Unmanned aerial vehicle low-altitude three-dimensional data acquisition method and system

Abstract

The invention discloses a low-altitude three-dimensional data acquisition method and system for an unmanned aerial vehicle, and particularly relates to the field of three-dimensional data acquisition, and the method comprises the steps: selecting a four-rotor unmanned aerial vehicle, carrying a laser radar, a high-resolution camera and high-precision positioning equipment, and carrying out the low-altitude flight; the laser radar collects surface information, and the camera shoots a ground image; then, processing the acquired data, including abnormal value removal, missing value supplementation, image denoising and image enhancement; thirdly, converting the processed data into three-dimensional point cloud data, and identifying an image to obtain an actual appearance and a name of a target object; and finally, performing high-precision topographic surveying and mapping and three-dimensional modeling according to the processed three-dimensional point cloud data to generate a final three-dimensional model.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional data acquisition, and more specifically, to a method and system for low-altitude three-dimensional data acquisition of an unmanned aerial vehicle. Background Art

[0002] Drones are widely used in military, civil and commercial fields. In the military field, drones can be used for reconnaissance, surveillance, target positioning, strike and other tasks; in the civil field, drones can be used for aerial photography, mapping, environmental monitoring, disaster relief, etc.; in the commercial field, drones can be used for logistics distribution, advertising, agricultural plant protection, etc. With the continuous development of technology, the performance of drones is constantly improving and the cost is gradually decreasing, making drones more and more widely used in various fields.

[0003] With the continuous advancement of science and technology and the rapid development of drone technology, three-dimensional data acquisition has been widely used in various fields. Traditional three-dimensional data acquisition methods include ground measurement, aerial photography or laser scanning, but in actual use, there are still some shortcomings, such as traditional data acquisition equipment is expensive, maintenance and operation costs are high; traditional methods have a long acquisition cycle and cannot respond to changes quickly. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method and system for collecting low-altitude three-dimensional data of an unmanned aerial vehicle to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: Step A1: Select a drone and carry it with a laser radar, a high-resolution camera, and a high-precision positioning device to fly at a low altitude in the target area; Step A2: After the drone takes off, it flies according to the preset path and maintains the predetermined flight altitude and speed, while starting the laser radar and high-resolution camera to collect data; Step A3: remove outliers and supplement missing values ​​from the data collected by the laser radar, and then perform image denoising and image enhancement on the images collected by the high-resolution camera; Step A4: converting the processed data into three-dimensional point cloud data; and identifying the image after image denoising and image enhancement to obtain the actual appearance and name of each target object in the image; Step A5: Perform high-precision terrain mapping and 3D modeling based on the processed 3D point cloud data.

[0006] Preferably, in step A1, the drone selects a quad-rotor drone, and the laser radar selects a LiDAR system with high precision, high scanning rate and wide viewing angle to obtain detailed terrain and object surface information; the high-resolution camera selects a micro single camera with a high resolution of 20MP or more, which can capture clear and detailed image information; the high-precision positioning equipment uses a real-time dynamic positioning (RTK) device, and the RTK GPS system provides centimeter-level positioning accuracy to ensure accurate recording of the drone flight trajectory data. The integrated IMU sensor is combined with the RTK device to realize real-time monitoring of the drone flight situation and improve the accuracy and reliability of the data.

[0007] Preferably, in step A2, before the flight, the starting position of the drone and the ending position of the drone are determined, and necessary intermediate path points are added between the starting point and the end point to form a complete flight trajectory; the Dijkstra algorithm is used to find the shortest path to reduce flight time and energy consumption; the method of the Dijkstra algorithm is specifically as follows: B1: Create a distance array d to store the shortest distances from the source node to all other nodes; set the distance of the source node to 0 and the distances of all other nodes to infinity; B2: Create a priority queue f to store the nodes to be processed and their distances; add the source node and its distance to the priority queue; B3: When the priority queue is not empty, perform the following operations: Take out the node g with the smallest current distance and its distance ; Traverse all nodes h adjacent to g and calculate the distance from the source node to j ;if Less than , then update for , and j and its distance Join the priority queue; B4: When the priority queue is empty, the algorithm ends. At this time, the d array stores the shortest distance from the source node to all other nodes; B6: Output the shortest distance from the source node to all other nodes according to the d array; B7: Ensure that the path meets safety requirements and avoids collision with obstacles.

[0008] Preferably, in step A3, the mean and standard deviation of the collected data are calculated. If the calculated mean and standard deviation exceed n times the set threshold, the data is recorded as an outlier, the outlier is removed, and the data excluding the outlier is supplemented with missing values, and the supplemented missing value is the calculated mean; The specific method of image denoising using mean is: C1: Determine the window size of the mean filter. The window size is an odd number to avoid the center offset problem. The larger the window, the stronger the smoothing effect. C2: Calculate the local mean: For each pixel in the image, select a window area centered on the pixel; C3: Calculate the sum of all pixel values ​​in the window and then divide it by the number of pixels in the window to get the local mean; C4: Replace the pixel value at the center of the window in the original image with the calculated local mean; C5: For pixels at the edge of the image, special treatment is required because they do not have enough neighbors. Common methods include copying edge values, mirroring edge values, or using a smaller filter window; C6: Repeat the above process until every pixel in the image is replaced by the newly calculated value; C7: Depending on the noise level and the amount of detail you want to preserve, you may need to adjust the filter window size; C8: De-noising effect was evaluated by visual inspection and quantitative indicators; the quantitative indicators were mean square error and peak signal-to-noise ratio.

[0009] Preferably, in step A3, the distance of the laser pulse return, i.e. the measured depth, is calculated by the time difference according to the known value of the speed of light; According to the distance from the laser radar to the target object and the angle information of the target object, the specific size and position of the target object can be calculated; Extract feature information related to the target object from the enhanced image; use Canny edge detection, shape analysis, and texture analysis techniques to extract features from the image; these feature information will be used for subsequent image classification and recognition.

[0010] According to the extracted feature information, the target objects in the image are classified and their specific names are identified; template matching is used for image classification and recognition; the extracted features are matched with the preset templates to find the most similar template, thereby determining the name of the target object; The target object in the image after template recognition is identified, and the identified target object is labeled with a name.

[0011] Preferably, in step A5, the three-dimensional point cloud data is subjected to Delaunay triangulation to generate a surface model, and the generation method is specifically as follows: F1: Construct a super triangle that is large enough to include all 3D point cloud data; F2: Insert each point in the 3D point cloud data into the current triangle mesh in turn; F3: For each newly inserted point, find all triangles containing the point. All triangles containing the point are called influencing triangles. F4: Delete the common edges of the affected triangles and connect the new points with all the vertices of the affected triangles to form a new triangle mesh; F4: Locally optimize the newly formed triangular mesh according to the empty circle characteristics and the maximum minimum angle characteristics of Delaunay triangulation; F5: By exchanging the diagonals of adjacent triangles, ensure that the circumscribed circle of each triangle does not contain other 3D point cloud data points.

[0012] F6: Repeat the above point-by-point insertion and local optimization process until all 3D point cloud data are inserted into the triangular mesh; F7: Finally, delete all triangles related to the super triangle to get the final Delaunay triangular mesh; F8: Check the generated Delaunay triangulated mesh for overlaps, cracks, or irregular triangles; further optimize the mesh as needed, including smoothing and detail enhancement. F9: Apply the generated Delaunay triangulated mesh to high-precision terrain mapping, 3D modeling, and numerical simulation; export the optimized triangulated mesh to OBJ format for subsequent application and processing.

[0013] The generated surface model is given color, texture and other attributes, including material and reflection, to enhance the visual effect; the surface model with attributes is converted into a 3D voxel model through 3D point cloud data using a voxelization algorithm, and the final 3D model is generated; High-precision topographic mapping performs point cloud registration on 3D point cloud data. The goal of point cloud registration is to convert 3D point cloud data in different coordinate systems into a unified coordinate system. Common methods include feature-based registration and marker-based registration. The 3D point cloud data is then segmented into ground points and non-ground points. Ground points are used to generate digital elevation models, while non-ground points can be used for other applications, such as building modeling. Generate digital elevation models from ground 3D point cloud data; this is the basis of terrain surveying and mapping; digital elevation models generate regular grid elevation data through interpolation algorithms; Use digital elevation model data to build a surface model; common surface models include triangular meshes and regular grids; triangular meshes form triangles by connecting adjacent elevation points and are suitable for irregular terrain; regular grids cover the entire area with regular grids and are suitable for flat or regular terrain; The generated digital elevation model is evaluated for accuracy to ensure that it meets the project requirements. The evaluation is performed using the measured values ​​of known control points and the model prediction values; the final topographic map is exported in OBJ format for further analysis or application.

[0014] Technical effects and advantages of the present invention: The present invention uses an unmanned aerial vehicle equipped with collection equipment to fly at low altitude in a target area and collect data along a preset path; the collected data is converted into three-dimensional point cloud data after being processed by outlier removal, missing value supplementation, image denoising and enhancement, and image recognition is performed; high-precision terrain mapping and three-dimensional modeling are performed based on the processed three-dimensional point cloud data to generate a final three-dimensional model; the maintenance cost of this collection method is lower than the operation and maintenance cost of traditional data collection equipment unmanned aerial vehicles, and the collection cycle is greatly shortened, which can respond to changes quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0016] Figure 2 It is a schematic diagram of module connection of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 As shown, the present invention provides a method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle, and the method is: Step A1: Select a drone and carry it with a laser radar, a high-resolution camera, and a high-precision positioning device to fly at a low altitude in the target area; In the step A1, the drone selects a quad-rotor drone, and the laser radar selects a LiDAR system with high precision, high scanning rate and wide viewing angle to obtain detailed terrain and object surface information; the high-resolution camera selects a micro single camera with a high resolution of 20MP or above, which can capture clear and detailed image information; the high-precision positioning equipment uses a real-time dynamic positioning (RTK) device, and the RTK GPS system provides centimeter-level positioning accuracy to ensure accurate recording of the drone's flight trajectory data. The integrated IMU sensor is then combined with the RTK device to achieve real-time monitoring of the drone's flight status and improve the accuracy and reliability of the data.

[0019] After the selection of drones, lidars, high-resolution cameras, and high-precision positioning equipment is completed, check the working status of each device to ensure that each device is in working condition.

[0020] Step A2: After the drone takes off, it flies according to the preset path and maintains the predetermined flight altitude and speed, while starting the laser radar and high-resolution camera to collect data; In step A2, before the flight, the starting position of the drone and the ending position of the drone are determined, and necessary intermediate path points are added between the starting point and the end point to form a complete flight trajectory; the Dijkstra algorithm is used to find the shortest path to reduce flight time and energy consumption; the specific method of the Dijkstra algorithm is: B1: Create a distance array d to store the shortest distances from the source node to all other nodes; set the distance of the source node to 0 and the distances of all other nodes to infinity; B2: Create a priority queue f to store the nodes to be processed and their distances; add the source node and its distance to the priority queue; B3: When the priority queue is not empty, perform the following operations: Take out the node g with the smallest current distance and its distance ; Traverse all nodes h adjacent to g and calculate the distance from the source node to j ;if Less than , then update for , and j and its distance Join the priority queue; B4: When the priority queue is empty, the algorithm ends. At this time, the d array stores the shortest distance from the source node to all other nodes; B6: Output the shortest distance from the source node to all other nodes according to the d array; B7: Ensure that the path meets safety requirements and avoids collision with obstacles.

[0021] During the flight, the LiDAR will periodically emit laser pulses to capture surface information; these pulses are emitted in the direction of the drone's flight; the radar system has a built-in high-frequency laser transmitter that can emit a large number of laser pulses in a short period of time; near-infrared lasers are usually used, and this wavelength has a good reflective effect on plants and ground materials; the LiDAR emits laser pulses at a frequency of hundreds of thousands of times per second to form dense three-dimensional point cloud data; After the laser pulse is emitted from the transmitter, it propagates in a straight line to the target object, which can be the ground, trees, or buildings. After encountering the target object, the laser pulse is reflected back to the sensor. Different materials and surfaces have different reflection characteristics for lasers, and the reflection intensity will also be different. The laser radar system has a built-in photodetector to receive the returned laser signal. The system records the time difference between the emission and reception of the laser pulse. The camera regularly captures ground images according to the set time interval and flight path; it takes a picture every 10m to ensure the details of the coverage area; Step A3: remove outliers and supplement missing values ​​from the data collected by the laser radar, and then perform image denoising and image enhancement on the images collected by the high-resolution camera; In step A3, the mean and standard deviation of the collected data are calculated. If the calculated mean and standard deviation exceed n times the set threshold, the data is recorded as an outlier, the outlier is removed, and the data after the outlier is removed is supplemented with missing values, and the supplemented missing value is the calculated mean; The specific method of image denoising using mean is: C1: Determine the window size of the mean filter. The window size is an odd number to avoid the center offset problem. The larger the window, the stronger the smoothing effect. C2: Calculate the local mean: For each pixel in the image, select a window area centered on the pixel; C3: Calculate the sum of all pixel values ​​in the window and then divide it by the number of pixels in the window to get the local mean; C4: Replace the pixel value at the center of the window in the original image with the calculated local mean; C5: For pixels at the edge of the image, special treatment is required because they do not have enough neighbors. Common methods include copying edge values, mirroring edge values, or using a smaller filter window; C6: Repeat the above process until every pixel in the image is replaced by the newly calculated value; C7: Depending on the noise level and the amount of detail you want to preserve, you may need to adjust the filter window size; C8: De-noising effect was evaluated by visual inspection and quantitative indicators; the quantitative indicators were mean square error and peak signal-to-noise ratio.

[0022] The specific method of image enhancement is: D1: Applies the selected Laplacian filter to the image, computing the difference between each pixel and its neighboring pixels.

[0023] D2: Perform a convolution operation on the image, that is, use the filter to multiply the local area of ​​the image element by element and sum them.

[0024] D3: Add the filtering result to the original image to enhance the edges and details. The calculation method for adding the filtering result to the original image is as follows: ,in, represents the enhanced image, Z represents the original image, It represents the amplification factor, G represents the selected Laplace operator filter; Represented as a convolution operation; D4: Ensure that the pixel values ​​of the enhanced image are within the valid range, which includes 0 to 255; if the calculated value exceeds this range, it needs to be cropped.

[0025] Step A4: converting the processed data into three-dimensional point cloud data; and identifying the image after image denoising and image enhancement to obtain the actual appearance and name of each target object in the image; In step A3, based on the known value of the speed of light, the distance of the laser pulse returning, i.e. the measured depth, is calculated by the time difference; The specific method for calculating the distance from the laser radar to the target object is: , where L represents the distance from the laser radar to the target object, c represents the speed of light, and t represents the time delay; The speed of light in a vacuum is about 300,000 km / s. The time delay represents the time from the emission of the laser beam to the reception of the reflected signal. At the same time, through the built-in rotation mechanism, the laser radar measures the angle information of the target object, including azimuth and pitch angle; The calculation method of the azimuth angle is as follows: ,in, Expressed as azimuth, the position coordinates of the target object are , the coordinates of the drone carrying the laser radar when flying are ; The specific calculation method of the pitch angle is: ,in, It is represented by the pitch angle, z is the height of the drone carrying the laser radar from the ground. Expressed as the height of the target object from the ground; According to the distance from the laser radar to the target object and the angle information of the target object, the specific size and position of the target object can be calculated; the width calculation method at the horizontal distance is as follows: ,in, It is expressed as the width on the horizontal distance, L is the distance from the laser radar to the target object, Expressed as azimuth; The specific method for calculating the height in vertical distance is: ,in, It is expressed as the height in vertical distance, L is the distance from the laser radar to the target object, Expressed as pitch angle; The calculated distance, azimuth and elevation angle of the laser radar to the target object are converted into a three-dimensional coordinate system. The specific calculation method is as follows: , where r represents the radius in the three-dimensional coordinate system, that is, the distance from the laser radar to the target object, and L represents the distance from the laser radar to the target object; , where a represents the horizontal axis in the three-dimensional coordinate system, and r represents the distance from the laser radar to the target object. Expressed as azimuth, Expressed as pitch angle; , where b represents the vertical axis in the three-dimensional coordinate system, and r represents the distance from the laser radar to the target object. Expressed as azimuth, Expressed as pitch angle; , where c represents the axis of the depth direction in the three-dimensional coordinate system, and r represents the distance from the laser radar to the target object. Expressed as azimuth; but Represented as converted 3D point cloud data.

[0026] Extract feature information related to the target object from the enhanced image; use Canny edge detection, shape analysis, and texture analysis techniques to extract features from the image; these feature information will be used for subsequent image classification and recognition.

[0027] According to the extracted feature information, the target objects in the image are classified and their specific names are identified; template matching is used for image classification and recognition; the extracted features are matched with the preset templates to find the most similar template, thereby determining the name of the target object; the specific method of using template matching is as follows: E1: Select a template; the template is usually a small area related to the target object; E2: Slide the template on the original image, traversing the entire image from left to right and from top to bottom; E3: For each position, calculate the similarity score between the template and the image region at that position; the similarity score is calculated using the sum of the absolute values ​​of the differences between the pixel values; The similarity score is calculated as follows: , where M represents the similarity score, W represents the width of the template image, I represents the height of the template image, R represents the input original image, V represents the template image, p represents the coordinate in the width direction of the template image, q represents the coordinate in the height direction of the template image, s represents the horizontal coordinate of the area in the original image that is currently matching with the template, and u represents the vertical coordinate of the area in the original image that is currently matching with the template; E4: Select the best matching position according to the calculated similarity score; obtain the best matching position through the threshold comparison algorithm, compare the calculated similarity score with the threshold, and if it is equal to the threshold, output a template matching success instruction; if it is not equal to the threshold, output a template matching failure instruction; E5: Output the best matching position.

[0028] The target object in the image after template recognition is identified, and the identified target object is labeled with a name.

[0029] Step A5: Perform high-precision terrain mapping and 3D modeling based on the processed 3D point cloud data; In step A5, the three-dimensional point cloud data is subjected to Delaunay triangulation to generate a surface model, and the specific generation method is as follows: F1: Construct a super triangle that is large enough to include all 3D point cloud data; F2: Insert each point in the 3D point cloud data into the current triangle mesh in turn; F3: For each newly inserted point, find all triangles containing the point. All triangles containing the point are called influencing triangles. F4: Delete the common edges of the affected triangles and connect the new points with all the vertices of the affected triangles to form a new triangle mesh; F4: Locally optimize the newly formed triangular mesh according to the empty circle characteristics and the maximum minimum angle characteristics of Delaunay triangulation; F5: By exchanging the diagonals of adjacent triangles, ensure that the circumscribed circle of each triangle does not contain other 3D point cloud data points.

[0030] F6: Repeat the above point-by-point insertion and local optimization process until all 3D point cloud data are inserted into the triangular mesh; F7: Finally, delete all triangles related to the super triangle to get the final Delaunay triangular mesh; F8: Check the generated Delaunay triangulated mesh for overlaps, cracks, or irregular triangles; further optimize the mesh as needed, including smoothing and detail enhancement. F9: Apply the generated Delaunay triangulated mesh to high-precision terrain mapping, 3D modeling, and numerical simulation; export the optimized triangulated mesh to OBJ format for subsequent application and processing.

[0031] The generated surface model is given color, texture and other attributes, including material and reflection, to enhance the visual effect; the surface model with attributes is converted into a 3D voxel model through 3D point cloud data using a voxelization algorithm, and the final 3D model is generated; High-precision topographic mapping performs point cloud registration on 3D point cloud data. The goal of point cloud registration is to convert 3D point cloud data in different coordinate systems into a unified coordinate system. Common methods include feature-based registration and marker-based registration. The 3D point cloud data is then segmented into ground points and non-ground points. Ground points are used to generate digital elevation models, while non-ground points can be used for other applications, such as building modeling. Generate digital elevation models from ground 3D point cloud data; this is the basis of terrain surveying and mapping; digital elevation models generate regular grid elevation data through interpolation algorithms; Use digital elevation model data to build a surface model; common surface models include triangular meshes and regular grids; triangular meshes form triangles by connecting adjacent elevation points and are suitable for irregular terrain; regular grids cover the entire area with regular grids and are suitable for flat or regular terrain; The generated digital elevation model is evaluated for accuracy to ensure that it meets the project requirements. The evaluation is performed using the measured values ​​of known control points and the model prediction values; the final topographic map is exported in OBJ format for further analysis or application.

[0032] See also Figure 2 As shown, in this embodiment, it should be specifically explained that the present invention provides a UAV low-altitude three-dimensional data acquisition system, including the following modules: UAV preparation module: used to select a UAV and carry it with laser radar, high-resolution camera, and high-precision positioning equipment to fly at low altitude in the target area; UAV data collection module: After the UAV takes off, it flies according to the preset path and maintains the predetermined flight altitude and speed, while starting the laser radar and high-resolution camera for data collection; Data processing module: used to remove outliers and supplement missing values ​​from the data collected by the lidar, and then perform image denoising and image enhancement on the images collected by the high-resolution camera; 3D point cloud data calculation module: used to convert the processed data into 3D point cloud data; and recognize the image after image denoising and image enhancement to obtain the actual appearance and name of each target object in the image; 3D modeling module: used for high-precision terrain mapping and 3D modeling based on processed 3D point cloud data.

[0033] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle, characterized in that: include: Step A1: Select a drone and carry it with a laser radar, a high-resolution camera, and a high-precision positioning device to fly at a low altitude in the target area; Step A2: After the drone takes off, it flies according to the preset path and maintains the predetermined flight altitude and speed, while starting the laser radar and high-resolution camera to collect data; Step A3: remove outliers and supplement missing values ​​from the data collected by the laser radar, and then perform image denoising and image enhancement on the images collected by the high-resolution camera; Step A4: converting the processed data into three-dimensional point cloud data; The denoised and enhanced images are then identified to obtain the actual appearance and name of each target object in the image; Step A5: Perform high-precision terrain mapping and 3D modeling based on the processed 3D point cloud data.

2. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: In the step A1, the drone selects a quad-rotor drone, and the laser radar selects a LiDAR system with high precision, high scanning rate and wide viewing angle to obtain detailed terrain and object surface information; the high-resolution camera selects a micro single camera with a high resolution of 20MP or above, which can capture clear and detailed image information; the high-precision positioning equipment uses a real-time dynamic positioning (RTK) device, and the RTKGPS system provides centimeter-level positioning accuracy to ensure accurate recording of the drone's flight trajectory data. The integrated IMU sensor is combined with the RTK device to realize real-time monitoring of the drone's flight status and improve the accuracy and reliability of the data.

3. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: In step A2, before the flight, the starting position of the UAV and the ending position of the UAV are determined, and necessary intermediate path points are added between the starting point and the end point to form a complete flight trajectory; the Dijkstra algorithm is used to find the shortest path to reduce flight time and energy consumption; during the flight, the lidar will periodically emit laser pulses to capture surface information; these pulses are emitted along the flight direction of the UAV; the radar system has a built-in high-frequency laser transmitter that can emit a large number of laser pulses in a short period of time; near-infrared lasers are usually used, and this wavelength has a good reflection effect on plants and ground materials; the lidar emits laser pulses at a frequency of hundreds of thousands of times per second to form dense three-dimensional point cloud data; the camera regularly captures ground images according to the set time interval and flight path; and a shot is taken every 10m to cover the details of the shooting area.

4. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: In step A3, the mean and standard deviation of the collected data are calculated. If the calculated mean and standard deviation exceed n times the set threshold, the data is recorded as an outlier, the outlier is removed, and the data after the outlier is removed is supplemented with missing values, and the supplemented missing value is the calculated mean; The specific method of image enhancement is: D1: Apply the selected Laplacian filter to the image and calculate the difference between each pixel and its neighboring pixels; D2: Perform convolution operation on the image, that is, use the filter to multiply the local area of ​​the image element by element and sum them; D3: Add the filtering result to the original image to enhance the edge and details; the calculation method for adding the filtering result to the original image is as follows: ,in, represents the enhanced image, Z represents the original image, It represents the amplification factor, G represents the selected Laplace operator filter; Represented as a convolution operation; D4: Ensure that the pixel values ​​of the enhanced image are within the valid range, which includes 0 to 255; if the calculated value exceeds this range, it needs to be cropped.

5. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: In step A3, based on the known value of the speed of light, the distance of the laser pulse returning, i.e. the measured depth, is calculated by the time difference; The specific method for calculating the distance from the laser radar to the target object is: , where L represents the distance from the laser radar to the target object, c represents the speed of light, and t represents the time delay; The speed of light in a vacuum is about 300,000 km / s. The time delay represents the time from the emission of the laser beam to the reception of the reflected signal. Through the built-in rotating mechanism, the LiDAR measures the angle information of the target object, including azimuth and pitch angle; The calculation method of the azimuth angle is as follows: ,in, Expressed as azimuth, the position coordinates of the target object are , the coordinates of the drone carrying the laser radar when flying are ; The specific calculation method of the pitch angle is: ,in, It is represented by the pitch angle, z is the height of the drone carrying the laser radar from the ground. Expressed as the height of the target object from the ground.

6. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 5, characterized in that: According to the distance from the laser radar to the target object and the angle information of the target object, the specific size and position of the target object can be calculated; the width calculation method at the horizontal distance is as follows: ,in, It is expressed as the width on the horizontal distance, L is the distance from the laser radar to the target object, Expressed as azimuth; The specific method for calculating the height in vertical distance is: ,in, It is expressed as the height in vertical distance, L is the distance from the laser radar to the target object, Expressed as pitch angle; The calculated distance, azimuth and elevation angle of the laser radar to the target object are converted into a three-dimensional coordinate system.

7. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: Extract feature information related to the target object from the enhanced image; use Canny edge detection, shape analysis, and texture analysis techniques to extract features from the image; These feature information will be used for subsequent image classification and recognition; Based on the extracted feature information, the target objects in the image are classified and their specific names are identified; template matching is used for image classification and recognition; the extracted features are matched with the preset template to find the most similar template, thereby determining the name of the target object.

8. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 7, characterized in that: The specific method of using template matching is: E1: Select a template; the template is usually a small area related to the target object; E2: Slide the template on the original image, traversing the entire image from left to right and from top to bottom; E3: For each position, calculate the similarity score between the template and the image region at that position; the similarity score is calculated using the sum of the absolute values ​​of the differences between the pixel values; The similarity score is calculated as follows: , where M represents the similarity score, W represents the width of the template image, I represents the height of the template image, R represents the input original image, V represents the template image, p represents the coordinate in the width direction of the template image, q represents the coordinate in the height direction of the template image, s represents the horizontal coordinate of the area in the original image that is currently matching with the template, and u represents the vertical coordinate of the area in the original image that is currently matching with the template; E4: Select the best matching position according to the calculated similarity score; obtain the best matching position through the threshold comparison algorithm, compare the calculated similarity score with the threshold, and if it is equal to the threshold, output a template matching success instruction; if it is not equal to the threshold, output a template matching failure instruction; E5: Output the best matching position; The target object in the image after template recognition is identified, and the identified target object is labeled with a name.

9. The method for collecting low-altitude three-dimensional data of an unmanned aerial vehicle according to claim 1, characterized in that: In the step A5, the three-dimensional point cloud data is subjected to Delaunay triangulation to generate a surface model; the generated surface model is given color, texture and other attributes, including material and reflection, to enhance the visual effect; the surface model after being given attributes is constructed into a three-dimensional voxel model through the three-dimensional point cloud data using a voxelization algorithm, and a final three-dimensional model is generated; High-precision topographic mapping performs point cloud registration on 3D point cloud data, and then divides the 3D point cloud data into ground points and non-ground points, and generates a digital elevation model from the ground 3D point cloud data; this is the basis of topographic mapping; the digital elevation model generates regular grid elevation data through interpolation algorithms; Use DEM data to build a surface model and evaluate the accuracy of the generated DEM to ensure that it meets project requirements; evaluate using the measured values ​​of known control points and the model's predicted values; and output the final topographic map in OBJ format for further analysis or application.

10. A UAV low-altitude 3D data acquisition system, using a UAV low-altitude 3D data acquisition method as claimed in any one of claims 1 to 9, characterized in that: UAV preparation module: used to select a UAV and carry it with laser radar, high-resolution camera, and high-precision positioning equipment to fly at low altitude in the target area; UAV data collection module: After the UAV takes off, it flies according to the preset path and maintains the predetermined flight altitude and speed, while starting the laser radar and high-resolution camera for data collection; Data processing module: used to remove outliers and supplement missing values ​​from the data collected by the lidar, and then perform image denoising and image enhancement on the images collected by the high-resolution camera; 3D point cloud data calculation module: used to convert the processed data into 3D point cloud data; The denoised and enhanced images are then identified to obtain the actual appearance and name of each target object in the image; 3D modeling module: used for high-precision terrain mapping and 3D modeling based on processed 3D point cloud data.

Images