An encoding target and decoding positioning method for unmanned aerial vehicle close-range photogrammetry

Abstract

The application provides a kind of encoding target and decoding positioning method for unmanned aerial vehicle close-range photogrammetry, it is related to target real-time decoding positioning technical field, the encoding target used includes target matrix, center positioning circle, multiple encoding circle slot, multiple encoding circle dot, center cross wire and numbered anchor point;Unmanned aerial vehicle original image containing encoding target is acquired;Target detection model is used to detect the encoding target area in original image, and target boundary box is obtained;The target boundary box is expanded outward, cropped and size normalized, and ROI image is obtained;Key point detection network is used to extract shared feature map from ROI image;The slot key point coordinates of each encoding circle slot are determined by ordered slot positioning branch;The slot binary state of each encoding circle slot is determined by slot state judging branch;Sub-pixel level ellipse fitting fine positioning is carried out on ROI image, and target center coordinates are obtained;Target identity code and target center coordinates are output.

Description

Technical Field

[0001] This invention relates to the field of real-time target decoding and positioning technology, and in particular to an coded target and decoding and positioning method for close-range photogrammetry using unmanned aerial vehicles. Background Technology

[0002] In applications such as UAV close-range photogrammetry, engineering facility inspection, building facade measurement, bridge and slope monitoring, 3D reconstruction of weakly textured scenes, and measurement of complex structures, artificial markers are typically deployed within the measured object or area to improve the reliability of image-based corresponding point matching and the accuracy of photogrammetric results. Artificial markers generally include non-coded targets and coded targets. Non-coded targets have a simple structure and high center positioning accuracy, but lack a unique identification number, often requiring manual intervention in multi-view image processing, resulting in low automation. Coded targets, on the other hand, have a unique identification number and can establish stable corresponding point relationships across multiple images, which is beneficial for improving the efficiency of automated UAV image processing.

[0003] Existing coded targets mainly include square markers, circular or ring-shaped targets, and point-distributed targets. Square markers are suitable for pose estimation, but in close-range photogrammetry, high-precision positioning of the target center usually relies more on circular or elliptical features. Circular markers are imaged as ellipses under perspective projection, and high-precision center positioning can be achieved by fitting the ellipse to the edge points. However, traditional circular or ring-shaped coded targets are usually limited in terms of coding capacity expansion by the target area and the complexity of the coding structure. Point-distributed targets have strong coding capabilities, but in UAV close-range photogrammetry, due to factors such as flight altitude, shooting distance, attitude changes, motion blur, illumination changes, and background interference, point-shaped coded features may occupy only a few pixels in the image, or even be lost, stuck together, or difficult to recognize stably, thus affecting the reliability of recognition and decoding.

[0004] UAV close-range photogrammetry differs from traditional ground-based close-range photogrammetry and conventional aerial photogrammetry. Its operating distance is typically greater than in ordinary close-range measurement scenarios, but the accuracy requirements are higher than in general low-altitude aerial surveying applications. In these scenarios, the target needs to possess a large encoding capacity, retain geometric features suitable for high-precision center positioning, and maintain good robustness in the face of challenges such as distant small targets, complex lighting, rotational attitude changes, local blurring, and background interference. Therefore, there is an urgent need for a UAV close-range photogrammetry coded target and its decoding and positioning method that combines high-precision positioning, large-capacity encoding, and automated decoding and positioning capabilities. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides an coded target and decoding positioning method for close-range photogrammetry using unmanned aerial vehicles (UAVs), comprising: The coded target used for close-range photogrammetry of UAVs includes a target base, a central positioning circle, multiple coded grooves, multiple coded dots, a central crosshair, and numbered anchor points. The central positioning circle is set in the central region of the target substrate to serve as a precise positioning feature for the target center; The coding grooves are distributed at equal angles around the central positioning circle and are used as slots for setting candidate positions of coding dots. The coded dots are set in the corresponding coded slots according to the target identity code. Each coded slot represents the corresponding binary state by having or not having a coded dot. The central crosshair consists of two mutually perpendicular thin lines whose intersection coincides with the center of the central positioning circle, and is used to accurately indicate the center of the target. The numbering anchor point is set in a preset area of ​​the target substrate. There is a preset fixed angle offset relationship between the numbering anchor point and the starting position of the target identification code, which is used to determine the starting position of the target identification code. The target identification code is formed by the starting position determined by the numbering anchor point and the preset reading direction.

[0006] Optionally, the number of the coded slots is determined based on the number of binary bits in the target identity code.

[0007] Optionally, the numbering anchor point is set in the numbering area at the lower left corner of the target base. The numbering anchor point is a manually readable target identification code and serves as a rotation reference anchor point for determining the starting position of the target identification code when the target undergoes planar rotation.

[0008] Optionally, the ratio of the radius of the central positioning circle, the shortest distance from the outer edge of the central positioning circle to the outer edge of the coding groove, and the radius of the coding groove is determined according to the number of coding grooves.

[0009] Optionally, the shortest distance between adjacent coding dots is set to be greater than or equal to the diameter of the coding dots.

[0010] This method for decoding and locating coded targets for close-range photogrammetry using unmanned aerial vehicles includes: S1. Acquire raw UAV images containing coded targets; S2. Use a target detection model to detect the coded target region in the original image and obtain the target bounding box; S3. Expand, crop, and normalize the size of the target bounding box to obtain the ROI image; S4. Use a keypoint detection network to extract shared feature maps from ROI images; S5. Determine the coordinates of key points of each coded circular slot through ordered slot positioning branches; S6. Determine the binary status of each coded circular slot by judging the slot status branch; S7. Perform subpixel-level ellipse fitting and fine localization on the ROI image to obtain the target center coordinates; S8. Determine the starting position of the target identity code based on the anchor point number and generate a binary code sequence according to the preset reading direction. Convert the binary code sequence into a decimal target identity code according to the preset high and low bit rules, and output the target identity code and target center coordinates.

[0011] Optionally, determining the coordinates of key points of each coded circular slot through ordered slot positioning branches includes: Based on the shared feature map, predict N ordered slot heatmaps that correspond one-to-one with multiple coded slots; For each slot, the response peak is found in the corresponding ordered slot heatmap, or the MSRA heatmap decoding method is used to recover the coordinates of the key points of the slot. The recovered key point coordinates of the slot are then mapped back to the original UAV image coordinate system through inverse affine transformation.

[0012] Optionally, determining the binary state of each encoded circular slot through the slot state judgment branch includes: The shared feature map is mapped to the projected feature map through 1×1 convolution; For each ordered slot heatmap, perform Softmax in the spatial dimension to obtain the attention weight of the corresponding slot; The projected feature map is weighted and aggregated using attention weights to obtain the slot embedding feature; By inputting the slot embedding feature into a multilayer perceptron with shared parameters, the probability of the existence of the encoded dot in the slot is obtained. The binary state of the slot is determined by comparing the preset threshold with the probability of the presence of the coded dots in the slot.

[0013] Optionally, the subpixel-level ellipse fitting and fine localization of the ROI image to obtain the target center coordinates includes: Perform preliminary ellipse fitting on the center positioning circle in the ROI image to obtain the approximate ellipse center of the center positioning circle; Starting from the approximate elliptical center of the central positioning circle, multiple radial rays are emitted outward at preset angular intervals, and gray values ​​are sampled on each ray. By searching for the position with the largest gray gradient amplitude, the initial edge point of the central positioning circle is obtained. A local grayscale window is constructed with each initial edge point of the centrally located circle as the center, low-order Zernike moments are calculated, and the sub-pixel offset of the edge points relative to the pixel center is estimated based on the Zernike moment features. By constraining the sub-pixel offset along the radial ray direction, a set of sub-pixel level edge points of the center positioning circle is obtained. Outlier points are removed from the subpixel-level edge point set, and the subpixel-level center coordinates of the center positioning circle are calculated using the least squares or RANSAC ellipse fitting method.

[0014] By adopting the above technical solution, the present invention has at least the following beneficial effects: 1. This invention adopts a composite target structure with a large central circle for positioning and a small outer circle for coding, which combines the advantages of high-precision central positioning of circular markers with the advantages of large-capacity coding of dot-shaped coding structures.

[0015] 2. This invention arranges multiple coding grooves at equal angles on the outer side of the central positioning circle, and uses the presence or absence of coding dots in the coding grooves to represent binary states. The coding method is simple, clear, and easy to expand.

[0016] 3. This invention uses numbered anchor points to determine the starting position of the encoding, so that the encoding sequence does not need to use cyclic shift equivalent encoding or rely on minimum Hamming distance constraints, thus preserving a high encoding capacity.

[0017] 4. By designing the ratio of the central positioning circle, the coding dots, and the spacing, this invention reduces the possibility of adhesion between the central positioning circle and the coding dots, and between adjacent coding dots, in the imaging of small targets at long distances by UAVs, thereby improving the applicability of UAVs in close-range photogrammetry scenarios.

[0018] 5. This invention employs a deep learning recognition process that combines target detection, ROI normalization, key point heatmap prediction, slot status judgment, and binary decoding to improve the automatic recognition and decoding capabilities under complex lighting, rotational posture, motion blur, scale changes, and long-distance imaging conditions.

[0019] 6. This invention uses the center coordinates of the center positioning circle obtained by deep learning methods as the initial value for fine positioning, combines radial ray search and Zernike moment sub-pixel edge positioning methods to extract the sub-pixel edge points of the center positioning circle, and obtains the target center coordinates through ellipse fitting, thereby improving the accuracy, stability and robustness of target center positioning under UAV close-range photogrammetry conditions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the overall structure of the coded target; Figure 2 This is a schematic diagram of the coding groove; Figure 3 This is a diagram illustrating the reading order of the encoded dots; Figure 4 A schematic diagram showing the radius of the center positioning circle, the shortest distance from the outer edge of the center positioning circle to the outer edge of the coding groove, and the ratio of the radius of the coding groove. Figure 5 Schematic diagrams of extended coding target structures for different coding bit lengths are shown. (a) is a schematic diagram of a 16-bit coding target structure, and (b) is a schematic diagram of a 20-bit coding target structure. Figure 6 A flowchart of a coding target decoding and positioning method for close-range photogrammetry using an unmanned aerial vehicle (UAV) is provided in an embodiment of the present invention. Figure 7 The original imagery of the drone containing coded targets; Figure 8 This is a map showing the detection results for the encoded target region; Figure 9 Image of the ROI of the encoded target; Figure 10 Heat map of ordered tank locations; Figure 11 The diagram shows the encoding slot state probability and binary determination results; Figure 12 The preliminary ellipse fitting result for the center-locating circle in the ROI image; Figure 13 Image showing the edge extraction results along multiple radial ray directions; Figure 14 This is a diagram showing the subpixel edge thinning results based on Zernike moments; Figure 15 Image showing the result of removing outliers from a subpixel-level edge point set; Figure 16 This is a diagram showing the ellipse fitting and localization results based on sub-pixel edge points; Figure 17 This is the final decoding and localization result image. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1As shown, an embodiment of the present invention provides an coded target for close-range photogrammetry using a UAV, comprising a target base, a central positioning circle, multiple coded grooves, multiple coded dots, a central crosshair, and numbered anchor points.

[0024] The target substrate can be square, rectangular, or other planar substrates that are easy to deploy, print, and recognize as drone images.

[0025] The center positioning circle is set in the central region of the target substrate to serve as a feature for precise target center positioning. In image perspective projection, the center positioning circle is typically projected as an ellipse; therefore, the target center coordinates can be obtained by extracting the edge points of the center positioning circle and performing ellipse fitting. Because the center positioning circle has a relatively large area, it can provide more continuous edge points, which is beneficial for improving the center positioning accuracy in UAV imagery.

[0026] like Figure 2 As shown, multiple coding slots are distributed at equal angles around the central positioning circle to serve as slots for setting candidate positions for coding dots. Specifically, there are 12 coding slots, which are distributed at equal angles along the outer side of the central positioning circle. Each coding slot corresponds to one binary bit, and the 12 coding slots together constitute a 12-bit binary code.

[0027] The coded dots are set in the corresponding coded slots according to the target identity code. Each coded slot can be in two states: if a coded dot is set in the coded slot, the slot represents binary state 1; if no coded dot is set in the coded slot, the slot represents binary state 0. The target identity code is formed by the combination of the states of multiple coded slots.

[0028] For example, when encoding dots are set in the 1st, 3rd, 6th and 10th encoding slots, and the remaining encoding slots are empty, a corresponding 12-bit binary encoding sequence 101001000100 can be formed.

[0029] The central crosshair consists of two mutually perpendicular thin lines whose intersection coincides with the center of the central positioning circle. It is used to accurately indicate the target center. Specifically, the length of the central crosshair is set to 3 / 5 of the diameter of the central positioning circle, and the line width can be set to a fixed 0.5mm. This design facilitates accurate aiming at the target center during on-site manual verification or observation by measuring equipment, while not significantly affecting the edge extraction of the central positioning circle.

[0030] Numbered anchor points are set in a predetermined area of ​​the target substrate to determine the starting position of the target identification code. Specifically, the numbered anchor points are set in the lower left corner of the target substrate's numbering area. These anchor points are manually readable target identification codes, facilitating on-site deployment and measurement recording. Simultaneously, the numbered anchor points serve as rotation reference anchor points. Due to the predetermined fixed angular offset between the numbered anchor points and the starting position of the target identification code, a fixed starting position can still be determined even when the target undergoes arbitrary planar rotation, thus allowing the code sequence to be read in a fixed direction. The starting position determined by the numbered anchor points is either the 0th or 1st bit, and the reading direction is counter-clockwise. Figure 3 As shown. During recognition and decoding, starting from the start bit, the system sequentially reads whether there is an coded dot in each of the 12 coded slots. The slot with a coded dot is marked as 1, and the slot without a coded dot is marked as 0, thus forming a binary sequence of length 12.

[0031] like Figure 4 As shown, let the radius of the central positioning circle be R, the radius of the encoding dot be r, and the shortest distance from the outer edge of the central positioning circle to the outer edge of the encoding dot be d. To ensure that the central positioning circle and the encoding dot do not stick together under long-range UAV imaging conditions, and at the same time ensure that the central positioning circle has a sufficiently large imaging area, in one design of a 12-bit coded target, the R:d:r ratio is 5:2:1. This ratio allows the central positioning circle to retain a large area, which is beneficial for ellipse fitting; at the same time, it maintains a sufficient distance between the encoding dot and the central positioning circle, so that even when the UAV is imaging a long-range or small target, the two are not likely to stick together in the image.

[0032] In addition, the minimum spacing between adjacent coded dots is set to be greater than or equal to the diameter of the coded dots to prevent adjacent coded dots from sticking together in the image.

[0033] Through the above proportional constraints, when the coded dots are too small to be stably imaged at long distances, the central positioning circle and the coded dots will not stick together first, thus avoiding interference from the coded area with the elliptical fitting of the central positioning circle.

[0034] like Figure 5As shown, in applications with higher encoding capacity requirements, the number of encoding slots can be expanded to 16, 20, or other numbers, thus forming a 16-bit, 20-bit, or other bit-based binary encoding scheme. Theoretically, 16-bit encoding can provide more identification numbers, while 20-bit encoding can further expand the encoding capacity, suitable for large-scale UAV close-range photogrammetry, multi-area joint measurement, or automatic identification of a large number of targets. When expanding the number of encoding bits, each encoding slot is still distributed at equal angles along concentric rings outside the central positioning circle, and each encoding slot still indicates the binary state by the presence or absence of an encoding dot. Because the number of encoding bits increases, the angular interval between adjacent slots decreases, so the radius r of the encoding dot can be reduced accordingly, and the radius R of the central positioning circle and the shortest distance d from the outer edge of the central positioning circle to the outer edge of the encoding slot can be adjusted, ensuring that the shortest distance between the central positioning circle and the encoding dot, and the shortest distance between adjacent encoding dots, still meet the anti-adhesion requirements. In the 16-bit and 20-bit encoding implementations, the shortest distance d from the outer edge of the center positioning circle to the outer edge of the encoding circle groove remains the diameter of the encoding point, i.e., d=2r; at the same time, the shortest distance between adjacent encoding points is not less than or slightly greater than the diameter of the encoding point. Since a smaller radius of the encoding point releases more target area, the area ratio of the center positioning circle can be appropriately increased to increase the number of edge points that can be used for ellipse fitting, thereby improving the center positioning stability in UAV imagery.

[0035] like Figure 6 As shown, this embodiment of the invention provides a method for decoding and locating an coded target for close-range photogrammetry using a UAV, employing a 12-bit coded target, including: S1. Acquire raw UAV images containing coded targets.

[0036] The raw images can be acquired by a visible light camera, oblique photography camera, or other imaging devices mounted on a drone, such as... Figure 7 As shown.

[0037] S2. Use a target detection model to detect the coded target region in the original image to obtain the target bounding box, such as... Figure 8 As shown.

[0038] S3. Expand, crop, and normalize the size of the target bounding box to obtain the ROI image.

[0039] Since the bounding box may only cover the central area or there may be detection deviations, in order to ensure that the coded circular slots and numbered anchor points are completely included, the target bounding box is expanded outward according to a preset ratio to obtain an extended bounding box.

[0040] Then, affine cropping and size normalization are performed on the corresponding regions of the expanded bounding box to obtain the ROI image. The ROI image can be normalized to a fixed size, such as 256×256×3, as shown below. Figure 9As shown, this is used to input the subsequent keypoint detection network.

[0041] S4. Use a key point detection network to extract shared feature maps from ROI images.

[0042] The keypoint detection network can be HRNet-W32, or other networks that can maintain spatial resolution and output keypoint heatmaps.

[0043] S5. Determine the coordinates of key points of each coded circular slot through ordered slot positioning branches.

[0044] Specifically, based on the shared feature map, 12 ordered slot heatmaps are predicted that correspond one-to-one with multiple coded slots, such as... Figure 10 As shown, each slot heatmap corresponds to a coded circular slot in a fixed order. Therefore, the channel order of the slot heatmap corresponds to the coded bit order. For each slot, the response peak is found in the corresponding ordered slot heatmap, or the MSRA heatmap decoding method is used to recover the coordinates of the key points of the slot. The recovered coordinates of the key points of the slot are mapped back to the original UAV image coordinate system through inverse affine transformation.

[0045] S6. Determine the binary status of each coded circular slot by judging the slot status branch.

[0046] Specifically, the shared feature map is mapped to a projected feature map using a 1×1 convolution; a softmax operation is performed on each ordered slot heatmap in the spatial dimension to obtain the attention weights for the corresponding slots; the projected feature maps are weighted and aggregated using these attention weights to obtain slot embedding features; the slot embedding features are input into a multilayer perceptron with shared parameters to obtain the probability of the presence of coded dots in the slots; a preset threshold of 0.8 is compared with the probability of the presence of coded dots in the slots to determine the binary state of the slots, such as... Figure 11 As shown.

[0047] S7. Perform subpixel-level ellipse fitting and fine localization on the ROI image to obtain the target center coordinates; Specifically, a preliminary ellipse fitting is performed on the center positioning circle in the ROI image to obtain an approximate elliptical center for the center positioning circle, such as... Figure 12 As shown; starting from the approximate elliptical center of the central positioning circle, multiple radial rays are emitted outward at preset angular intervals, and grayscale values ​​are sampled on each ray. By searching for the position with the largest grayscale gradient amplitude, the initial edge point of the central positioning circle is obtained, as shown. Figure 13As shown; a local grayscale window is constructed with each initial edge point of the central positioning circle as the center, low-order Zernike moments are calculated, and the sub-pixel offsets of the edge points relative to the pixel center are estimated based on the Zernike moment characteristics; the sub-pixel offsets are constrained along the radial ray direction to obtain the sub-pixel level edge point set of the central positioning circle, as shown. Figure 14 As shown; outlier removal is performed on the sub-pixel level edge point set, such as... Figure 15 As shown, the sub-pixel-level center coordinates of the center positioning circle are calculated using the least squares or RANSAC ellipse fitting method, as shown. Figure 16 As shown.

[0048] S8. Determine the starting position of the target identification code based on the anchor point number and generate a binary code sequence according to the preset reading direction. Convert the binary code sequence into a decimal target identification code according to the preset high and low bit rules, and output the target identification code and target center coordinates. Figure 17 As shown, this is used for the identification of control points, checkpoints, connection points, or constraint points in close-range photogrammetry using unmanned aerial vehicles (UAVs).

[0049] The present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A coded target for close-range photogrammetry using unmanned aerial vehicles (UAVs), characterized in that, It includes a target base, a central positioning circle, multiple coding grooves, multiple coding dots, a central crosshair, and numbered anchor points; The central positioning circle is set in the central region of the target substrate to serve as a precise positioning feature for the target center; The coding grooves are distributed at equal angles around the central positioning circle and are used as slots for setting candidate positions of coding dots. The coded dots are set in the corresponding coded slots according to the target identity code. Each coded slot represents the corresponding binary state by having or not having a coded dot. The central crosshair consists of two mutually perpendicular thin lines whose intersection coincides with the center of the central positioning circle, and is used to accurately indicate the center of the target. The numbering anchor point is set in a preset area of ​​the target substrate. There is a preset fixed angle offset relationship between the numbering anchor point and the starting position of the target identification code, which is used to determine the starting position of the target identification code. The target identification code is formed by the starting position determined by the numbering anchor point and the preset reading direction.

2. The coded target for close-range photogrammetry of unmanned aerial vehicles according to claim 1, characterized in that, The number of coded slots is determined based on the number of binary bits in the target identity code.

3. The coded target for close-range photogrammetry of unmanned aerial vehicles as described in 1, characterized in that, The numbering anchor point is set in the numbering area at the lower left corner of the target base. The numbering anchor point is a manually readable target identification code and serves as a rotation reference anchor point for determining the starting position of the target identification code when the target undergoes planar rotation.

4. The coded target for close-range photogrammetry of unmanned aerial vehicles according to claim 1, characterized in that, The radius of the central positioning circle, the shortest distance from the outer edge of the central positioning circle to the outer edge of the coding groove, and the ratio of the radius of the coding groove are determined according to the number of coding grooves.

5. The coded target for close-range photogrammetry of unmanned aerial vehicles according to claim 1, characterized in that, The minimum spacing between adjacent coding dots is set to be greater than or equal to the diameter of the coding dot.

6. A method for decoding and locating an coded target for close-range photogrammetry using unmanned aerial vehicles (UAVs), characterized in that, The method, employing the encoded target as described in any one of claims 1 to 5, comprises: S1. Acquire raw UAV images containing coded targets; S2. Use a target detection model to detect the coded target region in the original image and obtain the target bounding box; S3. Expand, crop, and normalize the size of the target bounding box to obtain the ROI image; S4. Use a keypoint detection network to extract shared feature maps from ROI images; S5. Determine the coordinates of key points of each coded circular slot through ordered slot positioning branches; S6. Determine the binary status of each coded circular slot by judging the slot status branch; S7. Perform subpixel-level ellipse fitting and fine localization on the ROI image to obtain the target center coordinates; S8. Determine the starting position of the target identity code based on the anchor point number and generate a binary code sequence according to the preset reading direction. Convert the binary code sequence into a decimal target identity code according to the preset high and low bit rules, and output the target identity code and target center coordinates.

7. The coded target decoding and positioning method for close-range photogrammetry of unmanned aerial vehicles according to claim 6, characterized in that, The step of determining the coordinates of key points of each coded circular slot through ordered slot positioning branches includes: Based on the shared feature map, predict N ordered slot heatmaps that correspond one-to-one with multiple coded slots; For each slot, the response peak is found in the corresponding ordered slot heatmap, or the MSRA heatmap decoding method is used to recover the coordinates of the key points of the slot. The recovered key point coordinates of the slot are then mapped back to the original UAV image coordinate system through inverse affine transformation.

8. The coded target decoding and positioning method for close-range photogrammetry of unmanned aerial vehicles according to claim 6, characterized in that, The step of determining the binary state of each coded circular slot through the slot state judgment branch includes: The shared feature map is mapped to the projected feature map through 1×1 convolution; For each ordered slot heatmap, perform Softmax in the spatial dimension to obtain the attention weight of the corresponding slot; The projected feature map is weighted and aggregated using attention weights to obtain the slot embedding feature; By inputting the slot embedding feature into a multilayer perceptron with shared parameters, the probability of the existence of the encoded dot in the slot is obtained. The binary state of the slot is determined by comparing the preset threshold with the probability of the presence of the coded dots in the slot.

9. The coded target decoding and positioning method for close-range photogrammetry of unmanned aerial vehicles according to claim 6, characterized in that, The subpixel-level ellipse fitting and fine localization of the ROI image to obtain the target center coordinates includes: Perform preliminary ellipse fitting on the center positioning circle in the ROI image to obtain the approximate ellipse center of the center positioning circle; Starting from the approximate elliptical center of the central positioning circle, multiple radial rays are emitted outward at preset angular intervals, and gray values ​​are sampled on each ray. By searching for the position with the largest gray gradient amplitude, the initial edge point of the central positioning circle is obtained. A local grayscale window is constructed with each initial edge point of the centrally located circle as the center, low-order Zernike moments are calculated, and the sub-pixel offset of the edge points relative to the pixel center is estimated based on the Zernike moment features. By constraining the sub-pixel offset along the radial ray direction, a set of sub-pixel level edge points of the center positioning circle is obtained. Outlier points are removed from the subpixel-level edge point set, and the subpixel-level center coordinates of the center positioning circle are calculated using the least squares or RANSAC ellipse fitting method.

Images