Road stake number determination method and device in unmanned aerial vehicle road inspection process

Abstract

The invention relates to the technical field of unmanned aerial vehicle road inspection, and particularly discloses a road stake number determination method and device in an unmanned aerial vehicle road inspection process, and the method comprises the following steps: S1, data collection; step S2, stake mark identification area feature extraction and identification; s3, acquiring and processing flight path data of the unmanned aerial vehicle; step S4, feature extraction and local model construction; step S5, global model integration and stake number determination; according to the method, various sensor data, including optical images, laser radar data, flight attitude data and the like, of the unmanned aerial vehicle are comprehensively utilized, and a unique space geometric model and a data fusion algorithm are constructed, so that the highway stake number is accurately determined; the data accuracy and integrity are improved through multi-sensor fusion, and the influence of environmental interference on a single sensor is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) highway inspection, and in particular to a method and device for determining a highway pile number during an unmanned aerial vehicle (UAV) highway inspection. Background Art

[0002] In the field of highway inspection, traditional methods of determining pile numbers mostly rely on manual field surveys or vehicle-mounted positioning equipment. The manual method is extremely inefficient and labor-intensive, and vehicle-mounted positioning is prone to signal loss or deviation in some complex road conditions or remote mountainous areas. Some of the existing methods for determining pile numbers by drone inspections rely solely on image recognition of pile number identification, which is greatly affected by weather and identification clarity; others rely solely on GPS positioning, which greatly reduces accuracy in areas with signal interference. Summary of the invention

[0003] In response to the technical problems raised in the background technology, the present invention provides a method and device for determining highway stake numbers during a highway inspection by an unmanned aerial vehicle.

[0004] The technical solution adopted by the present invention is: a method for determining a highway pile number during a highway inspection by an unmanned aerial vehicle, which comprises the following steps:

[0005] Step S1, data collection: The drone flies according to the preset inspection route. During the flight, the high-definition camera on board is used to shoot high-definition images and GPS data of the road surface and surrounding areas at a fixed frame rate (set to frames per second). Specifically, the high-definition optical camera, lidar sensor and inertial measurement unit (IMU) are turned on. The optical camera shoots highway images at a fixed time interval t1 to obtain image information of the highway surface, roadside facilities and possible pile number identification. The lidar emits a laser beam at a time interval t2 to scan the highway and surrounding terrain to obtain three-dimensional point cloud data and record the spatial coordinates x of each point. l ,y l ,z l ) and reflection intensity information, IMU records the flight attitude data of the drone in real time, including pitch angle α, yaw angle β, roll angle γ and acceleration information (a x ,a y ,a z );

[0006] Step S2, extraction and recognition of features of the pile number identification area: using the Canny edge detection algorithm to perform edge detection on the preprocessed image to extract edge information in the image;

[0007] Step S3, acquisition and processing of drone flight trajectory data: The drone has a built-in high-precision inertial measurement unit (IMU) and global positioning system (GPS). During the flight, it records its own position information (longitude lon) at a fixed time interval (set as Δt). i, latitude lat i 、Altitude h i ), attitude information (pitch angle φ i , yaw angle ψ i , roll angle θ i ) ; Smooth the acquired position data and use the Kalman filter algorithm to remove noise and outliers in the data; Assume that the system state equation is x k =Ax k-1 +Bu k +w k, The observation equation is z k =Hx k +v k , where x k is the system state vector, A is the state transfer matrix, B is the control input matrix, u k is the control input vector, w k is the system noise vector, z k is the observation vector, H is the observation matrix, v k To observe the noise vector, the state estimation value is continuously updated through the Kalman filter recursive formula to obtain the smoothed position data;

[0008] Step S4, feature extraction and local model construction: from the preprocessed optical image, edge features in the image are extracted using an edge detection algorithm, and then possible edge contours of highway edge lines, lane lines, and pile number markings are identified through shape analysis;

[0009] Step S5, global model integration and stake number determination: each local space geometric model is spliced ​​and integrated to form a global space model of the highway.

[0010] The present invention is further configured such that, in step S2, the specific method of image preprocessing is as follows:

[0011] Grayscale the collected image and convert the color image into a grayscale image;

[0012] Assume that the RGB values ​​of the pixel (x, y) in the original color image are R(x, y), G(x, y), and B(x, y);

[0013] Then the gray value I(x,y) calculation formula is: I(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y);

[0014] Perform Gaussian filtering on the grayscale image to remove noise interference in the image;

[0015] Suppose the filtered image is G(x,y) and the Gaussian filter function is h(x,y), then Among them, G(x,y) is the filtered image, I(x,y) is the original image, h(s,t) is the Gaussian filter function, k is the filter window radius, σ is the Gaussian standard deviation;

[0016] Filter the laser radar point cloud data to remove points below the ground and noise points;

[0017] The slope between adjacent points is calculated through the slope-based filtering algorithm, and points with abnormally large or small slopes are determined as noise points and removed. Suppose two adjacent points are P1(x1, y1, z1) and P2(x2, y2, z2);

[0018] For IMU data, a complementary filtering algorithm is used to fuse the accelerometer and gyroscope data to improve the accuracy of attitude angle measurement;

[0019] Assume the angular velocity measured by the gyroscope is ω g , the acceleration measured by the accelerometer is a a, The fused attitude angle θ f =(1-k)θ f +kθ g , where θ g is the attitude angle calculated by the gyroscope, and k is the fusion coefficient.

[0020] The present invention is further configured that, in step S3, the Canny edge detection algorithm is used to perform edge detection on the preprocessed image, and the specific method for extracting edge information in the image is as follows:

[0021] Assume that the image gradient amplitude is M(x,y), the gradient direction is θ(x,y), and calculate the gradient of each pixel in the image to determine the edge;

[0022] Construct feature templates based on prior knowledge of the shape, color, size, etc. of highway pile number signs;

[0023] The template matching algorithm is used to search for the area matching the feature template in the image after edge detection, and the area whose matching degree exceeds the set threshold (set as T) is determined as the suspected stake mark area;

[0024] Assume the template image is T(x,y), the image to be matched is S(x,y), and the matching degree is

[0025]

[0026] The present invention is further configured as follows: in step S4, the coordinates of the edge points of the image are set to (x e ,y e );

[0027] Extract the road surface point cloud from the LiDAR point cloud data, and cluster the point clouds belonging to the road surface together through a clustering algorithm;

[0028] Calculate the normal vector of the point cloud, and determine the point whose normal vector is close to the preset highway plane normal vector as a highway surface point;

[0029] Assume that the point cloud clustering algorithm uses the density-based DBSCAN algorithm. For a point P(x, y, z), if the number of points within its radius r exceeds the set threshold n t , then these points are classified into the same category;

[0030] Combine IMU attitude data and LiDAR point cloud data to build a local space geometry model;

[0031] Taking the current position of the drone as the origin, the laser radar point cloud coordinates are transformed into the local coordinate system according to the IMU attitude angle; let point P(x l ,y l ,z l ) in the local coordinate system (x ll ,y ll ,z ll )The conversion formula is:

[0032]

[0033] The present invention is further configured that, in step S5, each local space geometric model is spliced ​​and integrated to form a global space model of the highway. The specific method is as follows:

[0034] By finding the overlapping areas between adjacent local models, the adjacent models are aligned and spliced ​​using the feature point matching algorithm;

[0035] Assume that adjacent local models M1 and M2 are used, extract SIFT feature points in the overlapping area, and obtain matching point pairs (P m1 ,P m2 ), use the least squares method to calculate the transformation matrix T between the models, and transform M2 into a coordinate system consistent with M1;

[0036] Determine the position of the reference points with known pile numbers in the global model. By setting reference points with obvious features and known pile numbers on the road in advance, use the data collected by the drone to determine the coordinates (x rp ,y rp ,z rp );

[0037] For any position P(x, y, z) during the drone inspection process, calculate the spatial geometric relationship between it and the known reference point; let the reference point P rp (xrp ,y rp ,z rp ), calculate the spatial distance between two points and the angle θ between the connecting line and the direction of the kilometer road (obtained through vector calculation);

[0038] According to the known reference point stake number N rp , the actual road length L between the reference points rp And the calculated distance d and angle θ, determine the stake number N of the current position;

[0039] The formula is: Where ΔN is the difference in pile numbers between adjacent reference points.

[0040] The present invention is further configured as a device for determining a highway stake number during a highway inspection by an unmanned aerial vehicle, comprising:

[0041] Data collection module: the drone flies according to the preset inspection route and collects various data during the flight;

[0042] The pile number identification area feature extraction and recognition module uses edge detection algorithm to perform edge detection on the preprocessed image and extract edge information in the image;

[0043] A data processing module processes the data in the data acquisition module;

[0044] The feature extraction and local model building module uses edge detection algorithms to extract edge features from preprocessed optical images, and then identifies possible edge contours of highway edges, lane lines, and pile number marks through shape analysis for modeling.

[0045] Global model integration and pile number determination module: various local spatial geometric models are spliced ​​and integrated to form a global spatial model of the highway.

[0046] The beneficial effects of the present invention are as follows: in the present invention, a variety of sensor data of the UAV is comprehensively utilized, including optical images, laser radar data, and flight attitude data, etc., and a unique spatial geometric model and data fusion algorithm are constructed to achieve accurate determination of the highway pile number. First, different types of data collected by the UAV are preprocessed to extract key features; then local spatial geometric relationships are constructed based on feature points, and then gradually expanded to a global highway model; finally, the highway pile number is obtained by matching and calculating in combination with known reference point information, which can effectively overcome the shortcomings of the existing technology, improve the accuracy, reliability and adaptability of pile number determination, and provide strong technical support for highway inspection, maintenance and other work. The data accuracy and integrity are improved through multi-sensor fusion, and the influence of environmental interference on a single sensor is reduced. For example, in areas with weak GPS signals, laser radar and image features can assist in positioning. Secondly, the intelligent algorithm realizes automatic feature extraction and matching, reduces manual intervention, and improves efficiency. The constructed spatial model can adapt to highways of different terrains, and the pile number can be accurately determined whether it is in mountainous areas or plains. DETAILED DESCRIPTION

[0047] The following will be combined with 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.

[0048] In order to solve the problems existing in the background technology, the present application proposes the following technical solution: a method for determining a highway stake number during a drone highway inspection, comprising the following steps:

[0049] Step S1, data collection: The drone flies according to the preset inspection route. During the flight, the high-definition camera on board is used to shoot high-definition images and GPS data of the road surface and surrounding areas at a fixed frame rate (set to frames per second). Specifically, the high-definition optical camera, lidar sensor and inertial measurement unit (IMU) are turned on. The optical camera shoots highway images at a fixed time interval t1 to obtain image information of the highway surface, roadside facilities and possible pile number identification. The lidar emits a laser beam at a time interval t2 to scan the highway and surrounding terrain to obtain three-dimensional point cloud data and record the spatial coordinates x of each point. l ,y l ,z l ) and reflection intensity information, IMU records the flight attitude data of the drone in real time, including pitch angle α, yaw angle β, roll angle γ and acceleration information (a x ,a y ,a z );

[0050] Step S2, extraction and recognition of features of the pile number identification area: using the Canny edge detection algorithm to perform edge detection on the preprocessed image to extract edge information in the image;

[0051] Step S3, acquisition and processing of drone flight trajectory data: The drone has a built-in high-precision inertial measurement unit (IMU) and global positioning system (GPS). During the flight, it records its own position information (longitude lon) at a fixed time interval (set as Δt). i , latitude lat i 、Altitude h i ), attitude information (pitch angle φ i , yaw angle ψ i , roll angle θ i ) ; Smooth the acquired position data and use the Kalman filter algorithm to remove noise and outliers in the data; Assume that the system state equation is x k =Ax k-1 +Bu k +w k, The observation equation is z k =Hx k +v k , where x k is the system state vector, A is the state transfer matrix, B is the control input matrix, u k is the control input vector, w k is the system noise vector, z k is the observation vector, H is the observation matrix, v k To observe the noise vector, the state estimation value is continuously updated through the Kalman filter recursive formula to obtain the smoothed position data;

[0052] Step S4, feature extraction and local model construction: from the preprocessed optical image, edge features in the image are extracted using an edge detection algorithm, and then possible edge contours of highway edge lines, lane lines, and pile number markings are identified through shape analysis;

[0053] Step S5, global model integration and stake number determination: each local space geometric model is spliced ​​and integrated to form a global space model of the highway.

[0054] The present invention makes comprehensive use of various sensor data of drones, including optical images, laser radar data, and flight attitude data, etc., and realizes accurate determination of highway pile numbers by constructing a unique spatial geometric model and data fusion algorithm. First, different types of data collected by drones are preprocessed to extract key features; then local spatial geometric relationships are constructed based on feature points, and then gradually expanded to a global highway model; finally, combined with known reference point information, the highway pile number is obtained through matching and calculation, which can effectively overcome the shortcomings of existing technologies, improve the accuracy, reliability and adaptability of pile number determination, and provide strong technical support for highway inspection, maintenance and other work.

[0055] In the present invention, multi-sensor fusion is used to improve data accuracy and completeness, and the influence of environmental interference on a single sensor is reduced. For example, in areas with weak GPS signals, lidar and image features can assist in positioning. Secondly, intelligent algorithms can realize automatic feature extraction and matching, reduce manual intervention, and improve efficiency. Furthermore, the constructed spatial model can adapt to roads with different terrains, and can accurately determine the pile number regardless of whether it is mountainous areas or plains, providing a reliable basis for highway inspection, maintenance, planning and other tasks, improving the level of intelligent highway management, saving manpower and material costs, and enhancing the safety and reliability of highway operations.

[0056] In a further design, in step S2 of this embodiment, the specific method of image preprocessing is as follows:

[0057] Step 1: grayscale the collected image to convert the color image into a grayscale image;

[0058] Step 2: Assume that the RGB values ​​of the pixel point (x, y) in the original color image are R(x, y), G(x, y), and B(x, y);

[0059] Then the gray value I(x,y) calculation formula is: I(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y);

[0060] Step 3: Perform Gaussian filtering on the grayscale image to remove noise interference in the image;

[0061] Let the filtered image be G(x,y) and the Gaussian filter function be h(x,y), then Let the filtered image be G(x,y) and the Gaussian filter function be h(x,y), then Among them, G(x,y) is the filtered image, I(x,y) is the original image, h(s,t) is the Gaussian filter function, k is the filter window radius, σ is the Gaussian standard deviation;

[0062] Step 4: filter the laser radar point cloud data to remove points below the ground and noise points;

[0063] Step 5: Calculate the slope between adjacent points through the slope-based filtering algorithm, and determine the points with abnormally large or small slopes as noise points and remove them. Suppose two adjacent points are P1(x1, y1, z1) and P2(x2, y2, z2);

[0064] Step 6: For the IMU data, the accelerometer and gyroscope data are fused using a complementary filtering algorithm to improve the accuracy of attitude angle measurement;

[0065] Step 7: Assume the angular velocity measured by the gyroscope is ω g , the acceleration measured by the accelerometer is a a, The fused attitude angle θ f =(1-k)θ f +kθ g , where θ g is the attitude angle calculated by the gyroscope, and k is the fusion coefficient, which is adjusted according to the actual situation.

[0066] The above technical solution is explained as follows: First, grayscale processing and Gaussian filtering of the collected images can effectively remove noise interference in the images, make the images clearer, and help subsequent data analysis and processing. Through the precise grayscale value calculation formula, the accuracy of grayscale conversion is ensured, and the quality of image data is guaranteed; secondly, filtering and fusing the lidar point cloud data and IMU data can improve the accuracy and reliability of the data. Removing points and noise points below the ground can avoid the influence of invalid data on the results. The complementary filtering algorithm is used to fuse the accelerometer and gyroscope data, which can effectively improve the accuracy of attitude angle measurement and make the acquired data more reflective of the actual situation. Finally, through the slope-based filtering algorithm and the fused attitude angle calculation formula, the data can be accurately processed and more realistic results can be obtained. This multi-step, multi-algorithm combined data processing method has important application value in the relevant data processing and analysis fields.

[0067] In the further design, in step S3, the Canny edge detection algorithm is used to perform edge detection on the preprocessed image, and the specific method for extracting edge information in the image is as follows:

[0068] Step 1: Assume that the image gradient amplitude is M(x, y), the gradient direction is θ(x, y), and calculate the gradient of each pixel in the image to determine the edge;

[0069] Step 2: construct a feature template based on the prior knowledge of the shape, color, size, etc. of the highway pile number signs. For example, highway pile number signs are usually rectangular, mostly white or yellow, and have a specific font size range.

[0070] Step 3: Use the template matching algorithm to search for areas matching the feature template in the image after edge detection, and determine the areas where the matching degree exceeds a set threshold (set as T) as suspected stake mark areas;

[0071] Assume the template image is T(x,y), the image to be matched is S(x,y), and the matching degree is

[0072]

[0073] The above technical solution is explained as follows: First, the Canny edge detection algorithm is used to process the image, and the edge is determined by calculating the gradient of each pixel in the image. This method can accurately extract the edge information in the image and provide a clear outline basis for the subsequent identification of the pile number. Secondly, a feature template is constructed based on the prior knowledge of the shape, color and size of the highway pile number. This template construction method based on actual conditions enables more targeted identification of the pile number in the subsequent matching process, reducing the possibility of misjudgment. Finally, the template matching algorithm is used to search for areas that match the feature template in the image after edge detection, and the suspected pile number identification area is determined by setting the matching threshold. This method can efficiently screen out possible pile number identification areas in a large amount of image data, improving the accuracy and efficiency of recognition.

[0074] In the further design, in step S4, the coordinates of the edge points of the image are set to (x e ,y e );

[0075] Extract the road surface point cloud from the LiDAR point cloud data, and cluster the point clouds belonging to the road surface together through a clustering algorithm;

[0076] Calculate the normal vector of the point cloud, and determine the point whose normal vector is close to the preset highway plane normal vector as a highway surface point;

[0077] Assume that the point cloud clustering algorithm uses the density-based DBSCAN algorithm. For a point P(x, y, z), if the number of points within its radius r exceeds the set threshold n t , then these points are classified into the same category;

[0078] Combine IMU attitude data and LiDAR point cloud data to build a local space geometry model;

[0079] Taking the current position of the drone as the origin, the laser radar point cloud coordinates are transformed into the local coordinate system according to the IMU attitude angle; let point P(x l ,y l ,z l ) in the local coordinate system (xll ,y ll ,z ll )The conversion formula is:

[0080]

[0081] The above technical solution is explained as follows: By extracting the highway pavement point cloud from the LiDAR point cloud data and using a clustering algorithm for classification, the highway pavement points can be accurately identified, which helps to accurately analyze and judge the highway conditions in the future, such as pavement disease detection. Secondly, the density-based DBSCAN algorithm is used for point cloud clustering. This algorithm can effectively process irregularly distributed data points and can accurately group points belonging to the same category in complex point cloud data, reducing the possibility of misclassification. Finally, a local space geometric model is constructed by combining IMU attitude data and LiDAR point cloud data, and a transformation formula for points in the local coordinate system is given. This method can accurately establish a local space model and provide accurate spatial position information for subsequent highway inspections, positioning and other tasks.

[0082] In the further design, in step S5, each local space geometric model is spliced ​​and integrated to form a global space model of the highway. The specific method is as follows:

[0083] By finding the overlapping areas between adjacent local models, the adjacent models are aligned and spliced ​​using the feature point matching algorithm;

[0084] Assume that adjacent local models M1 and M2 are used, extract SIFT feature points in the overlapping area, and obtain matching point pairs (P m1 ,P m2 ), use the least squares method to calculate the transformation matrix T between the models, and transform M2 into a coordinate system consistent with M1;

[0085] Determine the position of the reference points with known pile numbers in the global model. By setting reference points with obvious features and known pile numbers on the road in advance, use the data collected by the drone to determine the coordinates (x rp ,y rp ,z rp );

[0086] For any position P(x, y, z) during the drone inspection process, calculate the spatial geometric relationship between it and the known reference point; let the reference point P rp (x rp ,y rp ,z rp ), calculate the spatial distance d between two points:

[0087] and the angle θ between the connecting line and the direction of the kilometer road (obtained through vector calculation);

[0088] According to the known reference point stake number N rp , the actual road length L between the reference points rp And the calculated distance d and angle θ, determine the stake number N of the current position;

[0089] The formula is: Where ΔN is the difference in pile numbers between adjacent reference points.

[0090] The above technical solution is explained as follows: By splicing and integrating each local spatial geometric model to form a global spatial model of the highway, the overall spatial structure of the highway can be fully reflected. Secondly, the feature point matching algorithm and the least squares method are used to align adjacent local models. This method can accurately integrate different local models together, avoid the misalignment problem in the model splicing process, and ensure the accuracy of the global model. Finally, the pile number is determined by calculating the spatial geometric relationship between the drone inspection position and the known pile number reference point.

[0091] A device for determining a highway pile number during a drone highway inspection process, comprising:

[0092] Data collection module: the drone flies according to the preset inspection route and collects various data during the flight;

[0093] The pile number identification area feature extraction and recognition module uses edge detection algorithm to perform edge detection on the preprocessed image and extract edge information in the image;

[0094] A data processing module processes the data in the data acquisition module;

[0095] The feature extraction and local model building module uses edge detection algorithms to extract edge features from preprocessed optical images, and then identifies possible edge contours of highway edges, lane lines, and pile number marks through shape analysis for modeling.

[0096] Global model integration and pile number determination module: various local spatial geometric models are spliced ​​and integrated to form a global spatial model of the highway.

[0097] In summary, the present invention comprehensively utilizes a variety of sensor data of drones, including optical images, laser radar data, and flight attitude data, etc., and realizes accurate determination of highway pile numbers by constructing a unique spatial geometric model and data fusion algorithm. First, different types of data collected by drones are preprocessed to extract key features; then local spatial geometric relationships are constructed based on feature points, and then gradually expanded to a global highway model; finally, the highway pile number is obtained by matching and calculating in combination with known reference point information, which can effectively overcome the shortcomings of existing technologies, improve the accuracy, reliability and adaptability of pile number determination, and provide strong technical support for highway inspection, maintenance and other work. Multi-sensor fusion improves data accuracy and integrity, and reduces the impact of environmental interference on a single sensor. For example, in areas with weak GPS signals, laser radar and image features can assist in positioning. Secondly, intelligent algorithms realize automatic feature extraction and matching, reduce manual intervention, and improve efficiency. The constructed spatial model can adapt to highways of different terrains, and can accurately determine pile numbers whether in mountainous areas or plains.

[0098] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for determining highway pile numbers during unmanned aerial vehicle highway inspection, characterized in that: The following steps are involved: Step S1, data collection: The UAV flies according to the preset inspection route. During the flight, the high-definition camera on board is used to shoot high-definition images and GPS data of the road surface and surrounding areas at a fixed frame rate (frames / second). Specifically, the high-definition optical camera, lidar sensor and inertial measurement unit (IMU) are turned on. The optical camera shoots highway images at a fixed time interval t1 to obtain image information of the road surface, roadside facilities and possible pile number identification. The lidar emits a laser beam at a time interval t2 to scan the highway and surrounding terrain to obtain three-dimensional point cloud data and record the spatial coordinates (x l ,y l ,z l As well as reflection intensity information, the IMU records the flight attitude data of the drone in real time, including pitch angle α, yaw angle β, roll angle γ and acceleration information (a x ,a y ,a z ); Step S2, extraction and recognition of features of the pile number identification area: using the Canny edge detection algorithm to perform edge detection on the preprocessed image to extract edge information in the image; Step S3, acquisition and processing of drone flight trajectory data: The drone has a built-in high-precision inertial measurement unit (IMU) and global positioning system (GPS). During the flight, it records its own position information (longitude lon) at a fixed time interval (set as Δt). i , latitude lat i 、Altitude h i ), attitude information (pitch angle φ i , yaw angle ψ i , roll angle θ i ) ; Smooth the acquired position data and use the Kalman filter algorithm to remove noise and outliers in the data; Assume that the system state equation is x k =Ax k-1 +Bu k +w k, The observation equation is z k =Hx k +v k , where x k is the system state vector, A is the state transfer matrix, B is the control input matrix, u k is the control input vector, w k is the system noise vector, z k is the observation vector, H is the observation matrix, v k To observe the noise vector, the state estimation value is continuously updated through the Kalman filter recursive formula to obtain the smoothed position data; Step S4, feature extraction and local model construction: from the pre-processed optical image, edge features in the image are extracted using an edge detection algorithm, and then possible edge contours of highway edge lines, lane lines, and pile number markings are identified through shape analysis; Step S5, global model integration and stake number determination: each local space geometric model is spliced ​​and integrated to form a global space model of the highway.

2. The method for determining highway pile numbers during unmanned aerial vehicle highway inspection according to claim 1 is characterized in that: In step S2, the specific method of image preprocessing is as follows: Grayscale the collected image and convert the color image into a grayscale image; Assume that the RGB values ​​of the pixel (x, y) in the original color image are R(x, y), G(x, y), and B(x, y); Then the gray value I(x,y) calculation formula is: I(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y); Perform Gaussian filtering on the grayscale image to remove noise interference in the image; Let the filtered image be G(x,y) and the Gaussian filter function be h(x,y), then Let the filtered image be G(x,y) and the Gaussian filter function be h(x,y), then Among them, G(x,y) is the filtered image, I(x,y) is the original image, h(s,t) is the Gaussian filter function, k is the filter window radius, σ is the Gaussian standard deviation; Filter the laser radar point cloud data to remove points below the ground and noise points; The slope between adjacent points is calculated through the slope-based filtering algorithm, and points with abnormally large or small slopes are determined as noise points and removed. Suppose two adjacent points are P1(x1, y1, z1) and P2(x2, y2, z2); For IMU data, a complementary filtering algorithm is used to fuse the accelerometer and gyroscope data to improve the accuracy of attitude angle measurement; Assume the angular velocity measured by the gyroscope is ω g , the acceleration measured by the accelerometer is a a, The fused attitude angle θ f =(1-k)θ f +kθ g , where θ g is the attitude angle calculated by the gyroscope, and k is the fusion coefficient.

3. The method for determining highway pile numbers during unmanned aerial vehicle highway inspection according to claim 2 is characterized in that: In step S3, the Canny edge detection algorithm is used to perform edge detection on the preprocessed image, and the specific method for extracting edge information in the image is as follows: Assume that the image gradient amplitude is M(x,y), the gradient direction is θ(x,y), and calculate the gradient of each pixel in the image to determine the edge; Construct feature templates based on prior knowledge of the shape, color, size, etc. of highway pile number signs; The template matching algorithm is used to search for the area matching the feature template in the image after edge detection, and the area whose matching degree exceeds the set threshold (set as T) is determined as the suspected stake mark area; Assume the template image is T(x,y), the image to be matched is S(x,y), and the matching degree is 4. The method for determining highway pile numbers during unmanned aerial vehicle highway inspection according to claim 3 is characterized in that: In step S4, let the image edge point coordinates be (x e ,y e ); Extract the road surface point cloud from the LiDAR point cloud data, and cluster the point clouds belonging to the road surface together through a clustering algorithm; Calculate the normal vector of the point cloud, and determine the point whose normal vector is close to the preset highway plane normal vector as a highway surface point; Assume that the point cloud clustering algorithm uses the density-based DBSCAN algorithm. For a point P(x, y, z), if the number of points within its radius r exceeds the set threshold n t , then these points are classified into the same category; Combine IMU attitude data and LiDAR point cloud data to build a local space geometry model; Taking the current position of the drone as the origin, the laser radar point cloud coordinates are transformed into the local coordinate system according to the IMU attitude angle; let point P(x l ,y l ,z l ) in the local coordinate system (x ll ,y ll ,z ll )The conversion formula is:

5. The method for determining highway pile numbers during unmanned aerial vehicle highway inspection according to claim 4, characterized in that: In step S5, each local space geometric model is spliced ​​and integrated to form a global space model of the highway. The specific method is as follows: By finding the overlapping areas between adjacent local models, the adjacent models are aligned and spliced ​​using the feature point matching algorithm; Assume that adjacent local models M1 and M2 are used, extract SIFT feature points in the overlapping area, and obtain matching point pairs (P m1 ,P m2 ), use the least squares method to calculate the transformation matrix T between the models, and transform M2 into a coordinate system consistent with M1; Determine the position of the reference points with known pile numbers in the global model. By setting reference points with obvious features and known pile numbers on the road in advance, use the data collected by the drone to determine the coordinates (x rp ,y rp ,z rp ); For any position P(x, y, z) during the drone inspection process, calculate the spatial geometric relationship between it and the known reference point; let the reference point P rp (x rp ,y rp ,z rp ), calculate the spatial distance d between two points: and the angle θ between the connecting line and the direction of the kilometer road (obtained through vector calculation); According to the known reference point stake number N rp , the actual road length L between the reference points rp And the calculated distance d and angle θ, determine the stake number N of the current position; The formula is: Where ΔN is the difference in pile numbers between adjacent reference points.

6. A device for determining highway pile numbers during drone highway inspection, characterized in that: include: Data collection module: the drone flies according to the preset inspection route and collects various data during the flight; The pile number identification area feature extraction and recognition module uses edge detection algorithm to perform edge detection on the preprocessed image and extract edge information in the image; A data processing module processes the data in the data acquisition module; The feature extraction and local model building module uses edge detection algorithms to extract edge features from preprocessed optical images, and then identifies possible edge contours of highway edges, lane lines, and pile number marks through shape analysis for modeling. Global model integration and pile number determination module: various local spatial geometric models are spliced ​​and integrated to form a global spatial model of the highway.

Images