of coordinates (x i ,y i ), i=1,2,3,...,K, use linear least squares to solve; after finding a,b,c, the center of the circle is: (4) constructing an error equation based on the virtual control points, and adjusting and solving the installation offset angle and the target elevation; the error equation is V=AΔ-L; wherein V, A, L, and Δ are the residual, coefficient matrix, constant term, and unknown correction number of the error equation, respectively; and solving the installation offset angle and the target elevation by the least squares method according to the error equation; (5) The center coordinates of the circle in step (3) are used as the plane coordinates of the target, and the target elevation in step (4) is used as the elevation coordinates of the target, and the three-dimensional coordinates of the target are output.

2. The method for improving target positioning accuracy based on drone video data according to claim 1, characterized in that: In step (5), the single image positioning is performed using the installation offset angle to update the three-dimensional coordinates of the target.

3. The method for improving target positioning accuracy based on drone video data according to claim 1 or 2, characterized in that: The algorithm for single-image positioning is: the coordinates (x, y, z) of the local rectangular coordinate system of the target are obtained by the following conversion: T : Among them, (u,v) T is the image pixel coordinate of the target, R1 is the rotation matrix composed of the image plane size and the pixel size, which converts the image point coordinates into physical coordinates; R2 is the camera focal length, as well as the rotation matrix composed of the target distance and scale parameters, which converts the position of the target from the two-dimensional physical coordinate system to the three-dimensional camera coordinate system; R3 is the rotation matrix composed of the camera angle, t1 represents the coordinates of the origin of the pod platform coordinate system in the UAV body coordinate system, R3 and t1 together convert the position of the target from the camera coordinate system to the UAV body coordinate system; S1 represents the rotation matrix composed of the yaw angle ψ, pitch angle θ, and roll angle φ of the UAV, t2 represents the coordinates of the UAV in the world coordinate system, S1 and t2 together convert the position of the target from the UAV body coordinate system to the world coordinate system; R x is the installation error of the pod, which is represented by the rotation matrix composed of the installation offset angle.

4. The method for improving target positioning accuracy based on drone video data according to claim 3, characterized in that: Pod installation error R x It is represented by a rotation matrix consisting of two installation offset angles (ω, k) or three installation offset angles (l, ω, k); Denote (u, v), R, R3, t1, S1, t2, where R represents the parameter matrix consisting of the camera focal length, image plane size, and resolution; z represents the target elevation; D represents the length of the laser ranging; when the installation offset angle is two, the error equation is constructed using the following formula: f3(ω,k,z,S1,R3,D,t2)=0 When the installation offset angle is three, the error equation is constructed by the following formula: f3(l,ω,k,z,S1,R3,D,t2)=0.

5. The method for improving target positioning accuracy based on drone video data according to claim 4, characterized in that: f 1,2 The derivation is as follows: Based on the single image positioning algorithm, the three-dimensional coordinates (x, y, z) of the target are T Convert to image plane coordinates (u,v) f3 is derived as follows: According to the direction of the laser direction in the world coordinate system Combining the elevation component z2 of the laser direction vector in the world coordinate system, the length D of the laser ranging, and the elevation coordinate t2(3) of the UAV in the world coordinate system, the target elevation z=Dz2+t2(3) is obtained, and the equation is converted to f3.

6. The method for improving target positioning accuracy based on drone video data according to claim 3, characterized in that: The calculated installation offset angle is fed back to the single-image positioning algorithm for positioning solution, and the plane coordinates of the target are obtained by solution. The obtained coordinates are then output to the error fitting algorithm and the error equation construction. During the flight and positioning process, the solution is performed alternately and iteratively.

Images