Image acquisition method for three-dimensional point cloud reconstruction and component identification driving of overhead line system

Abstract

The invention provides an image acquisition method for three-dimensional point cloud reconstruction and component identification driving of a contact network, and particularly relates to the field of image acquisition of the contact network, and the method comprises the steps: constructing a three-dimensional point cloud model of a contact network structure through a two-dimensional laser radar and a high-precision odometer which are installed on a pushing device; identifying a work support, a carrier cable, a dropper, an electric connection and a positioning point through clustering and principal component analysis, and obtaining key parameters such as a guide height, a pull-out value and a support column limit; and in combination with camera view field parameters, a geometric model is established, a pitch angle and a yaw angle are solved simultaneously, and the camera is controlled to complete image acquisition at the optimal angle and position. Through the method provided by the invention, self-adaptive shooting control under structure identification driving can be realized, complete and clear imaging of key components is ensured, blind shooting and repeated shooting are avoided, and the method has the advantages of high precision, high efficiency and strong real-time performance, and is suitable for a complex overhead line system inspection scene.

Description

Technical Field

[0001] The present invention relates to the field of contact network image acquisition, and in particular to an image acquisition method driven by contact network three-dimensional point cloud reconstruction and component recognition. Background Art

[0002] In rail transit electrification systems, image acquisition of the catenary structure is crucial for ensuring status awareness of power supply equipment and diagnosing structural defects. Existing technologies typically use high-definition cameras, binocular vision systems, or manual inspections to capture images of the catenary before a train passes or during downtime. These images are then combined with post-processing image analysis to identify the position and status of key components such as droppers, catenary cables, electrical connections, and arm systems. Sometimes, trackside-mounted LiDAR or cameras are also used for rough ranging, assisting with image positioning or modeling.

[0003] However, these existing methods commonly suffer from the following problems: First, due to the independent processes of image acquisition and structural recognition, there is a lack of real-time perception and guidance of the target component's structural features during the capture process, resulting in frequent image occlusion, off-focus shots, and missing key areas. Second, image acquisition relies on fixed or preset viewing angles, making it difficult to adaptively adjust the field of view and shooting posture based on the component's actual spatial state, affecting image clarity and effectiveness. Furthermore, when targeting complex structures (such as multiple types of pillars and unusually shaped mounting components), traditional methods exhibit significant lags in viewpoint regulation and composition control, resulting in technical bottlenecks such as a large number of redundant images, high post-processing pressure, and poor real-time performance. These issues hinder the improvement of image accuracy, the reduction of false positives, and the improvement of inspection efficiency, and a stable and effective closed-loop acquisition mechanism has yet to be established. Summary of the Invention

[0004] The present invention proposes an image acquisition method driven by three-dimensional point cloud reconstruction and component recognition of contact networks, aiming to improve the passive process in the existing technology of blindly taking images with a fixed perspective and then using image recognition to assist positioning.

[0005] Among them, a contact network three-dimensional point cloud reconstruction and component recognition driven image acquisition method includes the following steps: S1. Using a 2D laser radar installed on the pusher to scan the plane, combined with displacement information obtained by the odometer, the cross-sectional point clouds of consecutive frames are registered and superimposed according to the track's travel direction to construct 3D point cloud data of the contact network area. S2. Perform neighbor cluster analysis on multi-frame point cloud data, identify linear components based on principal component direction extraction (PCA), and then identify the work supports, catenary cables, droppers, and electrical connection components based on the results. The positioning points, along with their lead height and pullout values, are then identified based on their relative spatial characteristics to the track center. S3. Analyze the cluster distribution of points on the supporting supports and load-bearing cables. If a positioning point exists, obtain its coordinates based on its spatial distribution and connectivity in the point cloud. Determine the pillar type based on the continuity of the point cloud segments within the spatial segment where the positioning point is located. Based on this type identification, calculate the pillar lateral clearance distance using the point cloud morphology surrounding the positioning point. S4. Set the camera's viewing angle parameters, build a model from the camera's imaging center to the target structure based on spatial geometry, and calculate the pitch and yaw angles using this model combined with the pillar's side clearance distance. S5. Based on the calculated angle parameter, the distance between the camera and the cantilever system is calculated. Based on the calculated distance between the camera and the cantilever system, the propulsion device is controlled to move back or forward to acquire images.

[0006] The preferred implementation process of the above solution is as follows: 2D lidar is used to acquire a cross-sectional point cloud of the catenary structure on the track cross section. Successive point clouds are then registered and overlaid based on equipment displacement using an onboard high-precision odometry, thereby reconstructing the actual structure distribution of the catenary in three-dimensional space. Spatial clustering and principal component analysis are performed on the point cloud data to identify typical catenary components, such as work supports, load-bearing cables, droppers, and electrical connections. The location coordinates, lead-out values (vertical distance), and pull-out values (lateral distance) of the positioning points are extracted. The point cloud density and continuity characteristics are then combined to determine the support type and structural boundary. After obtaining the structural position parameters, a spatial geometric model is constructed, combining the camera's field of view angle, focal length, and image composition requirements. A mapping relationship is established between the camera's field of view and the target structure boundary in both the horizontal and vertical directions. This mapping relationship forms two constraint formulas: first, the sum of the support limit and lateral redundancy is equivalenced with the camera's field of view coverage in the yaw direction to derive the yaw angle; second, the lead-out height minus the underlying structure coverage requirement is equivalenced with the imaging range in the pitch direction to derive the pitch angle. By combining these two geometric constraint formulas, the pitch and yaw angles are jointly calculated. Based on the camera's calculated angle, the system inversely calculates the spatial distance between the camera and the target structure at the time of capture. It then controls the propulsion device to automatically retract to that distance, while simultaneously adjusting the camera's posture to complete the capture. This achieves full-process control, in which the 3D structure recognition results are used to reversely drive the camera's viewing angle adjustment and capture behavior.

[0007] Furthermore, the above scheme uses point cloud analysis results as a driving factor to reversely determine the shooting attitude and position, establishing a closed-loop control mechanism with structure as the driving factor and image as the response. Specifically, in steps S1 to S3, a dense 3D model of the catenary structure is dynamically constructed based on the 2D LiDAR transverse profile point cloud and the longitudinal displacement of the propulsion path. PCA is used to extract linear features and then integrate the spatial topology of the positioning points and the support context. This not only identifies the positional state of key components but also forms a precise quantitative representation of the target area's geometric parameters. This process ensures that the core variables of image acquisition no longer rely on visual perception or template matching, but instead establishes a coordinate control chain based on physical measurements of high-precision point clouds. In step S5, the spatial distance from the camera to the arm is derived using known viewing angle conditions, and combined with the odometry value to accurately retract the control device, achieving a combined closed-loop control of "attitude, position, and framing," eliminating the three types of systematic errors in traditional image acquisition: incomplete structure, angle offset, and distance error.

[0008] The beneficial effects of the invention are: (1) Based on the identified spatial parameters such as the positioning point height guide value, pull-out value, and pillar side limit, the present invention establishes a geometric mapping model between the camera imaging field of view and the target structure. The pitch angle and yaw angle are solved simultaneously using the simultaneous angle solution formula, so that the camera imaging focal plane can cover the target area to the greatest extent and fall into the optimal depth of field area, thereby ensuring image clarity and key structural integrity; (2) The method proposed in the present invention can automatically identify various pillar types and adjust the shooting redundant parameters according to the type changes. It has the ability to adapt to the structural morphology and is particularly suitable for the precise shooting needs of special-shaped contact network components in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A method flow chart of an image acquisition method driven by three-dimensional point cloud reconstruction and component recognition of a contact network provided by an embodiment of the present invention; Figure 2 A schematic diagram of the imaging focal plane and field of view provided by an embodiment of the present invention; Figure 3 A schematic diagram of the visible area and geometric dimensions of the cantilever system captured by a camera according to an embodiment of the present invention; Figure 4 A schematic diagram of a 3D space captured by a camera provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0011] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention and are not intended to limit the present invention. That is, the embodiments described herein are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and illustrated in the drawings herein may be arranged and designed in various different configurations.

[0012] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work shall fall within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0013] Furthermore, the terms "comprises," "comprising," or any other variations thereof are intended to cover a non-exclusive inclusion such that a process, method, article, or machine that comprises a list of elements includes not only those elements but also other elements not expressly listed or inherent to such process, method, article, or machine. In the absence of more limitations, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article, or machine that comprises the element.

[0014] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0015] Among them, such as Figure 1 , a contact network three-dimensional point cloud reconstruction and component recognition driven image acquisition method, comprising: S1. Using a 2D laser radar installed on the pusher to scan the plane, combined with displacement information obtained by the odometer, the cross-sectional point clouds of consecutive frames are registered and superimposed according to the track's travel direction to construct 3D point cloud data of the contact network area. S2. Perform neighbor cluster analysis on multi-frame point cloud data, identify linear components based on principal component direction extraction (PCA), and then identify the work supports, catenary cables, droppers, and electrical connection components based on the results. The positioning points, along with their lead height and pullout values, are then identified based on their relative spatial characteristics to the track center. S3. Analyze the cluster distribution of points on the supporting supports and load-bearing cables. If a positioning point exists, obtain its coordinates based on its spatial distribution and connectivity in the point cloud. Determine the pillar type based on the continuity of the point cloud segments within the spatial segment where the positioning point is located. Based on this type identification, calculate the pillar lateral clearance distance using the point cloud morphology surrounding the positioning point. S4. Set the camera's viewing angle parameters, build a model from the camera's imaging center to the target structure based on spatial geometry, and calculate the pitch and yaw angles using this model combined with the pillar's side clearance distance. S5. Based on the calculated angle parameter, the distance between the camera and the cantilever system is calculated. Based on the calculated distance between the camera and the cantilever system, the propulsion device is controlled to move back or forward to acquire images.

[0016] Specifically, a detailed implementation process for step S1 is proposed: Based on a 2D lidar, its laser beam non-contactly scans the surrounding contours within a plane. The distance to the object being measured is determined by the laser's emission-reception duration, or time-of-flight. Adding this to the laser beam's orientation yields a point in polar coordinates at the radar's center, creating a cross-sectional point cloud. A propulsion device travels on the track surface. Based on the coordinate relationship between the radar and the propulsion device, the center of the track surface can be converted to its origin. The device is equipped with a high-precision odometer that records the distance traveled along the track. By adding the distance traveled perpendicular to the cross-sectional direction to the two-dimensional cross-sectional point cloud, 3D point cloud data for the contact network space is generated.

[0017] Specifically, a detailed implementation process for step S2 is proposed: Based on the nearest neighboring points in the point cloud, multiple frames of points are clustered. The PCA algorithm is then used to calculate the linear relationship between the clustered points to determine whether they form a straight line. The direction vector is then determined to determine whether it is along the track. The pullout value and the height guide are then added to determine the catenary support and catenary cables. Furthermore, the point cloud data on the support and catenary cables is analyzed, and components such as droppers and electrical connections are identified based on the spatial distribution and characteristics of the points.

[0018] Specifically, for step S3, a detailed implementation process is proposed: analyze the cluster distribution of points on the support line. When a positioning point exists, the position of the positioning point can be determined based on the spatial distribution and connection relationship of the positioning point point cloud, thereby obtaining the positioning point coordinates. By comprehensively judging multiple consecutive frames near the positioning point, the side limit of the pillar, the arm base point and its coordinates can be obtained. At the same time, the pillar type can be determined based on the pillar cross-section. The calculated key points are visualized in two-dimensional projection space and 3D space. Because the pillar point cloud is continuous, this type of pillar is a narrow and long H-shaped steel column or concrete column. If the point cloud at the pillar is locally continuous and mesh-like, it is a steel frame structure, and the steel frame structure width is wider. In addition, the identification and calculation of the above-mentioned positioning points, suspension string electrical connection components, side limits, arm base and other key points, as well as the pillar type identification, only require the historical point cloud data of the current frame and the data within the next 5 frames, which is a real-time algorithm. In addition, the real-time algorithm meets the following requirements: the hardware configuration is a multi-core processor (≥4 cores, main frequency ≥2.4GHz) and memory ≥8GB; the data processing frame rate is ≥20fps, and the single-frame processing delay is ≤50ms.

[0019] Furthermore, the step S1 specifically includes the following sub-steps: S101. A two-dimensional laser radar is installed at the front end of the push device, emitting a laser beam in the horizontal plane at a fixed scanning frequency; S102. Obtaining a polar coordinate point cloud profile of the surrounding environment based on the flight time of the laser emission and the reflected echo; S103. The displacement data of the device along the track direction is recorded by a high-precision odometer installed on the pushing device, and the continuous frame cross-sectional point cloud is spatially registered and superimposed according to the displacement; S104. Reconstructing a 3D point cloud dataset of the catenary system in the track center coordinate system.

[0020] Furthermore, the step S2 specifically includes the following sub-steps: S201. Clustering the three-dimensional point cloud dataset according to spatial proximity to obtain multiple local structure fragments; S202 performs principal component analysis on each clustering result, extracts its main direction vector, and determines the structure type, which includes linear and nonlinear structures; S203. Obtain the height range and lateral position coordinates spanned by the linear structure identified as a result. Combined with the track center reference, extract the spatial pullout value and lead height value of the corresponding linear structure. Combined with the criteria for determining catenary support and catenary cables, determine that the linear structure that meets the criteria is a catenary support and catenary cable. S204. For nonlinear structures, by analyzing the minimum spatial distance between the center point of the corresponding nonlinear structure and the linear structure, the consistency of the connection direction, and the geometric regularity of the point cloud morphology, combined with the criteria for determining accessory components of the contact network, nonlinear structures that meet the criteria are determined to be accessory components. The accessory components include at least droppers and electrical connection components.

[0021] Furthermore, in step S3, the specific judgment process for judging the pillar type according to the continuity of the point cloud segment of the space segment where the positioning point is located is as follows: When the point cloud is dense and the pillar shape is narrow, long and continuous, it is determined to be an H-shaped steel column or a concrete column; When the point cloud is locally distributed in a wide grid pattern, it is determined to be a steel frame support.

[0022] Specifically, a judgment basis is proposed: the point cloud continuity threshold of the H-shaped steel column is ≥95%, and the width range is 0.2-0.5m; the steel frame structure must meet the grid density ≥10 points / m² and the node spacing ≤0.3m.

[0023] Furthermore, in step S4, the viewing angle parameters include at least the horizontal viewing angle, the vertical viewing angle, the imaging focal length, the field of view extension distance, and the downward distance, wherein the imaging focal length is the distance between the focal plane and the imaging center. Specifically, the downward distance refers to the target space distance extending downward from the identified positioning point, indicating the depth of the lower structure area that needs to be included in the image. The downward distance is used to determine the lower limit of the vertical coverage of the imaging field of view, which is used to calculate the pitch angle of the camera.

[0024] Furthermore, in step S4, the pitch angle is calculated and yaw angle The specific process is as follows: S401. The pillar side limit distance and the field of view extension distance are calculated and summed to obtain the side boundary distance; and calculate the horizontal projection of the camera focal plane center point and the horizontal boundary projection corresponding to half of the camera viewing angle; S402. The lateral limit distance is equivalent to: the value of the lateral projection of the center point of the camera focal plane on the horizontal plane, and the horizontal distance covered by the focal plane angle of view on one side, and the pitch angle relationship is obtained, and the pitch angle relationship is used as the lateral constraint condition; S403. Using the guide height value of the positioning point as a reference, extend downward a set distance, and use the difference between the guide height value and the distance as the minimum observation height that the camera needs to cover. S404. Based on the minimum observation altitude being equivalent to the difference between the vertical projection of the camera imaging focus in the direction of the principal axis and the projection width of the field of view at the lower boundary of the camera focal plane, a yaw angle relationship is obtained, and the yaw angle relationship is used as a vertical constraint. S405. According to the pitch angle relationship and the yaw angle relationship, the pitch angle is constrained by the joint constraint. and yaw angle Perform the solution.

[0025] Furthermore, the lateral constraint condition is specifically expressed as: ; Among them, the Indicates the side limit distance of the pillar, Indicates the extended field of view distance. represents the imaging focal length, represents the pitch angle, represents the yaw angle, Represents the horizontal viewing angle, that is, the horizontal viewing angle parameter of the camera lens, in degrees.

[0026] Furthermore, the vertical constraint condition is specifically expressed as: ; Among them, the Indicates the guide height value of the positioning point. Indicates the depth of the probe. Represents the vertical viewing angle, that is, the vertical viewing angle parameter of the camera lens, in degrees.

[0027] Furthermore, in step S5, the specific calculation process of the distance L0 between the camera and the arm system is expressed as follows: ; Among them, the Represents the distance between the camera and the arm system. represents the imaging focal length, represents the pitch angle, Indicates the yaw angle.

[0028] Furthermore, in step S5, the specific process of controlling the pushing device to retreat / advance and perform image acquisition is as follows: controlling the pushing device to retreat to the T-L0 position, where T is the track mileage value of the positioning point, and simultaneously driving the pan / tilt head to adjust the camera pitch angle and yaw angle to the target values and perform image acquisition operations to complete image acquisition of the structure on one side; after the acquisition is completed, the device continues to move forward to T+L0 and sets the yaw angle to 180°-θ to complete image acquisition of the structure on the other side.

[0029] Example 2 Furthermore, as a preferred implementation of the above embodiment, a specific control process for taking photos is proposed for steps S3-S4: 1. During travel, when components such as suspension strings and electrical connections are detected, the left and right cameras are triggered to take photos.

[0030] 2. During travel, the vehicle identifies the pillars and determines their type. Lateral clearances are calculated based on the inner surface of the pillars. A positioning point is also identified. If a positioning point is detected, the left and right cameras are triggered to take photos. If lateral clearances are not yet calculated, the vehicle continues driving for approximately one meter. If the pillars are properly suspended, lateral clearances are calculated.

[0031] The following parameters are set: the horizontal viewing angle hfov of the camera lens, the vertical viewing angle vfov, the pre-adjusted lens focal length, and the distance L between the focal plane and the imaging center. Figure 2 The side limit of the support where the arm to be measured is C, the guide height of the positioning point is H, and the pull-out value is S. The guide height H and the pull-out value S are the vertical and horizontal distances of the positioning point relative to the center of the track. Figure 3 To capture the entire support, the visible area needs to be extended outward by a distance of E. Assume that the camera's shooting angle parameters are: the yaw angle, that is, the horizontal swing angle of the camera facing the arm is , the pitch angle is the angle between the camera looking up and the horizontal plane. , see Figure 4 , where F is the vertical projection point of the center point in the Z-axis direction, which is used to define the guide height H (i.e., the vertical distance of the center point relative to the track centerline); A is the horizontal projection point of the center point after being projected onto the track centerline plane in the Z-axis direction, which is used to define the pull-out value S (i.e., the horizontal distance of the center point relative to the track centerline); the track centerline is the reference baseline for all parameter measurements; the camera is the measurement starting point; and the center point is the key feature point on the measured arm.

[0032] 3. Build the corresponding model: The center point position (x, y) is ( ) The width and height of the projection area of the focal plane on the arm plane are: ; Left and right fields of view: Refer to the diagram of the visible area and geometric dimensions of the cantilever system captured by the camera. The distance between the leftmost field of view and the center of the track is C+E, so: ; The field of view at the upper and lower ends, with the pitch angle adjusted according to the elevation guide of the positioning point, covers the upper and lower arms of the contact network. Usually, based on the elevation guide of the positioning point, the downward distance D (in practical applications, D is 1.5 meters) is: ; According to the above two equations, the pitch angle is solved , and then solve the yaw angle .

[0033] Furthermore, as a preferred implementation of the above embodiment, in actual use, because the cantilever system has certain specifications, D and E are usually 1.5 meters. For steel frame type pillars, the pillar volume is large, and E is 2 meters.

[0034] 4. After calculating the camera's pitch and yaw angles, further calculate the camera's shooting position from the arm. Right now: , according to the mileage T of the positioning point, the control device returns to Position, you can stop to shoot, to improve efficiency, while retreating, control the camera gimbal system to adjust the pitch and yaw angles to the corresponding values. After taking the photo, it will automatically move forward until the mileage is At this position, the pitch angle of the gimbal remains unchanged, while the yaw angle is set to 180°- , that is, the symmetrical position, to shoot the other side of the arm-cantilever system.

[0035] Furthermore, as a preferred implementation of the above embodiment, if part of the arm is installed through a sling or soft suspension, the side limit cannot be calculated. In this case, the default value of 3.1 meters (this value is based on the standard limit setting in TB / T 3329-2013 "Electrified Railway Contact Network Design Code", and technical personnel in this field can set this value according to actual conditions) can be used.

[0036] 4. LiDAR identifies and locates key components of the overhead catenary system, guiding and controlling the angle and position of the camera unit so that the entire object being photographed falls within the optimal depth of field on the focal plane, ensuring clarity. This also ensures that the camera's field of view encompasses the key areas of the arm and support, ensuring optimal framing.

[0037] Example 3 As a preferred implementation of the above embodiment, an image acquisition device driven by three-dimensional point cloud reconstruction and component recognition of a contact network is proposed, comprising: 2D laser radar, configured to scan the catenary outline in a plane in a non-contact manner, and obtain polar coordinate point cloud data of the object under test through laser time of flight; The push device is installed on the track surface and has a built-in high-precision odometer. It is used to move along the track direction and record the distance traveled. It converts the polar coordinate point cloud data of the lidar into a three-dimensional space point cloud with the center of the track surface as the origin. The data processing module is used to cluster multi-frame point clouds, calculate the linear relationship of cluster points through the principal component analysis (PCA) algorithm, determine whether they extend along the track direction, and identify the working support, load-bearing cables, droppers, and electrical connection components based on the pull-out value and height guide. The camera system, including left and right cameras and a gimbal, is configured to trigger shooting when a target component is identified, and dynamically adjust the camera's pitch and yaw angles based on the positioning point height guide, pull-out value, and pillar side limits so that the shooting area covers the key structures of the arm system; The control module is used to control the push device to retract to the shooting position according to the mileage of the positioning point, and synchronously adjust the pan / tilt angle to complete bilateral shooting.

[0038] Furthermore, the scanning frequency of the 2D laser radar is 10-50Hz, the angular resolution is ≤0.1°, the ranging accuracy is ±2mm, and the effective detection range is 0.1-30m; the installation height of the laser radar and the propulsion device is 1.2-1.8m, and the inclination angle is 0-5°.

[0039] Furthermore, the triggering mechanism of the camera system includes: triggering photo taking when the point cloud clustering confidence of the suspension string or electrical connection component is ≥90%; the positioning point recognition must include at least 3 consecutive frames of point cloud data, and the spatial distribution error must be ≤2cm; the camera exposure time is automatically adjusted according to the ambient light intensity, and a global shutter is used to eliminate motion blur.

[0040] Example 4 Based on the above embodiment, a preferred implementation method is proposed, which introduces the field of view redundancy coefficient and the safety shooting coefficient to enable the horizontal condition constraint and the vertical condition constraint to have a robust adjustment mechanism, effectively improving the edge fault tolerance and target integrity of image acquisition, and is suitable for complex scenes such as different pillar types, uneven point cloud density, and structural positioning errors. Specifically, by introducing the field of view redundancy coefficient and the safety shooting coefficient β, so that the horizontal condition constraints and vertical condition constraints are transformed into: ; ； in, , used to adapt to factors such as pillar position error, structural deviation, camera vibration, etc., by default =1.2, which can increase the horizontal fault tolerance bandwidth by 20%; , used to adjust the pitch field of view coverage height. A slightly larger value (such as 1.2) is more suitable for shooting in the interference area of ​​hanging strings and electrical connection points, ensuring that the lower boundary content is captured completely. and They are used to adjust the boundary redundancy of the imaging area to adapt to various pillar shapes, viewing angle deviations and acquisition jitter conditions.

[0041] The foregoing description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be construed as excluding other embodiments. Rather, the present invention can be used in various other combinations, modifications, and environments and can be modified within the scope of the concept described herein through the above teachings or techniques or knowledge in the relevant field. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention are intended to be protected by the appended claims.

Claims

1. A method for image acquisition driven by three-dimensional point cloud reconstruction and component recognition of a contact network, characterized in that: include: S1. Using a 2D lidar installed on the pusher to scan the surface, combined with displacement information obtained by the odometer, the cross-sectional point clouds of consecutive frames are registered and overlaid along the track's travel direction to construct 3D point cloud data of the catenary area. S2. Perform neighbor cluster analysis on multi-frame point cloud data, identify linear components based on principal component direction extraction (PCA), and then identify the work supports, catenary cables, droppers, and electrical connection components based on the results. The positioning points, along with their lead height and pullout values, are then identified based on their relative spatial characteristics to the track center. S3. Analyze the cluster distribution of points on the supporting supports and load-bearing cables. If a positioning point exists, obtain its coordinates based on its spatial distribution and connectivity in the point cloud. Determine the pillar type based on the continuity of the point cloud segments within the spatial segment where the positioning point is located. Based on this type identification, calculate the pillar lateral clearance distance using the point cloud morphology surrounding the positioning point. S4. Set the camera's viewing angle parameters, build a model from the camera's imaging center to the target structure based on the spatial geometry, and calculate the angle parameters using the model combined with the pillar's side limit distance. S5. Based on the calculated angle parameter, the distance between the camera and the cantilever system is calculated. Based on the calculated distance between the camera and the cantilever system, the propulsion device is controlled to move back or forward to acquire images.

2. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: The step S1 specifically includes the following sub-steps: S101. A two-dimensional laser radar is installed at the front end of the push device, emitting a laser beam in the horizontal plane at a fixed scanning frequency; S102. Obtaining a polar coordinate point cloud profile of the surrounding environment based on the flight time of the laser emission and the reflected echo; S103. The displacement data of the device along the track direction is recorded by a high-precision odometer installed on the pushing device, and the continuous frame cross-sectional point cloud is spatially registered and superimposed according to the displacement; S104. Reconstructing a 3D point cloud dataset of the catenary system in the track center coordinate system.

3. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S201. Clustering the three-dimensional point cloud dataset according to spatial proximity to obtain multiple local structure fragments; S202 performs principal component analysis on each clustering result, extracts its main direction vector, and determines the structure type, which includes linear and nonlinear structures; S203. Obtain the height range and lateral position coordinates spanned by the linear structure, combine it with the track center reference, extract the spatial pullout value and guide height value of the corresponding linear structure, and determine that the linear structure that meets the standards is the catenary support and catenary cable; S204. For nonlinear structures, by analyzing the minimum spatial distance between the center point of the corresponding nonlinear structure and the linear structure, the consistency of the connection direction, and the geometric regularity of the point cloud morphology, nonlinear structures that meet the standards are determined to be accessory components. The accessory components include at least droppers and electrical connection components.

4. The method for image acquisition driven by three-dimensional point cloud reconstruction and component recognition of a contact network according to claim 1, characterized in that: In step S3, the specific process of determining the pillar type based on the continuity of the point cloud segments of the spatial segment where the positioning point is located is as follows: When the point cloud is dense and the pillar shape is narrow, long and continuous, it is determined to be an H-shaped steel column or a concrete column; When the point cloud is locally distributed in a wide mesh pattern, it is determined to be a steel frame support.

5. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: In step S4, the viewing angle parameters include at least a horizontal viewing angle, a vertical viewing angle, an imaging focal length, a field of view extension distance, and a downward distance, wherein the imaging focal length is the distance between the focal plane and the imaging center.

6. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: In step S4, the pitch angle is calculated and yaw angle The specific process is as follows: S401. The pillar side limit distance and the field of view extension distance are calculated and summed to obtain the side boundary distance; and calculate the horizontal projection of the camera focal plane center point and the horizontal boundary projection corresponding to half of the camera viewing angle; S402. The lateral limit distance is equivalent to: the value of the lateral projection of the center point of the camera focal plane on the horizontal plane, and the horizontal distance covered by the focal plane angle of view on one side, and the pitch angle relationship is obtained, and the pitch angle relationship is used as the lateral constraint condition; S403. Using the guide height value of the positioning point as a reference, extend downward a set distance, and use the difference between the guide height value and the distance as the minimum observation height that the camera needs to cover. S404. Based on the minimum observation altitude being equivalent to the difference between the vertical projection of the camera imaging focus in the direction of the principal axis and the projection width of the field of view at the lower boundary of the camera focal plane, a yaw angle relationship is obtained, and the yaw angle relationship is used as a vertical constraint. S405. According to the pitch angle relationship and the yaw angle relationship, the pitch angle is constrained by the joint constraint. and yaw angle Perform the solution.

7. The method for image acquisition driven by three-dimensional point cloud reconstruction and component recognition of a contact network according to claim 6, characterized in that: The lateral constraint condition is specifically expressed as: ； Among them, the Indicates the side limit distance of the pillar, Indicates the extended field of view distance. represents the imaging focal length, represents the pitch angle, represents the yaw angle, Represents the horizontal viewing angle, that is, the horizontal viewing angle parameter of the camera lens, in degrees.

8. The method for image acquisition driven by three-dimensional point cloud reconstruction and component recognition of a contact network according to claim 6, characterized in that: The vertical constraint condition is specifically expressed as: ； Among them, the Indicates the guide height value of the positioning point. Indicates the depth of the probe. Represents the vertical viewing angle, that is, the vertical viewing angle parameter of the camera lens, in degrees.

9. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: In step S5, the specific calculation process of the distance L0 between the camera and the arm system is expressed as follows: ； Among them, the Represents the distance between the camera and the arm system. represents the imaging focal length, represents the pitch angle, Indicates the yaw angle.

10. The image acquisition method for catenary 3D point cloud reconstruction and component recognition drive according to claim 1, characterized in that: In step S5, the specific process of controlling the pushing device to retreat / advance and perform image acquisition is as follows: controlling the pushing device to retreat to the T-L0 position, where T is the track mileage value of the positioning point, and simultaneously driving the pan / tilt head to adjust the camera pitch angle and yaw angle to the target values and perform image acquisition operations to complete image acquisition of the structure on one side; after the acquisition is completed, the device continues to advance to T+L0 and sets the yaw angle to 180°-θ to complete image acquisition of the structure on the other side.

Images