Aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry, computer-readable storage medium and computer device

Abstract

The present application relates to aerial photograph screening method based on multi-lens camera tilt photogrammetry, which comprises the following steps: calculating the projection range of aerial photograph, selecting the aerial photograph with the projection range less than the set threshold as the quasi-redundant aerial photograph, selecting the aerial photograph with the projection range not less than the set threshold as the modeling aerial photograph, extracting the building features of the quasi-redundant aerial photograph, screening the quasi-redundant aerial photograph without building features as the redundant aerial photograph for elimination, screening the quasi-redundant aerial photograph with building features as the modeling aerial photograph, eliminating the redundant aerial photograph, reducing the number of aerial photographs participating in modeling, and improving the modeling efficiency, calculating the projection range of aerial photograph on the ground surface by using the interior and exterior orientation function of aerial photograph, combining with building feature recognition to retain the effective aerial photograph at the edge of the survey area to ensure the modeling quality, and adding building feature recognition to reduce the distortion of three-dimensional model, drawing and other phenomena.

Description

Technical Field

[0001] The present invention relates to the field of surveying and mapping technology, and in particular to an aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry, a computer-readable storage medium, and a computer device. Background Art

[0002] To significantly reduce aerial survey time and labor costs, the surveying and mapping industry has recently embraced the use of multi-eyepiece cameras (primarily pentaculars) for oblique photogrammetry. This technology has been widely used in areas such as urban 3D modeling, integrated real estate surveying, and power transmission line measurement. Compared to traditional vertical photogrammetry methods, pentacular oblique photogrammetry can rapidly capture information such as building form and structure, as well as high-precision facade textures, from multiple angles. This allows for accurate and clear 3D reconstruction of real-world models.

[0003] A major feature of multi-eye or penta-eye cameras is that they simultaneously collect aerial footage from multiple lens directions at the same coordinate position. While this reduces data collection time, it also generates a large amount of redundant data, which affects the efficiency of data processing and modeling. To improve data processing and modeling efficiency, redundant aerial footage needs to be eliminated.

[0004] Currently, there's no standardized method for removing redundant aerial photos. Some literature suggests using collinearity equations and the Digital Elevation Model (DEM) of the survey area to inversely calculate the coverage of each photo, thereby eliminating redundant photos. However, in practice, detailed ground DEM information is often unavailable. Other literature suggests calculating the orientation of aerial photos for screening, but this method only determines relative positional relationships and its applicability to more complex survey areas remains to be verified.

[0005] Therefore, the above methods all affect the efficiency and quality of modeling. Summary of the Invention

[0006] In order to improve the efficiency and quality of modeling, the present invention provides an aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry, a computer-readable storage medium and a computer device.

[0007] In a first aspect, the present invention provides an aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry, which adopts the following technical solutions:

[0008] The aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry includes the following steps:

[0009] Calculate the projection range of the aerial film, select the aerial film whose projection range is less than the set threshold as the quasi-redundant aerial film, and the aerial film whose projection range is not less than the set threshold as the modeling aerial film.

[0010] The building features of the quasi-redundant aerial photographs are extracted, the quasi-redundant aerial photographs without building features are screened as redundant aerial photographs and removed, and the quasi-redundant aerial photographs with building features are screened as modeling aerial photographs.

[0011] By the above technical solution, the redundant aerial photographs are removed, the number of aerial photographs participating in modeling is reduced, and the efficiency of modeling is improved. The projection range of the aerial photographs on the ground is calculated by using the interior and exterior orientation functions of the aerial photographs, and the effective aerial photographs at the edge of the survey area are reserved by combining the building feature recognition to ensure the modeling quality. After the building feature recognition is added, the distortion and pull of the three-dimensional model and other phenomena can be reduced.

[0012] The building feature extraction solves the misjudgment results caused by a single index (such as the overlap rate) in the traditional method. For example, at the edge of the building area to be reconstructed, although the aerial photograph has captured the building texture information, if the coverage range is less than the judgment threshold, the photograph will be removed, resulting in the edge of the modeling appearing the phenomenon of pull and the like. Although the photographs with the overlap rate exceeding 0 or more can be simply added to the screening result range, in fact, the probability of the photographs at the edge of the aerial area existing the building is small, which weakens the significance of the removal.

[0013] In addition, the present solution only needs to obtain the original interior parameter information of the camera, the Pos data when shooting, and the aerial area range information, without the need of obtaining the actual information of the survey area, and is convenient to operate.

[0014] Preferably, the calculation of the projection range of the aerial photograph comprises the following steps:

[0015] According to the exterior orientation elements of the aerial photograph, the spatial position of the image plane is restored,

[0016] According to the camera interior orientation elements, the imaging range of the aerial photograph is solved,

[0017] According to the aerial flight height, the intersection points of each vertex of the imaging range and the ground horizontal plane are calculated, the intersection points are sequentially connected to obtain the projection range of the aerial photograph.

[0018] The above technical solution does not need to obtain any supplementary data, the main parameters relied on can be obtained synchronously during the survey area planning and flight operation, the parameters required for the conversion of the projection range of the aerial photograph are less, the parameter acquisition method is simple, the operation processing speed is fast, and a high accuracy can be ensured, thereby improving the efficiency and quality of modeling.

[0019] Preferably, the method for selecting the aerial photograph with the projection range less than the set threshold as the quasi-redundant aerial photograph and the aerial photograph with the projection range not less than the set threshold as the modeling aerial photograph comprises the following steps:

[0020] The real-world coordinates of the aerial photograph imaging range and the vertex coordinates of the aerial photograph after projection are obtained, and the overlapping area between the projection polygon and the survey area polygon is calculated. The aerial photographs whose overlapping area accounts for a proportion of the projection polygon area that is less than a set threshold are selected as quasi-redundant aerial photographs, and the aerial photographs whose overlapping area accounts for a proportion of the projection polygon area that is not less than a set threshold are selected as modeling aerial photographs.

[0021] The above technical solution is derived from the mathematical model of projective camera imaging. The proportion of the overlapping part of the survey area to the entire projected polygon can be obtained through simple geometric calculations, thereby providing a reference standard for eliminating redundant aerial photos. Eliminating redundant aerial photos can improve modeling efficiency.

[0022] Preferably, the real-world coordinates of the aerial photograph imaging range are calculated by a projective camera model, and the projective camera model is as follows:

[0023] First, assuming the camera has not rotated, place the camera center at the origin of the world coordinate system. At this point, the camera's principal axis vector coincides with the negative Z-axis of the world coordinate system.

[0024] Secondly, a virtual imaging plane is constructed, extending one focal length from the camera center along the principal axis vector. Without considering distortion and principal point offset, the camera's imaging range at this plane is equal to the sensor size. The shape of the rectangle is rectangular, with the long side perpendicular to the north direction and the center of the rectangle located at the principal point of the imaging plane.

[0025] Finally, combined with the exterior orientation elements of the camera, the vertices of the imaging range are rotated and translated. Assume that the coordinates of the vertices of the imaging range before and after the change are (x i ,y i ,z i ) and (x' i ,y' i ,z' i ), the real-world coordinates of the aerial photograph imaging range can be obtained by the following formula:

[0026]

[0027] in,

[0028]

[0029]

[0030]

[0031] Among them, the coordinates (X i ,Y i ,Z i ) is the photography center P i The coordinates in the world coordinate system, is the Euler rotation angle of the photographic beam relative to the world coordinate system, ω is the pitch angle, is the roll angle, κ is the yaw angle, R κ R ω are the rotation matrices constructed based on the rotation angles.

[0032] Through the above technical solution, the real-world coordinates of the aerial photograph imaging range are calculated through the projective camera model. It has strong universality, is applicable to most pentacular lenses, and is easy to convert, which can improve modeling efficiency.

[0033] Preferably, the calculation method of the vertex coordinates after the aerial photo projection is as follows:

[0034] Assume that the coordinates of the vertices after projection are (l i ,m i ,n i ),but

[0035]

[0036] in, represents the Hadamard product, and

[0037]

[0038] Among them, t represents the intermediate parameter and h represents the set flight altitude.

[0039] Through the above technical solution, after calling the camera parameters and Pos file to build the camera model, the projection polygon of the aerial photo projection area on the ground can be obtained. The overlapping part formed by the projection polygon and the survey area polygon can be obtained through geometric calculation, and the proportion of the overlapping part to the entire projection polygon (i.e., the overlap rate) can be obtained, thereby providing a standard for eliminating redundant photos.

[0040] Preferably, the rotation angle is calculated as follows:

[0041] For point p i (X i ,Y i ,Z i ) and its next photo location point p i+1 (X i+1 ,Y i+1 ,Z i+1 ), align the aircraft's rotation axis with the vector The direction pointed to, and set p i The roll angle of the aircraft If it is 0°, the point p can be obtained according to the following formula i Yaw angle κ ati and pitch angle ω i :

[0042] κ i =atan2(X i -X i+1 ,Y i+1 -Y i )

[0043]

[0044] The above technical solution allows for the rotation angle of the imaging beam relative to the world coordinate system, as some camera models provide this information, while others do not. When the POS file does not contain this information, the POS coordinates can be used to roughly estimate the rotation angle. For drones, the orientation of the aircraft's roll axis can be determined using the POS coordinates before and after the point, allowing the above formula to be used to calculate the rotation angle. Without the inertial navigation system's pose information, the POS data alone can be used to estimate the rotation angle, resulting in highly efficient modeling.

[0045] Preferably, the method for correcting the rotation angle information of lenses in different directions is as follows:

[0046] The rotation angle of the orthographic lens remains unchanged, the pitch angle of the forward-tilted lens increases by degree A, the pitch angle of the backward-tilted lens decreases by degree A, the roll angle of the left-tilted lens decreases by degree A, and the roll angle of the right-tilted lens increases by degree A.

[0047] Through the above technical solution, after calculating the rotation angle information of the orthographic lens, the rotation angle information of other oblique lenses can be obtained based on the rotation angle information of the orthographic lens, which facilitates the calculation of the real-world coordinates of the aerial photograph imaging range of the oblique lens and the vertex coordinates after the aerial photograph is projected, and then performs aerial photograph screening.

[0048] Preferably, extracting architectural features from quasi-redundant aerial photographs includes:

[0049] (1) Vegetation coverage,

[0050] (2) Image entropy,

[0051] (3) The number, average length and maximum length of identifiable line segments.

[0052] Through the above technical solution, image features are extracted from the above three dimensions for building recognition: (1) vegetation coverage. For aerial photographs with buildings, the surrounding vegetation coverage is usually sparse; (2) image entropy. This value is usually used to characterize the degree of randomness in the image. Since buildings usually have relatively simple color changes, their information entropy is relatively low in the image; (3) the number, average length and maximum length of identifiable line segments. The boundaries of buildings are usually clear in the image and can be detected and extracted by the algorithm. The algorithm is ED-lines, which can obtain linear features in the image without a lot of parameter adjustments, and the recognition effect is better than other traditional algorithms such as Hough and LSD.

[0053] The advantage of selecting image features of the above three dimensions is that the selected features are understandable, have low dimensions, and have strong generalization learning capabilities.

[0054] Preferably, based on the extracted architectural features of the quasi-redundant aerial photographs, an image classifier is constructed using a support vector machine, and the image classifier is used to screen out aerial photographs that may have building textures as modeled aerial photographs.

[0055] Through the above technical solution, the principle of constructing an image classifier using a support vector machine is to find a hyperplane (a surface with more than three dimensions) that maximizes the interval between two types of features (any two types of features) to achieve the classification effect. The purpose of the largest interval is to make the two types of features as distinguishable as possible.

[0056] A classifier is constructed based on existing feature learning. When another set of features is input, the classification result can be determined based on which class the set of features is closer to.

[0057] Since the classification task here is to classify architectural images and non-architectural images, that is, a binary classification task, the use of support vector machine to construct an image classifier can maintain the maximum soft margin between the two types of images, thereby making the distinguishability between the two types of images more obvious.

[0058] Preferably, the image classifier includes an orthographic lens image classifier and an oblique lens image classifier, and the orthographic lens image classifier and the oblique lens image classifier respectively use two different sets of training data sets, each set of data sets contains a preset number of aerial photographs containing buildings and aerial photographs without buildings.

[0059] By the above technical scheme, since the interior and exterior orientation parameters and the collected texture features of the orthographic lens and the oblique lens aerial photographs are different, two sets of training data sets are constructed for the orthographic lens and the oblique lens, each of which contains a preset number (such as 500) of aerial photographs containing buildings and aerial photographs not containing buildings, different classifiers are trained to adapt to the needs of different lenses, and after training, aerial photographs that do not meet the overlap rate can be identified to retain aerial photographs that may contain building textures, so that the aerial photographs containing building textures can be more accurately screened out from the redundant aerial photographs as modeling aerial photographs, and the modeling quality is further improved.

[0060] In a second aspect, the present application provides a computer readable storage medium for storing a computer program, which, when invoked, performs the aerial photograph screening method based on multi-lens camera oblique photogrammetry according to any one of the above technical solutions.

[0061] In a second aspect, the present application provides a computer device comprising a memory and a processor, the memory being used to store a computer program, the computer program being invoked by the processor to perform the aerial photograph screening method based on multi-lens camera oblique photogrammetry according to any one of the above technical solutions.

[0062] In summary, the present application has at least one of the following beneficial technical effects:

[0063] (1) The present application eliminates redundant aerial photographs, which can reduce the number of aerial photographs participating in modeling, thereby improving the efficiency of modeling. The projection range of the aerial photograph on the ground surface is calculated by using the interior and exterior orientation function of the aerial photograph, and the effective aerial photograph at the edge of the survey area is retained by combining building feature recognition to ensure the modeling quality. After adding building feature recognition, the distortion and pull flower of the three-dimensional model can be reduced.

[0064] (2) The conversion of the projection range of the aerial photograph requires fewer parameters, and the parameters are easy to obtain, fast, and can ensure high accuracy, thereby improving the efficiency and quality of modeling.

[0065] (3) The proportion of the overlapping part to the entire projection polygon can be obtained by simple geometric calculation, thereby providing a standard for eliminating redundant photographs, and the modeling efficiency is high.

[0066] (4) The image features are extracted from three dimensions of vegetation coverage, image entropy, and the number of recognizable line segments, average length, and maximum length for building recognition. The beneficial effect is that the selected features are understandable, have low dimension, and have strong generalization learning ability.

[0067] (5) Different classifiers are trained to meet the needs of different lenses. After the training is completed, aerial photos that do not meet the overlap rate can be identified to retain aerial photos that may have building textures. Aerial photos with building textures can be more accurately screened out from redundant aerial photos as modeling aerial photos, further improving the modeling quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 This is a schematic diagram of a low-overlap aerial photograph with building textures.

[0069] Figure 2 It is a schematic diagram of the rotation axis of a drone in the aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry of the present invention.

[0070] Figure 3 It is a schematic diagram of the projection range of aerial photographs of the aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry of the present invention.

[0071] Figure 4 It is a schematic diagram of the overall process of the aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following Figures 1-4 The present invention is further described in detail with reference to the accompanying drawings and embodiments.

[0073] The aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry provided in this embodiment adopts the following technical solution: The aerial photograph screening method based on multi-eyepiece camera tilt photogrammetry includes the following steps:

[0074] Calculate the projection range of the aerial film, select the aerial film whose projection range is less than the set threshold as the quasi-redundant aerial film, and the aerial film whose projection range is not less than the set threshold as the modeling aerial film.

[0075] The architectural features of quasi-redundant aerial photographs are extracted, and the quasi-redundant aerial photographs without architectural features are screened as redundant aerial photographs and eliminated, while the quasi-redundant aerial photographs with architectural features are screened as modeling aerial photographs.

[0076] This embodiment eliminates redundant aerial photographs, which can reduce the number of aerial photographs involved in modeling and thus improve modeling efficiency. By using the internal and external orientation functions of the aerial photographs to calculate the projection range of the aerial photographs on the surface, and combining building feature recognition, valid aerial photographs at the edge of the survey area are retained to ensure modeling quality. After adding building feature recognition, the distortion and streaking of the three-dimensional model can be reduced, thereby improving modeling quality.

[0077] Extracting architectural features solves the problem of misjudgment caused by traditional methods that rely on a single indicator (such as overlap rate). For example, at the edge of the building area to be reconstructed, although the aerial photo has captured architectural texture information, if its coverage is less than the judgment threshold, this part of the photo will be eliminated, resulting in the appearance of streaks on the model edge (such as Figure 1 Although we can simply add all photos with an overlap rate exceeding 0 to the filter results, the probability of buildings existing in photos at the edge of the flight area is actually low, which weakens the significance of excluding them.

[0078] In addition, this method only needs to obtain the original internal parameter information of the camera, the Pos data during shooting, and the navigation area range information, without the need to obtain additional actual information of the measurement area, and is easy to operate.

[0079] The specific method is as follows:

[0080] Data preprocessing

[0081] After obtaining the POS file containing the aerial photograph's attitude information and the KML (Keyhole Markup Language) file representing the modeling range, their time and space references need to be unified for easy processing. The POS file can provide the exterior orientation elements of the aerial photograph, usually including the coordinates (X, Y, Z) of the photographic center in the world coordinate system. Some camera models also provide the rotation angle of the photographic beam relative to the world coordinate system. The boundary coordinates in the KML file also exist in the world coordinate system, so the coordinates can be unified to the same reference. In this embodiment, the spatial coordinates are unified to the local ENU (East-North-Up) coordinate system. The origin of the coordinate system is located at the geometric center of all aerial photo POS data, and the rotation angle information remains unchanged.

[0082] Only when the coordinates are unified can the POS positioning position of each photo be superimposed with the modeling range to determine whether the POS position of the aerial photo is within the modeling range.

[0083] When the pos file does not contain the rotation angle information, the rotation angle can be roughly restored by the pos coordinates. Figure 2 As shown, for a drone, the direction of the aircraft's rotation axis (RollAxis) can be restored by the information of the pos coordinates before and after the point coordinates, and then the rotation angle information can be calculated. is calculated as follows:

[0084] For point p i (X i ,Y i ,Z i ) and its next photo location point p i+1 (X i+1 ,Yi+1 ,Z i+1 ), align the aircraft's rotation axis with the vector The direction pointed to, and set p i The roll angle of the aircraft If it is 0°, the point p can be obtained according to the following formula i Yaw angle κ at i and pitch angle ω i :

[0085] κ i =atan2(X i -X i+1 ,Y i+1 -Y i ),

[0086]

[0087] For lenses in different directions, the rotation angle information needs to be further corrected.

[0088] The correction method for the rotation angle information of lenses in different directions is as follows:

[0089] The rotation angle of the orthographic lens remains unchanged, the pitch angle of the forward-tilted lens increases by degree A, the pitch angle of the backward-tilted lens decreases by degree A, the roll angle of the left-tilted lens decreases by degree A, and the roll angle of the right-tilted lens increases by degree A.

[0090] In this embodiment, the pitch angle of the forward-tilted lens increases by 50°, the pitch angle of the backward-tilted lens decreases by 50°, the roll angle of the left-tilted lens decreases by 50°, and the roll angle of the right-tilted lens increases by 50°.

[0091] The current common practice is to use an angle of 45°. In this embodiment, the left and right tilt angles of the lens are increased from 45° to 50°, which can further increase the lateral viewing angle, reduce the "eaves effect" of the pattern, and achieve a better lateral modeling effect.

[0092] Of course, the specific adjustment angle can be set according to actual needs, for example, A is in the range of 40° to 55°.

[0093] The calculation method of the projection range of the aerial photo is as follows:

[0094] The core criterion for selecting aerial photographs is the overlap ratio between the projection range of the aerial photograph and the modeling range. Figure 3 As shown, to obtain the projection range of the aerial photograph, this embodiment first restores the spatial position of its image plane based on the exterior orientation elements of the aerial photograph, and solves the imaging range of the aerial photograph based on the interior orientation elements of the camera. Then, based on the flight altitude, the intersection points of each vertex in the imaging range and the horizontal plane of the ground are calculated, and the intersection points are connected in sequence to obtain the projection range of the aerial photograph.

[0095] The real-world coordinates of the aerial photograph imaging range are calculated using a projective camera model, which is as follows:

[0096] For point p i , the imaging range of the aerial photograph can be calculated by the projective camera model: first, assuming that the camera has not rotated, the camera center is coincident with the origin of the world coordinate system. At this time, the principal axis vector of the camera coincides with the negative direction of the Z axis of the world coordinate system; secondly, a virtual imaging plane is constructed by extending a focal length from the center of the camera along the principal axis vector. Without considering the distortion and principal point offset, the imaging range of the camera at this plane is equal to the size of its sensor. It is rectangular in shape, with the long side perpendicular to the north direction and the center of the rectangle located at the principal point of the imaging plane; finally, combined with the exterior orientation elements of the camera, the vertices of the imaging range are rotated and translated. Assume that the coordinates of the vertices of the imaging range before and after the change are (x i ,y i ,z i ) and (x' i ,y' i ,z' i ), it can be obtained as follows:

[0097]

[0098] in,

[0099]

[0100]

[0101]

[0102] Among them, the coordinates (X i ,Y i ,Z i ) is the photography center P i The coordinates in the world coordinate system, is the Euler rotation angle of the photographic beam relative to the world coordinate system, ω is the pitch angle, is the roll angle, κ is the yaw angle, R κ R ω are the rotation matrices constructed based on the rotation angles.

[0103] After obtaining the real-world coordinates of the imaging range, it is necessary to further calculate the projection of the imaging range on the ground according to the flight altitude. Assume that the vertex coordinates after projection are (l i ,m i ,n i ), the calculation method of the vertex coordinates after the aerial photo projection is as follows:

[0104]

[0105] wherein, denotes a Hadamard product, and

[0106]

[0107] wherein, t represents an intermediate parameter, and h represents a set flight height.

[0108] After obtaining the projection coordinates, the overlapping area of the projection polygon and the survey area range polygon can be further calculated. When the proportion of the overlapping area to the area of the projection range polygon is less than a set threshold, the corresponding aerial photograph is considered to be a redundant aerial photograph and should be removed.

[0109]

[0110] There is currently no conventional algorithm to calculate the projection range of an aerial photograph. Some scholars believe that a DEM model can be added for interpolation calculation. Compared with this, the present embodiment has the advantages of requiring fewer prior parameters for conversion, obtaining parameters simply and quickly, and ensuring high accuracy.

[0111] The method for selecting an aerial photograph whose projection range is less than a set threshold as a quasi-redundant aerial photograph and an aerial photograph whose projection range is not less than a set threshold as a modeling aerial photograph includes the following steps:

[0112] The real-world coordinates of the aerial photograph imaging range and the vertex coordinates after projection of the aerial photograph are obtained. The overlapping area of the projection polygon and the survey area range polygon is calculated. An aerial photograph whose proportion of the overlapping area to the area of the projection polygon is less than a set threshold is selected as a quasi-redundant aerial photograph, and an aerial photograph whose proportion of the overlapping area to the area of the projection polygon is not less than a set threshold is selected as a modeling aerial photograph.

[0113] In the screening of aerial photographs, although the overlap of some aerial photographs is less than a set value, they may contain some texture information of buildings that need to be modeled. If only the overlap index is used for screening, the fine texture information of these aerial photographs cannot be effectively utilized, which may cause the model to have phenomena such as distortion and distortion after being built. In order to effectively solve this problem, the present embodiment introduces an aerial photograph building feature recognition mechanism to retain aerial photographs with lower overlap rates but higher probabilities of retaining building texture information.

[0114] The building features of the quasi-redundant aerial photograph include:

[0115] (1) vegetation coverage,

[0116] (2) image entropy,

[0117] (3) the number of identifiable line segments, the average length, and the maximum length.

[0118] Based on the extracted quasi-redundant aerial images of building features, an image classifier is built by using support vector machine, and the aerial images possibly containing building textures are screened out as modeling aerial images by using the image classifier.

[0119] The image classifier includes an orthographic lens image classifier and an oblique lens image classifier, and the orthographic lens image classifier and the oblique lens image classifier respectively adopt two different sets of training data sets, each of which contains a preset number of aerial images containing buildings and aerial images not containing buildings.

[0120] The method for extracting the building features of the quasi-redundant aerial images, screening out the quasi-redundant aerial images without building features as redundant aerial images, and screening out the quasi-redundant aerial images with building features as modeling aerial images is as follows:

[0121] According to the characteristics of building textures in images, the embodiment extracts image features from three dimensions for building recognition: (1) vegetation coverage, aerial images containing buildings usually have sparse vegetation coverage, and this value can be obtained by calculating the color index of the RGB image; (2) image entropy, which is usually used to represent the randomness of the image, and since the color change of the building is usually single, the information entropy of the building in the image is low; (3) the number of recognizable line segments, average length and maximum length, the boundary of the building in the image is usually clear, which can be detected and extracted by an algorithm.

[0122] After extracting the features, the embodiment selects a support vector machine (SVM) to build an image classifier, and the principle of the algorithm is to find a hyperplane that maximizes the margin between the two classes of features to achieve classification effect. Since there are certain differences in the interior and exterior orientation parameters and the collected texture features between the orthographic lens and the oblique lens aerial images, for this, the embodiment builds two sets of training data sets for the orthographic lens and the oblique lens, each of which contains 500 aerial images containing buildings and 500 aerial images not containing buildings, and trains different classifiers to adapt to the needs of different lenses. After the training is completed, the aerial images that do not meet the overlap rate can be identified (by extracting the same features and putting them into the trained image classifier for classification, and the output result is the identification result), so as to retain the aerial images that may contain building textures. The overall technical process is as shown in Figure 4 .

[0123] The existing commercial software generally does not remove redundant aerial images, and all aerial images are put into air triangulation and modeling, so that the operation time is very long and the efficiency is low. Although there are a few methods for aerial image screening, these methods for aerial image screening cannot effectively improve the modeling efficiency and modeling quality.

[0124] Compared with the prior art aerial photo screening method, the method of this embodiment has the following advantages:

[0125] The existing method for quickly screening effective drone aerial photographs only targets orthophotos and eliminates photos from a single orthophoto lens. Redundant aerial photographs are eliminated by calculating the center point photo and heading. Aerial photographs from the other four tilted lenses cannot be processed, and the improvement in modeling efficiency is very limited. The aerial photograph screening method of this embodiment is applicable to aerial photographs from five lenses, which can eliminate more redundant aerial photographs and better improve modeling efficiency.

[0126] A prior art method for screening aerial photographs requires first performing aerial triangulation to determine exterior orientation elements, but this embodiment does not require this. Redundancy is determined by calculating the intersection of extended vector lines with the survey area. This embodiment determines redundancy by calculating the intersection of the surface formed by the positions of each aerial photograph's vertices with the survey area. This method also incorporates altitude as a calculation parameter, making it more accurate and improving modeling quality.

[0127] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. Aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry, characterized by: The steps include: Calculate the projection range of the aerial film, select the aerial film whose projection range is less than the set threshold as the quasi-redundant aerial film, and the aerial film whose projection range is not less than the set threshold as the modeling aerial film. Extract architectural features from quasi-redundant aerial photos, filter out quasi-redundant aerial photos without architectural features as redundant aerial photos, and filter out quasi-redundant aerial photos with architectural features as modeling aerial photos. The calculation of the projection range of the aerial photograph comprises the following steps: According to the external orientation elements of the aerial photograph, the spatial position of the image plane is restored. Solve the imaging range of the aerial photo based on the camera's internal orientation elements. According to the flight altitude, calculate the intersection points of each vertex in the imaging range and the horizontal surface, connect the intersection points in order, and get the projection range of the aerial photograph. The method of selecting an aerial film whose projection range is smaller than a set threshold as a quasi-redundant aerial film and an aerial film whose projection range is not smaller than the set threshold as a modeling aerial film comprises the following steps: Obtain the real-world coordinates of the aerial photograph imaging range and the vertex coordinates of the aerial photograph after projection, calculate the overlapping area of ​​the projection polygon and the polygon of the survey area, select the aerial photograph whose overlapping area accounts for the ratio of the projection polygon area to the projected polygon area less than the set threshold as the quasi-redundant aerial photograph, and the aerial photograph whose overlapping area accounts for the ratio of the projection polygon area to the projected polygon area not less than the set threshold as the modeling aerial photograph, The extraction of architectural features of quasi-redundant aerial photos includes: (1) Vegetation coverage, (2) Image entropy, (3) The number, average length and maximum length of identifiable line segments.

2. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 1, characterized in that: The real-world coordinates of the aerial photograph imaging range are calculated using a projective camera model, which is as follows: First, assuming the camera has not rotated, place the camera center at the origin of the world coordinate system. At this point, the camera's principal axis vector coincides with the negative Z-axis of the world coordinate system. Secondly, a virtual imaging plane is constructed, extending one focal length from the camera center along the principal axis vector. Without considering distortion and principal point offset, the camera's imaging range at this plane is equal to the sensor size. The shape of the rectangle is rectangular, with the long side perpendicular to the north direction and the center of the rectangle located at the principal point of the imaging plane. Finally, combined with the exterior orientation elements of the camera, the vertices of the imaging range are rotated and translated. Assume that the coordinates of the vertices of the imaging range before and after the change are (x i ,y i ,z i ) and (x' i ,y' i ,z' i ), the real-world coordinates of the aerial photograph imaging range can be obtained by the following formula: in, Among them, the coordinates (X i ,Y i ,Z i ) is the photography center P i The coordinates in the world coordinate system, is the Euler rotation angle of the photographic beam relative to the world coordinate system, ω is the pitch angle, is the roll angle, κ is the yaw angle, R κ R ω are the rotation matrices constructed based on the rotation angles.

3. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 2, characterized in that: The calculation method of the vertex coordinates after the aerial photo projection is as follows: Assume that the coordinates of the vertices after projection are (l i ,m i ,n i ),but where o represents the Hadamard product, and Among them, t represents the intermediate parameter and h represents the set flight altitude.

4. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 2, characterized in that: The rotation angle is calculated as follows: For point p i (X i ,Y i ,Z i ) and its next photo location point p i+1 (X i+1 ,Y i+1 ,Z i+1 ), align the aircraft's rotation axis with the vector The direction pointed to, and set p i The roll angle of the aircraft If it is 0°, the point p can be obtained according to the following formula i Yaw angle κ at i and pitch angle ω i : κ i =atan2(X i -X i+1 ,Y i+1 -Y i ) 5. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 4, characterized in that: The correction method for the rotation angle information of lenses in different directions is as follows: The rotation angle of the orthographic lens remains unchanged, the pitch angle of the forward-tilted lens increases by degree A, the pitch angle of the backward-tilted lens decreases by degree A, the roll angle of the left-tilted lens decreases by degree A, and the roll angle of the right-tilted lens increases by degree A.

6. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 1, characterized in that: Based on the architectural features extracted from quasi-redundant aerial photographs, an image classifier is constructed using support vector machine, and the image classifier is used to screen out aerial photographs that may have building textures as modeling aerial photographs.

7. The aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to claim 6, characterized in that: The image classifier includes an orthographic image classifier and an oblique image classifier. The orthographic image classifier and the oblique image classifier use two different sets of training data sets, each of which contains a preset number of aerial photographs with buildings and aerial photographs without buildings.

8. A computer-readable storage medium for storing a computer program, characterized in that: When the computer program is called, the aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to any one of claims 1 to 7 is executed.

9. A computer device comprising a memory and a processor, wherein the memory is used to store a computer program, wherein: When the computer program is called by the processor, the aerial photograph screening method based on multi-eyepiece camera oblique photogrammetry according to any one of claims 1 to 7 is executed.

Images