A multi-period slope unmanned aerial vehicle image positioning and orientation method for establishing pseudo control points

Abstract

The application discloses a multi-period slope unmanned aerial vehicle image positioning and orientation method of establishing pseudo control points, and comprises the following steps: S1, acquiring multi-period unmanned aerial vehicle images and POS data of a highway slope; S2, acquiring DOM data and DSM data; S3, acquiring homonymic feature points between the multi-period DOM data; S4, creating a non-vegetation-covered area terrain mask; S5, creating a stable ground surface area terrain mask; S6, pseudo control point screening and extraction; and S7, repositioning and reorienting the unmanned aerial vehicle images based on the pseudo control points.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle aerial image positioning, and particularly relates to a multi-period slope unmanned aerial vehicle image positioning and orientation method for establishing pseudo control points. BACKGROUND

[0002] A slope is a free face with a certain inclination formed by rock mass, soil mass under the action of natural gravity or human factors. Generally, the slope has a large inclination, and when the tensile force in the longitudinal direction exceeds the tensile resistance of the slope, a natural disaster such as landslide or rockslide occurs. Therefore, it is crucial to monitor and prevent such slopes. The time series observation of the ground surface helps to quantify the geometric shape and motion state of the slope and landslide and provide early stability indicators, so that detailed and targeted disaster investigation and prevention can be carried out.

[0003] The unmanned aerial vehicle equipped with a miniature camera and the combination of motion recovery structure and multi-view stereo vision accelerate and simplify the establishment process of ground surface measurement products such as high-resolution dense point cloud and DSM, and become a time-saving, labor-saving, low-cost and simple and fast ground surface monitoring method. The method has been widely applied in the research of ground surface monitoring in various dynamic environments. The method mainly lays out uniform ground control points in the monitoring area, measures the geographic coordinates of the ground control points, and provides positioning and orientation reference for products such as dense point cloud and DSM, and then realizes multi-period ground surface investigation or monitoring.

[0004] However, in areas prone to geological disasters, laying out and maintaining ground control points is a high-cost and time-consuming activity, and it is difficult to ensure the stability and non-deformation of the ground control points. Therefore, the existing positioning and orientation of multi-period unmanned aerial vehicle image data only through ground control points has great limitations.

[0005] Therefore, the present application provides a multi-period slope unmanned aerial vehicle image positioning and orientation method for establishing pseudo control points, which can solve the above problems. SUMMARY

[0006] The technical problem to be solved by the present application is that in areas prone to geological disasters, laying out and maintaining ground control points is a high-cost and time-consuming activity, and it is difficult to ensure the stability and non-deformation of the ground control points. And the existing positioning and orientation of multi-period unmanned aerial vehicle image data only through ground control points has great limitations. Therefore, the present application provides a multi-period slope unmanned aerial vehicle image positioning and orientation method for establishing pseudo control points, which comprises:

[0007] S1, acquiring multi-period unmanned aerial vehicle images and POS (position and orientation system) data of a ground surface monitoring area;

[0008] S2, according to the POS data, the above-mentioned multi-period unmanned aerial image is respectively constructed into the DSM (Digital Surface Model) data and DOM (Digital Orthophoto Map) data of the monitoring area, and one period of data is selected as the reference data;

[0009] S3, for the multi-period DOM data constructed above, image feature detection method is used to extract candidate points which can be used as pseudo control points in the multi-period DSM data, and the candidate points extracted in each period are compared to screen out the same name pseudo control candidate points;

[0010] S4, for the multi-period DOM data constructed above, according to the spectral characteristics of the unmanned aerial image, the vegetation coverage area and the non-vegetation area of the monitoring area are extracted;

[0011] S5, for the multi-period DSM data constructed above, according to the elevation information in the DSM data, the detection area is divided into smooth surface area and rough surface area by using the elevation fluctuation, so as to establish the terrain mask of stable ground area and unstable ground area.

[0012] S6, the non-vegetation area and the stable ground area of the monitoring area extracted above are integrated to construct the pseudo control point candidate area of each period of unmanned aerial image DOM data, and the pseudo control point candidate points located in the area are extracted as the final pseudo control points;

[0013] S7, for the pseudo control points extracted above, except for the reference data, they are involved in the positioning and orientation process of other each period of unmanned aerial image, and the DSM data and DOM data of each period of unmanned aerial image are reconstructed, and the positioning and orientation reference consistent with the reference data is maintained.

[0014] Further, the step S2 described above according to the POS data to construct the multi-period unmanned aerial image into the DOM data and DSM data of the monitoring area, specifically includes:

[0015] According to the coordinate information and positioning information recorded in the POS data, the unmanned aerial image data is subjected to self-calibration block adjustment with additional parameters, image point cloud construction, dense matching, spatial interpolation, digital differential correction, splicing and inlay processing to obtain the DSM data and DOM data of each period of monitoring area;

[0016] Further, the same name pseudo control candidate point is obtained in the step S3 as follows:

[0017] The ORB algorithm is used to extract clear and easily identifiable edge feature points from the multi-period DOM data as pseudo control candidate points;

[0018] The corresponding point detection is performed on the edge feature points by using the RANSAC algorithm, and the false control candidate points without corresponding points in the multi-period DOM data are removed. The RANSAC algorithm achieves the target by repeatedly selecting a random subset of data, and the basic idea is as follows:

[0019] A group of points is randomly selected from the data set, and the current model relationship is calculated;

[0020] All other data points are tested by using the model relationship calculated above, if the error of each point to the model is less than the set threshold, the point is regarded as an "inlier", otherwise it is regarded as an "outlier";

[0021] If there are enough points classified as "inliers", the model relationship is reasonable enough, the model with the largest number of "inliers" so far is retained, and it is recorded as the best model relationship;

[0022] The above steps are repeated until the set number of iterations is reached, and the best model relationship is selected as the final model, so that the points outside the model are removed.

[0023] Further, the step S4 uses the spectral features of the unmanned aerial vehicle image to extract the vegetation coverage area and the non-vegetation coverage area, specifically including:

[0024] The V value index of each period of DOM data is calculated by formula , wherein R, G and B are the pixel values of each period of DOM at red, green and blue three bands respectively, and the value range of the V value index is [-1, 1].

[0025] According to the V value index calculated in the last step, it presents two wave peaks and one wave trough in the spatial distribution graph, the V value index at the wave trough is selected as the threshold, the vegetation and non-vegetation are segmented and extracted by using the gray scale segmentation method, and the non-vegetation coverage area is created as a terrain mask;

[0026] Further, the step S5 calculates the relief degree of elevation according to the elevation information in the DSM data, so as to extract and construct the terrain mask of the stable ground surface area and the unstable ground surface area, specifically including:

[0027] The maximum pixel value and the minimum pixel value in the 10*10 neighborhood of each pixel in the DSM data are counted respectively;

[0028] According to the maximum value and the minimum value obtained in the last step, the relief degree of elevation of each pixel in the DSM data is calculated by formula R = H max -H min , wherein H max and H minrespectively the maximum pixel value and the minimum pixel value in a 10*10 neighborhood of each pixel;

[0029] The value of the terrain relief of each period DSM data is taken as the judging basis of the stable area and the unstable area, the stable area and the unstable area are divided by setting a threshold value using the dichotomy, and the stable area extracted is created as a terrain mask;

[0030] Further, the step S6 creates a pseudo control point candidate area, and extracts a final pseudo control point, and specifically includes:

[0031] The terrain mask of the non-vegetation coverage area and the terrain mask of the stable area extracted are integrated, so that a pseudo control point candidate area in each period DOM data is extracted;

[0032] The pseudo control candidate point located in the pseudo control point candidate area is extracted as a final pseudo control point;

[0033] According to the final pseudo control point, the planar coordinates (X, Y) and the elevation coordinates (Z) of the pseudo control point are extracted from the DSM;

[0034] Further, the step S7 uses the pseudo control point to position and orient each period unmanned aerial vehicle image, and constructs new DSM data and DOM data, and specifically includes:

[0035] The pseudo control point is re-engaged in the aerial triangulation process of each period unmanned aerial vehicle image except the reference data in the form of a control point "sting point";

[0036] On the basis, after image point cloud construction, dense matching, spatial interpolation, digital differential rectification, splicing and inlaying processing of each period unmanned aerial vehicle image, the DSM data and the DOM data with consistent positioning and orientation of the reference data are obtained.

[0037] The present application has the following beneficial effects:

[0038] 1. The multi-period unmanned aerial vehicle image positioning and orientation method for establishing a pseudo control point of the present application constructs DSM and DOM through unmanned aerial vehicle images, extracts clear and easily recognizable edge points as pseudo control candidate points, creates a reliable stable area mask using the spectral characteristics and elevation characteristics of the images, and screens out reliable pseudo control points; the pseudo control points are re-engaged in the image aerial triangulation, DSM, DOM and other establishment processes, and in the absence of actual ground control points, the multi-period unmanned aerial vehicle image products are positioned and oriented to the same coordinate reference, so that geological survey or difference extraction can be carried out, and it has practical significance for surface monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1A whole flow chart of the multi-period unmanned aerial vehicle image positioning and orientation method of establishing pseudo control points provided by the present application is provided;

[0040] Figure 2 An unmanned aerial vehicle aerial image schematic diagram provided by the present application is provided, and the unmanned aerial vehicle aerial image is an RGB image;

[0041] FIG. 3 is a DSM data schematic diagram provided by the present application, wherein FIG. 3(a) is DSM data constructed by first-period aerial images, and FIG. 3(b) is DSM data constructed by second-period aerial images;

[0042] FIG. 4 is a DOM data schematic diagram provided by the present application, wherein FIG. 4(a) is DOM data constructed by first-period aerial images, and FIG. 4(b) is DOM data constructed by second-period aerial images;

[0043] Figure 5 A pseudo control candidate point schematic diagram provided by the present application is provided;

[0044] Figure 6 Vegetation coverage area and non-vegetation coverage area terrain masks provided by the present application are provided;

[0045] Figure 7 Stable ground surface area and unstable ground surface area terrain masks provided by the present application are provided;

[0046] Figure 8 A final pseudo control point selection schematic diagram provided by the present application is provided;

[0047] Figure 9 A schematic diagram of repositioning and orientation using pseudo control points provided by the present application is provided. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0049] EMBODIMENT

[0050] At present, the basic method of ground surface investigation or deformation monitoring based on unmanned aerial vehicle images is to obtain multi-period ground surface aerial data, to position and orient the multi-period image data to the same coordinate reference through ground control points, and then to construct DSM, DOM and other products to investigate and monitor the ground surface conditions. However, in the real situation, when monitoring special ground objects such as slopes, the ground control points are prone to displacement or damage with the change of the ground surface, and are difficult to be effectively utilized.

[0051] In order to solve the problem of positioning and orientation of multi-period unmanned aerial images which need actual ground control points, the application provides a multi-period slope unmanned aerial image positioning and orientation method of establishing pseudo control points. Based on unmanned aerial photography, obvious edge corner points which are easy to identify in the image itself are obtained, and candidate points suitable for being pseudo control points are screened out based on spectral information and elevation information of the image, so that the multi-period unmanned aerial images are positioned to the same coordinate reference without actual ground control points.

[0052] The application discloses a multi-period slope unmanned aerial image positioning and orientation method of establishing pseudo control points, and the application is further described in detail below in combination with the drawings and the specific embodiments.

[0053] Figure 1 The application provides a specific flowchart of the multi-period unmanned aerial image positioning and orientation method of establishing pseudo control points, and the application specifically includes the following steps:

[0054] Step S1: Obtain multi-period unmanned aerial images and POS data of a ground monitoring area, and the following example is set to mainly use two-period images.

[0055] Figure 2 The application provides a schematic diagram of an unmanned aerial photography image.

[0056] Step S2: According to the POS data, the multi-period unmanned aerial images are respectively constructed into DSM data and DOM data of the monitoring area, and the data of one period is selected as the reference data.

[0057] Step S2.1: According to the coordinate information and positioning information recorded in the POS data, the unmanned aerial image data is subjected to self-checking regional network adjustment with additional parameters, image point cloud construction and dense matching processing, so as to obtain point cloud data of each period of images.

[0058] Step S2.2: The point cloud data is established as a grid-shaped DSM, so as to accurately express the real ground form. FIG. 3 is a schematic diagram of DSM data provided by the application, wherein FIG. 3(a) is DSM data constructed by the first period of aerial photography images, and FIG. 3(b) is DSM data constructed by the second period of aerial photography images.

[0059] Step S2.3: The original unmanned aerial image is subjected to image orthographic correction in combination with the DSM data constructed above, and then the overlapping area of the orthographic image is cut and inlaid, so as to obtain complete DOM images of the research area. FIG. 4 is a schematic diagram of DOM data provided by the application, wherein FIG. 4(a) is DOM data constructed by the first period of aerial photography images, and FIG. 4(b) is DOM data constructed by the second period of aerial photography images.

[0060] Step S2.4. In the multi-period unmanned aerial vehicle image data, optionally one period is taken as a reference, and in the present example, the first period image data is taken as the reference data.

[0061] Step S3: For the multi-period DOM data constructed above, image feature detection method is used to extract candidate points that can be used as pseudo control points in the multi-period DSM data, and the candidate points extracted in each period are compared to screen out the same name pseudo control candidate points.

[0062] Step S3.1. Extract image candidate feature points using ORB algorithm.

[0063] Step S3.1.1. Construct Gaussian pyramid

[0064] The above original DOM data is continuously down-sampled, and Gaussian filtering is added on the basis of down-sampling to make the DOM image data more smooth. The images of each layer of the image pyramid are Gaussian blurred using different filters to construct a Gaussian image pyramid of 4 groups and 4 layers.

[0065] Step S3.1.2. Feature point detection

[0066] For each pixel P (the value of the pixel is I p ) in the above DOM image data, a circular region is constructed with its own position as the center and 3 as the radius;

[0067] The boundary of the circular region passes through 16 pixels, i.e. n |n=1,2,...,16};

[0068] The pixel value of pixel P is compared with the pixel values of the 16 pixels, and if the pixel value of pixel P is greater or smaller than the pixel values of the 16 pixels, the point P can be taken as a feature candidate point, and m can be usually taken as 9 or 12;

[0069] Among all the detected feature candidate points, adjacent candidate points are screened out. The difference sum of each point with the 16 pixels around it is calculated as the response size of the current candidate point;

[0070] The response size values of adjacent candidate points are compared, and the candidate point with smaller response size value is removed. The remaining points are the final feature points.

[0071] Step S3.1.3. Feature point main direction calculation

[0072] Each feature point and its 16 neighboring points are taken as a whole image block, and the moment of the whole image block is:

[0073]

[0074] In the above formula, x and y are the coordinates of the pixel, I(x,y) is the corresponding pixel gray value, and p and q take values ​​of [0,1].

[0075] Based on this, the centroid C of each image patch is calculated, i.e.:

[0076]

[0077] Connect the feature point to the calculated centroid C of the image patch to obtain the direction vector of the feature point, and calculate the principal direction of the feature point using the following formula:

[0078] θ = arctan(m) 01 ,m 10 );

[0079] Step S3.1.4 Feature Description

[0080] A local circular region is constructed with each of the above feature points as the center and a certain distance as the radius;

[0081] Within this circular area, 256 pairs of pixels are selected, and the pixel grayscale values ​​of each pair are compared and assigned binary values ​​as shown in the following formula:

[0082]

[0083] In the above formula, p(x) and p(y) are the pixel values ​​of the selected random points, respectively. After comparing and assigning values ​​to all the selected pixel values ​​as described above, a binary code is obtained, which is the feature descriptor of the feature point.

[0084] Step S3.1.5 Feature Matching

[0085] Based on the feature descriptor of each feature point, the similarity between pairwise feature points is calculated using Hamming distance. The two points with the smallest Hamming distance become a pair of corresponding feature points.

[0086] Step S3.2 Remove incorrect matching points and filter candidate points of the same name for pseudo-control points between multiple images.

[0087] Assume that the feature points extracted above are {Q} in the first DOM data and the second DOM data, respectively. ORB |(x i ,y i {i = 1, 2, ..., n} and {P} ORB |(x' i ,y' i ),i=1,2,…,n}, where n is the number of corresponding feature point pairs;

[0088] Step S3.2.1 Randomly select from QORB Four points not lying on the same straight line are selected from the sample, and the homography matrix h is calculated. In the formula, (x,y) represents Q. ORB A point in the point set, (x', y') is P. ORB The points corresponding to the point set;

[0089] Step S3.2.2: Set the feature point set Q at this time... ORB With P ORB Substitute all remaining points into the above formula and set a certain error threshold. Collect the feature point set that satisfies the current homography matrix and treat it as a "correctly corresponding pair," then calculate the current fitness function.

[0090] Step S3.2.3 After each iteration, the current fitness function is compared with the optimal fitness function. The calculation is repeated until the set maximum number of iterations is reached. The homography matrix corresponding to the optimal fitness function is used as the benchmark model, and then the "correctly corresponding point set" that satisfies the model is extracted from it.

[0091] Figure 5 This is a schematic diagram of feature point pairs that have a correct correspondence after being screened, as provided by the present invention.

[0092] Step S4: Based on the spectral characteristics of the UAV images, extract the vegetation cover area and non-vegetation area of ​​the monitored area from the multi-period DOM data constructed above.

[0093] Step S4.1 For each period of DOM data, according to the formula Calculate the V-value index for each period, where R, G, and B are the pixel values ​​of the DOM data in the red, green, and blue bands, respectively.

[0094] Step S4.2 Calculate the spatial distribution of the V value index of each period of DOM data. Use the V value at the trough between the two peaks as the threshold to segment the DOM data. Areas with a V value index less than the threshold are non-vegetation areas, and areas with a V value index greater than the threshold are vegetation areas.

[0095] Figure 6 The terrain mask for vegetation-covered and non-vegetation-covered areas provided in this invention example.

[0096] Step S5: Based on the elevation information in the constructed multi-period DSM data, the detection area is divided into smooth surface areas and rough surface areas using elevation undulation, thereby establishing terrain masks for stable and unstable surface areas.

[0097] Step S5.1 Neighborhood Statistics. For each cell in each period of DSM data, construct a local neighborhood region of the cell with the cell as the center and a side length of 10 cells, and count the maximum and minimum cell values ​​within the local region.

[0098] Step S5.2 Calculation of terrain relief. Based on the maximum and minimum pixel values ​​within the neighborhood of each pixel, the terrain relief is calculated using the formula R = H. max -H min Calculate the topographic relief value of this pixel, where H max and H min These are the maximum and minimum pixel values ​​in the neighborhood of that pixel, respectively.

[0099] Step S5.3 Terrain Segmentation. A binary classification method is used to segment stable and unstable surface regions, i.e., a threshold m is set, and if the terrain relief value of a pixel is less than m, it is considered to be located in a stable surface region;

[0100] Figure 7 The stable and unstable surface region terrain masks provided for examples of the present invention.

[0101] Step S6: Integrate the non-vegetated areas and stable surface areas of the monitoring area extracted above to construct the candidate regions of pseudo control points for each period of UAV image DOM data, and extract the candidate pseudo control points located in the region as the final pseudo control points.

[0102] Step S6.1 Candidate Region Integration. Overlay the above-mentioned non-vegetation-covered areas and stable surface areas, and extract the common area of ​​the two topographic layers, that is, the area that is both non-vegetation-covered and stable surface area, as the candidate area for pseudo-control point screening;

[0103] Step S6.2 False control point screening. Overlay the above-mentioned false control candidate points with the false control candidate region, and extract the false control candidate points located within the false control candidate region as the final false control points;

[0104] Figure 8 A schematic diagram illustrating the final pseudo-control point selection for an example of the present invention;

[0105] Step S6.3 Pseudo-control point information extraction. The final pseudo-control points are overlaid with the baseline DSM data, and the plane coordinates (X, Y) and elevation coordinates (Z) of the pseudo-control points are extracted from the baseline DSM.

[0106] Step S7: The extracted pseudo-control points, except for the baseline data, are used in the positioning and orientation process of each other UAV imagery to reconstruct the DSM data and DOM data of each UAV imagery, and to establish a positioning and orientation baseline that is consistent with the baseline data.

[0107] Step S7.1 Image "Point Striking". Perform aerial triangulation calculations on each UAV image except for the baseline data, treat the above pseudo control points as ground control points, and strike points on the image to provide ground control coordinates for aerial triangulation;

[0108] Figure 9 This is a schematic diagram illustrating the use of pseudo-control points for "spiking" and repositioning / orientation, as provided in this invention example.

[0109] Step S7.2 DOM and DSM Data Construction. After the above aerial triangulation operation, reconstruct the DOM and DSM data for each period with new positioning.

[0110] After the positioning and orientation of multiple UAV images are achieved through the above steps, the DOM data and DSM data constructed from each image can be overlaid to conduct surface change surveys and deformation analysis.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for positioning and orientation of slopes using UAV imagery from multiple periods to establish pseudo-control points, characterized in that, include: S1. Acquire multi-phase UAV imagery and POS data of the surface monitoring area; S2. Based on the POS data, construct the DSM data and DOM data of the monitoring area from the above multiple UAV images, and select the data from one period as the baseline data. S3. For the constructed multi-period DOM data, use image feature detection methods to extract candidate points that can be used as pseudo control points from the DSM data of multiple periods, and compare the candidate points extracted in each period to select pseudo control candidate points with the same name. S4. Based on the spectral characteristics of the constructed multi-period DOM data, extract the vegetation coverage area and non-vegetation area of ​​the monitoring area. S5. Based on the elevation information in the constructed multi-period DSM data, the detection area is divided into smooth surface area and rough surface area using the elevation undulation, thereby establishing a terrain mask for stable surface area and unstable surface area. S6. Integrate the extracted non-vegetated areas and stable surface areas of the monitoring area to construct the candidate regions of pseudo-control points for each period of UAV image DOM data, and extract the candidate pseudo-control points located in the region as the final pseudo-control points. S7. For the extracted pseudo-control points, in addition to the baseline data, they are used in the positioning and orientation process of each other UAV imagery to reconstruct the DSM data and DOM data of each UAV imagery, and to establish a positioning and orientation baseline that is consistent with the baseline data.

2. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, The process of constructing DSM and DOM data for the monitoring area based on POS data from the aforementioned multiple UAV images includes: Based on the coordinate and positioning information recorded in the POS data, the UAV image data is processed with additional parameters, including self-calibration regional network adjustment, image point cloud construction, dense matching, spatial interpolation, digital differential correction, stitching and mosaicking, to obtain the DSM and DOM data of each monitoring area.

3. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, The method for obtaining the pseudo-control candidate points with the same name in S3 is as follows: The ORB algorithm is used to extract clear and easily identifiable edge feature points from multiple periods of DOM data, which are then used as pseudo control candidate points. The RANSAC algorithm is used to detect corresponding points of the aforementioned edge feature points, eliminating false control candidate points that have no corresponding points in multiple periods of DOM data. The RANSAC algorithm achieves its goal by repeatedly selecting a random subset of the data; its basic idea is as follows: Randomly select a set of points from the dataset and calculate the current model relationships; The calculated model relationship is used to test all other data points. If the error between each point and the model is less than the set threshold, it is considered an "inner point"; otherwise, it is considered an "outer point". If enough points are classified as "interior points", then the model relationship is reasonable enough. Keep the model with the most "interior points" so far and record it as the best model relationship. Repeat the above steps until the set number of iterations is reached, and select the best model relationship as the final model.

4. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, Step S4 utilizes the spectral features of UAV imagery to extract vegetation-covered and non-vegetation-covered areas, including: From the formula Calculate the V-value index for each period of DOM data, where R, G, and B are the pixel values ​​of each period of DOM in the red, green, and blue bands, respectively, and the range of the V-value index is [-1, 1]. Based on the V-value index calculated in the previous step, the V-value index at the trough is selected as the threshold. Vegetation and non-vegetation are segmented and extracted by gray-scale segmentation, and the non-vegetation covered area is created as a terrain mask.

5. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, In step S5, the elevation relief is calculated based on the elevation information in the DSM data, thereby extracting and constructing terrain masks for stable and unstable surface areas, including: For each cell in the DSM data, calculate the maximum and minimum cell values ​​within its 10*10 neighborhood. Based on the maximum and minimum values ​​obtained in the previous step, from the equation R = H max -H min Calculate the topographic relief value for each cell in the DSM data, where H max and H min These are the maximum and minimum pixel values ​​within the 10*10 neighborhood of each pixel, respectively. The topographic relief value of DSM data for each period is used as the basis for judging stable and unstable regions. Stable and unstable regions are divided by setting a threshold and using a binary method. The extracted stable regions are then used as terrain masks.

6. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, In step S6, a candidate region for pseudo-control points is created, and the final pseudo-control points are extracted, including: The extracted terrain masks of non-vegetation areas and stable areas are integrated to extract candidate regions of pseudo-control points in each period of DOM data. The pseudo-control candidate points located in the pseudo-control point candidate region are extracted as the final pseudo-control points. Based on the final pseudo-control points, the plane coordinates X, Y and elevation coordinates Z of the pseudo-control points are extracted from the DSM.

7. The multi-phase slope UAV image positioning and orientation method for establishing pseudo-control points according to claim 1, characterized in that, In step S7, pseudo-control points are used to locate and orient the drone images for each period, and new DSM and DOM data are constructed, including: The pseudo-control points are reintegrated into the aerial triangulation process of each UAV image, except for the baseline data, in the form of control point "pricks". Based on the above, each UAV imagery is processed by image point cloud construction, dense matching, spatial interpolation, digital differential correction, stitching and mosaicking to obtain DSM data and DOM data with consistent positioning and orientation with the baseline data.

Images