CN107917699B - A method for improving the quality of aerial triangulation in oblique photogrammetry of mountainous landforms - Google Patents

Abstract

The present invention relates to a method for improving the aerial three-dimensional quality of oblique photogrammetric data of mountainous landforms. The steps are: 1) determining the absolute altitude of each route according to ground resolution requirements and mountain undulations; 2) obtaining mountainous areas through aerial remote sensing flight 3) Arrange the oblique image data, GNSS data and IMU data; 4) The overall air triangulation and the air triangulation are divided into blocks first, and then the air triads are combined to obtain the corresponding air triangulation results. and air-3 report; 5) Compare air-3 results and air-3 report under different conditions, select the best result, and use it for 3D modeling of mountain real scene to form clear and accurate 3D geographic information data. The present invention performs rapid post-processing on the oblique photogrammetry data of mountainous landforms, quantitatively analyzes the quality of aerial survey data products, judges and selects the optimal results in time, and solves the empty three problems caused by the mountainous landforms. Therefore, it can improve work efficiency and reduce losses. .

Description

A method for improving the quality of aerial triangulation in oblique photogrammetry of mountainous landforms

technical field

The invention relates to a method for improving the aerial triangulation quality of oblique image data of mountainous landforms, in particular to a method for improving the aerial triangulation quality of oblique photogrammetry of mountainous landforms.

Background technique

There are some problems in the empty analysis of oblique photogrammetry data of mountainous landforms, including:

1. If the mountains are undulating and the same ground resolution is guaranteed, the remote sensing aircraft needs to take oblique photography on routes with different absolute altitudes, which is prone to loopholes and changes in the image overlap rate, which is not conducive to aerial three; In the case of the same absolute altitude, the acquired oblique images have different ground resolutions, and the size of the same feature on adjacent images is inconsistent, which is not conducive to image matching;

2. Mountains are prone to winds with uncertain direction and speed, causing the plants on the mountain to sway, resulting in blurred images, which is not conducive to feature point extraction and image matching; in addition, winds with uncertain directions and speeds in the mountains will cause the aircraft to shake and tilt If it becomes larger, it may even cause the aircraft to hit a mountain or overturn, resulting in a larger image inclination and poorer accuracy, which is not conducive to aerial 3rd;

3. Mountainous landforms are prone to shadows. Different time periods and different dates have different light and shadow effects and different colors of images, which is not conducive to image matching;

4. Mountainous areas are generally covered by large areas of plants. Plants are objects with low texture and simple geometry. Similar plants are similar in images and are not easy to distinguish, which is not conducive to feature point extraction and image matching.

Therefore, for the oblique photogrammetry of mountainous landforms, it is more likely to cause the problem of empty space than the flat land and hills. If the quality of the space is poor, the later real scene 3D modeling may appear faults, distortions, stretching, blurring, and uneven color. and loopholes and other issues, resulting in job task failure.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for improving the aerial triangulation quality of oblique photogrammetry of mountainous landforms, perform aerial triangulation for oblique image data in mountainous areas, quantitatively analyze the quality of aerial survey data products, judge and select the optimal results in time, and solve the problem of Due to the air three problems caused by the mountainous landforms, it can improve work efficiency and reduce losses.

In order to achieve the above object, the present invention has the following technical solutions:

A method of the present invention for improving the quality of the oblique photogrammetry of mountainous landforms has the following steps:

1) Determine the absolute altitude of each route according to the ground resolution requirements and mountain undulations;

2) Obtain oblique image data, GNSS data and IMU data of mountainous areas through aerial remote sensing flight;

3) Sort out the oblique image data, GNSS data and IMU data;

4) The overall empty three and first block the empty three, and then merge the empty three, and obtain the corresponding empty three result and the empty three report;

5) Compare the results and reports of the three-dimensional space under different conditions, select the optimal result, and use it for the three-dimensional modeling of the real scene in the mountainous area to form clear and accurate three-dimensional geographic information data.

Wherein, the absolute altitude of each route in step 1) includes several identical altitudes, or includes several different altitudes, and the present invention can process both the data of the same absolute altitude and the data of different absolute altitudes; According to the ground resolution requirements in step 1) and the undulating situation of the mountainous area, the relative flight height and absolute flight height of each route can be obtained, wherein the relationship between the relative flight height and the ground resolution is:

In the formula, h is the relative flight height of the route, f is the focal length of the camera lens, GSD is the ground resolution, a is the pixel size of the camera CCD array, and f and a are constant values.

Among them, the relationship between absolute flight height and relative flight height is:

H=h+h′

In the formula, H is the absolute altitude of the route, h is the relative altitude of the route, and h' is the altitude of the mountain.

If the mountain area is small, that is, the height difference inside the mountain area is small, and the relative flight height is basically the same under the requirement of the same ground resolution, so that all routes are at the same absolute flight height;

If the mountainous area has large fluctuations, that is, the height difference within the mountainous area is large, and under the requirements of the same ground resolution, the route will be at different absolute altitudes;

Conversely, if all routes are required to be at the same absolute altitude, the data corresponding to mountainous areas with small undulations have the same ground resolution; the data corresponding to mountainous areas with large undulations have different ground resolutions.

Wherein, the oblique image data in the step 2) includes color digital image data from five angles of front view, rear view, left view, right view and downward view; the GNSS data in the step 2) includes longitude, latitude and The data of elevation, the IMU data in described step 2) comprises corner element Roll (Φ), Pitch (Θ), Heading (Ψ);

Wherein, the sorting in the step 3) refers to: according to the number of days of remote sensing flight, sorting out the positioning and attitude data corresponding to the images acquired every day and the positioning and attitude data corresponding to all the images;

Wherein, the overall blanking in the step 4) refers to: performing blanking on all the acquired data; first dividing the blanking and then merging blanking means: performing blanking on the data acquired every day, and then The results of the empty three are merged, and then the combined data is subjected to empty three; the overall empty three is carried out under the conditions of different phases and different inclination angles; the block empty three is carried out under the conditions of the same phase and the same inclination, so The block empty three is more stable than the overall empty three. After merging the block empty three and then performing the empty three, it is based on the result of the stable empty three. The result is also more stable than the overall empty three;

The overall empty three and the first block and combined empty three include SURF feature point detection, SURF feature point description and RANSAC accurate matching. The specific steps are:

1) SURF feature point detection: use the maximum value of the determinant of the Hessian matrix to detect feature points, assuming that I is an image, X(x, y) is a point in the image, and the scale is σ, then the Hessian matrix at point X is :

Among them, L _xx (X,σ) is the convolution of the second-order Gaussian differential at point X and the image I, and the other items have similar meanings;

The approximate Hessian matrix determinant is taken as the speckle response at X(x, y, σ) to speed up the operation efficiency. The specific formula is:

Det(H _approx )=D _xx D _yy -(0.9D _xx ) ²

Calculate the blob response value of each point in the image to form a response image, and judge whether a point is a candidate feature point by comparing the response value of a point in this scale space and the upper and lower scale spaces. If the threshold value is large or small, the point is used as the final candidate feature point, and its position and scale parameters are calculated;

2) Main direction assignment of SURF feature points: First, perform Haar wavelet operation. The specific parameters are: 6s is the radius, the feature point is the center, and the side length is 4s, and the Haar wavelet response value of the point in the x and y directions is obtained, where s is Then the Gaussian weighting operation is performed, and the specific parameters are: a fan-shaped sliding window with an opening angle of π/3, a step size of 0.2 radian sliding window, and the Haar wavelet response values dx and dy of the images in the window are accumulated to obtain a vector ( m _ω , θ _ω ):

Find the direction with the largest value in the multiple directions of the accumulated value of the Haar wavelet response value, and use this direction as the main direction of the feature point;

3) SURF feature point feature vector generation: construct a regular sub-window with 20s as the side length, the feature point as the center, the direction is consistent with the main direction of the feature point, and the size is 4×4. Use Haar wavelet with side length of 2σ to process the image to obtain the response values dx, dy in the x and y directions, and use Gaussian weighting to calculate the response value of each sub-window to obtain the feature vector of each sub-window:

υ _subwindow =[∑dx∑dy∑|dx|∑|dy]

A set of descriptor feature vectors contains a total of 4×4×4=64-dimensional feature vectors, and the complete information of a feature point can be obtained: spatial scale, coordinates, and 64-dimensional vector features;

4) RANSAC algorithm exact matching: match the right-view image according to the feature points of the left-view image, match the left-view image with the feature points of the right-view image, and then filter. The array of RANSAC model is estimated, and the correct matching point pair is judged, and n iterations are performed to obtain the final matching point and transformation matrix.

Among them, it also includes the following steps:

For the oblique image data, GNSS data and the combined navigation data with the corner elements Roll(Φ), Pitch(Θ), Heading(Ψ) as the initial values of the IMU data, carry out empty triangulation, and iteratively calculate the position information and attitude information of the oblique image data respectively. , remove the residuals and gross errors. When the optimal number of iterations is reached, if the airway set is not parallel to the X-Y plane, the iterative calculation will not continue, and the result will be judged to fail; if all the route sets are parallel to the X-Y plane , the iterative calculation is not continued, and the result is determined to be feasible; when the optimal number of iterations is not reached, if the airway set is not parallel to the X-Y plane in the air three results, the iterative calculation of the position information and attitude information of the oblique image data is continued. , remove the residuals and gross errors, until all the route sets are parallel to the X-Y plane, and the result is determined to be feasible; take the feasible empty triangulation result as the new initial value and continue to empty triangulation, and iteratively calculate the position information and attitude information of the oblique image data , remove the residuals and gross errors, observe the number of connection points after each calculation and whether the route set is parallel to the X-Y plane after each calculation. When the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, select this empty The third result is used as the new initial value and continues to empty the third, iteratively optimizes the position information and attitude information of the oblique image data, removes residuals and gross errors, and observes the number of connection points after each optimization and whether the route set and the X-Y plane after each calculation are Parallel, when the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, if the GNSS data does not use RTK measurement, the optimization will end, and the air-tripping result and air-tripping report will be obtained; if the GNSS data adopts RTK measurement, select this The second empty three results are used as the new initial value to continue the empty three, iteratively optimize the position information and attitude information of the oblique image data, remove the residual and gross errors, observe the number of connection points after each optimization and the route set and X-Y after each calculation. Whether the plane is parallel or not, when the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, select the result of the air three times as the final result, compare the final results of the two air three times, and select the one with more connection points and high accuracy for the real 3D construction. mold.

Among them, multiple iterative calculations and optimizations are used to eliminate residuals and gross errors, so that the position and posture of the photo are correct, and at the same time, the number of connection points is sufficient to ensure that the number of triangular meshes used for modeling is sufficient, so as to improve the three-dimensional model. the quality of.

Owing to having adopted the above technical scheme, the advantages of the present invention are:

1. The operation of the present invention is simple, and different strategies can be used for calculation, and then the optimal solution can be found from different results, so that the quality of the results is the best;

2 It is convenient and quick, can be quantitatively analyzed, has high precision, can improve work efficiency, and reduce repetitive work and work losses;

3 Able to solve problems caused by mountainous landforms and avoid mission failure.

Description of drawings

Fig. 1 is the flow chart of the present invention;

Fig. 2 is the hollow three flow chart of the present invention;

Fig. 3 is a schematic diagram of a hollow three-result existence route set that is not parallel to the X-Y plane of the present invention;

Fig. 4 is the schematic diagram of the normal result of hollow three of the present invention;

In the figure, 1. Route set 1; 2. Route set 2;

Detailed ways

The following examples are intended to illustrate the present invention, but not to limit the scope of the present invention.

Referring to accompanying drawing 1-FIG. 4, a method of the present invention for improving the quality of oblique photogrammetry of mountainous landforms has the following steps:

1) Determine the absolute altitude of each route according to the ground resolution requirements and mountain undulations;

2) Obtain oblique image data, GNSS data and IMU data of mountainous areas through aerial remote sensing flight;

3) Sort out the oblique image data, GNSS data and IMU data;

4) The overall empty three and first block the empty three, and then merge the empty three, and obtain the corresponding empty three result and the empty three report;

5) Compare the results and reports of the three-dimensional space under different conditions, select the optimal result, and use it for the three-dimensional modeling of the real scene in the mountainous area to form clear and accurate three-dimensional geographic information data.

Wherein, the absolute altitude of each route in step 1) includes several identical altitudes, or includes several different altitudes, and the present invention can process both the data of the same absolute altitude and the data of different absolute altitudes; According to the ground resolution requirements in step 1) and the undulating situation of the mountainous area, the relative flight height and absolute flight height of each route can be obtained, wherein the relationship between the relative flight height and the ground resolution is:

In the formula, h is the relative flight height of the route, f is the focal length of the camera lens, GSD is the ground resolution, a is the pixel size of the camera CCD array, and f and a are constant values.

Among them, the relationship between absolute flight height and relative flight height is:

H=h+h′

In the formula, H is the absolute altitude of the route, h is the relative altitude of the route, and h' is the altitude of the mountain.

If the mountain area is small, that is, the height difference inside the mountain area is small, and the relative flight height is basically the same under the requirement of the same ground resolution, so that all routes are at the same absolute flight height;

If the mountainous area has large fluctuations, that is, the height difference within the mountainous area is large, and under the requirements of the same ground resolution, the route will be at different absolute altitudes;

Conversely, if all routes are required to be at the same absolute altitude, the data corresponding to mountainous areas with small undulations have the same ground resolution; the data corresponding to mountainous areas with large undulations have different ground resolutions.

Wherein, the oblique image data in the step 2) includes color digital image data from five angles of front view, rear view, left view, right view and downward view; the GNSS data in the step 2) includes longitude, latitude and Elevation, the IMU data in described step 2) includes corner element Roll (Φ), Pitch (Θ), Heading (Ψ);

Wherein, the sorting in the step 3) refers to: according to the number of days of remote sensing flight, sorting out the positioning and attitude data corresponding to the images acquired every day and the positioning and attitude data corresponding to all the images;

Wherein, the overall blanking in the step 4) refers to: performing blanking on all the acquired data; first dividing the blanking and then merging blanking means: performing blanking on the data acquired every day, and then The results of the empty three are merged, and then the combined data is subjected to empty three; the overall empty three is carried out under the conditions of different phases and different inclination angles; the block empty three is carried out under the conditions of the same phase and the same inclination, so The block empty three is more stable than the overall empty three. After merging the block empty three and then performing the empty three, it is based on the result of the stable empty three. The result is also more stable than the overall empty three;

The overall empty three and the first block and combined empty three include SURF feature point detection, SURF feature point description and RANSAC accurate matching. The specific steps are:

1) SURF feature point detection: use the maximum value of the determinant of the Hessian matrix to detect feature points, assuming that I is an image, X(x, y) is a point in the image, and the scale is σ, then the Hessian matrix at point X is :

Among them, L _xx (X,σ) is the convolution of the second-order Gaussian differential at point X and the image I, and the other items have similar meanings;

The approximate Hessian matrix determinant is taken as the speckle response at X(x, y, σ) to speed up the operation efficiency. The specific formula is:

Det(H _approx )=D _xx D _yy -(0.9D _xx ) ²

Calculate the blob response value of each point in the image to form a response image, and judge whether a point is a candidate feature point by comparing the response value of a point in this scale space and the upper and lower scale spaces. If the threshold value is large or small, the point is used as the final candidate feature point, and its position and scale parameters are calculated;

2) Main direction assignment of SURF feature points: First, perform Haar wavelet operation. The specific parameters are: 6s is the radius, the feature point is the center, and the side length is 4s, and the Haar wavelet response value of the point in the x and y directions is obtained, where s is Then the Gaussian weighting operation is performed, and the specific parameters are: a fan-shaped sliding window with an opening angle of π/3, a step size of 0.2 radian sliding window, and the Haar wavelet response values dx and dy of the images in the window are accumulated to obtain a vector ( m _ω , θ _ω ):

Find the direction with the largest value in the multiple directions of the accumulated value of the Haar wavelet response value, and use this direction as the main direction of the feature point;

3) SURF feature point feature vector generation: construct a regular sub-window with 20s as the side length, the feature point as the center, the direction is consistent with the main direction of the feature point, and the size is 4×4. Use Haar wavelet with side length of 2σ to process the image to obtain the response values dx, dy in the x and y directions, and use Gaussian weighting to calculate the response value of each sub-window to obtain the feature vector of each sub-window:

υ _subwindow =[∑dx∑dy∑|dx|∑|dy]

A set of descriptor feature vectors contains a total of 4×4×4=64-dimensional feature vectors, and the complete information of a feature point can be obtained: spatial scale, coordinates, and 64-dimensional vector features;

4) RANSAC algorithm exact matching: match the right-view image according to the feature points of the left-view image, match the left-view image with the feature points of the right-view image, and then filter. The array of RANSAC model is estimated, and the correct matching point pair is judged, and n iterations are performed to obtain the final matching point and transformation matrix.

Among them, it also includes the following steps:

For the oblique image data, GNSS data and the combined navigation data with the corner elements Roll(Φ), Pitch(Θ), Heading(Ψ) as the initial values of the IMU data, carry out empty triangulation, and iteratively calculate the position information and attitude information of the oblique image data respectively. , remove the residuals and gross errors. When the optimal number of iterations is reached, if the airway set is not parallel to the X-Y plane, the iterative calculation will not continue, and the result will be judged to fail; if all the route sets are parallel to the X-Y plane , the iterative calculation is not continued, and the result is determined to be feasible; when the optimal number of iterations is not reached, if the airway set is not parallel to the X-Y plane in the air three results, the iterative calculation of the position information and attitude information of the oblique image data is continued. , remove the residuals and gross errors, until all the route sets are parallel to the X-Y plane, and the result is determined to be feasible; take the feasible empty triangulation result as the new initial value and continue to empty triangulation, and iteratively calculate the position information and attitude information of the oblique image data , remove the residuals and gross errors, observe the number of connection points after each calculation and whether the route set is parallel to the X-Y plane after each calculation. When the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, select this empty The third result is used as the new initial value and continues to empty the third, iteratively optimizes the position information and attitude information of the oblique image data, removes residuals and gross errors, and observes the number of connection points after each optimization and whether the route set and the X-Y plane after each calculation are Parallel, when the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, if the GNSS data does not use RTK measurement, the optimization will end, and the air-tripping result and air-tripping report will be obtained; if the GNSS data adopts RTK measurement, select this The second empty three results are used as the new initial value to continue the empty three, iteratively optimize the position information and attitude information of the oblique image data, remove the residual and gross errors, observe the number of connection points after each optimization and the route set and X-Y after each calculation. Whether the plane is parallel or not, when the number of connection points reaches the maximum and the route set is parallel to the X-Y plane, select the result of the air three times as the final result, compare the final results of the two air three times, and select the one with more connection points and high accuracy for the real 3D construction. mold.

Among them, the optimal number of iterations is 6-10 times.

Among them, multiple iterative calculations and optimizations are used to eliminate residuals and gross errors, so that the position and posture of the photo are correct, and at the same time, the number of connection points is sufficient to ensure that the number of triangular meshes used for modeling is sufficient, so as to improve the three-dimensional model. the quality of.

Table 1 is the empty triple precision report table of the embodiment of the present invention:

It can be seen from the above table that the accuracy of the Aerial 3 is basically unchanged and has become stable. It can be seen that the "connection points" has reached the maximum. At this time, it is necessary to observe whether the route set is parallel to the X-Y plane. If it is parallel, the Aerial 3 ends normally.

Absolute altitude: refers to the vertical distance from the surface of the ground or the sea, also known as the "altitude". That is, the vertical distance from the standard atmospheric sea level, or the mean sea level as the standard height. The terrain and the height of the objects marked on the navigation map are all calculated according to the absolute height.

Relative altitude: the vertical distance of the aircraft above a specified point.

SURF: The full name is Speed-up robust features, that is, the accelerated robust feature algorithm, which is a highly robust local feature point detector. The algorithm can be used for object recognition or 3D reconstruction in the field of computer vision.

RANSAC: is the abbreviation of Random Sample Consensus. It is an algorithm that calculates the mathematical model parameters of the data and obtains valid sample data according to a set of sample data sets containing abnormal data. The RANSAC algorithm is often used in computer vision. For example, in the field of stereo vision, the matching point problem of a pair of cameras and the calculation of the fundamental matrix are solved simultaneously.

X-Y plane: the plane parallel to the mean sea level.

Route set: It is a set composed of several oblique images.

Oblique photogrammetry: Oblique photogrammetry technology is a high-tech developed in the field of international surveying and mapping in recent years. It subverts the previous limitation that orthophotos can only be shot from a vertical angle. At the same time, images are collected from five different angles, including one vertical and four obliques, introducing users into a real and intuitive world that conforms to human vision. Aerial oblique images can not only truly reflect the situation of ground objects, but also greatly expand the application field of remote sensing images by adopting advanced positioning technology, embedding accurate geographic information, richer image information, and more advanced user experience. Make the industrial application of remote sensing images more in-depth. Since oblique images provide users with richer geographic information and a more user-friendly experience, this technology has been widely used in industries such as emergency command, homeland security, urban management, and real estate taxation in developed countries such as Europe and the United States.

Realistic 3D modeling: It refers to the automatic generation of high-resolution 3D models with realistic texture maps based on a series of 2D photos or a set of oblique images. If the oblique image has coordinate information, then the geographic location information of the model is also accurate. This model has realistic effects, comprehensive elements, and measurement accuracy, which not only brings an immersive feeling to the human body, but also can be used for surveying applications, which is a true restoration of the real world.

IMU: Inertial measurement unit, which is a device that measures the three-axis attitude angle (or angular rate) and acceleration of an object;

GNSS: The abbreviation of Global Navigation Satellite System, that is, the global navigation satellite system. As early as the mid-1990s, in order to break the monopoly of the United States in the satellite positioning, navigation, and timing markets, obtain huge market benefits, and increase employment opportunities for Europeans, the European Union has been working on the civil GNSS plan. Call it the Global Navigation Satellite System. The plan is implemented in two steps: the first step is to build a first-generation global navigation satellite system (then called GNSS-1, which was later built as EGNOS) that comprehensively utilizes the GPS system of the United States and the GLONASS system of Russia; the second step It is to establish a second-generation global navigation satellite system that is completely independent of the GPS system of the United States and the GLONASS system of Russia, that is, the Galileo satellite navigation and positioning system under construction. It can be seen that GNSS has not been a single constellation system since its inception, but a comprehensive constellation system including GPS, GLONASS, etc.; differential GNSS refers to the use of additional data from the reference GNSS receiver with known positions to reduce A technique for small positioning errors of GPS systems or GLONASS systems.

Integrated navigation data: refers to the integrated navigation data combined with satellite navigation data (GNSS data) and inertial navigation data (IMU data), including the position information and attitude information of the target.

Tiepoints: Within the overlapping range of the stereo pair, the constellation points of the same object point on different pictures are called image points with the same name, and a large number of automatically or manually generated image points with the same name are collectively called tie points.

Image matching: The process of identifying points with the same name between two or more images through a certain matching algorithm.

Air triangulation: aerial triangulation. Aerial triangulation is a measurement method of three-dimensional photogrammetry. According to a small number of field control points, the control points are encrypted indoors, and the elevation and plane position of the encrypted points are obtained. Its main purpose is to provide absolutely oriented control points for mapping areas lacking field control points. Aerial triangulation is generally divided into two types: analog aerial triangulation, which is optical-mechanical aerial triangulation; analytical aerial triangulation, commonly known as computer encryption. Simulated aerial triangulation is an aerial triangulation carried out on an all-purpose stereo measuring instrument (such as a multiplexer). It is to restore the three-dimensional model of the route that is similar to or corresponding to the photography on the instrument, select the encrypted points according to the needs of the mapping, and determine its elevation and plane position. In aerial photogrammetry, it is a method to encrypt control points indoors by using the inherent geometric characteristics of photos. That is, using continuously captured aerial photographs with a certain overlap, based on a small number of field control points, photogrammetry is used to establish a route model or an area network model (optical or digital) corresponding to the field, so as to obtain the plane coordinates of the encrypted points and Elevation. Mainly used for topographic maps.

RTK: The real-time dynamic differential method. This is a new and commonly used GPS measurement method. The previous static, fast static and dynamic measurements all need to be solved afterwards to obtain centimeter-level accuracy. RTK is a measurement method that can obtain centimeter-level positioning accuracy in real time in the field. It adopts the dynamic real-time differential method of carrier phase, which is a major milestone in GPS application. Its appearance brings a new dawn to engineering stakeout, topographic mapping, and various control measurements, and greatly improves the efficiency of field operations. RTK positioning technology is a real-time dynamic positioning technology based on carrier phase observations. RTK can provide real-time three-dimensional positioning results of stations in a specified coordinate system, and achieve centimeter-level accuracy. In RTK operation mode, the base station transmits its observations together with the station coordinate information to the rover through the data link. The rover not only receives data from the base station through the data link, but also collects GPS observation data, forms differential observations in the system for real-time processing, and gives centimeter-level positioning results, which lasts less than one second. The rover can be in a static state or in a moving state; it can be initialized at a fixed point before entering the dynamic operation, or it can be powered on directly under dynamic conditions, and complete the search and solution of the ambiguity of the whole week in a dynamic environment. After the solution of the unknowns for the whole week is fixed, the real-time processing of each epoch can be carried out. As long as the tracking of the phase observations of more than four satellites and the necessary geometric figures can be maintained, the rover can give centimeter-level positioning results at any time.

Roll, Pitch, Heading: namely the roll angle, pitch angle and declination angle, which are the angle element systems usually used by the inertial measurement unit (IMU) to describe the attitude of the sensor;

Obviously, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the embodiments of the present invention. For those of ordinary skill in the art, changes or modifications in other different forms can also be made on the basis of the above description. Not all implementations can be exhaustive here. Any obvious changes or changes derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims (3)

1. A method for improving the aerial three quality of mountain landform oblique photogrammetry is characterized by comprising the following steps:

1) determining the absolute flight height of each route according to the ground resolution requirement and the mountainous area fluctuation condition;

the absolute flight height of each route comprises a plurality of same heights or comprises a plurality of different heights; obtaining the relative altitude and the absolute altitude of each route according to the ground resolution requirement and the mountain area fluctuation condition in the step 1), wherein the relational expression of the relative altitude and the ground resolution is as follows:

in the formula, h is the relative altitude of a flight path, f is the focal length of a camera lens, GSD is the ground resolution, a is the pixel size of a camera CCD array, and f and a are constant values;

wherein, the relation between the absolute altitude and the relative altitude is as follows:

H＝h+h′

in the formula, H is the absolute flight height of the flight path, H is the relative flight height of the flight path, and H' is the altitude of the mountainous area;

if the fluctuation of the mountainous area is small, namely the height difference inside the mountainous area is small, the relative flight heights are basically the same under the requirement of the same ground resolution, so that all the flight lines are at the same absolute flight height;

if the mountainous area has large fluctuation, namely the height difference inside the mountainous area is large, the flight lines are at different absolute flight heights under the requirement of the same ground resolution;

on the contrary, if all the flight lines are required to be at the same absolute flight height, the data corresponding to the mountainous area with small fluctuation have the same ground resolution; the data corresponding to the mountainous area with large fluctuation have different ground resolutions;

2) acquiring tilt image data, GNSS data and IMU data of a mountain area through aerial remote sensing flight;

3) sorting the oblique image data, the GNSS data and the IMU data;

4) the whole empty three and the empty three are firstly partitioned and then merged to obtain corresponding empty three results and empty three reports;

the whole empty three in the step 4) refers to: performing null processing on all the acquired data; the first blocking and empty three and then merging empty three means: respectively carrying out space-three on the data acquired every day, then combining all the space-three results, and carrying out space-three on the combined data; the whole air separation is carried out under the conditions of different time phases and different inclination angles; the blocking empty three is carried out under the conditions of the same time phase and the same inclination angle, so the blocking empty three is more stable than the integral empty three, and the empty three is carried out again based on the stable empty three result after the blocking empty three is combined, and the result is also more stable than the integral empty three;

the integral three-in-one method and the method for partitioning three-in-one and combining three-in-one comprise SURF feature point detection, SURF feature point description and RANSAC accurate matching, and specifically comprise the following steps:

(1) SURF feature point detection: detecting the characteristic points by using the maximum value of a Hessian matrix determinant, and assuming that I is an image, X (X, y) is one point in the image, and the scale is sigma, the Hessian matrix at the point X is as follows:

wherein L is_xx(X, σ) is the convolution of the second-order derivative of Gaussian at point X with image I, with the remaining terms having similar meanings;

taking the determinant of the Hessian matrix as the response of the spot at X (X, y, sigma) to accelerate the operation efficiency, the specific formula is as follows:

Det(H_approx)＝D_xxD_yy-(0.9D_xx)²

calculating a speckle response value of each point in the image to form a response image, judging whether a certain point is a candidate feature point or not by comparing the response values of the certain point in the scale space and the upper and lower scale spaces, and if the response value is larger or smaller than 26 neighborhood values, taking the point as a final candidate feature point and calculating the position and scale parameters of the point;

(2) SURF feature point principal direction assignment: firstly, performing Haar wavelet operation, wherein the specific parameters are as follows: 6s is the radius, the characteristic point is the center, the side length is 4s, and Haar wavelet response values of the points in the x and y directions are obtained, wherein s is the space scale; and then carrying out Gaussian weighting operation, wherein the specific parameters are as follows: a fan-shaped sliding window with the opening angle of pi/3 and the step length of 0.2 radian sliding window, and the Haar wavelet response values dx and dy of the images in the window are accumulated to obtain a vector (m)_ω,θ_ω)：

Solving a direction with the maximum value in a plurality of directions of the accumulated value of the Haar wavelet response values, and taking the direction as the main direction of the feature point;

(3) SURF feature point feature vector generation: constructing a regular sub-window with the side length of 20s, the center of the feature point, the direction consistent with the main direction of the feature point and the size of 4 multiplied by 4; processing the image by using a Haar wavelet with the side length of 2 sigma to obtain response values dx and dy in x and y directions, and calculating the response value of each sub-window by using Gaussian weighting to obtain a feature vector of each sub-window:

υ_sub-window＝[∑dx∑dy∑|dx|∑|dy]

A set of descriptor feature vectors contains 4 × 4 × 4 ═ 64-dimensional feature vectors, and complete information of one feature point can be obtained: spatial scale, coordinates, 64-dimensional vector features;

(4) the RANSAC algorithm matches exactly: matching a right view image according to the characteristic points of the left view image, matching the right view image according to the characteristic points of the right view image, screening, storing the matching point pairs into a new array if the matching can be successful for two times, carrying out RANSAC model estimation, judging correct matching point pairs, and carrying out iterative computation for n times to obtain final matching points and a conversion matrix;

5) comparing the empty three results and the empty three reports under different conditions, and selecting the optimal result for mountain real-scene three-dimensional modeling to form clear and accurate three-dimensional geographic information data;

performing aerotriangulation on the oblique image data, the GNSS data and the combined navigation data taking angle elements of Roll (phi), Pitch (theta) and Heading (psi) as initial values of IMU data, respectively performing iterative computation on position information and attitude information of the oblique image data, removing residual errors and gross errors, and when the optimal iteration times are reached, if the aerotriangulation results are not parallel to an X-Y plane, stopping iterative computation and judging that the results fail; if all the aerial sets are parallel to the X-Y plane, the iterative computation is not continued, and the result is judged to be feasible; when the optimal iteration times are not reached, if the three results of the air exist and the aerial sets are not parallel to the X-Y plane, continuously performing iterative computation on position information and posture information of the oblique image data, eliminating residual errors and gross errors until all the aerial sets are parallel to the X-Y plane, and judging that the results are feasible; continuously performing space-three calculation by taking a feasible space-three result as a new initial value, iteratively calculating position information and attitude information of the oblique image data, removing residual errors and gross errors, observing the number of connection points after each calculation and whether a flight line set is parallel to an X-Y plane after each calculation, selecting the space-three result as a new initial value to continuously perform space-three when the number of the connection points reaches the maximum and the flight line set is parallel to the X-Y plane, iteratively optimizing the position information and the attitude information of the oblique image data, removing the residual errors and the gross errors, observing the number of the connection points after each optimization and whether the flight line set is parallel to the X-Y plane after each calculation, and finishing optimization if the GNSS data is not measured by RTK when the number of the connection points reaches the maximum and the flight line set is parallel to the X-Y plane to obtain a space-three result and a space-three report; if the GNSS data adopts RTK measurement, selecting the secondary air-three result as a new initial value to continue air-three, iteratively optimizing the position information and the attitude information of the oblique image data, eliminating residual errors and gross errors, observing the number of the optimized connection points and whether the airline set is parallel to the X-Y plane after each calculation, selecting the secondary air-three result as a final result when the number of the connection points reaches the maximum and the airline set is parallel to the X-Y plane, comparing the two secondary air-three final results, and selecting the connection points with more connection points and high precision for the live-action three-dimensional modeling.

2. The method of claim 1, wherein the method comprises the steps of: the oblique image data in the step 2) comprises color digital image data of five angles of front view, back view, left view, right view and downward view; the GNSS data in the step 2) comprises longitude, latitude and elevation, and the IMU data in the step 2) comprises angle elements Roll (phi), Pitch (theta) and Heading (psi).

3. The method of claim 1, wherein the method comprises the steps of: the finishing in the step 3) is as follows: and respectively sorting out positioning and attitude determining data corresponding to the images acquired every day and positioning and attitude determining data corresponding to all the images according to the number of days of remote sensing flight.

Images