An image stitching method for low-altitude unmanned aerial vehicle (UAV) monitoring networks

Abstract

This invention belongs to the field of image processing technology, specifically relating to an image stitching method for a low-altitude unmanned aerial vehicle (UAV) monitoring network. The method includes: acquiring a reference image and an image to be stitched; extracting matching point pairs between the two images; determining the static confidence level of the matching point pairs based on their disparity changes; extracting the static matching point pairs and performing multiple linear transformations to obtain multiple static projection matrices; selecting the optimal static projection matrix for image transformation; and extracting dynamic targets within the overlapping area. When generating the stitched image, the gray values ​​of the regions identified as dynamic targets in the reference image are replaced with the gray values ​​of the same regions in the image to be stitched, thus completing the image stitching. This invention, through a dynamic-static separation strategy, effectively avoids the contamination of geometric transformation calculations by dynamic targets, significantly improving stitching accuracy and visual quality.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to an image stitching method for a low-altitude unmanned aerial vehicle (UAV) monitoring network. Background Technology

[0002] With its high mobility and low cost, UAV low-altitude monitoring plays an important role in urban management and environmental protection. By continuously capturing high-resolution image sequences along a predetermined route and then using image stitching technology, these independent photos with limited perspectives are combined into a geographic information panoramic map. This ability to transform discrete snapshots into continuous spatial data is a key technology to overcome the limitations of single images and achieve comprehensive regional macro-situational awareness, providing a global perspective for subsequent decision analysis.

[0003] Existing image stitching techniques, such as phase correlation, are based on performing a Fourier transform on the global image and calculating the pixel translation between two images by analyzing the phase spectrum differences to achieve alignment stitching. This method is relatively accurate for ideal situations where the scene is completely static and there is only translational motion.

[0004] However, considering that there are dynamic targets such as moving vehicles and pedestrians in real urban and traffic monitoring scenarios, when dynamic targets are present, the feature points extracted from these targets will provide motion information that is completely inconsistent with the static background. This erroneous motion information is like data noise, which seriously pollutes the process of solving the translation. Image fusion based on this polluted translation, regardless of the fusion strategy adopted, will inevitably lead to obvious ghosting, blurring and other visual artifacts around the dynamic objects in the image stitching results, which seriously reduces the usability of the stitched image and makes it unable to meet the needs of high-precision monitoring and analysis. Summary of the Invention

[0005] To address the technical problem in existing technologies that fail to distinguish between static and dynamic features when processing images containing moving targets, leading to inaccurate translation calculations and ultimately stitching artifacts, this invention provides an image stitching method for low-altitude unmanned aerial vehicle (UAV) monitoring networks, comprising: acquiring a reference image. and the image to be stitched and extract and All matching point pairs are identified; the static confidence of each matching point pair is determined based on the difference between the disparity of each matching point pair and the mode of the disparities of all matching point pairs; the static confidence is clustered and filtered to obtain all static matching point pairs; the static matching point pairs are sampled multiple times and linearly transformed to obtain multiple static projection matrices; the reprojection distance of the corresponding matching point in each static projection matrix is ​​weighted based on the static confidence of each matching point pair to determine the reprojection error under each static projection matrix; the static projection matrix with the smallest reprojection error is taken as the optimal static projection matrix; based on the optimal static projection matrix, the reprojection distance of each matching point pair is calculated as follows: Transform to In the coordinate system, obtain Transformed image as well as and The overlapping region; based on each position within the overlapping region in and The difference in grayscale values ​​determines whether each location within the overlapping area is a dynamic target; and Overlapping placement, where The grayscale value of the position of each dynamic target in the image is replaced with... By matching the grayscale values ​​at the same positions, the images can be stitched together.

[0006] This invention analyzes the distribution characteristics of static confidence scores to extract all static matching point pairs, and performs multiple sampling and linear transformations to obtain multiple static projection matrices. Based on the static confidence scores, the reprojection distances of these multiple static projection matrices are weighted to select the optimal static projection matrix that reflects the motion of the static background. The optimal static projection matrix is ​​used to obtain overlapping regions, and dynamic targets are identified by the difference in grayscale values ​​at the same location within the overlapping regions. Targeted processing of dynamic targets solves the problem of interference from dynamic targets in the stitching process, ensuring the geometric accuracy and visual quality of the stitching results.

[0007] Preferably, the extraction and All matching point pairs, including: pairs and Perform grayscale conversion and use the scale-invariant feature transform algorithm for detection. and The key points in the grayscale image are extracted and descriptors are generated; based on the descriptors, matching is performed to obtain all matching point pairs.

[0008] Preferably, the static confidence level of each matching point pair satisfies the expression: In the formula, For the first The static confidence of each matching pair; For the first The disparity of the positions of each matching point pair; This represents the index value and total number of all matching point pairs; The mode function; To determine the absolute value sign; These are preset hyperparameters; It is a natural exponential function.

[0009] This invention transforms the concept of dynamic and static into a static confidence value by calculating the difference between the disparity of the matching point pair and its global disparity mode. This can effectively identify abnormal matches generated by dynamic targets and provide data support for subsequent dynamic target extraction.

[0010] Preferably, the disparity of the position of the matching point pair is equal to the Euclidean distance between the two matching points in the matching point pair.

[0011] Preferably, obtaining all static matching point pairs includes: using the static confidence scores of all matching point pairs as clustering samples, performing k-means clustering to obtain multiple clusters; calculating the average value of all static confidence scores within each cluster, and using the matching point pairs within the cluster with the largest average value as static matching point pairs.

[0012] Preferably, the step of performing a linear transformation on multiple samples of static matching point pairs to obtain multiple static projection matrices includes: performing multiple random samples on all static matching point pairs to obtain multiple sets of sub-matching point pairs; and using a direct linear transformation algorithm to calculate multiple static projection matrices for each set of sub-matching point pairs.

[0013] This invention generates a set of static projection matrices by sampling all static matching point pairs multiple times. This strategy of generating multiple matrices and selecting the best one avoids the risk of deviation caused by noise or bad samples in a single calculation, and enhances the accuracy of solving the transformation model.

[0014] Preferably, the reprojection error degree under each static projection matrix satisfies the expression: In the formula, For the first Reprojection error under a static projection matrix; For the first The static confidence of each matching pair; , For the first In each matching point pair corresponding to The position of the matching point and The position of the matching point; For the first One static projection matrix; For the first In each matching point pair corresponding to The location of the matching point, through the first Transform the static projection matrix to Predicted position in the coordinate system; To take the Euclidean distance function; This represents the index value and total number of all matching point pairs.

[0015] This invention introduces a static confidence weighting mechanism when evaluating the optimal static projection matrix. When calculating the total error, it can amplify the contribution of static points and suppress the influence of dynamic points, ensuring that the final selected optimal static projection matrix can accurately reflect the transformation of the static background.

[0016] Preferably, the acquisition Transformed image This includes: combining the optimal static projection matrix with... Multiply the positions of each pixel in the image to obtain the result. Transform to The coordinate system corresponds to the position of each pixel, and the gray value of the pixel corresponding to the position is filled with . The grayscale value of the corresponding pixel is obtained. .

[0017] Preferably, the and The method for obtaining the overlapping region is: traversing and In the context of all locations, positions where the same location appears repeatedly are marked as overlapping areas, resulting in... and The overlapping area.

[0018] Preferably, determining whether each position within the overlapping region is a dynamic target includes: determining whether the first position within the overlapping region is a dynamic target. Each position corresponds to and The grayscale value in the middle is denoted as and ,like Then the first This position is not a dynamic target; if Then the first Each position is a dynamic target.

[0019] This invention assesses whether a target is dynamic by comparing the grayscale values ​​of different images at the same location within an overlapping area to provide an accurate location basis for the final artifact-free stitching.

[0020] The beneficial effects of this invention are as follows:

[0021] (1) This invention distinguishes between dynamic and static information by identifying possible static matching point pairs, which solves the problem that traditional methods cannot identify dynamic noise sources, and enables subsequent processing to be based on reliable static features, thereby improving the quality of the stitched image;

[0022] (2) The present invention uses static confidence weighted reprojection error as the evaluation standard, which ensures that the calculation process of the optimal projection matrix is ​​completely dominated by a highly reliable static background, which can effectively shield the erroneous motion information of dynamic targets, so that the calculated transformation relationship can accurately reflect the camera attitude change, thereby achieving misaligned background stitching.

[0023] (3) By performing targeted processing on the dynamic target area, the present invention can completely eliminate the visual artifacts commonly found in traditional stitching and present the clear state of the dynamic target at the latest moment in the stitched image, thus ensuring the visual authenticity of the final panoramic image. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating an image stitching method for a low-altitude unmanned aerial vehicle (UAV) monitoring network according to the present invention;

[0025] Figure 2 This is a schematic representation of the reference image. ;

[0026] Figure 3 This is a schematic representation of the images to be stitched together. ;

[0027] Figure 4 This schematically illustrates the application of the traditional phase correlation method. Figure 2 and Figure 3 The splicing effect diagram;

[0028] Figure 5 This is an illustrative representation of the invention. Figure 2 and Figure 3 The spliced ​​effect diagram. Detailed Implementation

[0029] 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, not all, of the embodiments of the present invention. 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.

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] This invention discloses an image stitching method for a low-altitude unmanned aerial vehicle (UAV) monitoring network, referring to... Figure 1This includes steps S1 to S5:

[0032] S1. Obtain the reference image and the image to be stitched and extract and All matching point pairs.

[0033] It should be noted that, in order to avoid missing areas during the stitching of panoramic images, the stitching process usually starts from the first two adjacent frames captured. However, in high-resolution image sequences continuously captured by drones along a predetermined route, adjacent frames often have large overlapping areas, meaning that the feature points in the overlapping areas have a high degree of matching. Therefore, it is necessary to perform coarse matching on adjacent frames to filter out feature points that may be overlapping areas, thereby providing data for subsequent stitching analysis.

[0034] Specifically, this invention acquires a sequence of high-resolution images continuously captured by a drone along a predetermined flight path. An example analysis is performed on the first two adjacent frames; the first image in the temporal sequence of the first two adjacent frames is used as a reference image. The second image in the first two adjacent frames is used as the image to be stitched together. ;like Figure 2 For reference image , Figure 3 Images to be stitched .

[0035] Considering the common complexities such as rotation and scale changes in aerial photography, the Scale Invariant Feature Transform (SIFT) algorithm, which has good invariance to these transformations, is used for initial keypoint identification, thereby improving the accuracy of initial matching.

[0036] right and Perform grayscale conversion to obtain the grayscale value of each pixel in each image; simultaneously, it is also necessary to obtain the position of each pixel in each image. Specifically, take the bottom left corner pixel of each original image as the origin, and define the horizontal direction to the right from the origin as... The positive direction of the axis is defined by taking the vertically upward direction from the origin as... The positive direction of the axis is used to construct a Cartesian coordinate system; the position of each pixel in the original image is obtained in the Cartesian coordinate system, including... Coordinates in the axial direction and Coordinates along the axis.

[0037] Get and The matching point pairs are obtained in the following way: and The grayscale image is input into the SIFT algorithm, which detects stable keypoints and generates a 128-dimensional descriptor vector for each keypoint. This vector encapsulates the gradient direction and magnitude information of the neighborhood around the keypoint, thus obtaining all matching point pairs. It should be noted that the SIFT algorithm is an existing technology and will not be described in detail here.

[0038] Thus, we have obtained and All matching point pairs.

[0039] S2. Determine the static confidence level of each matching point pair based on the difference between the disparity of the position of each matching point pair and the mode of the disparity of the positions of all matching point pairs.

[0040] It should be noted that because there are moving vehicles, pedestrians and other dynamic targets in the drone aerial images, the matching degree of the points corresponding to these areas is relatively high. If the matching point pairs corresponding to dynamic and static target areas are not distinguished, the accuracy of the transformation will be affected by the uncertainty of the movement of dynamic targets when calculating the transformation parameters of the image to be stitched. Therefore, it is necessary to analyze each matching point pair and evaluate its static confidence to reduce the interference of dynamic targets.

[0041] Specifically, the disparity of the position of each matching point pair is calculated, where the disparity is equal to the Euclidean distance between the two matching points in each pair; the static confidence of each matching point pair is determined based on the difference between the disparity of the position of each matching point pair and the mode of the disparities of the positions of all matching point pairs; the static confidence satisfies the expression:

[0042]

[0043] In the formula, For the first The static confidence of each matching pair; For the first The disparity of the positions of each matching point pair; This represents the index value and total number of all matching point pairs; The mode function; To determine the absolute value sign; These are preset hyperparameters; It is a natural exponential function.

[0044] in, Reflecting the The value represents the degree of difference between the disparity of a matched point pair and the mode of disparity generated by most static background points. A larger value indicates a more significant difference between the motion pattern of the matched point pair and the motion pattern of most static background points in the scene. This means the matched point pair has a higher probability of being located on an independent dynamic target rather than a stable static background, and the static confidence of the matched point pair is considered low. Conversely, a smaller value indicates a high confidence level. The preset hyperparameters... It can be taken as 0.001 to avoid the case where the denominator is 0.

[0045] At this point, the static confidence scores for all matching point pairs have been obtained.

[0046] S3. Cluster and filter the static confidence scores to obtain all static matching point pairs and perform linear transformations on multiple samples to obtain multiple static projection matrices. Based on the static confidence scores of each matching point pair, weight the reprojection distance of the corresponding matching point in each static projection matrix to determine the reprojection error under each static projection matrix.

[0047] It should be noted that, considering the dynamic target is in real-time motion, meaning the corresponding dynamic target changes within the same coordinate system, it is possible to obtain all matching point pairs... and The optimal geometric mapping relationship of the matching points, such that The corresponding matching point is transformed to In the coordinate system, this can help to accurately distinguish dynamic targets. Since dynamic targets have a significant impact on the final stitching quality, it is more appropriate to select potentially static matching point pairs for analysis of the best geometric mapping relationship. Furthermore, dynamic targets are relatively small in images taken by drones, and their corresponding matching point pairs are relatively few compared to static targets. Therefore, we can extract the matching point pairs with higher static confidence from all matching point pairs and perform linear transformation to obtain the static geometric mapping relationship.

[0048] Specifically, static matching point pairs are obtained by using the static confidence scores of all matching point pairs as clustering samples and performing k-means clustering. The clustering results are obtained, and the average static confidence scores of all matching point pairs within each cluster are calculated. The matching point pair within the cluster corresponding to the maximum value of the average is taken as the static matching point pair.

[0049] The static projection matrix is ​​obtained by randomly selecting sub-matching point pairs from all static matching point pairs, and then performing a linear transformation on each selected sub-matching point pair set using the Direct Linear Transform (DLT) algorithm to obtain each static projection matrix. For example, the number of times the sub-matching point pair set is randomly selected can be preset to 20, resulting in 20 static projection matrices. The implementer can adjust this as needed. It should be noted that the direct linear transformation is a well-known technique and will not be described in detail here.

[0050] The reprojection error of each static projection matrix is ​​determined by weighting the reprojection distance of the corresponding matching points in each static projection matrix based on the static confidence score of each matching point pair; the reprojection error satisfies the expression:

[0051]

[0052] In the formula, For the first Reprojection error under a static projection matrix; For the first The static confidence of each matching pair; , For the first In each matching point pair corresponding to The position of the matching point and The position of the matching point; For the first One static projection matrix; For the first In each matching point pair corresponding to The location of the matching point, through the first Transform the static projection matrix to Predicted position in the coordinate system; To take the Euclidean distance function; This represents the index value and total number of all matching point pairs.

[0053] in, Reflecting the In each pair of matching points, the corresponding The location of the matching point, through the first Transform the static projection matrix to Predicted position in coordinate system and The reprojection distance of the original matching point in the coordinate system, considering that the independent movement of dynamic targets does not conform to the overall transformation law of the static background, the reprojection distance value of its corresponding matching point after transformation is usually large. In order to avoid these dynamic error values ​​polluting the overall evaluation results, it is necessary to reduce the contribution of matching points with low static confidence, i.e., suspected dynamic target matching points, to the total error. Therefore, by This weighting factor ensures that when calculating the reprojection error, more attention is paid to the alignment of static points, while ignoring the errors caused by dynamic points; that is, when When it is larger, The larger the value, the more likely it is that the first... When the static confidence of a matching point pair is high, its contribution to the total error is greater; conversely, when the static confidence of a matching point pair is low, its contribution to the total error is greater. The smaller, The smaller the value, the more likely the first... The lower the static confidence of a matching point pair, the smaller its contribution to the total error; finally, the weighted errors of all matching point pairs are summed up and averaged to obtain the result. The reprojection error of a static projection matrix transformation.

[0054] For example, in the matching point pair corresponding The position of the matching point in the middle is The static projection matrix is , two-dimensional coordinates Convert to homogeneous coordinates Perform matrix multiplication to obtain new homogeneous coordinates. Transforming homogeneous coordinates back to two-dimensional coordinates using the current projection matrix yields the result. The predicted position in the coordinate system is .

[0055] At this point, the reprojection error degree under each static projection matrix has been obtained.

[0056] S4. Determine the optimal static projection matrix; based on the optimal static projection matrix, ... Transform to In the coordinate system, obtain Transformed image as well as and The overlapping region; based on each position within the overlapping region in and The difference in grayscale values ​​is used to determine whether each location within the overlapping area is a dynamic target.

[0057] It should be noted that the reprojection error is calculated after comprehensively considering the static confidence of all matching points. Therefore, the smaller the value, the better the corresponding projection matrix can fit the geometric transformation relationship of static background points in the scene. This means that among all static projection matrices, the static projection matrix that minimizes the reprojection error is the one that can most accurately describe the geometric mapping relationship of the static background rigid transformation between the two images. Through transformation processing, the overlapping area in the two images can be obtained. Considering that the dynamic target is moving in real time, if there are still significant differences between pixels at the same position in the aligned overlapping area, such differences can only be caused by the movement of the dynamic target. Therefore, by obtaining the overlapping area and calculating the grayscale difference corresponding to each position in the overlapping area, the dynamic target can be identified, providing key data indicators for subsequent high-quality seamless stitching.

[0058] Specifically, the optimal static projection matrix is ​​obtained as follows: traverse all the reprojection error degrees under the static projection matrices, and extract the static projection matrix corresponding to the minimum reprojection error degree as the optimal static projection matrix.

[0059] Get Transformed image The method for obtaining it is as follows: Combine the optimal static projection matrix with... Multiply the positions of each pixel in the image to obtain the result. Transform to The coordinate system corresponds to the position of each pixel, and the gray value of the pixel corresponding to the position is filled with . The grayscale value of the corresponding pixel is obtained. The transformed image.

[0060] Considering that the optimal static projection matrix is ​​obtained through the static target transformation of two images, meaning that the overlapping areas of the two images in the same coordinate system are theoretically locations that appear twice, therefore, obtaining... and The overlapping region is obtained as follows: traverse... and In the entire dataset, positions where the same position appears twice are marked as overlapping regions. and The overlapping area.

[0061] Based on each location within the overlapping region and The difference in grayscale values ​​determines whether each location within the overlapping area is a dynamic target: if This indicates that within the overlapping region, the first... Each position corresponds to and If the grayscale values ​​are equal, it means that the location belongs to the static background, therefore, its dynamic target identifier is set to 0, meaning that the location is not a dynamic target; conversely, if... This indicates that within the overlapping region, the first... Each position corresponds to and The grayscale value in the image has changed. Since the static background is aligned, this difference can only be attributed to the independent movement of the dynamic target. Therefore, its dynamic target identifier is set to 1, indicating that this position is a dynamic target. For the overlapping region, the first Each position corresponds to grayscale values ​​in; For the overlapping region, the first Each position corresponds to The grayscale value in the image.

[0062] At this point, the positions of all dynamic targets within the overlapping area have been obtained.

[0063] S5, will and Overlapping placement, where The grayscale value of the position of each dynamic target in the image is replaced with... By matching the grayscale values ​​at the same positions, the images can be stitched together.

[0064] It should be noted that, considering the unpredictable motion information of actual dynamic targets, which can affect the consistency of the static background's translation, leading to contamination of the transformation process and a significant reduction in stitching quality; after obtaining the positional information of the dynamic targets within the overlapping area, considering... The dynamic target position ratio It can better reflect the true state, therefore, through the analysis of The location information of the dynamic target in the middle is masked, thereby significantly reducing the ghosting problem in the stitching.

[0065] Specifically, will and Overlapping placement, where The grayscale value at the corresponding position of each dynamic target in the image is replaced with... The image stitching result is obtained by taking the gray values ​​at the same positions in the image.

[0066] For example, Figure 4 For traditional phase correlation method Figure 2 and Figure 3 The stitched image clearly shows severe ghosting and blurring, with double images appearing for roads, building outlines, and vehicles. Figure 5 For the present invention to Figure 2 and Figure 3The stitched image shows that by separating moving and static targets, the shortcomings of traditional methods have been solved, ensuring the high quality and realism of the final stitched image.

[0067] This completes the work on... and Stitching; use the stitched image as the reference image for the new round, and use the image with the later time sequence in the next two adjacent frames as the image to be stitched in the new round. Repeat steps S1-S5 to obtain the stitched image for the new round, until all time sequence images of the UAV monitoring network are stitched together.

Claims

1. An image stitching method for low-altitude unmanned aerial vehicle monitoring network, characterized in that, The method comprises the following steps: Acquiring a reference image and the image to be stitched and extracting all matching point pairs of and determining a static confidence of each matching point pair according to a difference between a parallax of a position of each matching point pair and a mode of parallaxes of positions of all matching point pairs; clustering and screening the static confidence to obtain all static matching point pairs; linearly transforming multiple samplings of the static matching point pairs to obtain multiple static projection matrices; weighting a re-projection distance of a corresponding matching point of each matching point pair in each static projection matrix based on the static confidence of the matching point pair to determine a re-projection error degree under each static projection matrix; and taking a static projection matrix with the minimum re-projection error as an optimal static projection matrix. based on the optimal static projection matrix transform to the coordinate system of the transformed image and the overlapping area with ; according to the difference of the gray value of each position in and in the overlapping area, determine whether each position in the overlapping area is a dynamic target; The image is divided into a plurality of dynamic targets, and the image is divided into a plurality of dynamic targets. And The image is overlaid and placed, wherein the gray value of the position of each dynamic target in the image is replaced by the gray value of the same position in the image. The image is overlaid and placed, wherein the gray value of the position of each dynamic target in the image is replaced by the gray value of the same position in the image. The image is overlaid and placed, wherein the gray 2.The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, characterized in that, The extraction And All matching point pairs, including: To and grayscale, using scale invariant feature transform algorithm to detect and key points in the grayscale image and generate descriptors; based on the descriptors, matching is performed to obtain all matching point pairs. 3.The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, characterized in that, The static confidence of each matching point pair satisfies an expression as follows: ； wherein, is the static confidence of the th matching point pair; is the disparity of the position of the th matching point pair; is the index value and total number of all matching point pairs; is the mode function; is the absolute value sign; is the preset hyperparameter; is the natural exponential function.

4. The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 3, characterized in that, The parallax of the position of the matching point pair is equal to an Euclidean distance between two matching points in the matching point pair.

5. The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, characterized in that, The method for obtaining all static matching point pairs comprises the following steps: taking the static confidence of all matching point pairs as clustering samples to perform k-means clustering to obtain multiple clustering clusters; and calculating an average value of all static confidences in each clustering cluster, and taking matching point pairs in a clustering cluster with the maximum average value as static matching point pairs. 6.The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, wherein, The method for linearly transforming multiple samplings of the static matching point pairs to obtain multiple static projection matrices comprises the following steps: randomly sampling all static matching point pairs multiple times to obtain multiple sub-matching point pair sets; and using a direct linear transformation algorithm to calculate the multiple static projection matrices from each sub-matching point pair set.

7. The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, characterized in that, The re-projection error degree under each static projection matrix satisfies an expression as follows: ； In the formula, For the first Reprojection error under a static projection matrix; For the first The static confidence of each matching pair; , For the first In each matching point pair corresponding to The position of the matching point and The position of the matching point; For the first One static projection matrix; For the first In each matching point pair corresponding to The location of the matching point, through the first Transform the static projection matrix to Predicted position in the coordinate system; To take the Euclidean distance function; This represents the index value and total number of all matching point pairs. 8.The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, wherein, The acquisition Transformed image , comprising: multiply the optimal static projection matrix with the position of each pixel point in the coordinate system of the image, to obtain the position of each pixel point in the coordinate system of the image after the transformation transform the position of each pixel point in the coordinate system of the image to the coordinate system of the image, fill the gray value of the pixel point corresponding to the position as the gray value of the pixel point corresponding to the position in the coordinate system of the image, to obtain the image after the transformation​​​ 9.The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, wherein, The The overlapping region of The overlapping region of The overlapping region of The overlapping region of The overlapping region of The overlapping region of 10. The image stitching method for low-altitude unmanned aerial vehicle monitoring network according to claim 1, characterized in that, The method for determining whether each position in the overlapping area is a dynamic target comprises the following steps: Let the gray value of the position in the overlapping area in the first image be denoted as Let the gray value of the position in the overlapping area in the second image be denoted as Let the gray value of the position in the overlapping area in the third image be denoted as If , then the position is not a dynamic target; if , then the position is a dynamic target. ​​​​

Images