CN105222788B - The automatic correcting method of the matched aircraft Route Offset error of feature based - Google Patents

Abstract

The self-calibration method of aircraft route offset error based on feature matching belongs to the field of aircraft route position offset correction technology, which is divided into two parts: offline processing and online processing: the offline processing part uses satellites or drones to obtain the in- Prior image, use the SIFT algorithm soft method to obtain the SIFT feature point parameters of the prior image and the geographic location information of the landmark area to generate a prior database; the aircraft loads the prior database into the aircraft memory, and collects images during flight Use the SIFT algorithm software to obtain the SIFT feature point parameters, match with the prior image SIFT feature points to form a feature point pair, and then use the set fine matching threshold of feature point pairs and the set aircraft yaw accuracy respectively Using different offset calculation methods, compared with the real-time calculation method based on the single-point landmark aircraft space position, the present invention overcomes the influence of height and attitude angle information on the calculation accuracy.

Description

Self-calibration method of aircraft route offset error based on feature matching

technical field

The invention provides a feature-matching-based self-correction method for flight path offset errors, which belongs to the combined field of image information processing and navigation. According to the feature registration result, the path offset calculation method is adaptively selected, so as to obtain the position offset of the aircraft.

Background technique

The aircraft relies on the inertial navigation system (Inertial Navigation System, INS) for positioning during flight, but due to the cumulative error of the inertial device, the position coordinates output by the inertial navigation system will accumulate and diverge over time. In order to correct the cumulative error of inertial navigation, the output of INS is generally corrected by means of integrated navigation to improve the output accuracy of inertial navigation. INS/satellite positioning system is the most commonly used way of integrated navigation, such as Global Positioning System (Global Position System, GPS) or Beidou Navigation Satellite System (Beidou Navigation Satellite System, BDS).

However, in practical applications, the satellite positioning system is not available in real time and will be affected by the availability of satellite platforms and signal interference. When the satellite positioning system fails, other methods are needed to correct the output results of the inertial navigation. Due to the strong autonomy, high precision and good real-time performance of visual navigation, INS/visual navigation has become an important development direction of integrated navigation.

Ding Wenrui and Kang Chuanbo's patent (application publication number CN 103822635A) is based on the visual information-based real-time calculation method of the spatial position of the drone in flight. This patent uses a single-point landmark to calculate the spatial position of the aircraft in real time. Height information, which introduces the height error into the calculation result. However, the present invention can make full use of the information of the landmark area, and use SIFT matching feature points to perform offset calculation. When the number of matching pairs is greater than the set threshold N_fine, the method of step (5.2) or step (5.3) can be used for calculation. , the step (5.2) method eliminates the influence of the height error on the calculation accuracy, and the step (5.3) method eliminates the influence of the attitude angle and height information on the calculation accuracy.

Contents of the invention

The object of the present invention is to provide a method for calculating route offset based on feature registration using an image sensor. When the satellite positioning system fails, by collecting image information of known landmarks, the inertial navigation error of the aircraft is realized. Correction in real time.

The system is divided into two parts: offline processing and aircraft online processing.

The role of offline processing is to select landmark areas and establish a priori information database, including the following steps:

Step (1) The landmark area is selected and the prior image information of the landmark area is obtained.

Step (2) SIFT feature point extraction is performed on the prior image of the landmark area.

Step (3) Generate a priori database of the landmark area, including prior image information, feature point information, and geographic location information of the landmark area.

The role of online processing is to use the image sensor to collect images of landmark areas when the aircraft is flying, compare them with the prior database of landmark areas established offline, detect the landmark areas, and calculate the aircraft’s position based on the pixel coordinates of the landmark areas in the collected images. location information. Include the following steps:

Step (1) Store the prior database of landmark regions into the aircraft's memory.

Step (2) Take a real-time image of the selected landmark area and perform image preprocessing:

Due to the influence of the environment, sensors, and platform interference, the image data obtained by the aircraft will have noise and various forms of interference. In order to ensure the quality and performance of subsequent operations, the acquired images must first be preprocessed.

Step (3) extracts the SIFT feature points of the collected image:

The SIFT algorithm software is used to extract the SIFT feature points from the pre-processed collected images, the collected images of the landmark areas are used as the input of the SIFT algorithm, and the output is obtained to obtain a stable SIFT feature point parameter set.

Step (4) feature matching:

Feature matching is based on the SIFT feature parameter set of the prior image in the online data of the landmark area and the SIFT feature parameter set of the image collected by the image sensor. The feature matching is divided into two parts: rough matching and fine matching.

Step (4.1) rough feature matching:

Rough matching is to use the unique description vector (the Euclidean distance between the prior image of the landmark area and the description vector of the SIFT feature points of the collected image) to establish the mapping relationship between the prior image and the collected image, and use the Euclidean distance between the vectors to Perform coarse registration. The output result of the rough matching includes the feature point pairs that are successfully matched between the prior image and the collected image, and also includes the score of the matching feature point pair. The score is the square of the Euclidean distance of the feature point pair description vector.

Step (4.2) counts the results of rough matching:

If the successfully matched feature point pairs in the rough registration are "0", then output "failure to solve" and go to step (2). If the number of successfully matched feature point pairs in this step is greater than zero, count the number of successfully matched Logarithmic N.

Step (4.3) feature fine matching:

Set the threshold of feature fine matching to N_fine (N_fine>3), and set the required solution accuracy to DIST (unit is meter).

Feature fine matching is based on rough matching, using MLESAC (Maximum Likelihood Estimation by Sample and Consensus, random sampling maximum likelihood estimation) to further screen, remove feature matching pairs with matching errors, and transfer the remaining matching pairs to the step ( 5) Carry out the calculation of the position offset of the aircraft.

When the number of feature point pairs successfully registered is N≤N_fine, output the smallest feature point pair in the score among the N pairs of feature point pairs, and use the method of step (5.1) to solve. When the number of feature point pairs with successful coarse registration is N>N_fine, perform fine registration to filter out the wrong matching pairs and then output the correct feature point pairs. At the same time, judge the set solution accuracy DIST, if DIST≤10m , turn to step (5.3) for calculation, if DIST>10m, turn to step (5.2) for calculation.

Step (5) route offset calculation:

The route offset solution is to calculate the position offset of the aircraft according to the pixel coordinates of the landmark area in the collected image, which is the key point of the present invention.

The present invention provides three methods (a) (b) (c) for calculating the route offset, and select the corresponding method according to the number N of successfully matched feature point pairs in the fine matching and the set calculation accuracy DIST to solve.

The advantages of the present invention are:

(1) Good adaptability. The spatial position of the aircraft can be positioned by directly using the coordinates of the known landmark areas in the collected images without using GPS or Beidou information.

(2) Good stability. When calculating the position coordinates of the aircraft in the present invention, three strategies are specifically designed, and the combination of the three methods can make full use of the information of the landmark area and improve the calculation accuracy. At the same time, when the information near the local punctuation point is not perfect, It can also be solved by method (a), which has strong adaptability.

Description of drawings

Figure 1 is a flowchart of the off-line processing portion of the present invention.

Figure 2 is a flow diagram of the on-line processing portion of the present invention.

Figure 3 is the output result after rough matching during online processing.

Figure 4 is the output result after fine matching during online processing.

Detailed ways

The specific implementation manner of the present invention will be described in detail below according to the accompanying drawings.

The self-calibration method of the aircraft route offset error based on feature matching of the present invention, the flow chart of its offline processing part is shown in Figure 1, including the following steps:

Step (1) Select the landmark area and obtain the prior image of the landmark area:

The landmark area is selected according to the pre-planned flight route of the aircraft. The selected landmark areas should have rich, obvious image features and are not easy to be confused, such as skyscrapers, overpasses, rivers, etc., which have obvious characteristics and are difficult to replicate.

Use satellites or unmanned aerial vehicles to collect images of the set landmark area, and obtain the prior image of the landmark area and the geographic coordinate information of the landmark area.

Step (2) extracts the SIFT feature points of the prior image:

Use SIFT algorithm software to extract SIFT feature points from the prior image to obtain the description vector of its feature points A collection of prior image feature point description vectors N _c is the number of SIFT feature points of the prior image,

Step (3) Generate a prior database of landmark areas: store the image information obtained in steps (1) and (2), geographic coordinates of landmark areas, and SIFT feature point parameters into the prior database to complete the establishment of the prior database.

The self-calibration method of the aircraft offset error based on feature matching of the present invention, the flow chart of its online processing part is shown in Figure 2, including the following steps:

Step (1) Store the a priori database of the landmark area in the memory of the aircraft in advance.

Step (2) Capture images of selected landmark areas in real time, and perform image preprocessing:

Preprocessing is performed on the captured image obtained by the aircraft image sensor. The preprocessing includes denoising by median filter method, image enhancement by grayscale histogram correction, and lens distortion correction by using the internal parameters of the camera calibrated in advance.

Step (3) carries out SIFT feature extraction to the acquisition image of landmark area:

Use the SIFT algorithm software to extract the SIFT feature points from the preprocessed image and obtain the description vector of the feature points Generate a set V _p of the SIFT feature point description vectors of the acquired image, N _p is the number of SIFT feature points of the collected image.

Step (4) described prior image and acquisition image carry out SIFT feature point matching:

Feature matching is based on the SIFT feature parameter set of the prior image in the online data of the landmark area and the SIFT feature parameter set of the image collected by the image sensor. Feature matching is divided into two parts: rough matching and fine matching.

Step (4.1) rough feature matching:

Step (4.1.1), randomly select a description vector of a feature point in the collection image SIFT feature point set V _p Make it with the description vectors of all feature points in the set V _c of prior image SIFT feature point description vectors Follow the steps below to perform rough matching of feature points:

Step (4.1.1.1), calculate the following parameters:

Calculate the description vectors of all feature points in V _p and V _c respectively Euclidean distance and by the Euclidean distance The size of from small to large forms an ascending sequence of Euclidean distances, assuming that in the Euclidean distance sequence is the only minimum value in the sequence, is the second smallest value of the Euclidean distance sequence.

Step (4.1.1.2), judge no:

If: less than, the matching is successful, and a pair of feature points is obtained and will Output as the score score of the feature point pair, namely

If: greater than or equal to, the rough matching of this point fails, that is, the feature point in the collected image There are no corresponding feature points in the prior image that can be matched.

Step (4.1.1.3), performing step (4.1.1.1) to step (4.1.1.2) sequentially on the remaining feature points in the SIFT feature point set _Vp of the collected image until _Np SIFT feature points of the collected image All matched.

Step (4.2) statistics rough matching results:

If: if the number of successfully matched SIFT points is "0", the output solution fails and returns to step (2). If the number of successfully matched SIFT feature points is greater than zero, count the number of successfully matched pairs N.

Figure (3) is the result after rough matching, the red area is the prior image, the red circle represents the SIFT feature point of the prior image, the blue part is the collected image, and the green cross represents the SIFT feature point of the collected image . The matching pairs of coarse registration matching will have incorrect matching points, as shown in the figure, so fine registration is used for further screening.

Step (4.3), feature point fine matching:

Step (4.3.1), initialization:

Set fine matching threshold N_define=8, yaw accuracy DIST=20m.

Step (4.3.2) judge whether the number of rough matching successes N≤8 counted in step (4.2):

If: rough matching feature points N≤8, execute step (5-a),

If: rough matching feature points N>8, execute step (4.3.3),

Step (4.3.3), use MLESAC (Maximum Likelihood Estimation by Sample and Consensus, random sampling maximum likelihood estimation) for fine registration, analyze the consistency of the coordinate relationship between the matching pairs determined by the rough registration, and remove the features of matching errors Matching pairs, output the remaining correct matching feature points.

Figure (4) is the output of feature matching results after fine registration, and the wrong matching points have been removed.

Step (4.3.4) judge whether the set yaw accuracy DIST≤10m:

If: > 10m, then perform step (5-b),

If: ≤10m, execute step (5-c),

Step (5), calculating the position offset of the aircraft according to the pixel coordinates of the landmark area in the collected image, which is the key point of the present invention:

The data provided by the aircraft includes the current latitude and longitude coordinates (B, L) of the aircraft, the current altitude H, and the current attitude angle of the aircraft: yaw angle α, pitch angle β, and roll angle γ.

The specific implementation of the three methods for calculating the position offset of the aircraft is as follows:

Step (5.1), using the height information H, the three attitude angles α, β, γ, and the pixel coordinates of the landmark area M in the captured image Calculate the coordinates of the aircraft in the northeast sky coordinate system established with the position coordinates given by the aircraft inertial navigation as the origin.

Let the coordinates of any point m in the landmark area M be (x _m , y _m , z _m ) in the camera coordinate system, and the coordinates of the corresponding imaging point p in the imaging plane coordinate system be (p _x , p _y ), The coordinates of the imaging point p in the image coordinate system (uv coordinate system) are (u _p , v _p ). Then the relationship between the coordinates (up _, v _p ) in the image coordinate system and the coordinates (p _x , p _y ) in the imaging plane coordinate system is shown in Formula 1-1. According to the projection principle, the coordinates (x _m , y _m , z _m ) of m in the camera coordinate system and the coordinates (p _x , p _y ) of the imaging plane coordinate system are shown in formula 1-2. Sx and Sy are the number of pixels per unit physical size in the X-axis and Y-axis directions of the imaging plane, respectively, and f is the focal length of the camera.

From formula 1-1 and formula 1-2 can get

which is

The northeast sky coordinate system (X _w Y _w Z _w ) is established with the optical center O of the camera as the origin. The relationship between the camera coordinate system and the northeast sky coordinate system is shown in formula 1-4. R _w is the camera coordinate system and the camera Attitude rotation matrix for sky coordinates,

From formula 1-3 and formula 1-4,

When a landmark point m can be correctly identified, the three attitude angles at the moment, yaw angle α, pitch angle β, roll angle γ, and the shooting height H at this time can be obtained, z _w = H, and the landmark point can be obtained The z-axis coordinate z _m of m in the camera coordinate system, so as to obtain the coordinates (x _w , y _w , z _w ) of m in the northeast sky coordinate system of the camera.

This method needs to depend on the altitude H of the aircraft, the attitude angle and the coordinates of the landmarks in the image. The height H is measured by the height sensor, and there is an error, and the attitude angle also has an error. Using data that already has errors to calculate the position of the aircraft, the result will bring errors into the result.

Step (5.2), according to method (5-a), it can be seen that the coordinates (x _c , y _c , z _c ) in the northeast sky established with the aircraft navigation output position as the origin are assumed to be the real position of the aircraft , the pixel coordinates of the landmark point m in the landmark area in the acquisition image sensor are ( _up , v _p ), and the coordinates of the landmark point m in the northeast sky coordinate system established with the aircraft navigation output position as the origin are (x _m , y _m, z _m ), its imaging position ( _up , v _p ) in the image coordinate system can be determined by imaging equations 1-6.

u _p ＝[(cosαcosγ+sinαsinβsinγ)(x _m -x _c )-cosβsinγ(z _m -z _c )

-(cosγsinα-cosαsinβsinγ)(y _m -y _c )]fS _x

/[(cosαsinγ-cosγsinαsinβ)(x _m -x _c )

-(sinαsinγ+cosαcosγsinβ)(y _m -y _c )+cosβcosγ(z _m -z _c )]v _p ＝[sinβ(z _m -z _c )+cosαcosβ(y _m -y _c )+cosβsinα(x _m - x _c )]fS _y

/[(cosαsinγ-cosγsinαsinβ)(x _m -x _c )

-(sinαsinγ+cosαcosγsinγ)(y _m -y _c )+cosβcosγ(z _m -z _c )]

(1-6)

Among them, (x _c , y _c , h _c ) are the spatial coordinates of the camera; (α, β, γ) are the spatial attitude angle of the camera; (f, S _x , S _y ) are the imaging focal length and pixel distribution density of the camera , is a built-in parameter of the camera, which can be provided by the manufacturer or calibrated offline.

The above equations are about (x _c , y _c , z _c ) and (α, β, γ) non-homogeneous linear equations, using the least squares criterion to solve (x _c , y _c , z _c ), get The aircraft's current position coordinates (x _c , y _c , z _c ), that is, the offset of the aircraft's real position relative to the aircraft's inertial navigation output position coordinates, this method only uses the attitude angle information for calculation, eliminating the impact of the height H measurement error on the final impact on the outcome.

In step (5.3), when feature matching pairs N≥3, an optimization function can also be constructed, and an iterative method can be used to solve the optimization problem.

As formula 1-7 constructs the optimization function, The corresponding pixel coordinates of the fine-matched feature points in the collected image are calculated according to the position and attitude angle of the aircraft, and are calculated by the imaging equation, which are about (x _c , y _c , z _c ) and (α, β, γ) function, is the pixel coordinates of the feature points in the collected image that are fine-matched successfully.

m is the logarithm of feature points output after fine matching, i=1, 2, . . . m.

The optimization function J is iteratively obtained to obtain the real position coordinates and attitude angle of the aircraft, and the fminserch function that comes with matlab is used to iteratively solve it. The number of iterations is set to 10000, and the allowable error of the iteration is set to 1e-4. Find (x _c , y _c , z _c ). The position coordinates of the aircraft calculated by this method can eliminate the influence of height error and attitude angle error.

Among the three methods, the first method requires the use of attitude angle information and altitude information, and the calculated aircraft position coordinates introduces attitude angle errors and altitude errors. However, this method has lower requirements for matching feature pairs. In theory, as long as feature matching pairs N≥1, it can be used.

The second method theoretically requires feature matching pairs N≥3, and the corresponding threshold N_define can be set according to the actual system in actual design. This method does not need to use altitude information, but only needs to provide attitude angle information. The calculated position coordinate error of the aircraft is only affected by the attitude angle error, and the error is smaller than the first method.

The third method theoretically requires feature matching pairs N≥3, and the corresponding threshold N_define can be set according to the actual system in actual design. This method adopts an iterative method, neither height information nor attitude angle information is required, and the calculation accuracy can reach below 10m.

The combination of (a), (b) and (c) can make full use of the information of the landmark area and improve the accuracy of the calculation. At the same time, when the information near the landmark is not perfect, the method (a) can also be used. Solve the problem and have strong adaptability.

Claims (1)

1. the automatic correcting method of the matched aircraft Route Offset error of feature based, which is characterized in that be divided into ground and locate offline Reason and two parts of aircraft online processing：

Ground processed offline part, is divided into following steps：

Step (1) is chosen with obvious characteristic and is difficult to the region replicated as landmark region, obtained using unmanned plane or satellite The prior image of the landmark region；

Step (2) extracts the SIFT feature of prior image：

Using SIFT algorithm softwares to the description vectors of prior image extraction SIFT featureForm prior image feature The set of point description vectors N_cFor prior image SIFT feature is counted,

Step (3) generates the prior data bank of landmark region, including prior image, the ground of characteristic point information and landmark region Manage location information,

Aircraft online processing part, is divided into following steps：

Step (1), the prior data bank of landmark region is previously loaded into the memory of aircraft,

Step (2), shoots the realtime graphic of selected landmark region, and carries out image preprocessing：

The landmark region is shot during aircraft real-time flight and carries out Image Acquisition, then, successively using medium filtering denoising, ash Histogram adjusting is spent to enhance image, finally carries out lens distortion calibration with preset camera intrinsic parameter,

The SIFT feature of step (3) extraction acquisition image,

The description vectors of image zooming-out SIFT feature collected using SIFT algorithm softwaresGeneration acquisition image SIFT The set V of feature point description vector_p, N_pTo adopt Collect the SIFT feature points of image,

Step (4) carries out SIFT feature matching to the prior image and acquisition image：

Step (4.1) characteristic point slightly matches：

Step (4.1.1), in acquisition image SIFT feature point set V_pIn appoint and take the description vectors of a characteristic pointMake itself and elder generation Test the set V of image SIFT feature description vectors_cIn all characteristic points description vectorsCharacteristic point is carried out according to the following steps Thick matching：

Step (4.1.1.1) calculates following parameters：

V is calculated respectively_pWith V_cIn all characteristic points description vectorsEuclidean distanceAnd by the Euclidean away from FromSize from small to large, form the ascending sequence of Euclidean distance, it is assumed that in the Euclidean distance sequenceFor the only one minimum value in the sequence,For the sub-minimum of the Euclidean distance sequence,

Step (4.1.1.2) judgesIt is no：

If：It is less than, then successful match, obtains a pair of of characteristic point pairIt and willAs this feature point pair Score score exported, i.e.,

If：It is more than or equal to, then the thick this feature point that it fails to match, that is, acquires in image of the pointNo pair in prior image The characteristic point answered can match,

Step (4.1.1.3), in the acquisition image SIFT feature point set V_pIn each characteristic point of remainder perform step successively (4.1.1.1)~step (4.1.1.2), the N until acquiring image_pAll matching finishes a SIFT feature,

Step (4.2) counts thick matching result：

If：If the SIFT points of successful match are " 0 ", output resolves failure, return to step (2), if the SIFT of successful match is special Sign points are more than zero, then count the logarithm N of successful match,

Step (4.3), the matching of characteristic point essence：

Step (4.3.1), initialization：

The matched threshold value N_define of setting essence, and the N_define set should meet N_define>3, precision DIST is yawed, it is single Position is rice,

The thick successful match number N≤N_define counted in step (4.3.2) judgment step (4.2) is no：

If：Thick matching characteristic points N≤N_define, performs step (5.2),

If：Thick matching characteristic points N>N_define performs step (4.3.3),

Step (4.3.3), using MLESAC (Maximum Likelihood Estimation by Sample and Consensus, random sampling maximal possibility estimation) smart registration is carried out, to the matching that rough registration determines to the consistent of coordinate relationship Property is analyzed, and removes the characteristic matching pair of matching error, remaining correct matching characteristic point is exported,

Step (4.3.4) judges that yaw precision DIST≤10m of setting is no：

If：＞ 10m then perform step (5.2),

If：≤ 10m then performs step (5.3),

Step (5), according to the position offset of pixel coordinate calculating aircraft of the landmark region in image is acquired：

Step (5.1) aircraft acquires following parameter,

The current latitude and longitude coordinates of unmanned plane (B, L), present level H, unmanned plane current pose angle：Yaw angle α, pitch angle β, roll Angle γ,

Step (5.2), step is as follows：

Step (5.2.1) selects matching pair minimum in score, performs step (5.2.2),

Step (5.2.2), the successively position offset of calculating aircraft according to the following steps：

Step (5.2.2.1) establishes northeast day coordinate system (X by origin of the optical center O of video camera_wY_wZ_w),

Landmark point m is calculated as follows using the position coordinates that aircraft inertia navigation system provides as origin in step (5.2.2.2) Coordinate (x under the northeast day coordinate system of foundation_w,y_w,z_w)：

Wherein (u_p,v_p) it is pixel coordinates of the landmark point m in image is acquired,

z_mFor z-axis coordinates of the landmark point m in camera coordinate system, z_w=H is it is known that be obtained z_m, so as to find out x_w,y_w,

R_wIt is the posture spin matrix of camera coordinate system and video camera northeast day coordinate,It is R_wTransposed matrix,

Sx and Sy is respectively the pixel number acquired in image forming plane coordinate system X-axis and Y direction under unit physical size, For the parameter-embedded of video camera,

F is focal length of camera, is the parameter-embedded of video camera,

Step (5.3), step is as follows：

Step (5.3.1) is set：

Coordinate of the actual position of aircraft under the heaven-made coordinate system in northeast established using aircraft navigation outgoing position as origin (x_c,y_c,z_c), pixel coordinates of the landmark point m in imaging sensor is acquired in landmark region is (u_p,v_p), while landmark point m Coordinate under the northeast day coordinate system that aircraft navigation outgoing position is origin foundation is (x_m,y_m,z_m),

Step (5.3.1.1) imaging equation is：

u_p=[(cos α cos γ+sin α sin β sin γ) (x_m-x_c)-cosβsinγ(z_m-z_c)

-(cosγsinα-cosαsinβsinγ)(y_m-y_c)]fS_x

/[(cosαsinγ-cosγsinαsinβ)(x_m-x_c)

-(sinαsinγ+cosαcosγsinβ)(y_m-y_c)+cosβcosγ(z_m-z_c)]v_p=[sin β (z_m-z_c)+cosα cosβ(y_m-y_c)+cosβsinα(x_m-x_c)]fS_y

/[(cosαsinγ-cosγsinαsinβ)(x_m-x_c)

-(sinαsinγ+cosαcosγsinγ)(y_m-y_c)+cosβcosγ(z_m-z_c)]

Step (5.3.1.2), above-mentioned equation are about (x_c,y_c,z_c) with the Linear Equations of (α, beta, gamma), using most Small two, which multiply criterion, solves (x_c,y_c,z_c), obtain unmanned plane current position coordinates (x_c,y_c,z_c) namely aircraft actual position phase Offset for aircraft inertial navigation outgoing position coordinate,

Step (5.4), step is as follows：

Majorized function J is calculated as follows in step (5.4.1),

Wherein

M is the characteristic point logarithm that exports after essence matching, i=1,2 ... m,

The correspondence picture that the characteristic point of smart successful match is calculated according to the position and attitude angle of aircraft in acquisition image Plain coordinate, is calculated by imaging equation, is about (x_c,y_c,z_c) with the function of (α, beta, gamma),

It is pixel coordinate of the characteristic point of smart successful match in acquisition image in image is acquired,

Step (5.4.2) uses following Optimization Methods：

Solution is iterated using the fminseach function pairs J of matlab, iterations are set as 10000, the allowable error of iteration 1e-4 is set as, (x is obtained_c,y_c,z_c)。