A high-precision geographic information calculation method for adaptive scene labeling

Abstract

The present invention discloses a high-precision geographic information calculation method for adaptive scene marking, comprising the following steps: obtaining a static picture taken by a camera under a scene and high-precision longitude and latitude information corresponding to the scene; drawing single-side contour lines of each parallel lane line in the same direction as the camera optical axis direction and converging them; adjusting according to the convergence result of the single-side contour lines of the parallel lane lines; calibrating road lines that are perpendicular or approximately perpendicular to the camera optical axis direction according to the scene, generating multiple intersection points, i.e., reference points, and forming a grid map; finding the longitude and latitude information corresponding to all reference points in a high-precision map as the geographic information of the reference points; obtaining the corresponding circumscribed regular rectangular frame according to each point on the high-precision map corresponding to each reference point in the image, and constructing a projection transformation matrix from the image plane to the bird's-eye view image; obtaining the longitude and latitude of the image point according to the distance ratio of the projection point in the longitude and latitude directions, and completing the calculation of the geographic information.

Description

Technical Field

[0001] The present invention relates to the technical field of geographic location positioning, and in particular to a high-precision geographic information calculation method for adaptive scene marking. Background Art

[0002] Data show that remote sensing technology and LiDAR technology are developing rapidly, and both can obtain large-scale urban surface information, provide data sources for urban research, and are widely used in urban street landscape research. However, the surface landscape images obtained by remote sensing technology are from a bird's-eye view, and their two-dimensional characteristics cannot express the landscape preparation at the street level, let alone present the surface landscape from a human perspective. Although LiDAR technology obtains the three-dimensional coordinates of the objects and presents them in the form of discrete points, overcoming the two-dimensional defects of remote sensing images, optical scanning instruments are expensive and the cost of obtaining data is high. Only specific departments have the ability to scan and measure to obtain airborne or vehicle-mounted LiDAR data. In addition, LiDAR data has precise three-dimensional coordinates, and the data is confidential, so it is difficult for general researchers to obtain large-scale LiDAR data. These shortcomings limit large-scale urban street research. Therefore, it is a convenient and effective way to match the longitude and latitude coordinates through a more convenient two-dimensional directional map to provide a data source for urban research.

[0003] Chinese patent document CN112966064A discloses a "method and system for matching longitude and latitude coordinates of two-dimensional directional map images". The method includes: uploading a two-dimensional directional map image to be matched; obtaining a satellite map corresponding to the two-dimensional directional map image; selecting at least three key points on the two-dimensional directional map image; obtaining the longitude and latitude coordinates of the key points through the satellite map; calculating the coordinate transformation matrix of the two-dimensional directional map image according to the longitude and latitude coordinates of the key points; and calculating the longitude and latitude coordinates of the four vertices of the two-dimensional directional map image according to the coordinate transformation matrix. The present invention also discloses a system for matching longitude and latitude coordinates of two-dimensional directional map images. The above technical solution has high requirements for two-dimensional directional map images and is not applicable to images taken at daily angles. Summary of the invention

[0004] The present invention mainly solves the technical problem that the original technical solution has high requirements for two-dimensional directional map images and is not applicable to images taken at daily angles. It provides a high-precision geographic information calculation method for adaptive scene marking, which divides the picture into multiple non-overlapping quadrilateral grids by calibrating the positions and directions of road lane lines and auxiliary lane lines in the reference image, and then obtains the longitude and latitude information of the object by determining the position where a specific object appears in the picture and the grid to which it belongs. In order to solve the problem of increased calculation error of targets at long distances caused by the perspective relationship that the target is "larger near and smaller far away", the calculation error is controlled within 1m through adaptive and precise positioning of reference points in the early stage of scene calibration.

[0005] The above technical problem of the present invention is mainly solved by the following technical solution: The present invention comprises the following steps:

[0006] S1 obtains the static picture taken by the camera under the scene and the high-precision longitude and latitude information corresponding to the scene;

[0007] S2 draws the single-side contour lines of each parallel lane line in the same direction as the camera optical axis and converges them;

[0008] S3 is adjusted according to the convergence result of the single-side contour lines of the parallel lane lines;

[0009] S4 calibrates a road line that is perpendicular or approximately perpendicular to the camera optical axis according to the scene, generates multiple intersection points, i.e., reference points, and forms a grid map;

[0010] S5 finds the longitude and latitude information corresponding to all the reference points generated in the step in the high-precision map as the geographic information of the reference points;

[0011] S6 obtains the corresponding circumscribed rectangular frame according to the reference points in the image corresponding to the points on the high-precision map, and constructs a projection transformation matrix from the image plane to the bird's-eye view image;

[0012] S7 obtains the longitude and latitude of the image point according to the distance ratio of the projection point in the longitude and latitude directions, and completes the calculation of the geographic information.

[0013] Preferably, the step S1 obtains a static picture taken by the camera in the scene and marks three or more parallel lane lines or at least two points on the single-side contour of lines parallel to the lane lines, as well as straight lines that are not parallel to the lane lines but are convenient to mark in the image, manually or by detection algorithm.

[0014] Preferably, in step S2, if the point markings of multiple parallel lines are accurate, according to the properties of projective transformation, each parallel line should converge to a point in the image; if there are certain errors between the point markings of multiple parallel lines and the actual situation, then multiple parallel lane lines will disappear at multiple different but adjacent points, that is, multiple parallel lines intersect at multiple points.

[0015] Preferably, in step S3, when multiple parallel lines intersect at multiple points, an adaptive calibration adjustment is performed:

[0016] S3.1 Arbitrarily select a point from the original marked point sequence on each parallel line as one of the two fixed points S that determine this parallel line after adjustment;

[0017] S3.2 After determining a fixed point S of the parallel line, select points T around a point T in the original point sequence of this parallel line. i The selection of this point should satisfy: the angular difference between S and T and between S and T i is not greater than 2. The number of selected Ts i is generally less than 50, that is, i ≤ 50;

[0018] After determining the candidate adjustment points for each parallel lane line, the algorithm adaptively adjusts N parallel lane lines Li (0 <= i <= N - 1), and at the same time calculates the offset angle between each adjusted lane line and the original lane line. If the N adjusted parallel lane lines converge to a point and the variance of the angular deviation values of the N parallel lane lines is the smallest at this time, then the algorithm ends. Finally, the N parallel lane lines intersect at a point, that is, there is only one intersection point. At this time, the points (S, Ti) of each lane line are the final determined points after adjustment of this lane line.

[0019] Preferably, the specific steps of step S3.3 include:

[0020] S3.31 Loop through each parallel lane line L i (0 <= i <= N - 1) According to the constraint conditions described in the second item of step 3, obtain the possible lane line set Set(L i ) of each lane line after adjustment;

[0021] S3.32 Sequentially take out a lane line from the lane line set Set(L i )(0 < i <= N - 1) to form a lane line set Set_N(L ij , L i+1k , L i+2l ,..., L N-1p) There are N′ groups in total. Assuming that the number of original calibration points on the original lane line is 2, then N′=power(50x2,N), where 50 is the number of candidate points determined in the second item in step 3, and N is the number of parallel lane lines;

[0022] S3.33 N′ groups Set_N (L ij ,L i+1k ,L i+2l ,...,L N-1p ) is judged by the intersection of N lane lines in the group: three lane lines are randomly selected from the N lane lines, and the number of their intersections is calculated. If it is greater than 1, this group does not meet the requirements, the judgment is terminated, and the calculation of the next group is continued; if the number of intersections of the three randomly selected lane lines is 1, one of the remaining lane lines is added one by one, and its intersection with the first three lane lines is judged. If the number is greater than 1, the judgment is terminated. If the number of intersections of the N lane lines in the group is 1, then the point pairs on each lane at this time and the offset angle Ang of each lane line and its original lane line are recorded. Li ;

[0023] S3.34 Count the set Set_N (L ij ,L i+1k ,L i+2l ,...,L N-1p ), if Num=1, the lane line in the corresponding set is the lane line after adjustment we want; if Num>1, the variance of the angle offset between each lane line after change and the original calibrated lane line in each set of lane lines is calculated, and the set with the smallest variance is taken as the final adjustment result, that is:

[0024] Preferably, the input in step S3.3 is: a sequence of points on N parallel lane lines; the conditions are: N parallel lanes intersect at one point, and the variance of the angle deviation values ​​of the N parallel lane lines is the smallest; the output is: N parallel lane lines intersecting at one point.

[0025] Preferably, in step S4, the original scene is divided into a plurality of non-overlapping regular / irregular grids by coordinating parallel lines and road lines that are perpendicular or approximately perpendicular to the camera optical axis direction, and the intersection points between the grids are used as reference points.

[0026] Preferably, the step S6 measures the actual distance corresponding to the length and width of the circumscribed rectangle in meters, and converts the width and height distance values ​​of the circumscribed rectangle into smaller values ​​while ensuring that the aspect ratio remains unchanged, so that at least one of the corresponding values ​​of the width W and height H is a four-digit number, thereby determining the length and width values ​​of the projected image.

[0027] Preferably, the step S6 also stipulates that: the coordinates of the upper left corner point of the projected image are (0,0), the horizontal direction is the x-axis direction, the vertical direction is the y-axis direction, the coordinates of the lower right corner point are (W-1,H-1), and 4 or more reference points are randomly selected from the image to find their coordinates on the projected image, thereby generating multiple pairs of [image reference points, image reference points corresponding to projection coordinates], and thereby generating a projection matrix from the image to the avatar image, that is,

[0028]

[0029] in It is the projection transformation matrix from the image plane to the bird's-eye view image, that is, the homography matrix.

[0030] Preferably, the step S7 specifically includes:

[0031] S7.1 projects the projection point to be calculated onto the bird's-eye view image, and then determines in which grid the projection point is located;

[0032] S7.2, based on the coordinates and longitude and latitude information of the four points constituting the grid, respectively obtain the distance ratios of the projection points in the longitude and latitude directions;

[0033] S7.3 obtain two longitude and latitude values ​​respectively according to the ratio relationship between the projection points in the longitude and latitude directions;

[0034] S7.4 To reduce random errors, calculate the average of the two longitude and latitude values ​​of the projection point and use them as the longitude and latitude of the desired image point to complete the calculation of geographic information.

[0035] The beneficial effects of the present invention are as follows: by calibrating the positions and directions of the road lane lines and auxiliary lane lines in the reference image, the picture is divided into a plurality of non-overlapping quadrilateral grids, and then by determining the position where a specific object appears in the picture and the grid to which it belongs, the longitude and latitude information of the object is obtained. In order to solve the problem of increased calculation error of targets at long distances caused by the perspective relationship causing the target to be "larger near and smaller far away", the calculation error is controlled within 1m by adaptively and accurately positioning the reference point in the early stage of scene calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the present invention.

[0037] Figure 2 It is a bird's-eye view of parallel lane lines of the present invention.

[0038] Figure 3 In a video scene of the present invention, parallel lane lines intersect at a single point.

[0039] Figure 4In a video scene of the present invention, parallel lane lines intersect at a multi-point graph.

[0040] Figure 5 It is a schematic diagram of a lane line adjustment range of the present invention.

[0041] Figure 6 It is a schematic diagram of screen grid division under a video viewing angle of the present invention.

[0042] Figure 7 It is an external rectangular frame corresponding to a parallel lane line and other marking lines of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings. Embodiment: A high-precision geographic information calculation method for adaptive scene marking in this embodiment, such as Figure 1 As shown, the following steps are included:

[0044] S1 obtains a static image taken by a camera under the scene and high-precision latitude and longitude information corresponding to the scene. Step S1 obtains a static image taken by a camera under the scene and manually or by detection algorithm to mark three or more parallel lane lines or at least two points on the single-side contour of lines parallel to the lane lines, as well as straight lines that are not parallel to the lane lines but are convenient to mark in the image, such as Figure 2 shown.

[0045] S2 draws the single-side contour lines of each parallel lane line in the same direction as the camera optical axis and converges them. Figure 3 As shown in , if the points of multiple parallel lines are accurately marked, then according to the properties of the projection transformation, each parallel line should converge to one point in the image; Figure 4 As shown in FIG. 1 , if there is a certain error between the point marking of multiple parallel lines and the actual existence, the multiple parallel lane lines will disappear at multiple different but adjacent points, that is, the multiple parallel lines intersect at multiple points.

[0046] S3 is adjusted according to the convergence result of the single-side contour lines of the parallel lane lines. Figure 5 As shown, when multiple parallel lines intersect at multiple points, adaptive calibration adjustment is performed:

[0047] S3.1 Select any point from the original marked point sequence on each parallel line as one of the two points that determine the parallel line after adjustment;

[0048] S3.2 After determining a fixed point S of the parallel line, select T around a point T in the original point sequence of the parallel line. i The selection of this point should satisfy: between S and T and between S and T i The angle difference between them is not greater than 2, select Ti The quantity is generally less than 50, that is, i ≤ 50;

[0049] After determining the candidate adjustment points for each parallel lane line in S3.3, the algorithm adaptively adjusts the N parallel lane lines Li (0 <= i <= N - 1), and at the same time calculates the offset angle between each adjusted lane line and the original lane line. If the N adjusted parallel lane lines converge at a point and the variance of the angle deviation values of the N parallel lane lines is the smallest at this time, the algorithm ends. Finally, the N parallel lane lines intersect at a point, that is, there is only one intersection point. At this time, the points (S, Ti) of each lane line are the finally determined points after the adjustment of this lane line.

[0050] The specific process of the algorithm is described as follows:

[0051] Input: The point sequence on N parallel lane lines;

[0052] Satisfied conditions: The N parallel lane lines intersect at a point, and the variance of the angle deviation values of the N parallel lane lines is the smallest;

[0053] Output: N parallel lane lines intersecting at a point (vanishing point);

[0054] The specific steps include:

[0055] S3.31 Loop through each parallel lane line L i (0 <= i <= N - 1) According to the constraint conditions described in the second item of step 3, obtain the possible lane line set Set(L i ) of each lane line after adjustment;

[0056] S3.32 Sequentially take out one lane line from the lane line set Set(L i )(0 < i <= N - 1) to form a lane line set Set_N(L ij , L i+1k , L i+2l ,..., L N-1p ) with a number of N groups. Assume that the original number of calibration points on the original lane line is 2, then N' = power(50 x 2, N), where 50 is the number of candidate points determined in the second item of step 3, and N is the number of parallel lane lines;

[0057] S3.33 Sequentially for the N' groups of Set_N(L ij , L i+1k , L i+2l ,..., L N-1p) is judged by the intersection of N lane lines in the group: three lane lines are randomly selected from the N lane lines, and the number of their intersections is calculated. If it is greater than 1, this group does not meet the requirements, the judgment is terminated, and the calculation of the next group is continued; if the number of intersections of the three randomly selected lane lines is 1, one of the remaining lane lines is added one by one, and its intersection with the first three lane lines is judged. If the number is greater than 1, the judgment is terminated. If the number of intersections of the N lane lines in the group is 1, then the point pairs on each lane at this time and the offset angles of each lane line and its original lane line are recorded.

[0058] S3.34 Count the set Set_N (L ij ,L i+1k ,L i+2l ,...,L N-1p ), if Num=1, the lane line in the corresponding set is the lane line after adjustment we want; if Num>1, the variance of the angle offset between each lane line after change and the original calibrated lane line in each set of lane lines is calculated, and the set with the smallest variance is taken as the final adjustment result, that is:

[0059] S4 calibrates the road lines that are perpendicular or approximately perpendicular to the camera optical axis according to the scene, generates multiple intersection points, i.e., reference points, and forms a grid map, such as Figure 6 As shown in the figure, the traffic scene is divided into grids, each of which has a corresponding number. By combining parallel lines and road lines that are perpendicular or approximately perpendicular to the camera optical axis, the original scene is divided into multiple regular / irregular grids that do not overlap, and the intersections between the grids are used as reference points.

[0060] S5 finds the longitude and latitude information corresponding to all the reference points generated in the step in the high-precision map as the geographic information of the reference points;

[0061] S6 obtains the corresponding circumscribed rectangular frame according to the reference points in the image corresponding to the points on the high-precision map, and constructs the projection transformation matrix from the image plane to the bird's-eye view image, such as Figure 7As shown. Measure the actual distance corresponding to the length and width of the circumscribed rectangle in meters. While ensuring that the aspect ratio remains unchanged, convert the width and height distance values ​​of the circumscribed rectangle into smaller values ​​so that at least one of the corresponding values ​​of width W and height H is a four-digit number. The length and width values ​​of the projected image are determined. At the same time, it is stipulated that the coordinates of the upper left corner of the projected image are (0,0), the horizontal direction is the x-axis direction, the vertical direction is the y-axis direction, and the coordinates of the lower right corner are (W-1,H-1). Randomly select 4 or more reference points from the image to find their coordinates on the projected image, thereby generating multiple pairs of [image reference points, image reference points corresponding to projection coordinates], and use this to generate the projection matrix from the image to the avatar image, which is

[0062]

[0063] in It is the projection transformation matrix from the image plane to the bird's-eye view image, that is, the homography matrix.

[0064] S7 obtains the longitude and latitude of the image point according to the distance ratio of the projection point in the longitude and latitude directions, completing the calculation of the geographic information. Step S7 specifically includes:

[0065] S7.1 projects the projection point to be calculated onto the bird's-eye view image, and then determines in which grid the projection point is located;

[0066] S7.2, based on the coordinates and longitude and latitude information of the four points constituting the grid, respectively obtain the distance ratios of the projection points in the longitude and latitude directions;

[0067] S7.3 obtain two longitude and latitude values ​​respectively according to the ratio relationship between the projection points in the longitude and latitude directions;

[0068] S7.4 To reduce random errors, calculate the average of the two longitude and latitude values ​​of the projection point and use them as the longitude and latitude of the desired image point to complete the calculation of geographic information.

[0069] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0070] Although the term adaptive calibration adjustment, projection transformation, etc. is used more frequently in this article, the possibility of using other terms is not excluded. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional restrictions is contrary to the spirit of the present invention.

Claims

1. A high-precision geographic information calculation method for adaptive scene labeling, characterized in that: The following steps are involved: S1 obtains the static picture taken by the camera under the scene and the high-precision longitude and latitude information corresponding to the scene; S2 draws the single-side contour lines of each parallel lane line in the same direction as the camera optical axis and converges them. If the points of multiple parallel lines are accurately marked, each parallel line converges at one point. If there is an error in the point marking, multiple parallel lines intersect at multiple points. S3 is adjusted according to the convergence result of the single-side contour lines of the parallel lane lines. When multiple parallel lines intersect at multiple points, a point is randomly selected from the original marked point sequence on each parallel line as a fixed point S between the two points of the parallel line after adjustment; T is selected around a point T in the original point sequence of the parallel line. i , adaptively adjust the N parallel lane lines Li. If the adjusted N parallel lanes converge at one point and the variance of the angle deviation values ​​of the N parallel lane lines is the smallest, the algorithm ends; S4 generates reference points by calibrating a road line perpendicular to the camera optical axis according to the scene; S5 finds the longitude and latitude information corresponding to all reference points in the high-precision map; S6 obtains an external rectangular frame according to the correspondence between each reference point in the image and each point on the high-precision map, and constructs a projection transformation matrix from the image plane to the bird's-eye view image; S7 obtains the longitude and latitude of the image point according to the distance ratio of the projection point in the longitude and latitude directions, and completes the calculation.

2. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1 is characterized in that: The step S1 obtains a static picture taken by the camera in the scene and marks three or more parallel lane lines or at least two points on the single-side contour of lines parallel to the lane lines, as well as straight lines that are not parallel to the lane lines but are convenient to mark in the image, manually or by detection algorithm.

3. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1 is characterized in that: In step S2, if the point markings of multiple parallel lines are accurate, then according to the properties of the projection transformation, each parallel line should converge at one point in the image; if there is a certain error between the point markings of multiple parallel lines and the actual existence, then multiple parallel lane lines will disappear at multiple different but adjacent points, that is, multiple parallel lines intersect at multiple points.

4. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1 or 3, characterized in that: In step S3, when multiple parallel lines intersect at multiple points, adaptive calibration adjustment is performed: S3.1, a point is randomly selected from the original marked point sequence on each parallel line as a fixed point S between two points of the parallel line after adjustment; S3.2 After determining a fixed point S of the parallel line, select T around a point T in the original point sequence of the parallel line. i The selection of this point should satisfy: between S and T and between S and T i The angle difference between them is not greater than 2, select T i The number is generally less than 50, that is, i≤50; S3.3 After determining the candidate adjustment point for each parallel lane line, the algorithm performs adaptive adjustment on the N parallel lane lines Li, 0<=i<=N-1, and calculates the offset angle between each adjusted lane line and the original lane line. If the N parallel lanes converge at one point after adjustment, and the variance of the angle deviation values ​​of the N parallel lane lines is the smallest at this time, the algorithm ends. Finally, the N parallel lane lines intersect at one point, that is, there is only one intersection point. At this time, the point (S, Ti) of each lane line is the final determined point after the adjustment of this lane line.

5. The high-precision geographic information calculation method for adaptive scene labeling according to claim 4 is characterized in that: The specific steps of step S3.3 include: S3.31 loops through each parallel lane line L i , 0<=i<=N-1, according to the constraints described in step S3.2, the possible lane line set Set(L i ); S3.32 sequentially extracts a lane line from the lane line set Set(L i ), where 0 < i <= N - 1, to form a lane line set Set_N(L ij , L i+1k , L i+2l ,..., L N-1p ) with a total of N' groups. Assuming that the original number of calibration points on the lane line is 2, then N' = power(50 x 2, N), where 50 is the number of candidate points determined in the second item of step 3, and N is the number of parallel lane lines; S3.33 N′ groups Set_N (L ij ,L i+1k ,L i+2l ,...,L N-1p ) is judged by the intersection of N lane lines in the group: three lane lines are randomly selected from the N lane lines, and the number of their intersections is calculated. If it is greater than 1, this group does not meet the requirements, the judgment is terminated, and the calculation of the next group is continued; if the number of intersections of the three randomly selected lane lines is 1, one of the remaining lane lines is added one by one, and its intersection with the first three lane lines is judged. If the number is greater than 1, the judgment is terminated. If the number of intersections of the N lane lines in the group is 1, then the point pairs on each lane at this time and the offset angle Ang of each lane line and its original lane line are recorded. Li ; S3.34 Count the set Set_N (L ij ,L i+1k ,L i+2l ,...,L N-1p ), if Num=1, then the lane line in the corresponding set is the lane line after adjustment we want; if Num>1, then calculate the variance of the angle offset between each lane line after change and the original calibrated lane line in each set of lane lines, and take the set with the smallest variance as the final adjustment result, that is: D ang_dst =min(D angt ,D angt+1 ,...,D angNum-1 ).

6. The high-precision geographic information calculation method for adaptive scene labeling according to claim 4 is characterized in that: The input quantity in step S3.3 is: a sequence of points on N parallel lane lines; the conditions are: N parallel lanes intersect at one point, and the variance of the angle deviation values ​​of the N parallel lane lines is the smallest; the output quantity is: N parallel lane lines intersecting at one point.

7. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1 is characterized in that: In step S4, the original scene is divided into a plurality of non-overlapping regular / irregular grids by coordinating parallel lines and road lines that are perpendicular or approximately perpendicular to the camera optical axis, and the intersections between the grids are used as reference points.

8. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1 is characterized in that: The step S6 measures the actual distance corresponding to the length and width of the circumscribed rectangle in meters. While ensuring that the aspect ratio remains unchanged, the width and height distance values ​​of the circumscribed rectangle are converted into smaller values ​​so that at least one of the corresponding values ​​of the width W and height H is a four-digit number. The length and width values ​​of the projected image are determined.

9. The high-precision geographic information calculation method for adaptive scene labeling according to claim 8, characterized in that: The step S6 also stipulates that: the coordinates of the upper left corner point of the projected image are (0,0), the horizontal direction is the x-axis direction, the vertical direction is the y-axis direction, and the coordinates of the lower right corner point are (W-1,H-1), and four or more reference points are randomly selected from the image to find their coordinates on the projected image, thereby generating multiple pairs of [image reference points, image reference points corresponding to projection coordinates], and thereby generating the projection matrix of the image to the avatar image, that is, in It is the projection transformation matrix from the image plane to the bird's-eye view image, that is, the homography matrix.

10. The high-precision geographic information calculation method for adaptive scene labeling according to claim 1, characterized in that: The step S7 specifically includes: S7.1 projects the projection point to be calculated onto the bird's-eye view image, and then determines in which grid the projection point is located; S7.2 obtains the distance ratio of the projection point in the longitude and latitude directions according to the coordinates and longitude and latitude information of the four points constituting the grid; S7.3 obtains two longitude and latitude values ​​respectively according to the ratio of the projection point in the longitude and latitude directions; S7.4 calculates the average value of the two longitude and latitude values ​​of the projection point to reduce random errors, and uses them as the longitude and latitude of the desired image point to complete the calculation of geographic information.

Images