A UAV image processing method for terrain mapping

Abstract

The present application relates to the field of image processing technology, and specifically to a method for processing drone images for terrain mapping, the method comprising: obtaining a terrain mapping grayscale image; determining the terrain feature threshold of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains; obtaining the number of pyramid layers of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains in the refined feature image, and combining the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block, so as to splice the terrain mapping image. The present application improves the quality of image splicing by optimizing the selection and matching of terrain image feature points.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method for processing unmanned aerial vehicle images for terrain mapping. Background Art

[0002] Topographic surveying and mapping is an indispensable basic work in many engineering and scientific research activities, and the acquisition of terrain data is an important part of it. With the continuous advancement of drone technology, the ground observation capabilities of drones have gradually improved. Topographic surveying and mapping technology based on drone images has shown unique advantages and has been widely used in land and resources monitoring, flood control and drought relief, engineering construction, etc. In the actual surveying and mapping process, drones need to take multiple photos of the entire target area and use image stitching technology to stitch images with overlapping areas to generate the final topographic surveying and mapping image.

[0003] At present, the terrain mapping technology based on drone images mainly uses feature matching algorithms to stitch the acquired terrain images. The changes in light and shadow captured by drones during flight and the shaking of the onboard camera cause the image feature points selected by the classic feature matching algorithm to have defects such as poor stability and high fuzziness, which increases the deviation of image stitching and causes problems such as target misalignment or artifacts in the overlapping areas of the stitched images. Therefore, how to optimize the selection of drone image feature points in the process of terrain mapping is an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a UAV image processing method for terrain mapping to solve the existing problems.

[0005] The present invention discloses a method for processing unmanned aerial vehicle images for terrain mapping using the following technical solutions:

[0006] An embodiment of the present application provides a method for processing drone images for terrain mapping, the method comprising the following steps:

[0007] Get the terrain survey grayscale map;

[0008] The terrain surveying grayscale image is divided into a plurality of sub-blocks, and the edge of each sub-block is extracted to obtain the edge image of each sub-block;

[0009] Obtaining the connected domain in the edge image of each sub-block, and determining the terrain feature threshold of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains;

[0010] Eliminating the edges of the connected domains whose connected domain areas are larger than the terrain feature threshold in the edge images of each sub-block to obtain the feature images of each sub-block, and refining the feature images to obtain the refined feature images of each sub-block;

[0011] Based on the feature image of each sub-block and the area of ​​all connected domains in the refined feature image, combined with the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block, the number of pyramid layers of each sub-block is obtained to splice the topographic mapping image.

[0012] Preferably, the process of determining the terrain feature threshold is:

[0013] Determine a first threshold value for each sub-block based on the distribution of grayscale values ​​of all pixels in the edge image of each sub-block;

[0014] Determine a second threshold value for each sub-block based on the area of ​​all connected domains in the edge image of each sub-block;

[0015] The terrain feature threshold of each sub-block is determined based on the first threshold and the second threshold.

[0016] Preferably, the first threshold of each sub-block is the average value of the grayscale values ​​of all pixels in the edge image of each sub-block.

[0017] Preferably, the second threshold of each sub-block is the average value of the areas of all connected domains in the edge image of each sub-block.

[0018] Preferably, the expression of the terrain feature threshold of each sub-block is: ; In the formula, represents the terrain feature threshold of the i-th sub-block; represents the first threshold of the i-th sub-block; represents the second threshold of the i-th sub-block; norm() represents the normalization function.

[0019] Preferably, the method for obtaining the number of pyramid layers of each sub-block is:

[0020] Determine the curve characteristic value of each sub-block based on the characteristic image of each sub-block and the area of ​​each connected domain in the refined characteristic image;

[0021] The expression of the number of pyramid layers of each sub-block is: ; In the formula, represents the number of pyramid levels of the i-th sub-block; M is the preset maximum number of pyramid levels; represents the second threshold of the i-th sub-block; represents the curve eigenvalue of the i-th sub-block; ceil() represents the upward rounding function; norm() represents the normalization function.

[0022] Preferably, the method for obtaining the curve characteristic value of each sub-block is:

[0023] Determine a first eigenvalue and a second eigenvalue of each sub-block based on a difference between an area of ​​each connected domain in the feature image of each sub-block and an area of ​​a corresponding connected domain in the refined feature image;

[0024] The curve eigenvalue of each sub-block is determined based on the first eigenvalue and the second eigenvalue.

[0025] Preferably, the determining the first eigenvalue and the second eigenvalue of each sub-block includes:

[0026] The first eigenvalue of each sub-block is the cumulative sum of the ratios of the feature image of each sub-block to the area of ​​the corresponding connected domain in the refined image;

[0027] The second characteristic value of each sub-block is the cumulative sum of the ratios of the characteristic image of each sub-block to the perimeter of the corresponding connected domain in the refined image.

[0028] Preferably, the curve eigenvalue of each sub-block is the product of the first eigenvalue and the second eigenvalue of each sub-block.

[0029] Preferably, the step of stitching the topographic mapping images comprises:

[0030] The edge image and pyramid layer number of each sub-block are used as the input of the feature point extraction algorithm to obtain the feature points of each sub-block;

[0031] All feature points of all sub-blocks in two adjacent frames of terrain mapping images are used as inputs of the random sampling consistency algorithm to obtain the correct matching point pairs between the two adjacent frames and calculate the homography matrix;

[0032] The homography matrix is ​​used to transform the correct matching point pairs into one of the topographic mapping images. The line connecting all the correct matching points in the transformed topographic mapping image is used as the best stitching line. The topographic mapping images of two adjacent frames are spliced ​​and fused in combination with the best stitching line algorithm.

[0033] This application has at least the following beneficial effects:

[0034] This application divides the terrain mapping grayscale image into blocks according to the uneven distribution of image features of the terrain mapping grayscale image; determines the terrain feature threshold of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains to obtain the feature image of the sub-block, while retaining the image detail information, excluding the interference of pixels affected by light and shadow changes; based on the feature image of each sub-block and the area of ​​all connected domains in the refined feature image, combined with the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block, the number of pyramid layers of each sub-block is obtained, so as to enhance the feature point extraction capability in detail-rich areas, and avoid subsequent over-registration caused by extracting more unstable feature points in simple terrain. Finally, the terrain image feature points output by the improved SIFT algorithm are used to splice the terrain mapping images frame by frame to obtain a terrain mapping spliced ​​image of the target area. By optimizing the selection and matching of terrain image feature points, the overall quality of image splicing is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 is a flowchart of the steps of a method for processing UAV images for terrain mapping;

[0037] Figure 2 Schematic diagram of the pyramid layer extraction process;

[0038] Figure 3 A flow chart of the image stitching process for topographic mapping;

[0039] Figure 4 Schematic diagram of the topographic mapping image fusion process. DETAILED DESCRIPTION

[0040] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a drone image processing method for terrain mapping proposed in the present application in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0041] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0042] The following is a detailed description of a specific solution of a UAV image processing method for terrain mapping provided by the present application in conjunction with the accompanying drawings.

[0043] An embodiment of the present application provides a method for processing drone images for terrain mapping. Specifically, the following method for processing drone images for terrain mapping is provided. Figure 1 , the method comprises the following steps:

[0044] Step S1: photograph the terrain using the CMOS camera carried by the drone, and pre-process each frame of the image to obtain a terrain mapping image.

[0045] The terrain is photographed by the CMOS camera carried by the UAV to obtain an RGB spatial image with a size of W*H, where the shooting frame rate of the CMOS camera is F. In order to eliminate the influence of noise and some external interference and enhance the accuracy of subsequent analysis, a filtering algorithm is used to denoise each frame of the RGB spatial image, and each frame is recorded as a topographic mapping image.

[0046] It should be noted that the values ​​of the spatial impact dimensions W and H and the shooting frequency are all set manually. In this embodiment, W and H are 4096 and 2160 respectively, and F is 25. The implementer can set them according to the specific situation, and this embodiment does not impose any special restrictions.

[0047] It should be understood that there are many commonly used image filtering algorithms. In this embodiment, Gaussian filtering is used to reduce the noise of the image. Under the premise of ensuring the purpose of denoising, the implementer can also use image filtering algorithms such as median filtering or bilateral filtering in the prior art. This embodiment does not impose any special restrictions.

[0048] Among them, Gaussian filtering is a well-known technology in image processing technology, and the specific process of noise reduction is not further described in this embodiment.

[0049] The topographic mapping image is a three-channel RGB image, and it is not easy to detect the feature points in the image. Therefore, the topographic mapping image is converted into a grayscale image, which is recorded as a topographic mapping grayscale image.

[0050] Step S2: Determine the terrain feature threshold of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains.

[0051] S201: Divide the terrain mapping grayscale image into blocks.

[0052] The features of different terrains vary greatly, and the geomorphic features of different areas of the same terrain also vary in complexity. For example, hilly terrain has obvious ups and downs, while plain terrain is relatively flat, and the terrain features of the two are significantly different; in plain terrain, the image features of fields planted with the same crop are not obvious, while the images at the junction of different fields have more obvious features.

[0053] The complex and changeable characteristics of terrain lead to uneven distribution of image features in terrain mapping grayscale images. In order to better extract image features in terrain mapping grayscale images and enhance the stitching effect of subsequent terrain mapping images, the terrain mapping grayscale image is divided into a preset number of sub-blocks of equal size, so as to adaptively extract feature points in the sub-blocks according to the feature distribution of different sub-blocks in the image.

[0054] It should be noted that the value of the preset number is artificially set. In this embodiment, the value of the preset number is 200. The implementer can set it according to the specific situation. This embodiment does not impose any special restrictions.

[0055] S202: Determine the terrain feature threshold of each sub-block based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains.

[0056] The changes in ground light and shadow during the movement of the drone will increase the instability of the algorithm in selecting feature points. For example, the reflective points on the lake surface change greatly during movement, resulting in a large difference in the grayscale value of the reflective points in the terrain mapping grayscale image of two adjacent frames, making it a feature point in the previous frame of the image but not a feature point in the next frame of the image, thereby increasing the error of subsequent image stitching.

[0057] There are often curves with stable features in terrain images, such as roads in fields on plains, river banks, valleys and ridges in hills and mountains, etc. These characteristic curves are often very stable. Even after the image undergoes a certain degree of rotation, scaling or illumination changes, these points can still be accurately identified and can be used as good feature points in the image stitching process.

[0058] The smaller the area of ​​the connected domain in each sub-block of the topographic mapping image, the fewer the number of edges it contains, and the more uniform the distribution of the grayscale values ​​of the pixels, the more susceptible the topographic mapping image is to the influence of light and shadow. In order to remove the influence of the shadow formed by the light and shadow transformation on the edge confusion of the terrain, the edges that may be formed by the shadow in the sub-block are deleted by analyzing the area of ​​the connected domain in each sub-block and the distribution of the grayscale values ​​of the pixels. Specifically:

[0059] The edge of each sub-block is extracted to obtain the edge image of each sub-block, each sub-block is used as the input of the edge detection algorithm, and the edge of each sub-block is output to obtain the edge image of each sub-block.

[0060] It should be understood that there are many commonly used edge detection algorithms. In this embodiment, the Sobel algorithm is used to extract the edges in each sub-block. Under the premise of ensuring the purpose of obtaining the edges of each sub-block, the implementer may also use the Canny edge detection algorithm, Prewitt algorithm and other edge detection algorithms to achieve edge extraction. This embodiment does not impose any special restrictions.

[0061] Among them, the Sobel edge detection algorithm is a well-known technology for edge detection in image processing, and the specific process of edge detection is not described in detail.

[0062] The edge image of the sub-block is binarized, and the connected domain in the binarized edge image of each sub-block is extracted.

[0063] It should be noted that there are many commonly used image binarization methods and connected domain extraction algorithms. This embodiment uses the maximum entropy threshold segmentation algorithm and the region growing algorithm to binarize the edge image of each sub-block and extract the connected domain therein. The implementer can choose other image binarization methods and connected domain extraction algorithms according to the specific situation, and this embodiment does not impose any special restrictions.

[0064] The terrain feature threshold of each sub-block is determined based on the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains, specifically:

[0065] The first threshold value of each sub-block is the average of the gray values ​​of all pixels in the edge image of each sub-block, which is recorded as the first threshold value of each sub-block;

[0066] The second threshold of each sub-block is analyzed as the average of the areas of all connected domains in the edge image of each sub-block, which is recorded as the second threshold of each sub-block;

[0067] The expression of the terrain feature threshold of each sub-block is: ; In the formula, represents the terrain feature threshold of the i-th sub-block; represents the first threshold of the i-th sub-block; represents the second threshold of the i-th sub-block; norm() represents the normalization function.

[0068] The less obvious the image features in the sub-block, the more uniform the distribution of its pixel grayscale values, the smaller the first threshold, and the fewer the number of feature points it contains; the richer the image information in the sub-block, the more discrete the distribution of pixel grayscale values, the larger the first threshold, and the greater the probability of containing terrain feature curves.

[0069] Since the sub-block with a larger first threshold contains more detailed information, the characteristic curve in the corresponding terrain is more delicate; while the image edge in the sub-block with a smaller first threshold is more blurred, and the edge is more affected by the change of light and shadow.

[0070] Therefore, the edges in the connected domain whose area in the edge image is larger than the terrain feature threshold are eliminated to obtain a new binary image of the image, which is recorded as the feature image of each sub-block.

[0071] The richer the image detail information in the sub-block, the larger the first threshold, the larger the average area of ​​the detected image edge, the larger the calculated terrain feature threshold, and the more terrain feature curves retained; the more evenly distributed the grayscale values ​​of the pixels in the sub-block, the smaller the first threshold, and the smaller the corresponding terrain feature threshold, so as to delete the shadows formed by light and shadow changes in the sub-block, while retaining the image detail information, eliminating the interference of pixels affected by light and shadow changes.

[0072] Step S3: determining the first eigenvalue and the second eigenvalue of each sub-block based on the difference between the feature image of each sub-block and the area of ​​the corresponding connected domain in the refined feature image; and determining the curve eigenvalue of each sub-block based on the first eigenvalue and the second eigenvalue.

[0073] The jitter of the drone during flight causes the terrain feature curve in the terrain mapping grayscale image to become blurred, which will increase the deviation of feature point extraction between two adjacent frames. At the same time, the narrower the terrain feature curve, the less it is affected by light and shadow changes, and the stronger the stability in the image feature point extraction process. Therefore, in order to eliminate the interference of drone jitter, the curve feature value of each sub-block is determined based on the difference between the feature image of each sub-block and the area of ​​the corresponding connected domain in the refined feature image, specifically:

[0074] The feature image of each sub-block is used as the input of the thinning algorithm to obtain a binary image containing the thinned edge and record it as the thinned feature image of each sub-block. The feature image of each sub-block and the connected domain in the thinned image are in one-to-one correspondence.

[0075] It should be noted that there are many commonly used thinning algorithms. In this embodiment, the Zhang-Suen thinning algorithm is used to thin the feature image of each sub-block. The implementer may also select other thinning algorithms according to the specific situation, and this embodiment does not impose any special restrictions.

[0076] The Zhang-Suen thinning algorithm is a well-known technology in the field of image processing technology, and the specific process of image thinning is not described in detail.

[0077] Analyze the cumulative sum of the ratios of the feature image of each sub-block to the area of ​​the corresponding connected domain in the refined image, and record it as the first eigenvalue of each sub-block;

[0078] Analyze the cumulative sum of the ratios of the feature image of each sub-block to the perimeter of the corresponding connected domain in the refined image, and record it as the second eigenvalue of each sub-block;

[0079] The curve eigenvalue of each sub-block is the product of the first eigenvalue and the second eigenvalue of each sub-block.

[0080] On the one hand, the terrain characteristic curve with a smaller change in the area of ​​the connected domain indicates that it is less blurred by the shaking of the drone camera; on the other hand, the smaller the area change of the terrain characteristic curve, the smaller the probability that the characteristic curve is affected by the change of light and shadow. The smaller the ratio of the connected domain area of ​​the same terrain characteristic curve in the sub-block feature image and the sub-block refined feature image, the smaller the calculated curve characteristic value of the sub-block, and the stronger the stability of its pixel points as feature points extracted by subsequent algorithms. At the same time, the longer the length of the terrain characteristic curve, the more obvious the characteristics of its pixel points, the greater the reliability as the image splicing feature point, the larger the area of ​​the connected domain of the corresponding refined terrain characteristic curve, and the smaller the calculated curve characteristic value of the sub-block.

[0081] Step S4: based on the feature image of each sub-block and the area of ​​all connected domains in the refined feature image, combined with the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block, the number of pyramid layers of each sub-block is obtained.

[0082] The SIFT algorithm constructs a scale space through a Gaussian pyramid to extract feature points at different scales of the image. The complex terrain changes in the terrain mapping grayscale image cause the distribution of feature points of the image in different sub-blocks to be different. According to the characteristics of the sub-blocks of the terrain mapping grayscale image, the number of pyramid layers for each sub-block is calculated: ; Wherein, M is the preset maximum number of pyramid layers, and in this solution, M is 7; ceil() represents the upward rounding function; norm() represents the normalization function; represents the second threshold of the i-th sub-block; represents the curve eigenvalue of the i-th sub-block; Indicates the number of pyramid levels of the i-th sub-block.

[0083] The larger the second threshold of the grayscale image sub-block of terrain mapping and the smaller the curve characteristic value, the richer the detail information in the sub-block, and the stronger the reliability and stability of the terrain characteristic curve as the image stitching feature point, the larger the number of pyramid layers of the calculated sub-block, which provides more detailed scale changes for the sub-block, helps to detect feature points at different scales, and enhances the ability to extract feature points in areas with rich details. In addition, reducing the number of pyramid layers of sub-blocks with unclear image features can avoid subsequent over-registration caused by extracting more unstable feature points in simple terrain, while reducing the amount of algorithm calculation.

[0084] Each sub-block of the topographic mapping image and its pyramid layer number are used as the input of the SIFT algorithm, and feature points of each sub-block are extracted respectively to improve the accuracy and reliability of image feature point selection, and the feature points of each sub-block in the topographic mapping image are output.

[0085] The SIFT algorithm is a well-known technology in the field of image processing, and the specific process of extracting feature points is not described in detail.

[0086] Preferably, as an embodiment of the present application, the schematic diagram of the pyramid layer number extraction process is as follows: Figure 2 shown.

[0087] Step S5: Based on the output results of the image registration, the topographic mapping images are spliced ​​frame by frame using the best stitching line method to obtain a topographic mapping spliced ​​image of the target area, thereby completing the processing of the topographic mapping images taken by the drone.

[0088] According to the feature points of each sub-block in the terrain mapping grayscale image obtained in step S4, the terrain mapping image is spliced, specifically:

[0089] All feature points of all sub-blocks in two adjacent frames of topographic mapping images are used as inputs to the RANSAC random sampling consensus algorithm to obtain the correct matching point pairs between the two adjacent frames, and the homography matrix is ​​calculated. The homography matrix transforms these correct matching point pairs into one of the topographic mapping images, and the connecting line of these correct matching point pairs in the topographic mapping image is used as the best stitching line. Combined with the best stitching line algorithm, the topographic mapping images of two adjacent frames are spliced ​​and fused to obtain the spliced ​​topographic mapping image.

[0090] The stitched topographic mapping image is used as a new topographic mapping image to be stitched and fused with the next frame of topographic mapping image. All topographic mapping images are traversed and stitched and fused frame by frame to complete the processing of the topographic mapping impact of the drone photography.

[0091] Among them, the RANSAC algorithm, homography matrix and optimal seam line algorithm are well-known technologies in the field of image processing technology, and their specific implementation processes are not repeated here.

[0092] Preferably, as an embodiment of the present application, the topographic mapping image stitching process flow chart is as follows: Figure 3 shown.

[0093] Preferably, as an embodiment of the present application, a topographic mapping image fusion process diagram is as follows: Figure 4 shown.

[0094] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0095] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0096] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for processing unmanned aerial vehicle images for terrain mapping, characterized in that: The method comprises the following steps: Get the terrain survey grayscale map; The terrain surveying grayscale image is divided into a plurality of sub-blocks, and the edge of each sub-block is extracted to obtain the edge image of each sub-block; Obtaining a connected domain in the edge image of each sub-block, determining a terrain feature threshold of each sub-block based on the distribution of gray values ​​of all pixels in the edge image of each sub-block and the area of ​​all connected domains, determining a first threshold of each sub-block based on the distribution of gray values ​​of the pixels, and determining a second threshold of each sub-block based on the area of ​​the connected domain; Eliminating the edges of the connected domains whose connected domain areas are larger than the terrain feature threshold in the edge images of each sub-block to obtain the feature images of each sub-block, and refining the feature images to obtain the refined feature images of each sub-block; Based on the area of ​​all connected domains in the feature image and refined feature image of each sub-block, combined with the distribution of the grayscale values ​​of all pixels in the edge image of each sub-block, the number of pyramid layers of each sub-block is obtained to splice the topographic mapping image; The method for obtaining the number of pyramid layers of each sub-block is: The accumulated sum of the ratios between the areas of each connected domain in the feature image of each sub-block and the corresponding connected domain in the refined feature image and the accumulated sum of the ratios between the perimeters are used as the first eigenvalue and the second eigenvalue of each sub-block; and the curve eigenvalue of each sub-block is obtained based on the first eigenvalue and the second eigenvalue; The expression of the number of pyramid layers of each sub-block is: Where M i represents the number of pyramid levels of the i-th sub-block; M is the preset maximum number of pyramid levels; E i represents the second threshold of the i-th sub-block; ratio i represents the curve eigenvalue of the i-th sub-block; ceil() represents the upward rounding function; norm() represents the normalization function.

2. The method for processing unmanned aerial vehicle images for terrain mapping according to claim 1, characterized in that: The process of determining the terrain feature threshold is as follows: Determine a first threshold value for each sub-block based on the distribution of grayscale values ​​of all pixels in the edge image of each sub-block; Determine a second threshold value for each sub-block based on the area of ​​all connected domains in the edge image of each sub-block; The terrain feature threshold of each sub-block is determined based on the first threshold and the second threshold.

3. The method for processing unmanned aerial vehicle images for terrain mapping as claimed in claim 2, characterized in that: The first threshold of each sub-block is the average value of the grayscale values ​​of all pixels in the edge image of each sub-block.

4. The method for processing unmanned aerial vehicle images for terrain mapping as claimed in claim 2, characterized in that: The second threshold of each sub-block is the average value of the areas of all connected domains in the edge image of each sub-block.

5. The method for processing unmanned aerial vehicle images for terrain mapping as claimed in claim 2, characterized in that: The expression of the terrain feature threshold of each sub-block is: Where Th i represents the terrain feature threshold of the i-th sub-block; represents the first threshold of the i-th sub-block; E i represents the second threshold of the i-th sub-block; norm() represents the normalization function.

6. The method for processing unmanned aerial vehicle images for terrain mapping according to claim 1, characterized in that: The curve eigenvalue of each sub-block is the product of the first eigenvalue and the second eigenvalue of each sub-block.

7. The method for processing unmanned aerial vehicle images for terrain mapping according to claim 1, characterized in that: The method of splicing topographic mapping images comprises: The edge image and pyramid layer number of each sub-block are used as the input of the feature point extraction algorithm to obtain the feature points of each sub-block; All feature points of all sub-blocks in two adjacent frames of terrain mapping images are used as inputs of the random sampling consistency algorithm to obtain the correct matching point pairs between the two adjacent frames and calculate the homography matrix; The homography matrix is ​​used to transform the correct matching point pairs into one of the topographic mapping images. The line connecting all the correct matching points in the transformed topographic mapping image is used as the best stitching line. The topographic mapping images of two adjacent frames are spliced ​​and fused in combination with the best stitching line algorithm.

Images