Multi-point synchronous accurate hoisting and positioning method and device for bridge precast beam

Abstract

The invention relates to the technical field of data processing, and discloses a multi-point synchronous accurate hoisting positioning method and device for a bridge precast beam. The method comprises the following steps: optimizing three-dimensional scene data to obtain target environment data, performing path planning to obtain multi-point hoisting path data, and performing optimization processing on the path data to obtain an optimized hoisting path and attitude change data; performing real-time visual tracking processing on the precast beam to obtain real-time position and attitude data, and performing filtering processing on the real-time position and attitude data to obtain filtered pose data; and performing comparative analysis on the pose data, the optimized hoisting path data and the optimized pose change data to obtain control instruction data, and performing cooperative control on the multi-point hoisting system according to the control instruction data. The posture of the large precast beam can be accurately and efficiently controlled, the positioning precision is improved, and the requirement for rapid and efficient hoisting is met.

Description

Technical Field

[0001] The present application relates to the field of image data processing, and in particular to a method and device for multi-point synchronous precise lifting and positioning of prefabricated bridge beams. Background Art

[0002] In recent years, the application of precast bridge beams has become increasingly widespread, and their installation process requires extremely high precision. Traditional methods of hoisting precast bridge beams mainly rely on manual operation and simple mechanical equipment such as cranes and slings. These methods usually use single-point or double-point hoisting, and the operator controls the position and posture of the precast beam through vision and experience. In some complex hoisting environments, measuring instruments such as total stations or GPS devices are also used to assist in positioning.

[0003] However, these traditional methods have many shortcomings. First, manual operation is easily affected by subjective factors, and it is difficult to ensure high accuracy and consistency. Second, simple mechanical equipment is difficult to cope with complex lifting environments, such as small spaces or sites with obstacles. Furthermore, the single-point or double-point lifting method is not accurate enough for the posture control of large prefabricated beams, which can easily cause the prefabricated beams to shake and deviate. Finally, although measuring instruments can improve positioning accuracy, their use process is cumbersome and lacks real-time performance, making it difficult to meet the needs of fast and efficient lifting. Summary of the invention

[0004] The present application provides a method and device for multi-point synchronous precise lifting and positioning of prefabricated bridge beams, which is used to improve the technical problems such as the difficulty in ensuring high precision and consistency through manual operation in traditional methods, the lack of precise posture control of large prefabricated beams by single-point or double-point lifting methods, which easily causes shaking and deviation of prefabricated beams, and the cumbersome use of measuring instruments, lack of real-time performance, and low efficiency.

[0005] In the first aspect, the present application provides a method for the multi-point synchronous precise lifting and positioning of a bridge precast beam, the method comprising: collecting image data of the bridge precast beam and its surrounding environment, and preprocessing the image data to obtain preprocessed image data; performing three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and accurately calibrating the three-dimensional data to obtain three-dimensional data of the target precast beam; performing scene reconstruction processing on the preprocessed image data to obtain three-dimensional scene data of the lifting environment, and optimizing the three-dimensional scene data to obtain target environment data. data; performing path planning through the target precast beam three-dimensional data and the target environment data to obtain path data for multi-point lifting, and optimizing the path data to obtain optimized lifting path and posture change data; performing real-time visual tracking processing on the precast beam to obtain real-time position and posture data of the precast beam, and filtering the real-time position and posture data to obtain filtered posture data; comparing and analyzing the posture data with the optimized lifting path data and the posture change data to obtain control instruction data, and performing collaborative control on the multi-point lifting system according to the control instruction data.

[0006] In a second aspect, the present application provides a multi-point synchronous precise hoisting and positioning device for a prefabricated bridge beam, the multi-point synchronous precise hoisting and positioning device for a prefabricated bridge beam comprising:

[0007] An acquisition module is used to acquire image data of the precast bridge beam and its surrounding environment, and preprocess the image data to obtain preprocessed image data;

[0008] A calibration module, used for performing three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and accurately calibrating the three-dimensional data to obtain three-dimensional data of the target precast beam;

[0009] A reconstruction module, used for performing scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimizing the three-dimensional scene data to obtain target environment data;

[0010] A planning module, used to perform path planning through the target precast beam three-dimensional data and the target environment data, obtain path data for multi-point hoisting, and optimize the path data to obtain optimized hoisting path and posture change data;

[0011] A tracking module is used to perform real-time visual tracking processing on the prefabricated beam to obtain real-time position and posture data of the prefabricated beam, and filter the real-time position and posture data to obtain filtered position and posture data;

[0012] The control module is used to compare and analyze the posture data with the optimized lifting path data and the posture change data to obtain control instruction data, and to coordinately control the multi-point lifting system according to the control instruction data.

[0013] In the technical solution provided by the present application, by collecting image data of the precast bridge beam and its surrounding environment and performing preprocessing, high-quality preprocessed image data can be obtained, laying the foundation for subsequent 3D reconstruction and scene analysis, and improving the data accuracy of the entire lifting process; the preprocessed image data is subjected to 3D reconstruction processing to obtain 3D data of the precast beam and perform precise calibration, so as to obtain high-precision 3D data of the target precast beam, and provide reliable geometric information for lifting path planning and posture control; the preprocessed image data is subjected to scene reconstruction processing to obtain 3D scene data of the lifting environment and perform optimization, so as to generate accurate target environment data, which is helpful to identify potential obstacles and plan the optimal lifting path; the path planning is performed through the 3D data of the target precast beam and the target environment data, so as to obtain the path data of multi-point lifting. And through optimization processing, the optimized lifting path and posture change data are obtained, which can not only ensure the safety of the lifting process, but also improve the lifting efficiency; real-time visual tracking processing is performed on the precast beams to obtain the real-time position and posture data of the precast beams, and the filtered posture data is obtained through filtering processing, which can realize real-time and accurate monitoring of the position and posture of the precast beams, and provide timely and accurate feedback information for the control system; the posture data is compared and analyzed with the optimized lifting path data and posture change data to obtain the control command data, and the multi-point lifting system is coordinated and controlled according to the control command data, realizing closed-loop control of the lifting process, greatly improving the lifting accuracy and stability, realizing the intelligence and precision of the lifting process of precast bridge beams, significantly improving the efficiency and safety of lifting operations, and reducing human errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0015] Figure 1 This is a schematic diagram of an embodiment of a method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a multi-point synchronous precise lifting and positioning device for precast bridge beams in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiment of the present application provides a method and device for the precise lifting and positioning of a prefabricated bridge beam at multiple points in synchronization. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams includes:

[0019] Step S101, collecting image data of a precast bridge beam and its surrounding environment, and preprocessing the image data to obtain preprocessed image data;

[0020] Step S102, performing three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and accurately calibrating the three-dimensional data to obtain three-dimensional data of the target precast beam;

[0021] Step S103, performing scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimizing the three-dimensional scene data to obtain target environment data;

[0022] Step S104: performing path planning through the target precast beam three-dimensional data and the target environment data to obtain the path data of multi-point hoisting, and optimizing the path data to obtain the optimized hoisting path and posture change data;

[0023] Step S105: performing real-time visual tracking processing on the prefabricated beam to obtain real-time position and posture data of the prefabricated beam, and filtering the real-time position and posture data to obtain filtered position and posture data;

[0024] Step S106: compare and analyze the position and posture data with the optimized lifting path data and posture change data to obtain control instruction data, and coordinately control the multi-point lifting system according to the control instruction data.

[0025] It is understandable that the execution subject of the present application can be a multi-point synchronous precise hoisting and positioning device for prefabricated bridge beams, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0026] Specifically, image data of the bridge precast beam and its surrounding environment are collected, and a high-resolution camera array is used to take all-round photos of the precast beam and the surrounding environment. The acquired raw image data undergoes preprocessing steps such as denoising, contrast enhancement, and distortion correction to obtain high-quality preprocessed image data. During the preprocessing process, an adaptive median filter is used to remove noise, image clarity is improved by histogram equalization, and geometric correction is performed using pre-calibrated camera parameters. The preprocessed image data is subjected to three-dimensional reconstruction. This step uses the structured beam method for sparse reconstruction, and then a dense point cloud is generated by a multi-view stereo algorithm. The point cloud data is converted into a triangular mesh model through meshing, and then the surface is smoothed to obtain the preliminary three-dimensional data of the precast beam. To improve the accuracy, this three-dimensional data is compared and calibrated with the CAD design drawings of the precast beam, and finally high-precision three-dimensional data of the target precast beam is obtained.

[0027] The preprocessed image data is reconstructed. The key elements in the scene, including terrain, obstacles, and lifting equipment, are identified using a deep learning-based instance segmentation algorithm. The ground is separated from the point cloud using the RANSAC plane fitting algorithm, and the non-ground part is clustered to identify the main obstacles. The obstacle model is merged with the ground data, and the graph cut algorithm is used for global optimization to obtain accurate target environment data. Multi-point lifting path planning is performed based on the target precast beam 3D data and target environment data. First, the optimal lifting point position is determined, and the configuration space is constructed and discretized into a grid map. The improved rapidly expanding random tree (RRT*) algorithm is used to search for feasible paths in the configuration space, considering factors such as path length, stability, and safety margin. The initial path is smoothed using a B-spline curve, and the posture change of the precast beam during the lifting process is calculated. At the same time, an anti-sway algorithm based on model predictive control is designed to obtain the optimized lifting path and posture change data.

[0028] During the hoisting process, the prefabricated beams are tracked visually in real time. The improved FAST corner detection algorithm is used to detect feature points, the feature descriptors are extracted by the ORB algorithm, and the KNN algorithm is used for feature matching. The RANSAC algorithm is used to eliminate false matches and improve tracking stability. The 6-DOF pose of the prefabricated beam is estimated in real time based on the EPnP algorithm combined with the Kalman filter. The stable tracking of large-scale motion is achieved by constructing a multi-scale image pyramid, and the local bundle adjustment optimization is periodically performed to correct the accumulated error, and finally the filtered high-precision pose data is obtained. The pose data obtained in real time is compared and analyzed with the optimized hoisting path data and pose change data. The position and pose deviations are calculated to construct the state error vector. The error state is analyzed using the model predictive control algorithm, and the nonlinear control law is designed. The torque distribution of each lifting point is optimized by the quadratic programming algorithm to ensure load balance. The torque distribution data is dynamically constrained to obtain the final control instruction data, and the coordinated control of the multi-point hoisting system is realized.

[0029] For example, in the process of hoisting a prefabricated beam with a span of 50 meters, five high-definition cameras are first used to collect image data from different angles, with each camera collecting 100 images, for a total of 500 original images. After preprocessing, 500 high-quality images are obtained after denoising and correction. In the 3D reconstruction process, about 1 million feature points are extracted, and about 500,000 sparse point clouds are obtained by the structured beam method. After dense reconstruction, a dense point cloud of 3 million points is obtained. After gridding, a model containing 1 million triangular facets is generated. After comparison and calibration with the CAD drawing, a 3D model of the prefabricated beam with millimeter-level accuracy is finally obtained. During the scene reconstruction process, the ground, four major obstacles and two hoisting equipment are identified. In the path planning stage, the configuration space is discretized into a 100x100x100 three-dimensional grid map, and the RRT* algorithm is used to generate the initial path within 0.5 seconds. After optimization, a smooth hoisting path is obtained, which contains 1,000 path points, each of which contains position and posture information.

[0030] During the real-time tracking process, about 1,000 feature points were detected in each frame of the image, and the matching success rate reached 85%. Through the EPnP algorithm and Kalman filtering, millimeter-level position accuracy and 0.1-degree attitude accuracy were achieved. Finally, through comparative analysis and collaborative control, the precise positioning of the precast beams was achieved, and the positioning error of the entire hoisting process was controlled within ±5 mm, and the attitude error was controlled within ±0.5 degrees, greatly improving the accuracy and efficiency of the hoisting of precast bridge beams.

[0031] In the embodiment of the present application, by collecting image data of the precast bridge beam and its surrounding environment and performing preprocessing, high-quality preprocessed image data can be obtained, laying the foundation for subsequent 3D reconstruction and scene analysis, and improving the data accuracy of the entire lifting process; the preprocessed image data is subjected to 3D reconstruction processing to obtain 3D data of the precast beam and perform accurate calibration, so as to obtain high-precision 3D data of the target precast beam, and provide reliable geometric information for lifting path planning and posture control; the preprocessed image data is subjected to scene reconstruction processing to obtain 3D scene data of the lifting environment and optimize it, so as to generate accurate target environment data, which is helpful to identify potential obstacles and plan the optimal lifting path; path planning is performed through the 3D data of the target precast beam and the target environment data, so as to obtain path data for multi-point lifting, and The optimized lifting path and posture change data are obtained through optimization processing, which can not only ensure the safety of the lifting process, but also improve the lifting efficiency; the precast beams are subjected to real-time visual tracking processing to obtain the real-time position and posture data of the precast beams, and the filtered posture data are obtained through filtering processing, which can realize real-time and accurate monitoring of the position and posture of the precast beams, and provide timely and accurate feedback information for the control system; the posture data are compared and analyzed with the optimized lifting path data and posture change data to obtain the control command data, and the multi-point lifting system is coordinated and controlled according to the control command data, realizing closed-loop control of the lifting process, greatly improving the lifting accuracy and stability, realizing the intelligence and precision of the lifting process of precast bridge beams, significantly improving the efficiency and safety of lifting operations, and reducing human errors.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) photographing the precast bridge beam at multiple angles to obtain image data, and denoising the image data to obtain denoised image data;

[0034] (2) performing contrast enhancement processing on the denoised image data to obtain enhanced image data, and performing distortion correction on the enhanced image data to obtain corrected image data;

[0035] (3) extracting feature points from the corrected image data to obtain feature point data, and performing feature matching on the feature point data to obtain matched feature point pair data;

[0036] (4) using the RANSAC algorithm to remove outliers from the matched feature point pair data to obtain screened feature point pair data, and registering the corrected image data based on the screened feature point pair data to obtain registered image data;

[0037] (5) performing color balancing processing on the registered image data to obtain color-balanced image data, and performing image fusion on the color-balanced image data to obtain fused panoramic image data;

[0038] (6) Segmenting the fused panoramic image data using an adaptive threshold segmentation algorithm to obtain segmented image region data, and performing edge detection on the segmented image region data to obtain edge feature data;

[0039] (7) performing shape analysis on the edge feature data to obtain target contour data, and performing straight line and circle detection on the target contour data using a Hough transform algorithm to obtain geometric feature data;

[0040] (8) performing target recognition on the fused panoramic image data according to the geometric feature data to obtain target area data, and performing size measurement on the target area data to obtain target size data;

[0041] (9) performing texture analysis on the target area data to obtain texture feature data, and performing material identification on the target area data based on the texture feature data and the target size data to obtain material information data;

[0042] (10) Semantically annotating the fused panoramic image data according to the material information data, target size data, and geometric feature data to obtain preprocessed image data.

[0043] Specifically, the prefabricated beams are photographed from multiple angles, and the image data of the prefabricated beams are captured from different angles using a high-resolution camera array. The acquired raw image data is then denoised, and an adaptive median filter is used to remove noise to improve image quality. Next, the denoised image data is contrast enhanced, and the histogram equalization technique is used to improve the overall clarity of the image. The enhanced image data also needs to be distorted, and the deformation caused by the wide-angle lens is eliminated using pre-calibrated camera parameters to obtain the rectified image data. Feature point extraction of the rectified image data is the next key step, and the SIFT (Scale Invariant Feature Transform) algorithm is used to detect and describe local features in the image. The SIFT algorithm is characterized by good invariance to image scaling, rotation, and brightness changes. The extracted feature points are then feature matched, and the nearest neighbor ratio method is used to match the feature points between different images to obtain matched feature point pair data.

[0044] In order to improve the accuracy of feature matching, the RANSAC (Random Sample Consistency) algorithm is used to remove outliers from the matched feature point pair data. The RANSAC algorithm randomly selects sample sets through iterations, builds models and verifies them, and finally finds the optimal model and inlier set. After removing outliers, the remaining feature point pair data are used to align the corrected image data to ensure the accurate correspondence between images from different perspectives. The aligned image data is color balanced, and the gray world algorithm is used to adjust the overall tone of the image to eliminate color differences under different lighting conditions. The color-balanced image data is fused, and the multi-resolution spline fusion algorithm is used to seamlessly stitch multiple images to generate high-quality panoramic image data.

[0045] The panoramic image data is then processed by an adaptive threshold segmentation algorithm, which dynamically adjusts the threshold according to the grayscale distribution of the local area of ​​the image to effectively segment the target area in the image. The segmented image area data is edge detected, and the Canny edge detection algorithm is used to extract edge features to obtain clear edge feature data. Shape analysis is performed on the edge feature data to extract the target contour data, and the Hough transform algorithm is used to detect geometric features such as straight lines and circles. The Hough transform effectively identifies the geometric shapes in the image by converting the image space into the parameter space to obtain geometric feature data.

[0046] Based on the obtained geometric feature data, the fused panoramic image data is used for target recognition, and the position of the precast beam in the image is accurately located. The size of the identified target area data is measured, and the actual size data of the target is obtained by using the conversion relationship from pixels to actual size. The target area data also needs to be texture analyzed, and the gray level co-occurrence matrix (GLCM) is used to extract texture features, including statistics such as energy, contrast, and correlation. Combining the texture feature data and the target size data, the target area data is used for material recognition, and the support vector machine (SVM) classifier is used to distinguish different materials to obtain material information data.

[0047] Finally, the material information data, target size data and geometric feature data are integrated to semantically annotate the fused panoramic image data, give semantic information to each area in the image, and finally obtain the preprocessed image data.

[0048] For example, in the process of hoisting a prefabricated beam with a span of 40 meters, six high-definition cameras were first used to shoot from different angles, and each camera collected 50 images with a 4K resolution (3840x2160 pixels). In the denoising process, the window size of the adaptive median filter was set to dynamically adjust between 3x3 and 7x7, effectively removing 95% of the salt and pepper noise. In the contrast enhancement stage, histogram equalization expanded the dynamic range of the image from the original [0,220] to [0,255], improving the image contrast by about 20%. In the feature point extraction process, an average of about 8,000 SIFT feature points were extracted from each image. In the feature matching stage, the nearest neighbor ratio threshold of 0.7 was used, and about 3,000 pairs of feature points were initially matched. The RANSAC algorithm was iterated 200 times, successfully eliminating about 500 pairs of erroneous matches and retaining 2,500 pairs of high-quality matching points. These matching points are used for image registration, and the registration accuracy reaches the sub-pixel level.

[0049] Color balancing processing adjusted the average grayscale value of the image to 128, eliminating about 15% of the color difference under different viewing angles. Image fusion generated a high-resolution panoramic image of 12800x7200 pixels. The adaptive threshold segmentation algorithm used a local window of 11x11 to successfully segment the contour of the precast beam with a segmentation accuracy of 98%. The low and high thresholds of Canny edge detection were set to 50 and 150 respectively, extracting clear edge features. The accumulator threshold of Hough transform was set to 100 in parameter space, and 20 main edge lines and 8 circular lifting holes of the precast beam were detected. Target recognition accurately located the precast beam, and the measurement error was controlled within ±5 mm. In texture analysis, the step size of GLCM was set to 1, and the directions were 0°, 45°, 90° and 135°, extracting 10 texture features such as energy, contrast, and correlation. The SVM classifier used the RBF kernel function to successfully identify the concrete material of the precast beam with an accuracy of 99%. The final semantic annotation provides accurate semantic information for key elements such as precast beams, lifting equipment, obstacles, etc. in the panoramic image.

[0050] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0051] (1) Extracting feature points from the preprocessed image data to obtain feature point cloud data, and performing feature matching on the feature point cloud data to obtain matched feature point pair data;

[0052] (2) Sparsely reconstructing the matched feature point pair data through the structured beam method to obtain sparse point cloud data, and densely reconstructing the sparse point cloud data to obtain dense point cloud data;

[0053] (3) Meshing the dense point cloud data to obtain triangular mesh model data, and performing surface smoothing on the triangular mesh model data to obtain three-dimensional data of the precast beam;

[0054] (4) calibrating the dimensions of the three-dimensional data of the precast beam according to the design drawing data of the precast beam to obtain calibrated three-dimensional data, and optimizing the surface details of the calibrated three-dimensional data to obtain refined three-dimensional data;

[0055] (5) Texture mapping is performed on the refined three-dimensional data to obtain three-dimensional data with texture, and lighting simulation is performed on the three-dimensional data with texture to obtain three-dimensional data of the target precast beam.

[0056] Specifically, feature points are extracted from the preprocessed image data, and the SIFT (Scale Invariant Feature Transform) algorithm is used to detect and describe local features in the image. The SIFT algorithm constructs a Gaussian difference pyramid, detects extreme points in different scale spaces, and calculates the main direction of the feature points, and finally generates a 128-dimensional feature descriptor. These feature points constitute feature point cloud data, which are then matched. The nearest neighbor ratio method is used to match feature points between different images to obtain matched feature point pair data. The matched feature point pair data is sparsely reconstructed by the structured beam method. The structured beam method is a technology that simultaneously optimizes camera parameters and three-dimensional point coordinates, and its core is to minimize the reprojection error. The reprojection error is minimized through an iterative optimization algorithm (such as the Levenberg-Marquardt algorithm) to obtain sparse point cloud data. The sparse point cloud data is then densely reconstructed using the Multi-View Stereo (MVS) algorithm. The MVS algorithm calculates the disparity map between images and uses the triangulation principle to restore three-dimensional information. A commonly used MVS method is Patch-based Multi-View Stereo (PMVS), the core idea of ​​which is to iteratively expand and optimize small plane patches in three-dimensional space.

[0057] The dense point cloud data is meshed and the triangular mesh model data is generated using the Poisson surface reconstruction algorithm. The algorithm converts the point cloud into an implicit function by solving the Poisson equation, and then uses an isosurface extraction algorithm (such as MarchingCubes) to generate a triangular mesh. The triangular mesh model data is then surface smoothed, and the Laplace smoothing algorithm is used to eliminate the noise and irregularities on the mesh surface to obtain the preliminary three-dimensional data of the precast beam. To improve the accuracy, the three-dimensional data of the precast beam is compared and calibrated with the design drawing data of the precast beam. The calibration process uses the Iterative Closest Point (ICP) algorithm to achieve alignment by minimizing the distance between the three-dimensional data and the design drawing. The calibrated three-dimensional data is then optimized for surface details, and the subdivision surface algorithm (such as Loop subdivision) is used to increase the detail level of the model to obtain refined three-dimensional data.

[0058] Finally, texture mapping is performed on the refined 3D data to map the texture information in the original image to the surface of the 3D model. This process involves the calculation of UV coordinates and the generation of texture images. After texture mapping, the textured 3D data is subjected to lighting simulation, and a global illumination algorithm (such as ray tracing or radiosity) is used to simulate the appearance of the precast beam under real lighting conditions, ultimately obtaining highly realistic 3D data of the target precast beam.

[0059] For example, in the 3D reconstruction process of a precast beam with a span of 50 meters, SIFT feature points were first extracted from high-resolution images (8192x6144 pixels each) taken from 10 different angles. On average, about 15,000 feature points were extracted from each image, totaling about 150,000 feature points. The feature matching stage used a nearest neighbor ratio threshold of 0.8 and successfully matched about 60,000 pairs of feature points. The structured beam optimization process was iterated 50 times, and the final reprojection error was reduced to less than 0.5 pixels, resulting in about 20,000 sparse 3D points. The PMVS algorithm was used in the dense reconstruction stage, and the minimum number of patch images was set to 3, and about 5 million dense point clouds were finally generated. The meshing process used Poisson surface reconstruction with an octree depth of 10 to generate an initial mesh model containing about 1 million triangles. Laplace smoothing was iterated 10 times with a weight factor of 0.5, which effectively eliminated surface noise.

[0060] When compared with the design drawings, the ICP algorithm was iterated 30 times, and the final average alignment error was less than 2 mm. The surface detail optimization used the Loop subdivision algorithm for 2 iterations, increasing the number of triangles to about 4 million, greatly improving the detail performance of the model. Texture mapping uses a 4096x4096 resolution texture map, and UV unfolding uses a minimum stretching algorithm to ensure uniform texture distribution. Lighting simulation uses physically based rendering (PBR) technology, taking into account the material's diffuse reflection, highlight reflection, and roughness properties, and simulates the appearance of precast beams under different time and weather conditions. The final target precast beam 3D data has reached a high level in both geometric accuracy and visual realism, providing a reliable digital foundation for subsequent precise lifting and positioning.

[0061] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0062] (1) performing semantic segmentation on the preprocessed image data to obtain scene element segmentation data, and performing instance segmentation on the scene element segmentation data to obtain scene instance data;

[0063] (2) performing three-dimensional reconstruction on the scene instance data to obtain three-dimensional scene data of the hoisting environment, and performing plane extraction on the three-dimensional scene data of the hoisting environment to obtain ground plane data;

[0064] (3) performing scene stratification on the three-dimensional scene data of the hoisting environment according to the ground plane data to obtain stratified scene data, and performing cluster analysis on the non-ground part of the stratified scene data to obtain obstacle clustering data;

[0065] (4) performing three-dimensional modeling on the obstacle clustering data to obtain obstacle three-dimensional model data, and merging the obstacle three-dimensional model data with the ground plane data to obtain merged scene data;

[0066] (5) Perform global optimization on the merged scene data to obtain the target environment data.

[0067] Specifically, semantic segmentation is performed on the preprocessed image data, and deep learning models such as DeepLab v3+ are used to classify the images at the pixel level. The model uses an encoder-decoder architecture, combined with dilated convolution and dilated spatial pyramid pooling techniques to effectively capture multi-scale contextual information. Semantic segmentation divides each pixel in the image into predefined categories, such as ground, obstacles, prefabricated beams, hoisting equipment, etc., to obtain scene element segmentation data. Subsequently, instance segmentation is performed on the scene element segmentation data, and the Mask R-CNN algorithm is used to identify and segment independent object instances in the scene to obtain scene instance data. The scene instance data is reconstructed in three dimensions using the Structure from Motion (SfM) technology based on multi-view geometry. The SfM algorithm estimates the camera pose and three-dimensional point coordinates by matching feature points in images from different perspectives to generate a sparse point cloud. The Multi-View Stereo (MVS) algorithm is then used to encrypt the sparse point cloud to obtain dense three-dimensional scene data of the hoisting environment. Plane extraction is performed on these three-dimensional scene data, and the RANSAC (Random Sample Consensus) algorithm is used to fit the plane model. RANSAC randomly samples data points, iteratively fits the plane model and evaluates the number of inliers to finally obtain the best fitting ground plane data.

[0068] According to the ground plane data, the three-dimensional scene data of the hoisting environment is layered. This process calculates the distance from each point to the ground plane and sets a threshold to distinguish between ground points and non-ground points to obtain layered scene data. Cluster analysis is performed on the non-ground part of the layered scene data, and the density clustering algorithm DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is used. DBSCAN clusters based on the density distribution of points. It does not require the number of clusters to be specified in advance. It can effectively identify clusters of any shape and finally obtain obstacle clustering data. The obstacle clustering data is three-dimensionally modeled, and the Alpha Shapes algorithm is used to reconstruct the geometric shape of the obstacle. AlphaShapes is a generalization of Delaunay triangulation. By adjusting the α parameter to control the degree of reconstruction, it can effectively capture the outline and detailed features of the obstacle and obtain the obstacle three-dimensional model data. These obstacle three-dimensional model data are merged with the previously extracted ground plane data to obtain complete scene data including the ground and obstacles.

[0069] Finally, the merged scene data is globally optimized. This step uses graph optimization technology to construct a factor graph to represent the spatial relationship and constraints between the elements in the scene. The g2o (General Graph Optimization) library is used to implement nonlinear least squares optimization, while considering multiple factors such as camera pose, 3D point coordinates, and plane parameters, minimizing reprojection errors and geometric constraint errors to obtain highly accurate target environment data.

[0070] For example, at a hoisting site for a 60-meter-span precast beam, 200 environmental images with 4K resolution (3840x2160) were first collected. The DeepLab v3+ model was used in the semantic segmentation stage, pre-trained on the Cityscapes dataset, and then fine-tuned on the labeled hoisting scene dataset. The model outputs 14 categories, including ground, precast beams, cranes, brackets, buildings, etc. The average pixel accuracy of semantic segmentation reached 95%. Mask R-CNN was used for instance segmentation, pre-trained on the COCO dataset, and fine-tuned on the hoisting scene data, successfully identifying 20 independent object instances in the scene, including 2 cranes, 1 precast beam, 5 brackets, and 12 debris. During the 3D reconstruction process, the SfM algorithm processed 200 images, extracted about 1 million SIFT feature points, successfully matched 800,000 pairs of feature points, and reconstructed 300,000 sparse 3D points. The MVS algorithm encrypted the point cloud to 10 million points. The plane was extracted using the RANSAC algorithm, with the distance threshold set to 0.1 m and 1000 iterations. The ground plane was successfully extracted with a fitting error of less than 0.05 m.

[0071] The scene layering divides the point cloud into two layers: ground and non-ground, and sets the height threshold to 0.3 meters. DBSCAN clustering is performed on the non-ground point cloud, with the neighborhood radius ε = 0.5 meters and the minimum number of points MinPts = 100, and 23 obstacle clusters are successfully identified. Each obstacle cluster is 3D modeled using the Alpha Shapes algorithm, and the α value is dynamically adjusted according to the size of the obstacle, ranging from 0.1 to 1 meter. The final global optimization constructs a factor graph containing 200 camera pose nodes, 300,000 3D point nodes, and 23 obstacle model nodes. The g2o library is used for nonlinear least squares optimization, with the maximum number of iterations set to 100 and the convergence threshold set to 1e-6. The average reprojection error of the optimized scene data is reduced to 0.8 pixels, and the average position error of the 3D points is less than 1 cm. This high-precision environment reconstruction provides a reliable digital foundation for subsequent path planning and hoisting operations, significantly improving the accuracy and safety of precast beam hoisting.

[0072] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0073] (1) Determine the optimal lifting point position of multi-point lifting according to the three-dimensional data of the target precast beam, obtain the lifting point position data, and construct the configuration space according to the lifting point position data and the target environment data to obtain the discretized grid map data;

[0074] (2) Define the path cost function, obtain the cost function parameter data, and use the improved fast expansion random tree algorithm to perform path search on the discretized grid map data to obtain the path data of multi-point lifting;

[0075] (3) performing collision detection on the path data of multi-point lifting to obtain collision-free path data, and smoothing the collision-free path data using a B-spline curve to obtain smoothed path data;

[0076] (4) calculating the posture change of the precast beam during the hoisting process according to the smooth path data to obtain posture change data, and designing an anti-sway control strategy according to the posture change data to obtain anti-sway control parameter data;

[0077] (5) The smooth path data, posture change data and anti-sway control parameter data are combined to obtain the optimized lifting path and posture change data.

[0078] Specifically, the finite element analysis method is used to consider the weight distribution, structural characteristics and stress conditions of the precast beam, and the hanging point position that can minimize stress concentration and deformation is calculated to obtain the hanging point position data. Subsequently, the configuration space is constructed based on the hanging point position data and the target environment data, and the three-dimensional space is discretized into a grid map. The resolution of the grid map is dynamically adjusted according to the size of the precast beam and the complexity of the environment, usually set to 0.1-0.5 meters, to obtain the discretized grid map data.

[0079] Next, we define the path cost function, taking into account factors such as path length, smoothness, and safety margin. The path cost function can be expressed as:

[0080] J=αL+βS+γC

[0081] Among them, J is the total cost, L is the path length, S is the path smoothness, C is the minimum distance to the obstacle, and α, β, and γ are weight coefficients. By adjusting these parameters, the cost function parameter data is obtained. Then, the improved rapidly expanding random tree (RRT*) algorithm is used to search the path of the discretized raster map data. The core improvement of the RRT* algorithm is the introduction of heuristic search and dynamic adjustment of the sampling area. The search process can be expressed as:

[0082] T=(V,E)

[0083] V={x init ,...,x goal}

[0084] E={(x i ,x j )|x i ,x j ∈V,x i ≠x j}

[0085] Among them, T is the constructed tree, V is the node set, E is the edge set, x init is the starting position, x goal is the target position. The algorithm generates random sampling points iteratively, connects the nearest nodes, and performs local rewiring and optimization to finally obtain the path data of multi-point lifting. Collision detection is performed on the path data of multi-point lifting, and the hierarchical bounding box (AABB tree) is used to accelerate the collision detection process. The collision detection takes into account the actual geometric shape of the precast beam and the range of motion of the lifting equipment to ensure that the path does not collide with obstacles during the entire process and obtain collision-free path data. Subsequently, the collision-free path data is smoothed by the B-spline curve. The expression of the B-spline curve is:

[0086]

[0087] Among them, C(t) is the point on the curve, N i,p (t) is the p-order B-spline basis function, P i is the control point. By adjusting the control point position and the basis function order, the smooth path data is obtained. The attitude change of the precast beam during the hoisting process is calculated based on the smooth path data. The attitude change q is expressed by quaternion:

[0088] q=[q w ,q x ,q y ,q z ]

[0089] Among them, q w is the real part, q x ,q y ,q z is the imaginary part. By interpolating the posture of discrete points on the path, continuous posture change data is obtained. Based on the posture change data, an anti-sway control strategy is designed. The anti-sway control adopts the model predictive control (MPC) method to predict the swing state in the future and optimize the control input to minimize the swing amplitude to obtain the anti-sway control parameter data. Finally, the smooth path data, posture change data and anti-sway control parameter data are combined to form a complete lifting execution plan, and the optimized lifting path and posture change data are obtained.

[0090] For example, in the process of hoisting a precast beam with a span of 80 meters and a weight of 500 tons, the four best hoisting point positions were first determined by finite element analysis, with the distances between the hoisting points being 25 meters, 30 meters, and 25 meters, respectively. The constructed configuration space uses a grid map with a resolution of 0.2 meters, and a total of 400x200x100 grid cells are generated. In the path cost function, the weight coefficients α=0.6, β=0.3, and γ=0.1 are set to balance the path length, smoothness, and safety. The RRT* algorithm sets the maximum number of iterations to 10,000 and the search radius to 5 meters. The algorithm generates a search tree containing 2856 nodes in 3.5 seconds and finally finds an initial path with a total length of 112 meters. The collision detection uses an AABB tree to simplify the precast beam into 50 AABB boxes, and the environmental obstacles are represented by 200 AABB boxes. The detection process takes 0.08 seconds to ensure that the minimum distance between the path and the obstacle is not less than 1 meter.

[0091] The B-spline curve smoothing uses cubic basis functions and selects 50 control points. The total length of the smoothed path increases to 115 meters, but the maximum curvature decreases by 40%. The attitude change calculation uses a sampling frequency of 100Hz to generate 11,500 attitude data points. In the anti-sway control strategy, the prediction time domain of MPC is set to 5 seconds, the control time domain is 1 second, and the optimization goal is to minimize the weighted sum of the swing angle and angular velocity. The final optimized lifting scheme includes 115 meters of smooth path data, 11,500 attitude data points and 5 sets of anti-sway control parameters (corresponding to 5 key stages). The entire lifting process is expected to take 23 minutes, and the maximum swing angle of the precast beam is controlled within ±0.5 degrees, which greatly improves the accuracy and safety of the lifting. This highly optimized path planning method provides reliable technical support for the precise lifting of precast bridge beams and significantly improves the installation efficiency and quality of large precast components.

[0092] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0093] (1) Setting visual marking points on the precast beams to obtain marking point data, and performing feature point detection on the real-time image using the improved FAST corner point detection algorithm to obtain real-time feature point data;

[0094] (2) extracting descriptors from real-time feature point data using the ORB algorithm to obtain feature descriptor data, and performing feature matching between the feature descriptor data and the marker point data using the KNN algorithm to obtain matching feature point data;

[0095] (3) Using the RANSAC algorithm to remove external points from the matching feature point data, the screened matching point data is obtained, and the pose estimation of the screened matching point data is performed based on the EPnP algorithm to obtain the real-time position and pose data of the precast beam;

[0096] (4) constructing a multi-scale image pyramid according to the real-time position and posture data of the precast beam to obtain multi-scale image data, and performing local bundle adjustment optimization on the multi-scale image data and the screened matching point data to obtain optimized posture data;

[0097] (5) The optimized posture data and the real-time position and posture data of the precast beam are fused and filtered through an extended Kalman filter to obtain filtered posture data.

[0098] Specifically, visual markers are set on the prefabricated beams. These markers are usually high-contrast, easily recognizable patterns, such as circles, squares, or specially designed QR codes. The positions of the markers are accurately measured and recorded to form marker data. Subsequently, the feature points of the real-time acquired images are detected by the improved FAST (Features from Accelerated Segment Test) corner detection algorithm. The improvement of the FAST algorithm is mainly reflected in the adaptive threshold setting and non-maximum suppression method, which improves the stability and distribution uniformity of the feature points. The detection process quickly identifies the corner points by comparing the brightness difference between the pixel point and the 16 pixels around it, and obtains the real-time feature point data. Next, the ORB (Oriented FAST and Rotated BRIEF) algorithm is used to extract the descriptor of the real-time feature point data. The ORB algorithm combines the improved FAST corner detection and the rotation-invariant BRIEF (Binary Robust Independent Elementary Features) descriptor, and has the characteristics of high computational efficiency and good rotation invariance. The ORB descriptor is a 256-bit binary string, which is generated by comparing the pixel pairs around the feature point to obtain the feature descriptor data. Then, the KNN (K-Nearest Neighbors) algorithm is used to perform feature matching on the feature descriptor data and the pre-set marker point data. The KNN algorithm searches for the K nearest neighbors of each query point in the feature space, usually K is set to 2, and the matching points are screened by comparing the distance ratio between the nearest neighbor and the next nearest neighbor to obtain the matching feature point data. In order to improve the accuracy of matching, the RANSAC (Random Sample Consensus) algorithm is used to remove outliers from the matching feature point data. The RANSAC algorithm uses the principle of random sampling consistency to iteratively select the minimum sample set, fit the model and evaluate the number of inliers, and finally find the optimal model and inlier set. This process effectively removes erroneous matching points and obtains the screened matching point data. Subsequently, the pose estimation of the screened matching point data is performed based on the EPnP (Efficient Perspective-n-Point) algorithm. The EPnP algorithm is an efficient PnP (Perspective-n-Point) problem solving method. By expressing 3D points as linear combinations of 4 control points, the problem is converted into a linear equation system for solution, and the real-time position and pose data of the precast beam are quickly obtained.

[0099] According to the real-time position and attitude data of the precast beam, a multi-scale image pyramid is constructed. Image pyramid is a multi-resolution image representation method obtained by downsampling the original image step by step. The image size of each layer is 1 / 4 of the previous layer. Usually 3-4 layers of pyramid are constructed to obtain multi-scale image data. The multi-scale image data and the filtered matching point data are used for local bundle adjustment optimization. Bundle adjustment is a technology that optimizes camera parameters and three-dimensional point coordinates at the same time, and improves the accuracy of pose estimation by minimizing the reprojection error. Local bundle adjustment only optimizes the data of the most recent frames, balancing computational efficiency and accuracy to obtain optimized pose data. Finally, the optimized pose data and the real-time position and attitude data of the precast beam are fused and filtered by the extended Kalman filter. The extended Kalman filter is a state estimation method for nonlinear systems, which includes two steps: prediction and update. The prediction step uses a motion model to predict the state at the next moment, and the update step uses observation data to correct the prediction result. In this way, the filter can smooth the pose estimation results, suppress noise and mutations, and obtain the final filtered pose data.

[0100] For example, during the hoisting process of a precast beam with a span of 90 meters and a weight of 600 tons, 20 visual markers were first set on the surface of the precast beam. These markers were black and white circular patterns with a diameter of 20 cm. A high-speed camera with a resolution of 4K (3840x2160 pixels) and a frame rate of 60fps was used to capture real-time images of the precast beam. The improved FAST corner detection algorithm detected an average of 2000 feature points in each frame of the image. The detection threshold was adaptively set between 10-30 pixel differences, and the non-maximum suppression window size was 3x3. The ORB algorithm extracted descriptors for these 2000 feature points, each of which was a 256-bit binary string. When performing KNN feature matching, the K value was set to 2, the distance ratio threshold was set to 0.7, and about 500 pairs of feature point pairs were initially matched. The RANSAC algorithm set the maximum number of iterations to 1000 and the inlier threshold to 2 pixels, and finally screened out 400 pairs of reliable matching points. The EPnP algorithm uses these 400 pairs of matching points to perform pose estimation. The solution process takes an average of 5 milliseconds, the position estimation accuracy reaches the centimeter level, and the pose estimation accuracy reaches 0.1 degrees.

[0101] The constructed image pyramid contains 4 layers with resolutions of 3840×2160, 1920×1080, 960×540 and 480×270. The local bundle adjustment optimization considers the latest 10 frames of data and uses the Ceres solver for nonlinear optimization. The number of iterations is set to 20 and the convergence threshold is 1e -6, the optimized reprojection error is reduced by 30% on average. The state vector of the extended Kalman filter contains position (x, y, z), attitude (quaternion representation) and their first-order derivatives, a total of 13 dimensions. The process noise covariance matrix and the observation noise covariance matrix are obtained by offline calibration. The filter runs at a frequency of 100 Hz, which effectively smooths the pose estimation results, reduces the position jitter to the millimeter level, and reduces the attitude jitter to within 0.01 degrees.

[0102] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0103] (1) Comparing the posture data with the optimized lifting path data to obtain position deviation data, and comparing the posture data with the attitude change data to obtain attitude deviation data;

[0104] (2) constructing a state error vector according to the position deviation data and the attitude deviation data to obtain error state data, and analyzing the error state data through a model predictive control algorithm to obtain predictive control data;

[0105] (3) designing a nonlinear control law based on the predictive control data to obtain control law parameter data, and optimizing the torque distribution according to the control law parameter data to obtain torque distribution data;

[0106] (4) Dynamically constrain the torque distribution data to obtain control instruction data, and coordinately control the actuators of each lifting point according to the control instruction data.

[0107] Specifically, the real-time posture data is compared with the optimized lifting path data to calculate the position deviation. This process can be expressed as:

[0108] ΔP=P r -P p

[0109] Where ΔP is the position deviation vector, P r is the actual position vector, P p is the position vector on the planned path. At the same time, the attitude information in the posture data is compared with the predetermined attitude change data to obtain the attitude deviation data. The attitude deviation is usually expressed by quaternion:

[0110]

[0111] Here, ΔQ is the attitude deviation quaternion, Q r is the actual attitude quaternion, Q p is the planned attitude quaternion, Represents quaternion multiplication. According to the position deviation data and attitude deviation data, the state error vector is constructed. The state error vector contains the position error, attitude error and their derivatives, which can be expressed as:

[0112]

[0113] Among them, Δω T is the transposed matrix of the equivalent angular velocity of the attitude error, and are the transposed matrices of the derivatives of the position error and attitude error, ΔP T is the transposed matrix of the position deviation vector. This state error vector constitutes the error state data. The error state data is analyzed using the model predictive control (MPC) algorithm. The core idea of ​​MPC is to predict the future state of the system in the rolling time domain and optimize the control input to minimize the predefined objective function. The objective function of MPC can be expressed as:

[0114]

[0115] Here, N p For the prediction time domain, is the transposed matrix of the state error at time k, e k is the state error at time k, is the transposed matrix of the control input, u k is the control input, Q and R are the weight matrices of the state error and control input respectively. By solving this optimization problem, the predictive control data is obtained. Based on the predictive control data, a nonlinear control law is designed. Considering the nonlinear dynamic characteristics of the precast beam hoisting process, the feedback linearization method is used to design the control law u:

[0116] u=g(x)+f(x)v

[0117] Among them, x is the system state, g(x) and f(x) are nonlinear functions, and v is the new control input. In this way, the nonlinear system is converted into a linear system, which is convenient for controller design. The control law parameter data contains the specific forms of g(x) and f(x). According to the control law parameter data, the torque distribution is optimized. The multi-point lifting system is an overdriven system, and the torque of each lifting point needs to be reasonably distributed. The torque distribution problem can be expressed as a quadratic programming problem:

[0118]

[0119] Here, τ is the moment vector of each suspension point, W is the weight matrix, c is the cost vector, A and b define the force balance constraint, τ min and τ maxis the torque limit. Solve this optimization problem and get the torque distribution data. Finally, perform dynamic constraint processing on the torque distribution data. Considering the dynamic characteristics and safety limits of the actuator, smooth and limit the torque to get the final control command data. These control command data are used to drive the actuators of each suspension point to achieve precise coordinated control of the precast beam.

[0120] For example, in the process of hoisting a precast beam with a span of 100 meters and a weight of 700 tons, six hoisting points are used for synchronous control. The real-time posture tracking system provides posture data at a frequency of 100Hz with an accuracy of millimeters. At a certain moment, the position deviation of the precast beam was detected to be [0.05m, -0.03m, 0.02m] and the posture deviation was 0.5 degrees. The state error vector is constructed as a 12-dimensional vector, including 3D position error, 3D posture error and its derivative. The MPC algorithm sets the prediction time domain to 20 steps (i.e. 0.2 seconds) and the control time domain to 5 steps. In the objective function, the position error weight is set to 1000, the posture error weight is set to 500, and the control input weight is set to 0.1. By solving the quadratic programming problem, the optimal control sequence within the next 0.2 seconds is obtained.

[0121] In the design of nonlinear control law, g(x) includes the gravity compensation term, and f(x) takes into account the inertia tensor of the precast beam and the geometric distribution of the suspension points. In the torque distribution optimization problem, the weight matrix W is set to the unit matrix, and the cost vector c is set according to the distance from each suspension point to the center of mass of the precast beam to minimize the balance and make the load evenly distributed. The torque limit is set according to the actuator specifications, such as τ min =-50kN·m, τ max =50kN·m.

[0122] The generated control command data is the torque sequence of the six hanging points in the next 50ms, with 5 torque values ​​for each hanging point. These torque values ​​are smoothed to ensure the continuity of the control command. For example, the torque sequence of a certain hanging point may be [10.5, 11.2, 11.8, 12.3, 12.7] kN·m. This fine torque control ensures the stability and accuracy of the precast beam during the hoisting process, so that the maximum position deviation is controlled within ±2cm and the posture deviation is controlled within ±0.2 degrees, greatly improving the efficiency and safety of the installation of precast beams for large bridges.

[0123] The above describes the method for multi-point synchronous precise lifting and positioning of prefabricated bridge beams in the embodiment of the present application. The following describes the multi-point synchronous precise lifting and positioning device for prefabricated bridge beams in the embodiment of the present application. Figure 2 In the embodiment of the present application, one embodiment of the multi-point synchronous precise lifting and positioning device for prefabricated bridge beams includes:

[0124] The acquisition module 201 is used to acquire image data of the precast bridge beam and its surrounding environment, and preprocess the image data to obtain preprocessed image data;

[0125] The calibration module 202 is used to perform three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and to accurately calibrate the three-dimensional data to obtain three-dimensional data of the target precast beam;

[0126] The reconstruction module 203 is used to perform scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimize the three-dimensional scene data to obtain target environment data;

[0127] The planning module 204 is used to perform path planning based on the target precast beam three-dimensional data and the target environment data to obtain multi-point hoisting path data, and optimize the path data to obtain optimized hoisting path and posture change data;

[0128] The tracking module 205 is used to perform real-time visual tracking processing on the precast beam to obtain real-time position and posture data of the precast beam, and filter the real-time position and posture data to obtain filtered position and posture data;

[0129] The control module 206 is used to compare and analyze the posture data with the optimized lifting path data and the posture change data to obtain control instruction data, and to coordinately control the multi-point lifting system according to the control instruction data.

[0130] Through the coordinated cooperation of the above-mentioned components, by collecting and preprocessing the image data of the bridge precast beam and its surrounding environment, high-quality preprocessed image data can be obtained, laying the foundation for subsequent 3D reconstruction and scene analysis, and improving the data accuracy of the entire lifting process; by performing 3D reconstruction processing on the preprocessed image data, obtaining the 3D data of the precast beam and performing accurate calibration, high-precision 3D data of the target precast beam can be obtained, providing reliable geometric information for lifting path planning and posture control; by performing scene reconstruction processing on the preprocessed image data, obtaining the 3D scene data of the lifting environment and optimizing it, accurate target environment data can be generated, which is helpful to identify potential obstacles and plan the optimal lifting path; by performing path planning through the 3D data of the target precast beam and the target environment data, path data for multi-point lifting can be obtained. The data is obtained by optimizing the lifting path and posture change data, which can not only ensure the safety of the lifting process, but also improve the lifting efficiency; the precast beams are subjected to real-time visual tracking to obtain the real-time position and posture data of the precast beams, and the filtered posture data is obtained through filtering, which can realize real-time and accurate monitoring of the position and posture of the precast beams, and provide timely and accurate feedback information for the control system; the posture data is compared and analyzed with the optimized lifting path data and posture change data to obtain the control command data, and the multi-point lifting system is coordinated and controlled according to the control command data, realizing closed-loop control of the lifting process, greatly improving the lifting accuracy and stability, realizing the intelligence and precision of the lifting process of precast bridge beams, significantly improving the efficiency and safety of lifting operations, and reducing human errors.

[0131] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams, characterized in that: The multi-point synchronous precise hoisting and positioning method for prefabricated bridge beams comprises: Collecting image data of the precast bridge beam and its surrounding environment, and preprocessing the image data to obtain preprocessed image data; Performing three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and accurately calibrating the three-dimensional data to obtain three-dimensional data of the target precast beam; Performing scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimizing the three-dimensional scene data to obtain target environment data; Performing path planning through the target precast beam three-dimensional data and the target environment data to obtain multi-point hoisting path data, and optimizing the path data to obtain optimized hoisting path and posture change data; Performing real-time visual tracking processing on the prefabricated beam to obtain real-time position and posture data of the prefabricated beam, and performing filtering processing on the real-time position and posture data to obtain filtered position and posture data; The posture data is compared and analyzed with the optimized lifting path data and the posture change data to obtain control instruction data, and the multi-point lifting system is collaboratively controlled according to the control instruction data.

2. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The collecting of image data of the prefabricated bridge beam and its surrounding environment and preprocessing the image data to obtain preprocessed image data includes: Shooting the prefabricated bridge beam at multiple angles to obtain image data, and performing denoising on the image data to obtain denoised image data; Performing contrast enhancement processing on the denoised image data to obtain enhanced image data, and performing distortion correction on the enhanced image data to obtain corrected image data; Extracting feature points from the corrected image data to obtain feature point data, and performing feature matching on the feature point data to obtain matched feature point pair data; Eliminating outliers from the matched feature point pair data by using a RANSAC algorithm to obtain filtered feature point pair data, and registering the corrected image data according to the filtered feature point pair data to obtain registered image data; Performing color balancing processing on the registered image data to obtain color-balanced image data, and performing image fusion on the color-balanced image data to obtain fused panoramic image data; The fused panoramic image data is segmented by an adaptive threshold segmentation algorithm to obtain segmented image region data, and edge detection is performed on the segmented image region data to obtain edge feature data; Performing shape analysis on the edge feature data to obtain target contour data, and performing straight line and circle detection on the target contour data using a Hough transform algorithm to obtain geometric feature data; Performing target recognition on the fused panoramic image data according to the geometric feature data to obtain target area data, and performing size measurement on the target area data to obtain target size data; Performing texture analysis on the target area data to obtain texture feature data, and performing material identification on the target area data according to the texture feature data and the target size data to obtain material information data; The fused panoramic image data is semantically annotated according to the material information data, the target size data and the geometric feature data to obtain the preprocessed image data.

3. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The three-dimensional reconstruction process is performed on the preprocessed image data to obtain the three-dimensional data of the prefabricated beam, and the three-dimensional data is accurately calibrated to obtain the three-dimensional data of the target prefabricated beam, including: Extracting feature points from the preprocessed image data to obtain feature point cloud data, and performing feature matching on the feature point cloud data to obtain matched feature point pair data; Performing sparse reconstruction on the matched feature point pair data by using a structured beam method to obtain sparse point cloud data, and performing dense reconstruction on the sparse point cloud data to obtain dense point cloud data; Performing meshing processing on the dense point cloud data to obtain triangular mesh model data, and performing surface smoothing processing on the triangular mesh model data to obtain three-dimensional data of the prefabricated beam; Performing dimension calibration on the three-dimensional data of the prefabricated beam according to the design drawing data of the prefabricated beam to obtain calibrated three-dimensional data, and performing surface detail optimization on the calibrated three-dimensional data to obtain refined three-dimensional data; Texture mapping is performed on the refined three-dimensional data to obtain three-dimensional data with texture, and illumination simulation is performed on the three-dimensional data with texture to obtain the three-dimensional data of the target prefabricated beam.

4. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The performing scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimizing the three-dimensional scene data to obtain target environment data, includes: Performing semantic segmentation on the preprocessed image data to obtain scene element segmentation data, and performing instance segmentation on the scene element segmentation data to obtain scene instance data; Performing three-dimensional reconstruction on the scene instance data to obtain three-dimensional scene data of the hoisting environment, and performing plane extraction on the three-dimensional scene data of the hoisting environment to obtain ground plane data; Performing scene layering on the three-dimensional scene data of the hoisting environment according to the ground plane data to obtain layered scene data, and performing cluster analysis on the non-ground part of the layered scene data to obtain obstacle clustering data; Performing three-dimensional modeling on the obstacle cluster data to obtain obstacle three-dimensional model data, and merging the obstacle three-dimensional model data with the ground plane data to obtain merged scene data; The merged scene data is globally optimized to obtain the target environment data.

5. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The path planning is performed through the target precast beam three-dimensional data and the target environment data to obtain the path data of multi-point hoisting, and the path data is optimized to obtain the optimized hoisting path and posture change data, including: Determine the optimal lifting point position of multi-point lifting according to the three-dimensional data of the target prefabricated beam, obtain lifting point position data, and construct a configuration space according to the lifting point position data and the target environment data to obtain discretized grid map data; A path cost function is defined to obtain cost function parameter data, and a path search is performed on the discretized grid map data by using an improved fast expansion random tree algorithm to obtain multi-point hoisting path data; Performing collision detection on the path data of the multi-point hoisting to obtain collision-free path data, and smoothing the collision-free path data using a B-spline curve to obtain smoothed path data; Calculating the posture change of the prefabricated beam during the hoisting process according to the smooth path data to obtain posture change data, and designing an anti-sway control strategy according to the posture change data to obtain anti-sway control parameter data; The smooth path data, the posture change data and the anti-sway control parameter data are combined to obtain the optimized lifting path and posture change data.

6. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The real-time visual tracking process of the prefabricated beam is performed to obtain the real-time position and posture data of the prefabricated beam, and the real-time position and posture data are filtered to obtain the filtered posture data, including: Set visual marking points on the prefabricated beams to obtain marking point data, and use the improved FAST corner point detection algorithm to detect feature points on the real-time image to obtain real-time feature point data; Extract descriptors from the real-time feature point data using an ORB algorithm to obtain feature descriptor data, and perform feature matching between the feature descriptor data and the marker point data using a KNN algorithm to obtain matching feature point data; The matching feature point data is subjected to external point elimination by using the RANSAC algorithm to obtain screened matching point data, and the screened matching point data is subjected to posture estimation based on the EPnP algorithm to obtain the real-time position and posture data of the precast beam; Constructing a multi-scale image pyramid according to the real-time position and posture data of the precast beam to obtain multi-scale image data, and performing local bundle adjustment optimization on the multi-scale image data and the screened matching point data to obtain optimized posture data; The optimized posture data and the real-time position and posture data of the prefabricated beam are subjected to fusion filtering processing through an extended Kalman filter to obtain the filtered posture data.

7. The method for multi-point synchronous precise hoisting and positioning of prefabricated bridge beams according to claim 1 is characterized in that: The step of comparing and analyzing the posture data with the optimized lifting path data and the posture change data to obtain control instruction data, and performing collaborative control on the multi-point lifting system according to the control instruction data includes: Comparing the position and posture data with the optimized lifting path data to obtain position deviation data, and comparing the position and posture data with the posture change data to obtain posture deviation data; Constructing a state error vector according to the position deviation data and the posture deviation data to obtain error state data, and analyzing the error state data through a model predictive control algorithm to obtain predictive control data; Designing a nonlinear control law based on the predictive control data to obtain control law parameter data, and optimizing torque distribution according to the control law parameter data to obtain torque distribution data; Dynamic constraint processing is performed on the torque distribution data to obtain control instruction data, and each suspension point actuator is cooperatively controlled according to the control instruction data.

8. A multi-point synchronous precise hoisting and positioning device for a precast bridge beam, used to implement the multi-point synchronous precise hoisting and positioning method for a precast bridge beam as claimed in any one of claims 1 to 7, characterized in that: The multi-point synchronous precise hoisting and positioning device for prefabricated bridge beams comprises: An acquisition module is used to acquire image data of the precast bridge beam and its surrounding environment, and preprocess the image data to obtain preprocessed image data; A calibration module, used for performing three-dimensional reconstruction processing on the preprocessed image data to obtain three-dimensional data of the precast beam, and accurately calibrating the three-dimensional data to obtain three-dimensional data of the target precast beam; A reconstruction module, used for performing scene reconstruction processing on the pre-processed image data to obtain three-dimensional scene data of the hoisting environment, and optimizing the three-dimensional scene data to obtain target environment data; A planning module, used to perform path planning through the three-dimensional data of the target precast beam and the target environment data, obtain path data for multi-point hoisting, and optimize the path data to obtain optimized hoisting path and posture change data; A tracking module is used to perform real-time visual tracking processing on the prefabricated beam to obtain real-time position and posture data of the prefabricated beam, and filter the real-time position and posture data to obtain filtered position and posture data; The control module is used to compare and analyze the posture data with the optimized lifting path data and the posture change data to obtain control instruction data, and to coordinately control the multi-point lifting system according to the control instruction data.

Images