A three-dimensional positioning method and system for bridge damage areas based on drones and neural networks
By combining drones and neural networks, a three-dimensional model of a bridge is generated and the defect area is identified, which solves the problems of low efficiency and difficulty in accurate positioning of bridge defects in existing technologies, and realizes efficient and accurate three-dimensional positioning and visualization of bridge defect areas.
Patent Information
- Application Number
- CN202311670410.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-07
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-12-07
AI Technical Summary
Current bridge surface defect detection relies on manual inspection, which is inefficient, costly, and yields uncertain results, making it impossible to accurately locate the defective areas on a three-dimensional model.
The method uses drones to collect low-resolution images of bridges to generate 3D models, plans inspection routes, identifies diseased areas through neural networks, determines boundaries using polygon fitting algorithms, and combines incremental motion recovery structure algorithms to achieve precise positioning of diseased areas on the 3D model.
It has improved the efficiency and informatization of bridge defect detection, realized intelligent identification and precise three-dimensional positioning of bridge defect areas, ensured the collection of defect area information without blind spots, and improved the accuracy and visualization effect of detection.
Smart Images

Figure CN117710810B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to bridge defect area detection, and in particular, to a three-dimensional positioning method and system for bridge defect areas based on unmanned aerial vehicles and neural networks. Background Art
[0002] As the service life of bridges increases, the impact of bridge surface defects (cracks, spalling, honeycomb surface, exposed steel bars, etc.) on the bearing capacity and structural stability of bridges will gradually increase. Bridge surface defect inspections should be carried out regularly to provide a basis for the formulation of bridge maintenance plans. At present, bridge surface inspections mainly rely on manual inspections. For example, the inspection process of bridge surface defects requires staff to measure and record the defects. This method is inefficient, time-consuming, and costly. In addition, manual inspections are greatly affected by factors such as the environment and the professional and technical literacy of the inspectors, and the inspection results are uncertain.
[0003] Although drone technology and some intelligent recognition algorithms have begun to be applied to bridge surface inspection, a complete technical system has not yet been formed, and it is impossible to accurately locate the surface defects of bridges on the three-dimensional model of bridges. This application proposes a three-dimensional positioning method and system for bridge defect areas based on drones and machine vision, which realizes intelligent recognition of bridge defect areas and accurate positioning on the entire three-dimensional model of the bridge, so that the distribution of defects on the entire bridge can be intuitively displayed through the three-dimensional model of the bridge. Summary of the invention
[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a three-dimensional positioning method for bridge defect areas based on drones and neural networks. Through this method, the efficiency and informatization of inspection work are improved, the risk of inspection work is reduced, and the intelligent identification of bridge surface defects and accurate three-dimensional positioning are achieved.
[0005] In order to achieve the above object, the present invention is implemented by the following technical solution: a three-dimensional positioning method of a bridge defect area based on an unmanned aerial vehicle and a neural network, comprising the following steps:
[0006] Step 1: Use the camera on the drone to take aerial photos of the bridge to be inspected, collect multiple low-resolution images of the bridge, and generate a first three-dimensional model of the bridge based on the multiple low-resolution images; plan the inspection path of the drone and multiple viewpoints on the inspection path based on the generated first three-dimensional model of the bridge;
[0007] Step 2: The UAV flies according to the inspection path and collects multiple high-resolution images of the bridge at multiple viewpoints; a second bridge three-dimensional model is generated from the multiple high-resolution images through a three-dimensional reconstruction algorithm;
[0008] Step 3: Input the high-resolution image into the trained neural network to identify the diseased area in the high-resolution image; fit the boundary of each diseased area by a polygon fitting algorithm, and determine the boundary coordinates P of each diseased area based on the pixel coordinate system. i (u,v), the boundary of each diseased area is composed of multiple boundary coordinates P i (u,v), the subscript i represents the sequence number of the coordinate;
[0009] Step 4: Multiple boundary coordinates P of the diseased area i (u, v) is mapped to the second three-dimensional model of the bridge to locate the boundary of the damaged area on the second three-dimensional model of the bridge.
[0010] Furthermore, the step 4 is specifically as follows:
[0011] Step 4.1: Calculate the rotation matrix R and translation matrix t between the multiple high-resolution images obtained in step 2 through the incremental motion recovery structure algorithm, place the second bridge 3D model, the rotation matrix R and the translation matrix t between the images in the same coordinate system, and complete the restoration of the entire inspection scene. At this time, the translation matrix t is the coordinate of the camera optical center; use the following formula (1) to convert the defect boundary coordinate P in the pixel coordinate system i (u,v) is converted to world coordinates P w (x w ,y w ,z w ), subscript w is the serial number of the coordinate;
[0012]
[0013] Among them, f, dx, dy, u0, and v0 are the focal length, physical pixel length, physical pixel width, abscissa of the imaging plane center, and ordinate of the imaging plane center of the drone’s camera lens; q is the object distance parameter, which indicates the distance from the optical center of the camera to the photographed point;
[0014] Step 4.2: By moving from the camera optical center coordinate t to the world coordinate P w The ray L(s) is obtained by calculating the starting point and direction vector of L(s). The ray L(s) is shown in the following formula (2):
[0015] L(s)=s*(P w -t)+t (2);
[0016] Where s is a real number, which indicates the set distance moved along the ray direction from the coordinate t of the camera optical center; calculate the intersection point P of the ray L(s) and the second bridge 3D model r (x r ,y r ,zr ), subscript r is the number of the coordinate; P r (x r ,y r ,z r ) are the coordinates of the boundary of the damaged area on the three-dimensional model of the second bridge, thereby realizing the positioning of the boundary of the damaged area on the three-dimensional model of the second bridge.
[0017] Furthermore, the intersection point P of the calculation ray L(s) and the second bridge three-dimensional model is r (x r ,y r ,z r ), specifically:
[0018] Extract all the three-dimensional vertex coordinates of the second bridge three-dimensional model to form a three-dimensional vertex coordinate set A, project the three-dimensional vertex coordinate set A and the coordinate t of the camera optical center onto a plane perpendicular to the ray L(s), and obtain the corresponding plane coordinate set A' and plane coordinate t' after projection. The coordinates in the plane coordinate set A' correspond to the coordinates in the three-dimensional vertex coordinate set A one-to-one, and find the point set D' in the plane coordinate set A' that is located in the neighborhood of the set radius of t'; the point set D' corresponds to the subset D in the three-dimensional vertex coordinate set A, and the three-dimensional vertex coordinates closest to the camera optical center in the subset D are obtained. The three-dimensional vertex coordinates are the intersection point P of the ray L(s) and the second bridge three-dimensional model r (x r ,y r ,z r ).
[0019] Furthermore, the neural network may be a semantic segmentation network.
[0020] Furthermore, after the diseased area is identified through the semantic segmentation network, discrete noise points can be removed from the diseased area.
[0021] Furthermore, the method also includes: Step 5: The defective area can be drawn on the three-dimensional model of the second bridge according to the positioning of the defective area on the three-dimensional model of the second bridge through BIM modeling software.
[0022] The present application also proposes a three-dimensional positioning system for bridge damage areas based on an unmanned aerial vehicle and a neural network. The system includes an unmanned aerial vehicle, a processor, and a memory. The system is characterized in that the memory stores computer-executable instructions, and when the processor executes the instructions, the steps of the method described in any one of the above embodiments are implemented.
[0023] Beneficial effects of the present invention: (1) The present invention first uses a drone to photograph the bridge to be inspected to generate a first three-dimensional model of the bridge, plans an inspection route and multiple viewpoints based on the first three-dimensional model of the bridge, and uses the drone to photograph high-resolution images of the bridge at multiple viewpoints to establish a second three-dimensional model of the bridge and identify the defective areas in the bridge. The second three-dimensional model of the bridge established in this way and the high-resolution images obtained can ensure that all information on the bridge surface is collected without blind spots, preventing the defective areas from being missed, and the second three-dimensional model of the bridge established has higher accuracy, which is more conducive to the precise positioning of the defective areas.
[0024] (2) The present invention uses a neural network to identify the defective area of the collected high-resolution image, and determines the boundary of each defective area through a polygon fitting algorithm. Through steps 4.1 and 4.2, a mapping relationship between the two-dimensional points of the image and the three-dimensional points of the model is established, thereby achieving accurate three-dimensional positioning of the defective area on the three-dimensional model of the second bridge. Compared with other three-dimensional positioning methods in the prior art, this three-dimensional positioning method has higher accuracy, allowing the staff to intuitively see the distribution of the defective area from the three-dimensional model of the bridge. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0026] Figure 1 It is a flow chart of a three-dimensional positioning method of bridge damage area based on UAV and neural network;
[0027] Figure 2 This is the recognition result diagram of the diseased area on the image;
[0028] Figure 3 The coordinate system involved in locating the diseased area;
[0029] Figure 4 Schematic diagram of the principles involved in locating the diseased area. DETAILED DESCRIPTION
[0030] The present application will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present application and are not intended to limit the scope of the present application. In addition, it should be understood that after reading the content taught in the present application, those skilled in the art may make various changes or modifications to the present application, and these equivalent forms also fall within the scope limited by the present application.
[0031] A three-dimensional positioning method for a bridge damage area based on an unmanned aerial vehicle and a neural network includes the following steps 1-5.
[0032] Step 1: Use the camera on the drone to take aerial photos of the bridge to be inspected, collect multiple low-resolution images of the bridge, and generate a first three-dimensional model of the bridge based on the multiple low-resolution images; plan the drone's inspection path and multiple viewpoints on the inspection path based on the generated first three-dimensional model of the bridge.
[0033] Step 2: The UAV flies according to the inspection path and collects multiple high-resolution images of the bridge at multiple viewpoints; and generates a second bridge three-dimensional model from the multiple high-resolution images through a three-dimensional reconstruction algorithm.
[0034] There are a variety of three-dimensional reconstruction algorithms in the prior art, and those skilled in the art can select one according to actual needs.
[0035] The specific implementation of the above step 1 can be found in Figure 1 In the bridge surface image collection part, the drone is used to take aerial photos of the bridge to be tested. After the aerial photos are completed, the first bridge 3D model is generated using DJI Zhitu. It is a rough bridge model (a 3D model with relatively low resolution and coverage), and the model type is selected in point cloud format; the first bridge 3D model is only used to plan the refined inspection path. The 3D photogrammetry (tic-tac-toe flight) planning method can be used in the above aerial photos, and the lateral overlap rate can be set to 70%, and the heading overlap rate can be set to 80%.
[0036] The first bridge three-dimensional model is generated through point cloud slicing and clustering, convex hull generation and offset, and inspection path generation to plan the UAV's refined inspection path.
[0037] 1. Point cloud slicing and clustering
[0038] The point cloud model is sliced at equal intervals along the z-axis direction of the point cloud. The spacing of each layer of point cloud is determined by the lateral overlap rate. Because each layer of point cloud subsequently generates different layers of routes, the spacing between two adjacent layers of routes can be calculated based on the lateral overlap rate, thereby determining the spacing when slicing the point cloud. Then the DBSCAN density clustering algorithm is used to cluster the sliced point cloud, and each layer of the sliced point cloud is given the same label for subsequent use.
[0039] 2. Convex hull generation and offset
[0040] The fast convex hull method is used to generate the convex hull of each layer of point cloud (in a two-dimensional plane, the convex hull is a convex polygon formed by connecting the outermost points). The convex hull is offset outward according to the safe shooting distance to ensure the safety of the drone flight. The viewpoint of each layer of the route is planned using the heading overlap rate.
[0041] 3. Inspection route generation
[0042] After generating the viewpoints, use the genetic algorithm or ant colony algorithm to connect the viewpoints and generate the drone inspection path. Set the drone action for each viewpoint to take a photo, and set the safe shooting distance to 2.5m.
[0043] This method of fine-grained inspection path planning is only one implementation path. There are many methods for planning fine-grained paths, and technicians in this field can also choose other implementation paths according to actual needs. The fine-grained inspection path is planned based on the first bridge three-dimensional model generated in step 2. The difference from step one is that the fine-grained inspection requires obtaining clear high-resolution images of the bridge, which is conducive to improving the recognition accuracy of the diseased area and establishing a second bridge three-dimensional model corresponding to the collected high-resolution image. The second bridge three-dimensional model established in this way and the high-resolution image obtained can ensure that all information on the bridge surface is collected without blind spots, prevent the omission of the diseased area, and the second bridge three-dimensional model established has higher accuracy, which is more conducive to the precise positioning of the diseased area.
[0044] Step 3: Input the high-resolution image into the trained neural network to identify the diseased area in the high-resolution image; fit the boundary of each diseased area by a polygon fitting algorithm, and determine the boundary coordinates P of each diseased area based on the pixel coordinate system. i (u,v), the boundary of each diseased area is composed of multiple boundary coordinates P i (u,v), where the subscript i represents the ordinal number of the coordinate.
[0045] The specific implementation of step 3 above can be found in Figure 1 The part of intelligent identification of bridge apparent defects. Figure 2 It is the recognition result map of the diseased area on the image. Each diseased area is regarded as a polygon, and the boundary of the diseased area is fitted by a polygon fitting algorithm. The boundary coordinates P of the diseased area are determined based on the pixel coordinate system. i The boundary coordinates of the disease are used to indicate the specific position and shape of the disease in the image. The boundary coordinates of the diseased area in each image are saved separately. The boundary of the diseased area is generally represented by multiple boundary coordinates P i To express.
[0046] Furthermore, the neural network may be a semantic segmentation network. The semantic segmentation network may be constructed based on the deep learning framework pytorch.
[0047] Furthermore, after the diseased area is identified through the semantic segmentation network, discrete noise points can be removed from the diseased area.
[0048] Step 4: Multiple boundary coordinates P of the diseased area i(u, v) is mapped to the second three-dimensional model of the bridge to locate the boundary of the damaged area on the second three-dimensional model of the bridge.
[0049] Step 3 obtains the specific position and shape of the bridge defect area in the image, which is two-dimensional and cannot determine the position of the defect area on the second bridge three-dimensional model. Step 4 realizes the three-dimensional positioning of the boundary of the defect area on the second bridge three-dimensional model by mapping the two-dimensional points in the image with the three-dimensional points on the model. Figure 3 As shown in the figure, four coordinate systems are involved in disease positioning: pixel coordinate system, image coordinate system, camera coordinate system, and world coordinate system.
[0050] (1) Pixel coordinate system
[0051] The origin of the pixel coordinate system is located in the upper left corner of the image, the u-axis is parallel to the image plane and points horizontally to the right, and the v-axis is perpendicular to the u-axis and points downward. The unit is pixels.
[0052] (2) Image coordinate system
[0053] The origin of the image coordinate system is located at the intersection of the camera optical axis and the imaging plane, usually the midpoint of the imaging plane. The x-axis and y-axis are parallel to the u-axis and v-axis of the pixel coordinate system. The unit of the image coordinate system is millimeter, which is a physical unit.
[0054] (3) Camera coordinate system
[0055] The camera coordinate system uses the optical center of the camera as the origin of the coordinate system. The optical center of the camera can be understood as the geometric center of the camera lens. The Xc axis and Yc axis are parallel to the x-axis and y-axis of the image coordinate system, and the optical axis of the camera is the Zc axis. The camera coordinate system satisfies the right-hand rule. The optical center of the camera is a three-dimensional coordinate.
[0056] (4) World coordinate system
[0057] The world coordinate system is used to represent the absolute coordinates of objects in space. The world coordinate system is a special coordinate system that uses X W , Y W , Z W It establishes a reference frame needed to describe other coordinate systems. In a non-technical sense, the world coordinate system establishes the largest coordinate system we care about, which does not necessarily have to be the entire world.
[0058] The relationship diagram of the four coordinate systems is as follows Figure 3 As shown, the real world point, the imaging point and the camera optical center are collinear. The world coordinates of the real world point correspond to the corresponding coordinates in the camera coordinate system, the image coordinate system and the pixel coordinate system.
[0059] like Figure 3-4, the following steps 4.1 and 4.2 are used to achieve three-dimensional positioning of the diseased area.
[0060] The step 4 includes step 4.1 and step 4.2.
[0061] Step 4.1: Calculate the rotation matrix R and translation matrix t between the multiple high-resolution images obtained in step 2 through the incremental motion recovery structure algorithm, place the second bridge 3D model, the rotation matrix R and the translation matrix t between the images in the same coordinate system, and complete the restoration of the entire inspection scene. At this time, the translation matrix t is the coordinate of the camera optical center; use the following formula (1) to convert the defect boundary coordinate P in the pixel coordinate system i (u,v) is converted to world coordinates P w (x w ,y w ,z w ), subscript w is the serial number of the coordinate;
[0062]
[0063] Among them, f, dx, dy, u0, and v0 are the focal length, physical pixel length, physical pixel width, abscissa of the imaging plane center, and ordinate of the imaging plane center of the drone’s camera lens; q is the object distance parameter, which indicates the distance from the optical center of the camera to the shooting point.
[0064] The second bridge 3D model, the rotation matrix R between images, and the translation matrix t are placed in the same coordinate system to complete the restoration of the entire inspection scene; in this way, the specific position and angle of the camera relative to the bridge can be known when each bridge appearance picture is taken during the drone detailed inspection. The coordinate system can be the world coordinate system.
[0065] The incremental Structure from Motion (SFM) algorithm can estimate the camera motion of each image relative to the previous image, including translation and rotation. Through these camera motion estimates, the trajectory and motion path of the camera can be obtained, from which the rotation matrix R and translation matrix t between images can be obtained, and both R and t are three-dimensional coordinates.
[0066] When the drone performs detailed inspection, it is impossible to collect accurate object distance information; in this case, the q parameter can be set to f, and the world coordinate P calculated by formula 1 is w It falls on the symmetry plane of the camera sensor plane about the camera optical center.
[0067] Step 4.2: By moving from the camera optical center coordinate t to the world coordinate P w The ray L(s) is obtained by calculating the starting point and direction vector of L(s). The ray L(s) is shown in the following formula (2):
[0068] L(s)=s*(P w -t)+t (2);
[0069] Where s is a real number, which indicates the set distance moved along the ray direction from the coordinate t of the camera optical center; calculate the intersection point P of the ray L(s) and the second bridge 3D model r (x r ,y r ,z r ), subscript r is the number of the coordinate; P r (x r ,y r ,z r ) are the coordinates of the boundary of the damaged area on the three-dimensional model of the second bridge, thereby realizing the positioning of the boundary of the damaged area on the three-dimensional model of the second bridge.
[0070] Furthermore, the intersection point P of the calculation ray L(s) and the second bridge three-dimensional model is r (x r ,y r ,z r ), specifically: extract all the three-dimensional vertex coordinates of the second bridge three-dimensional model to form a three-dimensional vertex coordinate set A, project the three-dimensional vertex coordinate set A and the coordinate t of the camera optical center onto a plane perpendicular to the ray L(s), and obtain the corresponding plane coordinate set A' and plane coordinate t' after projection. The coordinates in the plane coordinate set A' correspond to the coordinates in the three-dimensional vertex coordinate set A one-to-one, and find the point set D' in the plane coordinate set A' that is located in the neighborhood of the set radius of t'; the point set D' corresponds to the subset D in the three-dimensional vertex coordinate set A, and the three-dimensional vertex coordinates in the subset D that are closest to the camera optical center are obtained, and the three-dimensional vertex coordinates are the intersection point P of the ray L(s) and the second bridge three-dimensional model r (x r ,y r ,z r ).
[0071] From the above steps, it can be seen that the present application first identifies the diseased area on each image through the trained neural network, fits the boundary of the diseased area using a polygon fitting algorithm, and determines the boundary coordinates P of the diseased area based on the pixel coordinate system. i The boundary of the diseased area is generally defined by multiple boundary coordinates P i Then, through steps 4.1 and 4.2, multiple boundary coordinates P representing the boundary of the diseased area are i Map it to the three-dimensional model of the second bridge, so as to locate the damaged area on the three-dimensional model of the second bridge.
[0072] Furthermore, the method also includes: Step 5: The defective area can be drawn on the three-dimensional model of the second bridge according to the positioning of the defective area on the three-dimensional model of the second bridge through BIM modeling software.
[0073] The present application also proposes a three-dimensional positioning system for bridge damage areas based on an unmanned aerial vehicle and a neural network. The system includes an unmanned aerial vehicle, a processor, and a memory. The system is characterized in that the memory stores computer executable instructions, and when the processor executes the instructions, the steps of the method described in any one of the above embodiments are implemented.
[0074] The above description is only a specific embodiment of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A three-dimensional positioning method for bridge damage areas based on drones and neural networks, comprising the following steps: Step 1: Use the camera on the drone to take aerial photos of the bridge to be inspected, collect multiple low-resolution images of the bridge, and generate a first three-dimensional model of the bridge based on the multiple low-resolution images; plan the inspection path of the drone and multiple viewpoints on the inspection path based on the generated first three-dimensional model of the bridge, where the viewpoint is the position where the camera of the drone takes pictures; Step 2: The UAV flies according to the inspection path and collects multiple high-resolution images of the bridge at multiple viewpoints; a second bridge three-dimensional model is generated from the multiple high-resolution images through a three-dimensional reconstruction algorithm; Step 3: inputting the high-resolution image into a trained neural network to identify the diseased area in the high-resolution image; The boundary of each diseased area is fitted by a polygon fitting algorithm, and the boundary coordinates P of each diseased area are determined based on the pixel coordinate system. i (u,v), the boundary of each diseased area is composed of multiple boundary coordinates P i (u,v), the subscript i represents the sequence number of the coordinate; Step 4: Multiple boundary coordinates P of the diseased area i (u, v) is mapped onto the second three-dimensional model of the bridge to locate the boundary of the damaged area on the second three-dimensional model of the bridge; The step 4 is specifically as follows: Step 4.1: Calculate the rotation matrix R and translation matrix t between the multiple high-resolution images obtained in step 2 through the incremental motion recovery structure algorithm, place the second bridge 3D model, the rotation matrix R and the translation matrix t between the images in the same coordinate system, and complete the restoration of the entire inspection scene. At this time, the translation matrix t is the coordinate of the camera optical center; use the following formula (1) to convert the defect boundary coordinate P in the pixel coordinate system i (u,v) is converted to world coordinates P w (x w ,y w ,z w ), subscript w is the serial number of the coordinate; Among them, f, dx, dy, u0, and v0 are the focal length, physical pixel length, physical pixel width, abscissa of the imaging plane center, and ordinate of the imaging plane center of the drone’s camera lens; q is the object distance parameter, which indicates the distance from the optical center of the camera to the photographed point; Step 4.2: By moving from the camera optical center coordinate t to the world coordinate P w The ray L(s) is obtained by calculating the starting point and direction vector of L(s). The ray L(s) is shown in the following formula (2): L(s)=s*(P w -t)+t (2); Where s is a real number, which indicates the set distance moved along the ray direction from the coordinate t of the camera optical center; calculate the intersection point P of the ray L(s) and the second bridge 3D model r (x r ,y r ,z r ), subscript r is the number of the coordinate; P r (x r ,y r ,z r ) are the coordinates of the boundary of the defective area on the three-dimensional model of the second bridge, thereby realizing the positioning of the boundary of the defective area on the three-dimensional model of the second bridge; The intersection point P of the calculation ray L(s) and the second bridge three-dimensional model r (x r ,y r ,z r ), specifically: Extract all the three-dimensional vertex coordinates of the second bridge three-dimensional model to form a three-dimensional vertex coordinate set A, project the three-dimensional vertex coordinate set A and the coordinate t of the camera optical center onto a plane perpendicular to the ray L(s), and obtain the corresponding plane coordinate set A' and plane coordinate t' after projection. The coordinates in the plane coordinate set A' correspond to the coordinates in the three-dimensional vertex coordinate set A one-to-one, and find the point set D' in the plane coordinate set A' that is located in the neighborhood of the set radius of t'; the point set D' corresponds to the subset D in the three-dimensional vertex coordinate set A, and the three-dimensional vertex coordinates closest to the camera optical center in the subset D are obtained. The three-dimensional vertex coordinates are the intersection point P of the ray L(s) and the second bridge three-dimensional model r (x r ,y r ,z r ); The neural network is a semantic segmentation network.
2. According to the three-dimensional positioning method of bridge damage areas based on drones and neural networks described in claim 1, after the damage area is identified by the semantic segmentation network, discrete noise points are removed from the damage area, and then the boundaries of each damage area are fitted by a polygon fitting algorithm.
3. The three-dimensional positioning method of bridge damage area based on drone and neural network according to claim 1, the method further comprises: Step 5: Using the BIM modeling software, draw the defective area on the three-dimensional model of the second bridge according to the location of the defective area on the three-dimensional model of the second bridge.
4. A three-dimensional positioning system for bridge damage areas based on drones and neural networks, the system comprising drones, processors, and memory, characterized in that: The memory stores computer executable instructions, and when the processor executes the instructions, the steps of the method according to any one of claims 1 to 3 are implemented.
