Mountain area canyon high-precision horizontal coordinate measurement method based on light beam adjustment

Abstract

The invention provides a mountainous area canyon high-precision horizontal coordinate measurement method based on light beam adjustment, which comprises the following steps: arranging ground control points by aerial photography of an unmanned aerial vehicle, carrying out SfM modeling, taking points at places where virtual control points need to be arranged, recording the current coordinates of the points as a reference, respectively offsetting a certain distance in a plurality of directions, as an input coordinate of an error correction method, a scene setting of a ground control point with a certain initial error is formed; performing light beam adjustment on each scene, recording the output position of each scene after light beam adjustment, and making a ray, namely an error correction vector; and then solving intersection points of two adjacent rays, and taking an average value of the intersection points as a calculation result, namely virtual control point coordinates optimized through the error correction effect. According to the method provided by the invention, centimeter-level coordinate positioning can be obtained in an area without ground control points, and decimeter-meter-level geographic positioning errors caused by lack of ground control points in traditional unmanned aerial vehicle aerial photography are reduced to centimeter level.

Description

Technical Field

[0001] The present invention belongs to the fields of remote sensing, surveying and mapping, monitoring technology, etc., and particularly relates to a high-precision horizontal coordinate measurement method for mountainous valleys based on bundle adjustment. Background Art

[0002] At present, terrain mapping based on advanced remote sensing technologies such as unmanned aerial vehicle (UAV) aerial photography and three-dimensional laser scanning has been gradually popularized. However, in high mountain and valley areas such as the southwestern mountainous regions, the above technologies face problems such as traffic restrictions and line-of-sight blockages during application, resulting in deviations in data integrity or accuracy. For example, UAV aerial photography requires the arrangement of sufficient ground control points to ensure high accuracy in the area covered by the control points. Once far away from the control points, the data accuracy will be significantly reduced (decimeter to meter level). In contrast, the three-dimensional laser scanning technology can directly obtain millimeter-level point clouds within the visible range by independently emitting high-precision lasers. However, it is easily blocked by the terrain in mountainous valleys and it is difficult to obtain complete terrain information.

[0003] Therefore, the present invention mainly aims at obtaining high-precision terrain information (i.e., geographical coordinates) in mountainous valley environments under scenarios such as traffic restrictions and line-of-sight blockages, and has developed a high-precision surface coordinate measurement method for mountainous valleys based on bundle adjustment. Summary of the Invention

[0004] The present invention proposes a high-precision horizontal coordinate measurement method for mountainous valleys based on bundle adjustment, which can calculate the geographical coordinates of any ground point through the "error correction effect" in bundle adjustment only based on UAV photogrammetry and a limited on-site operation space. The specific technical solution is as follows:

[0005] A high-precision horizontal coordinate measurement method for mountainous valleys based on bundle adjustment, comprising the following steps:

[0006] (1) According to the conventional UAV aerial photography method, arrange ground control points within the passable and visible range and perform SfM modeling, and take a point X at the place where virtual control points need to be arranged k+1 , that is, the point to be optimized. Let the distance from it to the nearest control point be d, and record its current coordinate as , that is, the optimization target;

[0007] (2) Taking as a reference, offset a certain distance e in multiple directions respectively as the input coordinates of the error correction method to form a scenario setting of "ground control points with a certain initial error";

[0008] (3) Perform bundle adjustment on each scenario respectively, record its output position after bundle adjustment Then taking Generate rays, i.e., error correction vectors; then solve for the intersection points of two adjacent rays, a total of 4, and use the average value of these 4 intersection points as the calculation result. This value is the coordinate of the virtual control point optimized through the error correction effect.

[0009] It also includes calculating 3 to 4 different positions after additional error correction to eliminate possible systematic errors.

[0010] The specific method is as follows:

[0011] For each point to be measured, first use bundle adjustment to solve for the 4 output positions after the error correction effect occurs, namely upper L (0,+) , lower L (0,-) , left L (-,0) , right L (+,0) . Through modifying the number of automatically connected points, 4 iterative calculations are respectively performed on each output position, and output positions with different error correction intensities are obtained;

[0012] Then, based on the starting position, each error correction vector is fitted by the least squares method; the graphical method only calculates the intersection point of the extension lines of two adjacent error correction vectors, and there will only be four intersection points between the error correction vectors in the same group;

[0013] Finally, use the geometric center of the quadrilateral formed by the above four intersection points as the calculated position of the point to be measured:

[0014]

[0015] In the formula: I (+，+) +I (-，+) +I (-，-) +I (+，-) respectively represent the intersection point coordinates in the first, second, third, and fourth quadrants of the point to be measured after the error correction vectors intersect; I c represents the calculated value of the horizontal coordinate of the point to be measured.

[0016] The method provided by the present invention can obtain centimeter-level coordinate positioning in areas without ground control points, reducing the decimeter - meter-level geolocation error caused by the lack of ground control points in traditional UAV aerial photography to the centimeter level. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic diagram of the incremental bundle adjustment of the present invention;

[0018] Figure 2 is a flowchart of the present invention;

[0019] (a) According to the conventional UAV aerial photography method, set artificial control points within the accessible and visible range and perform modeling. Manually mark the location where the virtual control point Xk+1 needs to be set, and record its coordinates at this time as

[0020] (b) Based on , offset by a distance e in 4 different directions to obtain the new coordinates as the input coordinates of this virtual control point;

[0021] (c) Assume that the error correction effect can occur. Then, after the next bundle adjustment, its output coordinates will shift towards its true coordinates, denoted as By solving the intersection center of the 4 rays pointing from to , the calculated coordinates of the virtual control point Xk+1 are obtained;

[0022] Figure 3 is the flow chart of the graphical method for the horizontal coordinates of the points to be measured in the present invention;

[0023] (a) Construct 4 error correction vectors and their extension lines, and solve the horizontal coordinates of the points to be measured through the intersection of the 4 straight lines;

[0024] (b)-(c) Fitting results of the error correction vectors L(0,-) and L(0,+);

[0025] Figure 4 is the schematic diagram of the test arrangement in the present invention;

[0026] (a) shows the distribution positions of the ground control points and the experimental point groups;

[0027] (b) shows the experimental point groups, which are composed of 1 center point and 12 surrounding points distributed in a cross shape;

[0028] (c) shows the conventional ground control points;

[0029] Figure 5 is the visualization result diagram of the error correction effect in the present invention;

[0030] Taking the experimental point group P11 as an example, under different automatic connection points, the positional relationship between the marked position, the starting position (the position corresponding to the input coordinates), and the output position after bundle adjustment is shown. Detailed implementation manner

[0031] The error correction effect described in the present invention is as follows:

[0032] In the SfM algorithm, as Figure 1, first, the image pair with more registration points and a large baseline will be selected as the initial model, and then new images will be added successively, and global optimization will be performed using bundle adjustment. Assume that when the P n+2 th image is added to the initial model, a total of k (k≥3) accurate ground control points are involved in the calculation, and in the P n+2 th and P n+3 th images, the pixel points corresponding to point k + 1 can be observed simultaneously. Then, the position of point k + 1 in space can be calculated (i.e., the observed value) after the P n+2 th and P n+3 th images are added to the initial model, and its error depends on the distance from point X k+1 to the nearest control point. When point X k+1 participates in the SfM calculation as the ground control point GCP k+1 , if the initial error of this point is δ=(e x , e y , e z ), and is greater than the average error of the last bundle adjustment iteration, then the bundle adjustment will inevitably readjust the optimization results of all other control points and tie points to obtain the optimal solution between the observed values and the input values of GCP k+1 . Since the number of accurate control points exceeds the number of inaccurate control points, during the global geometric consistency optimization process of the bundle adjustment, the observed value of GCP k+1 in the initial model (i.e., the coordinates inferred by the last bundle adjustment iteration) will have a higher weight compared to its incorrect input value. At this time, the result of the bundle adjustment is actually the "optimal solution" considering the error of GCP k+1 . For GCP k+1 , its "optimal solution" will be far from its input position and close to its position observed in the last global geometric consistency calculation. Assume that the accuracy of other control points except GCP k+1 is relatively high, then the observed value of point X k+1 in the last bundle adjustment iteration will be very close to its true position. Then, the "optimal solution" of the bundle adjustment can be understood as making the inaccurate control points with low weight shift towards their true positions, that is, the so-called "error correction effect".

[0033] A high-precision horizontal coordinate measurement method for mountainous canyons based on bundle adjustment in this implementation, the theoretical process of using the error correction effect to infer the horizontal coordinates of any point is as Figure 2 shown, mainly including 3 steps:

[0034] (1) According to the conventional UAV aerial photography method, arrange ground control points within the accessible and visible range and perform SfM modeling. Take point X at the place where virtual control points need to be arranged k+1(i.e., the point to be optimized), let the distance from it to the nearest control point be d, and record its current coordinates as (the target to be optimized), such as Figure 2 a in. The purpose of this step is to obtain the approximate coordinates of point X k+1 to avoid the control points being excluded as outliers by the bundle adjustment due to excessive initial errors.

[0035] (2) Taking as the reference, offset a certain distance e in multiple directions respectively to form the input coordinates for the error correction method to create the scenario of "ground control points with certain initial errors", such as Figure 2 b in.

[0036] (3) Perform bundle adjustment on each scenario respectively, and record its output position after bundle adjustment Then, taking as the ray (i.e., the error correction vector), and then solve the intersection points of two adjacent rays (a total of 4), and take the average value of these 4 intersection points as the calculation result This value is the coordinates of the virtual control point optimized through the error correction effect, such as Figure 2 c in.

[0037] In the actual application process, the number of automatically connected points will directly affect the strength of the error correction effect, which will be verified in the subsequent case analysis. Therefore, on the basis of Figure 2 c, it is necessary to calculate 3 - 4 additional positions after different error corrections to eliminate possible systematic errors, specifically as Figure 3 :

[0038] Among them, for each point to be measured, first use bundle adjustment to solve the 4 output positions after the error correction effect occurs, namely the upper L (0,+) , the lower L (0,-) , the left L (-,0) , the right L (+,0) , such as Figure 3 a in, and this step is the content shown in Figure 2 a - b in. Further, here, 4 iterative calculations are respectively performed on each output position by modifying the number of automatically connected points, and the output positions with different error correction strengths are obtained, such as Figure 3 b - c in. Then, taking the starting position as the reference, the error correction vectors are fitted by the least squares method. The 4 error correction vectors and their extension lines can theoretically produce 6 intersection points on the horizontal plane, but since the intersection points of the extension lines of two opposite error correction vectors (for example, L (+,0) and L (-,0) ) sometimes intersect at infinity, so the graphical method only calculates the intersection points of the extension lines of two adjacent error correction vectors. For example, L(+,0) Only calculate the intersections with L (0,+) and L (0,-) At this time, there will only be 4 intersections between the error correction vectors in the same group. Finally, use the geometric center of the quadrilateral formed by the above four intersections as the calculated position of the point to be measured:

[0039]

[0040] In the formula: I (+，+) +I (-，+) +I (-，-) +I (+，-) respectively represent the intersection coordinates in the first, second, third, and fourth quadrants of the point to be measured after the error correction vectors intersect; I c represents the calculated value of the horizontal coordinate of the point to be measured.

[0041] The specific content of this embodiment is as follows:

[0042] 1. Field test and layout:

[0043] To verify the technical solution of the present invention, the valley of Xiameng Township and the gentle slope terrace of Qingliu Village are selected as the research areas here. Among them, the relative height difference in the valley area is less than 10m, the vegetation is sparse, while the maximum height difference on both sides of the mountain slope is greater than 400m, and the vegetation is dense. On March 4, 2023, unmanned aerial vehicle (UAV) aerial images were collected using a Pegasus D2000 quadrotor UAV and a Sony Dop-3000 five-lens oblique camera. The average ground sampling distance is about 5cm / px, and the forward overlap rate and side overlap rate are 80% and 70% respectively.

[0044] First, 4 standard L-shaped control points were arranged at the southernmost side of the valley area in the research area to construct an accurate control area, that is, the first k accurate control points (k = 4) in the previous text. Then, 14 experimental point groups were manually arranged in the research area. Each experimental point group contains a center point and 12 surrounding points arranged at equal intervals in a cross shape, as Figure 4 shown in a.

[0045] All these points are square tiles of 0.3m×0.3m, and the distance between the center points of two adjacent squares is 0.5m, as Figure 4 shown in b. In bundle adjustment, the center of each experimental point group is used as the point X k+1 to be optimized, and the 12 surrounding points are respectively used as 12 different starting positions

[0046] 2. Verification of error correction effect and solution of horizontal coordinate

[0047] Figure 5 shows the error correction effect of the experimental point group P11 after bundle adjustment. The specific method is as follows:

[0048] ① In any UAV aerial photography modeling software, set the Figure 4 4 L-shaped control points as ground control points according to the conventional method;

[0049] ② Mark the center of P11 as a control point, but its input coordinates are the coordinates offset in 4 directions. 3 points are set at intervals of 0.5 m in each direction, that is, Figure 5 the center of the red square in it (i.e., the input position);

[0050] ③ For the above 4×3 working conditions, use UAV modeling software to perform aerial triangulation (i.e., SfM algorithm), and the output position after error correction can be obtained, that is, Figure 5 the green dots in it;

[0051] ④ In order to avoid systematic errors, on the basis of step ③, the number of automatic connection points for each working condition is adjusted (i.e., from low connection points to high connection points), and then aerial triangulation is performed again. The results show that Figure 5 the output positions (green dots) in it do shift to varying degrees, but their shift directions always point to the manually marked positions, proving that the error correction effect in bundle adjustment does exist;

[0052] ⑤ Taking the input position as the starting point, combined with its corresponding output position, fit a straight line by the least squares method to obtain the error correction vector, and calculate the central coordinates of the figure enclosed by the intersection points of the error correction vectors in 4 directions, which is the horizontal coordinate of the final point to be measured.

Claims

1. A high-precision horizontal coordinate measurement method for mountainous canyons based on bundle adjustment, characterized in that It includes the following steps: (1) According to the conventional UAV aerial photography method, ground control points are arranged and SfM modeling is carried out within the accessible and visible range, and a point X is taken at the place where virtual control points need to be arranged k+1 , that is, the point to be optimized. Let the distance from it to the nearest control point be d, and record its current coordinates as that is, the target to be optimized; (2) Taking as a reference, offset a certain distance e in multiple directions respectively, serving as the input coordinates for the error correction method to form the scenario setting of "ground control points with a certain initial error"; (3) Beam adjustment is performed on each scenario respectively, and its output position after beam adjustment is recorded. Then, A ray is made, that is, the error correction vector; then the intersection points of two adjacent rays are solved, a total of 4, and the average value of these 4 intersection points is used as the calculation result. This value is the virtual control point coordinates optimized through the error correction effect.

2. A high-precision horizontal coordinate measurement method for mountainous canyons based on bundle adjustment according to claim 1, characterized in that It also includes calculating 3 to 4 additional positions after error correction to eliminate possible systematic errors.

3. A high-precision horizontal coordinate measurement method for mountainous canyons based on bundle adjustment according to claim 2, characterized in that, The specific method is as follows: For each point to be measured, first use bundle adjustment to solve the four output positions after the error correction effect occurs, namely the upper L (0,+) , the lower L (0,-) , the left L (-,0) , the right L (+,0) . Four iterative calculations were respectively performed on each output position by modifying the number of automatically connected points, and output positions with different error correction intensities were obtained; Then, taking the starting position as a reference, the error correction vectors are fitted by the least squares method; in the graphical method, only the intersection point of the extension lines of two adjacent error correction vectors is calculated, and there will be only four intersection points among the error correction vectors in the same group; Finally, the geometric center of the quadrilateral formed by the above four intersection points is used as the calculated position of the point to be measured: Where: I (+，+) +I (-，+) +I (-，-) +I (+，-) respectively represent the intersection coordinates in the first, second, third, and fourth quadrants of the intersection of the error correction vectors after intersection; I c represents the calculated value of the horizontal coordinate of the point to be measured.

Images