Real-time Inkjet Coding and Position Re-measurement Method and System of a Fixed Measurement Robot during Signal Construction

Abstract

The present invention provides a method and system for real-time inkjet coding and position remeasurement of a surveying robot during signal construction, which relates to the technical field of robot surveying. The method includes obtaining three-dimensional point cloud data of a construction area and generating a grid map; identifying positioning marker points and planning an optimal motion trajectory; controlling the movement of the surveying robot and keeping the inkjet coding device horizontal; adjusting the height of the nozzle and performing inkjet coding when reaching a preset position; collecting inkjet coding images and extracting contour features; comparing the actual inkjet coding with a preset template and calculating the deviation; optimizing motion parameters when the deviation exceeds a threshold; uploading deviation data and generating a quality report. This method improves the accuracy and automation level of signal construction.

Description

Technical Field

[0001] The present invention relates to the technical field of robot fixed measurement, and particularly relates to a method and system for real-time inkjet coding and position remeasurement of a fixed measurement robot in signal construction. Background Art

[0002] With the continuous expansion of the scale of railway construction, the fixed measurement marks in signal engineering construction are important benchmarks to ensure the track alignment and the installation accuracy of signal equipment. The inkjet coding operation of the fixed measurement marks directly affects the positioning accuracy and construction quality of subsequent construction links. At present, the railway engineering construction has put forward higher requirements for the position accuracy and construction efficiency of the fixed measurement marks, and the automated and intelligent fixed measurement inkjet coding technology has gradually become the development trend of the industry.

[0003] The existing fixed measurement mark inkjet coding mainly uses a manual hand-held inkjet printer for operation, or uses a simple automatic inkjet coding device. These methods have many deficiencies: the accuracy of manual inkjet coding depends on the experience of the operator, and it is difficult to maintain a stable inkjet coding posture and a uniform spraying effect; the existing automatic inkjet coding device lacks intelligent navigation and attitude adjustment functions and cannot adapt to complex construction environments; at the same time, there is a lack of real-time quality inspection and data management means, resulting in difficult-to-guarantee construction quality and increased later rework and maintenance costs.

[0004] In summary, there is an urgent need to develop an intelligent fixed measurement system integrating robot technology, computer vision, and artificial intelligence to achieve automatic surveying and mapping and path planning of the construction area, precise inkjet coding position control, real-time attitude adjustment, and online detection of inkjet coding quality, improve the construction efficiency and quality of the fixed measurement marks, and enhance the automation level and quality management ability of railway engineering construction. The present invention can solve the problems in the prior art. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for real-time inkjet coding and position remeasurement of a fixed measurement robot in signal construction, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention,

[0007] A method for real-time inkjet coding and position remeasurement of a fixed measurement robot in signal construction is provided, including:

[0008] Scanning the construction area by a fixed measurement robot to obtain three-dimensional point cloud data of the construction area, rasterizing the construction area based on the three-dimensional point cloud data to generate a raster map, collecting image information of the construction area by the fixed measurement robot, identifying positioning mark points in the construction area through an image recognition algorithm, mapping the spatial coordinates of the positioning mark points to the raster map, determining the distribution of the positioning mark points in the raster map, and using a time series reinforcement iterative neural network algorithm to plan the optimal movement trajectory of the fixed measurement robot;

[0009] Control the surveying robot to move along the optimal motion trajectory, collect the surveying attitude data in real time through the inertial measurement unit of the surveying robot, and based on the surveying attitude data, control the inkjet device of the surveying robot to maintain a horizontal state. When the surveying robot moves to the preset inkjet position, measure and automatically adjust the vertical distance between the nozzle of the inkjet device and the ground. Collect the image information of the preset inkjet position through the vision positioning system, and combine it with the preset inkjet template to control the inkjet device to perform inkjet operations;

[0010] After the inkjet operation is completed, collect the high-definition image of the inkjet area, use the image processing algorithm to extract the contour features of the inkjet pattern, compare the contour features with the preset inkjet template, calculate the actual deviation of the inkjet position. When the actual deviation exceeds the preset threshold, based on the actual deviation, use the deep learning model to adaptively optimize the motion parameters of the surveying robot, and upload the actual deviation to the construction management system in real time to generate a construction quality report.

[0011] In an alternative embodiment,

[0012] Scan the construction area through the surveying robot to obtain the three-dimensional point cloud data of the construction area, and perform rasterization processing on the construction area based on the three-dimensional point cloud data to generate a raster map including:

[0013] Scan the construction area through the surveying robot to obtain the three-dimensional point cloud data of the construction area, calculate the statistical mean and standard deviation of the three-dimensional point cloud data, determine the effective data range, remove the abnormal data outside the effective data range to obtain the filtered three-dimensional point cloud data, divide the filtered three-dimensional point cloud data into voxels according to the preset voxel size, calculate the centroid coordinates of the three-dimensional points included in each voxel, and use the centroid coordinates as the downsampled points of the corresponding voxels to obtain the downsampled three-dimensional point cloud data;

[0014] Select the neighboring three-dimensional points within the corresponding radius range of each three-dimensional point in the downsampled three-dimensional point cloud data according to the preset radius, construct a covariance matrix based on the neighboring three-dimensional points, calculate the normal vector of each three-dimensional point through the covariance matrix, determine the vector difference between the normal vectors, use the median of the vector difference as the Gaussian kernel parameter, construct a Gaussian kernel function based on the Gaussian kernel parameter, and calculate the significance score of each three-dimensional point;

[0015] Sort the significance scores in descending order, select the top 30% of the three-dimensional points with the sorted significance scores as candidate feature points, judge the distance between the candidate feature points, remove the candidate feature points with a distance less than the preset distance threshold to obtain a set of feature points; calculate the point feature histogram descriptor of each feature point in the set of feature points and perform feature matching to obtain the initial registration result;

[0016] Calculate the distance values and normal angle values between corresponding feature points in the initial registration result, take the median of the distance values as the distance parameter, and the median of the normal angle values as the angle parameter; construct a distance weight term based on the distance parameter, construct an angle weight term based on the angle parameter, multiply the distance weight term and the angle weight term to obtain an adaptive weight, and use the adaptive weight for iterative optimization to obtain an accurate registration result;

[0017] Divide the construction area into grids of a preset size, calculate the distance between each grid and the three-dimensional point cloud data in the accurate registration result, calculate the observation likelihood of the Gaussian distribution, multiply the observation likelihood by the prior probability of the corresponding grid to obtain a probability product, calculate the reciprocal of the cumulative sum of the probability products as the normalization factor, and use the normalization factor to normalize the probability product to obtain the state probability of each grid. Mark each grid as an occupied state, a free state, or an unknown state based on the state probability, and generate a grid map of the construction area.

[0018] In an alternative embodiment, identify the positioning marker points in the construction area through an image recognition algorithm, map the spatial coordinates of the positioning marker points to the grid map, and determine the distribution of the positioning marker points in the grid map, including:

[0019] Input the image collected by the surveying robot into the backbone feature network. The backbone feature network extracts image features through residual modules. In the residual modules, use the spatial attention mechanism and the channel attention mechanism to perform weighted reconstruction on the spatial dimension and the channel dimension of the feature map respectively to obtain the first-layer feature map; input the first-layer feature map into the feature pyramid structure, and fuse features of different scales through upsampling and downsampling operations to obtain a multi-level feature map;

[0020] Input the multi-level feature map into the detection head network. The detection head network outputs the class scores and position coordinates of the positioning marker points. Based on the class scores and the position coordinates, determine the two-dimensional pixel coordinates of the positioning marker points in the image; perform distortion correction on the image based on the internal parameter matrix of the binocular camera, calculate the disparity information of the left and right images, and combine the external parameter matrix of the binocular camera. Convert the two-dimensional pixel coordinates to depth information through the principle of triangulation, and combine the two-dimensional pixel coordinates and the depth information to obtain the three-dimensional spatial coordinates of the positioning marker points;

[0021] Extract the corresponding feature points in the coordinate system corresponding to the camera coordinate system and the grid map, construct feature point matching pairs, calculate the rotation matrix and the translation vector, and use a non-linear optimization algorithm to optimize the rotation matrix and the translation vector to obtain a coordinate transformation matrix. Map the three-dimensional spatial coordinates to the coordinate system of the grid map through the coordinate transformation matrix to obtain the spatial distribution of the positioning marker points.

[0022] In an alternative embodiment,

[0023] Adopting the time-series reinforcement iterative neural network algorithm, planning the optimal motion trajectory of the final measurement robot includes:

[0024] Construct a time-series reinforcement iterative neural network, form a state vector by combining the real-time position information of the final measurement robot, the spatial distribution of positioning marker points, and the local information of the grid map, and form an action vector by combining the discretized linear velocity and angular velocity. The time-series reinforcement iterative neural network includes an active action network and a target value network;

[0025] The active action network selects an action vector based on the current state vector, executes the action vector to obtain the next state vector, and the target value network estimates the value based on the next state vector; calculate the path length value, motion smoothness value, and marker point coverage value of the action vector, and combine them to form a reward value;

[0026] Construct state transition data with the current state vector, action vector, reward value, and next state vector, calculate the time-series difference error of the state transition data, assign priorities to the state transition data based on the time-series difference error, and store the state transition data with different priorities in the experience replay pool; sample the state transition data from the experience replay pool according to the priority to train the active action network, and update the parameters of the active action network to the target value network according to the preset number of steps, repeat the iteration until convergence, and output the optimal motion trajectory for the final measurement robot to visit the positioning marker points.

[0027] In an alternative embodiment,

[0028] Control the final measurement robot to move along the optimal motion trajectory, collect the final measurement attitude data in real time through the inertial measurement unit of the final measurement robot. Based on the final measurement attitude data, control the inkjet device of the final measurement robot to maintain a horizontal state. When the final measurement robot moves to the preset inkjet position, measure and automatically adjust the vertical distance between the nozzle of the inkjet device and the ground, collect the image information of the preset inkjet position through the vision positioning system, and combine the preset inkjet template to control the inkjet device to perform inkjet operations, including:

[0029] Collect the acceleration component and angular velocity component through the inertial measurement unit of the final measurement robot, estimate the attitude quaternion through the extended Kalman filter algorithm, and calculate the roll angle, pitch angle, and yaw angle according to the attitude quaternion;

[0030] Based on the attitude quaternion, establish a kinematic model of the inkjet device, calculate the target angles of the first and second rotation axes of the two-degree-of-freedom pan-tilt, and control the stepping motors of the first and second rotation axes to rotate respectively through a PID controller, and set the inkjet device to a horizontal state;

[0031] When reaching the preset inkjet coding position, the vertical distance between the nozzle and the ground is measured by a laser ranging sensor, and the vertical distance is compared with the preset vertical distance boundary to obtain a distance deviation value and a distance deviation change rate. Fuzzy control rules are established and input into a fuzzy PID controller to adaptively adjust the PID parameters and control the rotational speed of the DC servo motor of the lifting mechanism to adjust the height of the nozzle.

[0032] Collect the image of the preset inkjet coding position, perform grayscale conversion and edge detection processing on the image to obtain contour features. Match the preset inkjet coding template with the contour features, and based on the matching result, determine the inkjet coding position coordinates and the spraying direction angle, and generate a spraying instruction sequence. The spraying instruction sequence includes spraying instructions containing nozzle position information, spraying pressure information, and spraying time information, and sends the spraying instruction sequence to the inkjet coding controller through the fieldbus.

[0033] During the inkjet coding process, the spraying image is collected in real time, and the detection parameters of the spraying lines in the spraying image are analyzed. The detection parameters include width value, continuity value, and clarity value, and an analysis result is generated. When any one of the detection parameters is lower than the corresponding preset threshold, the spraying pressure information and the spraying time information are adaptively adjusted based on the analysis result until each detection parameter reaches the corresponding preset threshold to complete the inkjet coding operation.

[0034] In an optional embodiment,

[0035] The acceleration components and angular velocity components are collected through the inertial measurement unit of the measuring robot, the attitude quaternion is estimated by the extended Kalman filter algorithm, and the roll angle, pitch angle, and yaw angle are calculated according to the attitude quaternion, including:

[0036] The acceleration components of the gravitational acceleration in three orthogonal axes are collected through the three-axis acceleration sensor of the inertial measurement unit, and the angular velocity components of the robot rotating around the three orthogonal axes are collected through the three-axis gyroscope. The acceleration components and the angular velocity components are sampled and filtered through the signal acquisition module.

[0037] Based on the angular velocity components, a system state equation is established, the attitude quaternion is set as the state vector, and the angular velocity is set as the system input to construct a nonlinear state transition model. Based on the acceleration components, an observation equation is established, and the projection of the gravity vector in the body coordinate system is set as the observed quantity, and a nonlinear observation model is constructed according to the relationship between the attitude quaternion and the gravity vector.

[0038] The nonlinear state transition model is expanded by the first-order Taylor expansion to obtain the state Jacobian matrix, and the nonlinear observation model is expanded by the first-order Taylor expansion to obtain the observation Jacobian matrix.

[0039] The attitude is estimated using the extended Kalman filter algorithm, including: performing state prediction using the state Jacobian matrix and the non-linear state transition model to obtain the predicted attitude quaternion and the predicted state covariance; calculating the Kalman gain using the observation Jacobian matrix, the non-linear observation model, and the predicted state covariance, multiplying the Kalman gain by the observation residual to obtain the state correction amount, adding the state correction amount to the predicted attitude quaternion to obtain the estimated attitude quaternion, and simultaneously updating the state covariance;

[0040] Normalize the estimated quaternion by dividing each component by its corresponding norm length to obtain the unit attitude quaternion; calculate three rotation angles describing the robot's attitude, including the roll angle about the X-axis, the pitch angle about the Y-axis, and the yaw angle about the Z-axis, according to the four components of the unit attitude quaternion through the conversion formula from the unit attitude quaternion to Euler angles.

[0041] In an alternative embodiment,

[0042] After the inkjet coding operation is completed, a high-definition image of the inkjet coding area is collected, the contour features of the inkjet coding pattern are extracted using an image processing algorithm, the contour features are compared with the preset inkjet coding template, the actual deviation of the inkjet coding position is calculated, and when the actual deviation exceeds the preset threshold, the motion parameters of the fixed measurement robot are adaptively optimized based on the actual deviation, and the actual deviation is uploaded to the construction management system in real time to generate a construction quality report including:

[0043] Collect the spectral image of the inkjet coding area through a hyperspectral camera, project a grid pattern through a structured light projector and collect the structured light image, input the spectral image and the structured light image into a synchronous trigger circuit for timing alignment to obtain standard dual-modal image data;

[0044] Input the standard dual-modal image data into a spectral-spatial joint feature extraction network. The spectral attention module of the spectral-spatial joint feature extraction network processes the spectral image to obtain spectral features, the spatial pyramid pooling layer processes the structured light image to obtain spatial features, and the spectral features and the spatial features are fused through a cross-modal feature fusion module to obtain the fusion features of the inkjet coding area;

[0045] Perform edge detection on the fusion features to obtain the edge point set of the inkjet coding pattern, input the edge point set into a dynamic graph convolutional network, extract local structure information through an adaptive graph convolutional layer, and use a graph attention mechanism to enhance the weight of the local structure information to obtain the contour features of the inkjet coding pattern;

[0046] Match the contour features with a preset inkjet coding template, obtain the initial pose deviation through one-stage registration by a graph matching network, and based on the initial pose deviation, perform two-stage registration through the iterative closest point algorithm to obtain the actual pose deviation of the inkjet coding pattern;

[0047] Input the actual pose deviation into a hierarchical reinforcement learning network. The high-level policy network of the hierarchical reinforcement learning network generates a parameter optimization strategy based on the actual pose deviation, and the low-level execution network adjusts the motion parameters of the surveying robot based on the parameter optimization strategy;

[0048] Input the motion parameters into a hybrid parameter optimizer. The hybrid parameter optimizer uses Bayesian optimization to determine the parameter search space, and iteratively optimizes the motion parameters within the parameter search space through an evolutionary algorithm to obtain optimized motion parameters;

[0049] Based on a pre-constructed uncertainty evaluation model, input the optimized motion parameters into the uncertainty evaluation model, calculate the measurement uncertainty introduced by the measurement process, the model uncertainty introduced by the model prediction, and the parameter uncertainty introduced by the parameter optimization respectively, and input the measurement uncertainty, the model uncertainty, and the parameter uncertainty into the evidence theory model to obtain the credibility evaluation value of the parameter optimization result.

[0050] In the second aspect of the embodiments of the present invention,

[0051] Provide a real-time inkjet coding and position remeasurement system for a surveying robot in signal construction, including:

[0052] The first unit is used to scan the construction area through the surveying robot, obtain the three-dimensional point cloud data of the construction area, perform rasterization processing on the construction area based on the three-dimensional point cloud data to generate a raster map, collect the image information of the construction area by the surveying robot, identify the positioning marker points in the construction area through an image recognition algorithm, map the spatial coordinates of the positioning marker points to the raster map, determine the distribution of the positioning marker points in the raster map, and adopt a time-series reinforcement iterative neural network algorithm to plan the optimal motion trajectory of the surveying robot;

[0053] The second unit is used to control the surveying robot to move along the optimal motion trajectory, collect the surveying attitude data in real time through the inertial measurement unit of the surveying robot, based on the surveying attitude data, control the inkjet device of the surveying robot to maintain a horizontal state, when the surveying robot moves to a preset inkjet position, measure and automatically adjust the vertical distance between the nozzle of the inkjet device and the ground, collect the image information of the preset inkjet position through a visual positioning system, and combine with a preset inkjet coding template to control the inkjet device to perform inkjet coding operations;

[0054] The third unit is used to collect high-definition images of the inkjet printing area after the inkjet printing operation is completed, extract the contour features of the inkjet printing pattern using image processing algorithms, compare the contour features with the preset inkjet printing template, calculate the actual deviation of the inkjet printing position, and when the actual deviation exceeds the preset threshold, use a deep learning model based on the actual deviation to adaptively optimize the motion parameters of the surveying and fixing robot, upload the actual deviation to the construction management system in real time, and generate a construction quality report.

[0055] In the third aspect of the embodiments of the present invention,

[0056] a kind of electronic device is provided, including:

[0057] a processor;

[0058] a memory for storing instructions executable by the processor;

[0059] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0060] In the fourth aspect of the embodiments of the present invention,

[0061] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0062] In the embodiments of the present invention, by scanning the construction area to generate a grid map, using image recognition algorithms to identify positioning marker points, and combining with the time series reinforcement iterative neural network algorithm to plan the optimal motion trajectory, the working efficiency of the surveying and fixing robot can be greatly improved. At the same time, the surveying and fixing attitude data is collected in real time and the inkjet printing device is automatically adjusted to ensure the accuracy of the inkjet printing operation; after the inkjet printing operation is completed, the contour features of the inkjet printing pattern are extracted by image processing algorithms and compared with the preset template to calculate the actual deviation. When the deviation exceeds the preset threshold, the system will adaptively optimize the motion parameters of the surveying and fixing robot, thereby continuously improving the construction quality. This real-time monitoring and self-optimization mechanism greatly enhances the control ability of the construction quality; the actual deviation data is uploaded to the construction management system in real time, and a construction quality report is generated, providing accurate and timely data support for construction management. This intelligent data collection and analysis method helps construction management personnel quickly understand the construction quality status, make timely decisions, and thus improve the overall intelligent level of construction management. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic flowchart of the method for real-time inkjet printing and position remeasurement of the surveying and fixing robot in signal construction of the embodiments of the present invention;

[0064] Figure 2This is a schematic structural diagram of the real-time inkjet coding and position remeasurement system for the fixed measurement robot in signal construction of the embodiments of the present invention. Specific embodiments

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0067] Figure 1 This is a schematic flowchart of the real-time inkjet coding and position remeasurement method for the fixed measurement robot in signal construction of the embodiments of the present invention. As Figure 1 shown, the method includes:

[0068] S101. Scan the construction area through the fixed measurement robot to obtain the three-dimensional point cloud data of the construction area. Based on the three-dimensional point cloud data, rasterize the construction area to generate a raster map. Use the fixed measurement robot to collect the image information of the construction area. Identify the positioning marker points in the construction area through an image recognition algorithm, and map the spatial coordinates of the positioning marker points to the raster map to determine the distribution of the positioning marker points in the raster map. Adopt a time-series enhanced iterative neural network algorithm to plan the optimal movement trajectory of the fixed measurement robot;

[0069] S102. Control the fixed measurement robot to move along the optimal movement trajectory. Collect the fixed measurement attitude data in real time through the inertial measurement unit of the fixed measurement robot. Based on the fixed measurement attitude data, control the inkjet device of the fixed measurement robot to maintain a horizontal state. When the fixed measurement robot moves to the preset inkjet position, measure and automatically adjust the vertical distance between the nozzle of the inkjet device and the ground. Collect the image information of the preset inkjet position through the visual positioning system, and combine it with the preset inkjet template to control the inkjet device to perform inkjet operations;

[0070] S103. After the inkjet operation is completed, collect the high-definition image of the inkjet area. Use an image processing algorithm to extract the contour features of the inkjet pattern. Compare the contour features with the preset inkjet template, calculate the actual deviation of the inkjet position. When the actual deviation exceeds the preset threshold, adopt a deep learning model based on the actual deviation to adaptively optimize the movement parameters of the fixed measurement robot, and upload the actual deviation to the construction management system in real time to generate a construction quality report.

[0071] First, the surveying robot conducts a full - range scan of the construction area through the carried 3D laser scanner to obtain high - precision 3D point cloud data. The voxel filtering algorithm is used to downsample and denoise the point cloud data to improve the data quality. Then, the octree algorithm is used to divide the processed point cloud data into grid cells of uniform size, and the size of each grid cell is set to 0.1m x 0.1m. Statistical analysis is performed on the point cloud within each grid cell to extract feature information such as the elevation and flatness of the grid, and a 2D grid map is generated.

[0072] Next, the surveying robot uses the carried high - definition camera to collect image information of the construction area. The YOLOv5 object detection algorithm is used to process the image to identify the positioning marker points in the construction area. The positioning marker points are usually black - and - white checkerboard patterns with a size of 20cm x 20cm. The YOLOv5 algorithm can quickly and accurately detect the positions and contours of these marker points. After obtaining the image coordinates of the positioning marker points, using the internal and external parameters of the camera, they are converted into 3D space coordinates. Then, these 3D coordinates are mapped onto the previously generated grid map to obtain the precise position distribution of the positioning marker points in the grid map.

[0073] Based on the grid map and the distribution information of the positioning marker points, the Sequential Temporal Reinforcement Iterative Neural network (STRIN) algorithm is used to plan the optimal motion trajectory of the surveying robot. The STRIN algorithm combines the advantages of reinforcement learning and recurrent neural networks and can effectively handle sequential decision - making problems. First, the grid map and the positions of the positioning marker points are used as state inputs, and the action space is defined as the moving direction and speed of the robot. Then, a reward function is designed, comprehensively considering factors such as path length, energy consumption, and obstacle avoidance. Through multiple iterative trainings, the STRIN algorithm can learn the optimal strategy and generate a smooth and efficient motion trajectory.

[0074] Along the planned optimal trajectory, the surveying robot is controlled to move. The inertial measurement unit (IMU) carried by the surveying robot real - time collects the attitude data of the robot, including three - axis acceleration, angular velocity, and magnetic field strength. The extended Kalman filtering algorithm is used to fuse the IMU data and the odometer data to obtain a high - precision attitude estimation result. Based on the attitude estimation result, the attitude of the chassis of the surveying robot is adjusted through the PID control algorithm to ensure that the ink - jetting device always maintains a horizontal state and improves the ink - jetting accuracy.

[0075] When the fixed measurement robot moves to the preset inkjet coding position, activate the laser ranging sensor on the robot to measure the vertical distance between the nozzle of the inkjet coding device and the ground. Set the target distance to 30 cm, and control the lifting of the inkjet coding device through a stepper motor to achieve precise adjustment of the nozzle height. At the same time, start the visual positioning system of the robot, which includes two orthogonally installed industrial cameras, to collect the stereo image of the inkjet coding area. Using the SIFT feature matching and triangulation principle, calculate the precise three-dimensional coordinates of the inkjet coding position.

[0076] Combined with the pre-designed inkjet coding template, control the inkjet coding device to perform the inkjet coding operation. The inkjet coding template contains information such as inkjet coding content, font, size, etc. Use a vector graphics library to convert the template into an inkjet coding instruction sequence to precisely control the opening and closing and movement of the nozzle. The inkjet coding device uses piezoelectric inkjet technology and can achieve a spraying accuracy of 0.1 mm. By adjusting the ink droplet size and spraying frequency, ensure that the inkjet coding pattern is clear and uniform.

[0077] After the inkjet coding operation is completed, the fixed measurement robot uses a high-resolution camera to collect the image of the inkjet coding area, and the image resolution is not less than 4K. Use an adaptive threshold segmentation algorithm to extract the contour of the inkjet coding pattern, and then use morphological operations to thin and smooth the contour. Register the processed contour features with the preset inkjet coding template, and use the iterative closest point (ICP) algorithm to calculate the deviation between the actual inkjet coding position and the ideal position. Set the deviation threshold to 2 mm, and when the actual deviation exceeds the threshold, trigger the adaptive optimization process.

[0078] The adaptive optimization uses a method based on deep reinforcement learning. Construct a multi-layer perceptron network, the input includes the current motion parameters, environmental features and deviation data, and the output is the optimized motion parameters. Through interacting with the environment and the guidance of the reward function, the network can gradually learn a better motion strategy and automatically adjust parameters such as the speed and acceleration of the robot to adapt to different ground conditions and external interferences.

[0079] Finally, upload the actual deviation data and the optimized parameters to the construction management system in real time. The system generates a detailed construction quality report based on these data, including the inkjet coding position distribution map, deviation statistical analysis, optimization effect evaluation, etc. Construction management personnel can timely understand the construction quality situation through the report and make corresponding adjustments and improvements.

[0080] In this embodiment, the whole process automation from area scanning, trajectory planning to inkjet coding operation is realized by the surveying robot, greatly reducing the need for manual operation. At the same time, by using high-precision sensors and advanced algorithms, the accuracy of the inkjet coding position is ensured, significantly improving the construction quality; through continuous position remeasurement and deviation analysis, the system can timely detect and correct construction errors. The introduction of the deep learning model enables the robot to automatically adjust the motion parameters according to the actual situation, improving the adaptability and robustness of the system; all construction data are uploaded and analyzed in real time to generate a comprehensive quality report. This not only facilitates construction supervision and acceptance, but also provides important data support for subsequent maintenance and optimization, which is conducive to long-term improvement of construction efficiency and quality.

[0081] In an alternative embodiment, the construction area is scanned by the surveying robot to obtain the three-dimensional point cloud data of the construction area, and the construction area is rasterized based on the three-dimensional point cloud data. The generated raster map includes:

[0082] The construction area is scanned by the surveying robot to obtain the three-dimensional point cloud data of the construction area. The statistical mean and standard deviation of the three-dimensional point cloud data are calculated to determine the effective data range, and the abnormal data outside the effective data range are removed to obtain the filtered three-dimensional point cloud data. The filtered three-dimensional point cloud data are divided into voxels according to the preset voxel size, and the centroid coordinates of the three-dimensional points included in each voxel are calculated. The centroid coordinates are used as the downsampled points of the corresponding voxels to obtain the downsampled three-dimensional point cloud data;

[0083] According to the preset radius, the neighboring three-dimensional points within the corresponding radius range of each three-dimensional point in the downsampled three-dimensional point cloud data are selected. A covariance matrix is constructed based on the neighboring three-dimensional points, and the normal vector of each three-dimensional point is calculated through the covariance matrix. The vector difference between the normal vectors is determined, and the median of the vector difference is used as the Gaussian kernel parameter. A Gaussian kernel function is constructed based on the Gaussian kernel parameter, and the significance score of each three-dimensional point is calculated;

[0084] The significance scores are sorted in descending order, and the top 30% of the three-dimensional points in the significance score ranking are selected as candidate feature points. The distances between the candidate feature points are judged, and the candidate feature points with distances less than the preset distance threshold are removed to obtain a set of feature points; The point feature histogram descriptor of each feature point in the set of feature points is calculated and feature matching is performed to obtain the initial registration result;

[0085] Calculate the distance values and normal angle values between corresponding feature points in the initial registration result, take the median of the distance values as the distance parameter, and the median of the normal angle values as the angle parameter; construct a distance weight term based on the distance parameter, construct an angle weight term based on the angle parameter, multiply the distance weight term by the angle weight term to obtain an adaptive weight, and use the adaptive weight for iterative optimization to obtain an accurate registration result;

[0086] Divide the construction area into grids of a preset size, calculate the distance between each grid and the three-dimensional point cloud data in the accurate registration result, calculate the observation likelihood of the Gaussian distribution, multiply the observation likelihood by the prior probability of the corresponding grid to obtain a probability product, calculate the reciprocal of the sum of the probability products as the normalization factor, use the normalization factor to normalize the probability product to obtain the state probability of each grid, and mark each grid as an occupied state, a free state, or an unknown state based on the state probability to generate a grid map of the construction area.

[0087] Specifically, use a topographic survey robot to scan the construction area to obtain three-dimensional point cloud data of the construction area. To improve the data quality, it is necessary to preprocess the original point cloud data. Calculate the statistical mean and standard deviation of the three-dimensional point cloud data, and set the effective data range to be three times the standard deviation plus or minus the mean. Remove the outlier points outside this range to obtain the filtered three-dimensional point cloud data.

[0088] Next, perform downsampling on the filtered point cloud data. Divide the three-dimensional space into cubic voxels with a side length of 10 cm. For each voxel, calculate the centroid coordinates of all three-dimensional points inside it, and use the centroid coordinates as the downsampled point of the voxel. In this way, the amount of point cloud data can be greatly reduced while retaining the geometric features of the original point cloud.

[0089] Then, perform feature extraction on the downsampled point cloud. Set the search radius to 30 cm, and search for the neighboring points within this radius for each three-dimensional point. Based on these neighboring points, construct a 3x3 covariance matrix, and obtain the normal vector of each point through eigenvalue decomposition. Calculate the angle between the normal vectors of adjacent points, and take the median as the Gaussian kernel parameter σ. Use this parameter to construct a Gaussian kernel function: exp(-d 2 / (2σ 2 )), where d is the distance between two points. Calculate the sum of the Gaussian kernel function values between each point and its neighboring points to obtain the significance score of the point.

[0090] Sort the significance scores of all points in descending order, and select the top 30% of the points as candidate feature points. To make the distribution of feature points uniform, set the minimum spacing threshold to 5 cm, and remove the candidate points with a spacing less than this threshold to finally obtain a set of feature points.

[0091] For each point in the feature point set, calculate its Point Feature Histogram (PFH) descriptor. The PFH descriptor characterizes local features by statistically analyzing the geometric relationships between point pairs in the neighborhood of the feature point. After the calculation, use a KD tree to match the feature points to obtain an initial point cloud registration result.

[0092] To further optimize the registration result, an Iterative Closest Point (ICP) algorithm with adaptive weights is introduced. First, calculate the Euclidean distance and normal angle between corresponding feature point pairs in the initial registration result. Take the median of the distances as the distance parameter d0, and the median of the normal angles as the angle parameter θ0. Construct the distance weight term wd = exp(-d 2 / d0 2 ) and the angle weight term wθ = exp(-θ 2 / θ0 2 ). Multiply the two weight terms to obtain the adaptive weight w = wd * wθ. During the ICP iteration process, use this adaptive weight to weight the error, which can effectively suppress the influence of abnormal matching pairs and improve the registration accuracy.

[0093] Finally, generate a grid map based on the accurately registered point cloud data. Divide the construction area into square grids with a side length of 20 cm. For each grid, calculate the distance d from its center point to the nearest point cloud point. Assuming that the measurement error follows a Gaussian distribution with a mean of 0 and a standard deviation of σ, the observation likelihood can be expressed as p(z|x) = exp(-d 2 / (2σ 2 ). Multiply the observation likelihood by the prior probability p(x) of the grid to obtain the posterior probability p(x|z) ∝ p(z|x) * p(x). Normalize the posterior probabilities of all grids to obtain the state probability of each grid.

[0094] Mark the grids as three states according to the state probability: occupied, free, or unknown. Specifically, when the state probability is greater than 0.7, mark it as the occupied state; when it is less than 0.3, mark it as the free state; and when it is between 0.3 and 0.7, mark it as the unknown state. In this way, finally generate a probability grid map of the construction area.

[0095] In this embodiment, through preprocessing steps such as filtering and downsampling, abnormal points and redundant data are effectively removed, improving the quality and processing efficiency of the point cloud data; introducing the ICP algorithm with adaptive weights can effectively suppress the influence of abnormal matching and significantly improve the accuracy and robustness of point cloud registration; adopting the probability grid mapping method, considering measurement uncertainty, can more accurately represent the environmental state and provide a reliable basis for subsequent path planning and navigation decision-making.

[0096] In an alternative embodiment, positioning marker points within the construction area are identified through an image recognition algorithm, and the spatial coordinates of the positioning marker points are mapped to a grid map. Determining the distribution of the positioning marker points in the grid map includes:

[0097] Input the image collected by the surveying robot into the backbone feature network. The backbone feature network extracts image features through residual modules. In the residual modules, the spatial attention mechanism and the channel attention mechanism are used to perform weighted reconstruction on the spatial dimension and the channel dimension of the feature map respectively, obtaining the first-layer feature map. Input the first-layer feature map into the feature pyramid structure, and fuse features of different scales through upsampling and downsampling operations to obtain a multi-level feature map;

[0098] Input the multi-level feature map into the detection head network. The detection head network outputs the class scores and position coordinates of the positioning marker points. Based on the class scores and the position coordinates, determine the two-dimensional pixel coordinates of the positioning marker points in the image. Perform distortion correction on the image based on the internal parameter matrix of the binocular camera, calculate the disparity information of the left and right images, and combine the external parameter matrix of the binocular camera. Convert the two-dimensional pixel coordinates into depth information through the triangulation principle, and combine the two-dimensional pixel coordinates and the depth information to obtain the three-dimensional spatial coordinates of the positioning marker points;

[0099] Extract the corresponding feature points in the coordinate system corresponding to the camera coordinate system and the grid map, construct feature point matching pairs, calculate the rotation matrix and the translation vector, and optimize the rotation matrix and the translation vector using a non-linear optimization algorithm to obtain a coordinate transformation matrix. Map the three-dimensional spatial coordinates to the coordinate system of the grid map through the coordinate transformation matrix to obtain the spatial distribution of the positioning marker points.

[0100] First, input the image collected by the surveying robot into the backbone feature network. This backbone feature network adopts the ResNet-50 structure and extracts image features through residual modules. In each residual module, a spatial attention mechanism and a channel attention mechanism are introduced to perform weighted reconstruction on the feature map. The spatial attention mechanism generates a spatial weight map through convolution operations to weight each spatial position of the feature map. The channel attention mechanism uses global average pooling and fully connected layers to generate a channel weight vector to weight each channel of the feature map. After feature extraction and attention weighting in multiple residual modules, the first-layer feature map is obtained.

[0101] Next, input the first-layer feature map into the feature pyramid structure. The feature pyramid structure contains 5 scale levels, corresponding to the sizes of 1 / 4, 1 / 8, 1 / 16, 1 / 32, and 1 / 64 of the original image respectively. Through upsampling and downsampling operations, features of different scales are fused. Specifically, the feature map of a smaller scale is upsampled by a factor of 2 and fused with the feature map of a larger scale through element-wise addition; the feature map of a larger scale is downsampled by a factor of 2 and fused with the feature map of a smaller scale through element-wise addition. This multi-scale feature fusion can effectively extract semantic information of different scales and obtain a multi-level feature map.

[0102] Then, input the multi-level feature map into the detection head network. The detection head network contains a classification branch and a regression branch. The classification branch outputs the class scores of each candidate box through convolutional layers and the Sigmoid activation function, and the regression branch outputs the offset of the position coordinates of each candidate box through convolutional layers. According to the preset confidence threshold (such as 0.5) and non-maximum suppression threshold (such as 0.3), the final detection results are filtered to determine the two-dimensional pixel coordinates of the positioning marker points in the image. For example, the two-dimensional pixel coordinates of a detected positioning marker point are (320, 240).

[0103] Next, based on the internal parameter matrix of the binocular camera, the image is corrected for distortion. The internal parameter matrix contains parameters such as focal length, principal point coordinates, and distortion coefficients. Using these parameters, the distortion is corrected through a polynomial model to obtain the corrected left and right images. Then, the disparity information of the left and right images is calculated. A stereo matching algorithm based on local block matching is used to calculate the disparity value for each pixel of the left and right images. Combining with the external parameter matrix of the binocular camera (including the rotation matrix and translation vector between the cameras), the two-dimensional pixel coordinates are converted into depth information through the principle of triangulation. For example, the depth value of a certain positioning marker point is 5 meters. The two-dimensional pixel coordinates are combined with the depth information to obtain the three-dimensional spatial coordinates of the positioning marker point, such as (1.6m, 1.2m, 5m).

[0104] Finally, extract the corresponding feature points in the camera coordinate system and the coordinate system corresponding to the grid map, and construct feature point matching pairs. Select at least 3 pairs of non-collinear feature point matching pairs, such as the points (0, 0, 0), (1, 0, 0), (0, 1, 0) in the camera coordinate system and the points (10, 10, 0), (11, 10, 0), (10, 11, 0) in the grid map coordinate system. Based on these matching pairs, use the SVD decomposition method to calculate the initial rotation matrix and translation vector. Then, use the Levenberg-Marquardt algorithm to perform nonlinear optimization on the rotation matrix and translation vector to minimize the reprojection error and obtain an accurate coordinate transformation matrix. Map the three-dimensional spatial coordinates of the positioning marker points to the coordinate system of the grid map through this coordinate transformation matrix to obtain the spatial distribution of the positioning marker points in the grid map. For example, the coordinates of a certain positioning marker point in the grid map are (15.6m, 12.2m).

[0105] In this embodiment, by introducing the spatial attention and channel attention mechanisms, the effectiveness of feature extraction is enhanced, and the detection accuracy of the positioning marker points is improved. Multi-scale feature fusion can effectively extract semantic information of different scales and enhance the adaptability to positioning marker points of different sizes; by using a binocular camera and the principle of triangulation, the accurate acquisition of the three-dimensional coordinates of the positioning marker points is realized, and the accuracy of the depth information is improved compared with the monocular camera scheme. The coordinate transformation method based on nonlinear optimization improves the accuracy of coordinate mapping; the overall scheme realizes the full-process automatic processing from image acquisition to grid map mapping, improves the efficiency and accuracy of obtaining the spatial distribution of the positioning marker points, and provides reliable basic data for subsequent robot positioning and navigation.

[0106] In an alternative embodiment, the optimal motion trajectory of the surveying and mapping robot is planned by using the time-series reinforcement iterative neural network algorithm, including:

[0107] Construct a time-series reinforcement iterative neural network, form a state vector by combining the real-time position information of the surveying and mapping robot, the spatial distribution of the positioning marker points, and the local information of the grid map, and form an action vector by combining the discretized linear velocity and angular velocity. The time-series reinforcement iterative neural network includes an active action network and a target value network;

[0108] The active action network selects an action vector based on the current state vector, executes the action vector to obtain the next state vector, and the target value network estimates the value based on the next state vector; calculate the path length value, motion smoothness value, and marker point coverage value of the action vector, and combine them to form a reward value;

[0109] Construct state transition data from the current state vector, action vector, reward value, and next state vector, calculate the temporal difference error of the state transition data, assign priorities to the state transition data based on the temporal difference error, and store state transition data with different priorities in the experience replay pool; sample state transition data from the experience replay pool according to the priorities to train the active action network, and update the parameters of the active action network to the target value network according to a preset number of steps, repeat the iteration until convergence, and output the optimal motion trajectory for the fixed measurement robot to access the positioning marker points.

[0110] Specifically, the state vector contains the following information: the current position coordinates (x, y, θ) of the robot, where x and y are planar coordinates and θ is the orientation angle; the set of coordinates of unvisited positioning marker points {(x1, y1), (x2, y2),..., (xn, yn)}; a local grid map centered on the robot, with a size of 20×20, and the value of each grid is 0 (idle) or 1 (obstacle). The action vector contains the linear velocity v and the angular velocity ω. The value range of v is [0, 0.5] m / s, which is discretized into 5 values; the value range of ω is [-π / 4, π / 4] rad / s, which is discretized into 7 values. Therefore, the size of the action space is 5×7 = 35.

[0111] The active action network adopts a fully connected neural network structure. The number of neurons in the input layer is the same as the dimension of the state vector. The hidden layer contains two layers, with 128 neurons in each layer. The number of neurons in the output layer is 35, corresponding to 35 discrete actions. The structure of the target value network is the same as that of the active action network, but the output layer has only 1 neuron, which is used to estimate the state value. The activation function of both networks uses the ReLU function.

[0112] The active action network selects the action vector a based on the current state vector s t Specifically, input s t into the active action network to obtain the Q-value estimates of 35 actions. The ε-greedy strategy is used to select actions, that is, select the action with the maximum Q value with a probability of 1 - ε, and randomly select an action with a probability of ε. The initial value of ε is 0.9 and linearly decays to 0.1 as the training progresses. Execute the selected action a t and simulate the robot's movement for 0.5 seconds to obtain the next state s t t+1 t+1 .

[0113] The target value network estimates the value V(s t+1 ) based on the next state vector s t+1 Specifically, input s t+1 into the target value network to directly obtain the value estimate of this state.

[0114] Next, calculate the action a tReward value r t , including three parts: path length value r l , motion smoothness value r s and marker point coverage value r c . The path length value r l = -d, where d is the distance the robot moves after executing action a t . The motion smoothness value r s = -|ω t - ω t-1 |, where ω t and ω t-1 are the angular velocities at the current moment and the previous moment respectively. The marker point coverage value rc = n / N, where n is the number of marker points currently visited and N is the total number of marker points. The final reward value r t = w1 * r l + w2 * r s + w3 * r c , where w1, w2, w3 are weight coefficients, taking 0.3, 0.3, 0.4 respectively.

[0115] Construct the state transition data (s t , a t , r t , s t+1 + 1) from the current state s t , action a t , reward r t , and the next state s t . Calculate the temporal difference error δ t = r t + γV(s t+1 ) - Q(s t , a t ), where γ is the discount factor, taking 0.99; V(s t+1 ) is the next state value estimate output by the target value network; Q(s t , a t ) is the Q-value estimate of the current state-action pair output by the active action network. Based on |δ t |, assign a priority p = |δ t | + ε to the state transition data, where ε is a small positive number to prevent the priority from being 0. Store the data and its priority (s t , a t , r t , s t+1 , p) in the experience replay pool.

[0116] The experience replay pool is implemented using a priority queue data structure with a capacity of 10,000. When the pool is full, new data replaces the oldest data with the lowest priority. A mini-batch (size 64) is sampled from the experience replay pool according to the probability distribution of priority p to train the active action network. Specifically, the TD error of the sampled data is calculated as δ = r + γV(s') - Q(s, a), where V(s') and Q(s, a) are calculated by the target value network and the active action network respectively. Then the loss function L = (δ - Q(s, a)) 2 , and the parameters of the active action network are updated through backpropagation.

[0117] Every 100 updates of the active action network, its parameters are softly updated to the target value network: θ' = τθ + (1 - τ)θ', where θ and θ' are the parameters of the active action network and the target value network respectively, and τ is the soft update coefficient, taking 0.01. The above training process is repeated until convergence or the maximum number of iterations (10,000 times) is reached.

[0118] After training is completed, the trained active action network is used to plan the motion trajectory of the inspection robot. Starting from the starting position, the following steps are repeated until all marked points are visited: 1) Construct the current state vector s t ; 2) Input st into the active action network and select the action a with the largest Q value t ; 3) Execute a t and update the position of the robot. The finally obtained trajectory is the optimal motion trajectory.

[0119] In this embodiment, the time-series reinforcement iterative neural network algorithm is adopted, which can adaptively learn the optimal decision-making strategy in complex environments, and has stronger generalization ability and environmental adaptability compared with traditional path planning algorithms; the priority experience replay mechanism is introduced, which improves the learning efficiency of key experiences, accelerates the convergence speed of the network, and also improves the efficiency of the algorithm; considering multiple objectives such as path length, motion smoothness and marked point coverage, the obtained trajectory can achieve a better balance in multiple indicators and meet the actual application requirements.

[0120] In an alternative embodiment, the inspection robot is controlled to move along the optimal motion trajectory, and the inspection attitude data is collected in real time through the inertial measurement unit of the inspection robot. Based on the inspection attitude data, the inkjet device of the inspection robot is controlled to maintain a horizontal state. When the inspection robot moves to the preset inkjet position, the vertical distance between the nozzle of the inkjet device and the ground is measured and automatically adjusted. Through the vision positioning system, the image information of the preset inkjet position is collected, and combined with the preset inkjet template, the inkjet device is controlled to perform inkjet operations including:

[0121] Collect the acceleration component and angular velocity component through the inertial measurement unit of the fixed measurement robot, estimate the attitude quaternion through the extended Kalman filter algorithm, and calculate the roll angle, pitch angle and yaw angle according to the attitude quaternion;

[0122] Based on the attitude quaternion, establish the kinematic model of the inkjet device, calculate the target angles of the first rotating shaft and the second rotating shaft of the two-degree-of-freedom gimbal, control the rotation of the stepping motors of the first rotating shaft and the second rotating shaft respectively through the PID controller, and set the inkjet device to be in a horizontal state;

[0123] When reaching the preset inkjet position, measure the vertical distance between the nozzle and the ground through the laser range finder, compare the vertical distance with the preset vertical distance boundary to obtain the distance deviation value and the distance deviation change rate, establish the fuzzy control rule, input the fuzzy control rule into the fuzzy PID controller, adaptively adjust the PID parameters, control the rotation speed of the DC servo motor of the lifting mechanism, and adjust the height of the nozzle;

[0124] Collect the image of the preset inkjet position, perform grayscale and edge detection processing on the image to obtain the contour features, match the preset inkjet template with the contour features, and based on the matching result, determine the inkjet position coordinates and the spraying direction angle, generate the spraying instruction sequence, and the spraying instruction sequence includes the spraying instruction sequence of the nozzle position information, spraying pressure information and spraying time information, and send the spraying instruction sequence to the inkjet controller through the fieldbus;

[0125] During the inkjet process, collect the spraying image in real time, analyze the detection parameters of the spraying lines in the spraying image, the detection parameters include the width value, continuity value and clarity value, generate the analysis result, and when any one of the detection parameters is lower than the corresponding preset threshold, adaptively adjust the spraying pressure information and the spraying time information based on the analysis result until each detection parameter reaches the corresponding preset threshold to complete the inkjet operation.

[0126] This embodiment provides a fixed measurement robot inkjet method, and the specific implementation steps are as follows:

[0127] First, control the movement of the fixed measurement robot along the preset optimal movement trajectory. The optimal movement trajectory can be generated by a path planning algorithm, such as the A* algorithm or the RRT algorithm, considering factors such as obstacle avoidance and movement smoothness. The fixed measurement robot can adopt a differential drive structure and realize movement by controlling the rotation speeds of the left and right wheel motors.

[0128] During the movement of the surveying robot, the surveying attitude data is collected in real time through the inertial measurement unit (IMU) on the robot. The IMU includes a three-axis accelerometer and a three-axis gyroscope, and the sampling frequency can be set to 100 Hz. The accelerometer measures the acceleration components in three directions, and the gyroscope measures the angular velocity components of three axes.

[0129] The acceleration and angular velocity data collected by the IMU are input into the extended Kalman filter (EKF) algorithm to estimate the attitude quaternion. The EKF algorithm includes two steps: prediction and update. The prediction step predicts the state based on the system model, and the update step corrects the prediction result by combining the observed data. The optimal estimated attitude quaternion is obtained through iterative calculation.

[0130] The roll angle, pitch angle, and yaw angle of the surveying robot are calculated based on the attitude quaternion. Specifically, it can be achieved through the conversion formula from quaternion to Euler angle. For example, the roll angle is equal to the arctangent function of the four components x, y, z, and w in the quaternion.

[0131] Based on the calculated attitude quaternion, a kinematic model of the inkjet device is established. The inkjet device is installed on a two-degree-of-freedom cloud platform, and the two rotating shafts of the cloud platform are respectively driven by stepping motors. Through the forward kinematic solution, the target angles of the two rotating shafts are calculated to keep the inkjet device horizontal.

[0132] For example, when the attitude quaternion indicates that the robot tilts forward by 5 degrees, the first rotating shaft (pitch axis) needs to rotate backward by 5 degrees for compensation. The stepping motors of the two rotating shafts are respectively controlled by a PID controller to rotate to the target angles. The proportional, integral, and derivative parameters of the PID controller can be determined initially by the Ziegler-Nichols tuning method and then fine-tuned according to the actual response characteristics.

[0133] When the surveying robot moves to the preset inkjet position, the vertical distance between the nozzle and the ground is measured by the laser range finder installed on the robot. The measurement range of the laser range finder is 0.1 - 10 m, and the accuracy is ±1 mm. The measured actual distance is compared with the preset ideal distance to obtain the distance deviation value.

[0134] At the same time, the change rate of the distance deviation is calculated, that is, the change amount of the distance deviation per unit time. The distance deviation value and the change rate are used as inputs to establish fuzzy control rules. The fuzzy control rules can adopt the IF-THEN form. For example: IF the deviation is positive and the change rate is positive THEN the output is large positive.

[0135] Input the fuzzy control rules into the fuzzy PID controller to achieve the adaptive adjustment of PID parameters. The fuzzy PID controller looks up the rule table based on the fuzzy quantities of the deviation and the change rate to obtain the adjustment amount of the PID parameters, thereby dynamically adjusting the PID parameters. The output of the controller acts on the DC servo motor of the lifting mechanism to adjust the motor speed, and then adjust the height of the nozzle.

[0136] Collect the image of the preset coding position through the vision positioning system. The vision system includes an industrial camera and an LED light source. The camera resolution is 1920x1080, and the frame rate is 60fps. Perform grayscale processing on the collected image to convert the RGB three-channel image into a single-channel grayscale image.

[0137] Then perform Canny edge detection on the grayscale image to extract the contour features in the image. The Canny algorithm includes steps such as Gaussian filtering, calculating the gradient magnitude and direction, non-maximum suppression, and double-threshold detection. The high and low thresholds for edge detection can be set to 50 and 150.

[0138] Match the feature points of the pre-stored coding template with the extracted contour features. Use the SIFT algorithm to extract the feature points and descriptors, and then use the FLANN algorithm for nearest neighbor matching. Calculate the affine transformation matrix according to the matching results to determine the coding position coordinates and the direction angle.

[0139] Generate a spraying instruction sequence based on the matching results, including information such as the nozzle position, spraying pressure, and spraying time. For example, the nozzle position can be represented as relative coordinates (x, y, z), the spraying pressure range is 0.2 - 0.8MPa, and the spraying time is 0.1 - 2s. Send the instruction sequence to the coding controller through the CAN bus for execution.

[0140] During the coding process, collect and analyze the spraying image in real time. Extract parameters such as the width, continuity, and clarity of the spraying line. The width can be measured through morphological operations, the continuity is evaluated by detecting the number of breakpoints, and the clarity is analyzed through edge gradients.

[0141] Compare the detected parameters with the preset thresholds. For example, the width threshold is 1 - 3mm, the continuity threshold is 95%, and the clarity threshold is 0.8. When any parameter is lower than the threshold, adaptively adjust the spraying pressure and time according to the analysis results. The pressure can be adjusted within the range of ±0.1MPa, and the time can be adjusted within the range of ±0.2s. Through multiple iterations of optimization, until all parameters reach the preset thresholds, complete the high-quality coding operation.

[0142] In this embodiment, through inertial measurement and attitude estimation, combined with a two-degree-of-freedom pan-tilt, the automatic horizontal holding of the inkjet printing device is achieved, improving the inkjet printing accuracy and quality; the fuzzy PID control algorithm is used to adaptively adjust the height of the nozzle, improving the robustness and adaptability of the system, and it can cope with different ground conditions; based on visual positioning and real-time image analysis, the precise positioning of the inkjet printing position and the dynamic optimization of the spraying parameters are realized, ensuring the consistency and reliability of the inkjet printing effect.

[0143] In an alternative embodiment, the acceleration components and angular velocity components are collected through the inertial measurement unit of the surveying robot, the attitude quaternion is estimated through the extended Kalman filter algorithm, and the roll angle, pitch angle, and yaw angle calculated according to the attitude quaternion include:

[0144] The acceleration components of the gravitational acceleration in three orthogonal axes are collected through the three-axis acceleration sensor of the inertial measurement unit, the angular velocity components of the robot rotating around three orthogonal axes are collected through the three-axis gyroscope, and the acceleration components and the angular velocity components are sampled and filtered through the signal acquisition module;

[0145] Based on the angular velocity components, a system state equation is established, the attitude quaternion is set as the state vector, the angular velocity is set as the system input, and a nonlinear state transition model is constructed; based on the acceleration components, an observation equation is established, the projection of the gravitational vector in the body coordinate system is set as the observable quantity, and a nonlinear observation model is constructed according to the relationship between the attitude quaternion and the gravitational vector;

[0146] The nonlinear state transition model is expanded by the first-order Taylor expansion to obtain the state Jacobian matrix, and the nonlinear observation model is expanded by the first-order Taylor expansion to obtain the observation Jacobian matrix;

[0147] The extended Kalman filter algorithm is used to estimate the attitude, including: using the state Jacobian matrix and the nonlinear state transition model for state prediction to obtain the predicted attitude quaternion and the predicted state covariance; using the observation Jacobian matrix, the nonlinear observation model, and the predicted state covariance to calculate the Kalman gain, multiplying the Kalman gain by the observation residual to obtain the state correction amount, adding the state correction amount to the predicted attitude quaternion to obtain the estimated attitude quaternion, and updating the state covariance at the same time;

[0148] The estimated quaternion is normalized, and each component is divided by the corresponding modulus length to obtain the unit attitude quaternion; according to the four components of the unit attitude quaternion, through the conversion formula from the unit attitude quaternion to the Euler angles, three rotation angles describing the robot's attitude are calculated, including the roll angle around the X axis, the pitch angle around the Y axis, and the yaw angle around the Z axis.

[0149] Specifically, the acceleration components and angular velocity components are collected by the inertial measurement unit of the survey robot. The inertial measurement unit includes a three-axis accelerometer and a three-axis gyroscope. The three-axis accelerometer is used to measure the components of the gravitational acceleration in the three orthogonal x, y, and z axes, with a sampling frequency of 100 Hz. The three-axis gyroscope is used to measure the angular velocity components of the robot rotating around the three orthogonal x, y, and z axes, with a sampling frequency of 200 Hz. The collected acceleration and angular velocity signals are filtered through a low-pass filter with a cut-off frequency of 20 Hz to eliminate high-frequency noise.

[0150] The filtered acceleration and angular velocity signals are input into an extended Kalman filter for attitude estimation. First, establish the system state equation, set the attitude quaternion q = [q0, q1, q2, q3] as the state vector, and the angular velocity ω = [ωx, ωy, ωz] as the system input, and construct a non-linear state transition model f(q, ω). Then establish the observation equation, set the projection of the gravity vector g in the body coordinate system as the observable quantity, and construct a non-linear observation model h(q) according to the relationship between the attitude quaternion and the gravity vector.

[0151] Perform a first-order Taylor expansion on the non-linear state transition model f(q, ω) to obtain a 4×4 state Jacobian matrix F. Perform a first-order Taylor expansion on the non-linear observation model h(q) to obtain a 3×4 observation Jacobian matrix H.

[0152] Use the state Jacobian matrix F and the non-linear state transition model f(q, ω) for state prediction to obtain the predicted attitude quaternion q_pred and the predicted state covariance P_pred. Use the observation Jacobian matrix H, the non-linear observation model h(q), and the predicted state covariance P_pred to calculate the Kalman gain K. Multiply the Kalman gain K by the observation residual to obtain the state correction Δq, add the state correction Δq to the predicted attitude quaternion q_pred to obtain the estimated attitude quaternion q_est, and update the state covariance P at the same time.

[0153] Normalize the estimated attitude quaternion q_est, divide each component by the modulus of the quaternion to obtain the unit attitude quaternion q_unit = [q0, q1, q2, q3]. According to the four components of the unit attitude quaternion q_unit, through the conversion relationship from quaternion to Euler angles, calculate three rotation angles describing the robot's attitude: the roll angle φ around the x-axis, the pitch angle θ around the y-axis, and the yaw angle ψ around the z-axis.

[0154] The specific implementation steps are as follows:

[0155] First, collect acceleration and angular velocity signals through the inertial measurement unit. The three-axis accelerometer measures the acceleration component ax = 0.1 m / s 2 , ay = -0.2 m / s2 where \(a_z = 9.8\ m / s\) 2 The three-axis gyroscope measures the angular velocity components \(\omega_x = 0.01\ rad / s\), \(\omega_y=-0.02\ rad / s\), and \(\omega_z = 0.03\ rad / s\).

[0156] Then, a low-pass filter is applied to the original signal. A second-order Butterworth low-pass filter with a cut-off frequency of 20 Hz is used. After filtering, we obtain \(a_{x\_f}=0.09\ m / s\) 2 where \(a_{y\_f}=-0.19\ m / s\) 2 where \(a_{z\_f}=9.79\ m / s\) 2 where \(\omega_{x\_f}=0.009\ rad / s\), \(\omega_{y\_f}=-0.018\ rad / s\), and \(\omega_{z\_f}=0.028\ rad / s\).

[0157] Next, a non-linear state transition model \(f(q,\omega)\) and a non-linear observation model \(h(q)\) are constructed. The state transition model describes the variation of the attitude quaternion \(q\) over time, and the observation model describes the relationship between the projection of the gravity vector \(g\) in the body coordinate system and the attitude quaternion \(q\).

[0158] The first-order Taylor expansion of the non-linear model is performed to obtain the state Jacobian matrix \(F\) and the observation Jacobian matrix \(H\). The \(F\) matrix reflects the influence of small changes in the state vector \(q\) on the state transition, and the \(H\) matrix reflects the influence of small changes in the state vector \(q\) on the observed quantity.

[0159] State prediction is carried out. Using the estimated attitude quaternion \(q\) at the previous moment and the measured angular velocity \(\omega\), the predicted attitude quaternion \(q_{pred}=[0.998, 0.005, -0.01, 0.015]\) is calculated through the non-linear state transition model \(f(q,\omega)\). At the same time, the predicted state covariance \(P_{pred}\) is updated using the state Jacobian matrix \(F\).

[0160] The observation residual is calculated. The predicted observation value is calculated using the non-linear observation model \(h(q)\) and compared with the actual measured acceleration value to obtain the observation residual.

[0161] The Kalman gain \(K\) is calculated. Based on the predicted state covariance \(P_{pred}\), the observation Jacobian matrix \(H\), and the observation noise covariance \(R\), the Kalman gain \(K\) is calculated.

[0162] The state estimate is updated. The Kalman gain \(K\) is multiplied by the observation residual to obtain the state correction \(\Delta q=[0.001, -0.002, 0.003, -0.004]\), which is added to the predicted attitude quaternion \(q_{pred}\) to obtain the estimated attitude quaternion \(q_{est}=[0.999, 0.003, -0.007, 0.011]\). At the same time, the state covariance \(P\) is updated.

[0163] Normalize the estimated attitude quaternion \(q_{est}\) to obtain the unit attitude quaternion \(q_{unit}=[0.9990, 0.0030, -0.0070, 0.0110]\).

[0164] Finally, according to the unit attitude quaternion \(q_{unit}\), the Euler angles are calculated as follows: roll angle \(\varphi = 0.35^{\circ}\), pitch angle \(\theta=-0.80^{\circ}\), yaw angle \(\psi = 1.26^{\circ}\). These three angles describe the current attitude of the robot.

[0165] In this embodiment, the extended Kalman filter algorithm is used to fuse the accelerometer and gyroscope data, making full use of the complementary characteristics of the two sensors. It not only retains the advantage of short-term stability of the gyroscope but also uses the long-term stability characteristic of the accelerometer to correct the cumulative error, improving the accuracy and reliability of attitude estimation. By constructing a nonlinear state space model and performing linearization processing, this method can effectively handle the nonlinear problems in attitude estimation and has better performance than the traditional linear Kalman filter. Using quaternions to represent the attitude avoids the singularity problem in the Euler angle representation and can achieve stable estimation within the full attitude range, which is suitable for attitude measurement in complex motion scenarios of robots.

[0166] In an alternative embodiment, after the inkjet coding operation is completed, a high-definition image of the inkjet coding area is collected. Using an image processing algorithm, the contour features of the inkjet coding pattern are extracted, and the contour features are compared with the preset inkjet coding template to calculate the actual deviation of the inkjet coding position. When the actual deviation exceeds the preset threshold, based on the actual deviation, a deep learning model is used to adaptively optimize the motion parameters of the measuring robot, and the actual deviation is uploaded to the construction management system in real time to generate a construction quality report including:

[0167] Collect the spectral image of the inkjet coding area through a hyperspectral camera, project a grid pattern through a structured light projector and collect the structured light image. The spectral image and the structured light image are input into a synchronous trigger circuit for timing alignment to obtain standard dual-modal image data;

[0168] Input the standard dual-modal image data into a spectral-spatial joint feature extraction network. The spectral attention module of the spectral-spatial joint feature extraction network processes the spectral image to obtain spectral features, and the spatial pyramid pooling layer processes the structured light image to obtain spatial features. The spectral features and the spatial features are fused through a cross-modal feature fusion module to obtain the fusion features of the inkjet coding area;

[0169] Perform edge detection on the fusion features to obtain the edge point set of the inkjet coding pattern. Input the edge point set into a dynamic graph convolutional network, extract local structure information through an adaptive graph convolutional layer, and use a graph attention mechanism to enhance the weight of the local structure information to obtain the contour features of the inkjet coding pattern;

[0170] Match the contour features with a preset inkjet coding template, and obtain the initial pose deviation through one-stage registration by a graph matching network. Based on the initial pose deviation, perform two-stage registration through the iterative closest point algorithm to obtain the actual pose deviation of the inkjet coding pattern;

[0171] Input the actual pose deviation into a hierarchical reinforcement learning network. The high-level policy network of the hierarchical reinforcement learning network generates a parameter optimization strategy based on the actual pose deviation, and the low-level execution network adjusts the motion parameters of the metrology robot based on the parameter optimization strategy;

[0172] Input the motion parameters into a hybrid parameter optimizer. The hybrid parameter optimizer uses Bayesian optimization to determine the parameter search space, and iteratively optimizes the motion parameters within the parameter search space through an evolutionary algorithm to obtain the optimized motion parameters;

[0173] Based on a pre-constructed uncertainty evaluation model, input the optimized motion parameters into the uncertainty evaluation model, calculate the measurement uncertainty introduced by the measurement process, the model uncertainty introduced by the model prediction, and the parameter uncertainty introduced by the parameter optimization respectively, and input the measurement uncertainty, the model uncertainty, and the parameter uncertainty into the evidence theory model to obtain the credibility evaluation value of the parameter optimization result.

[0174] After the inkjet coding operation is completed, first collect the high-definition image of the inkjet coding area. Specifically, use a high-resolution industrial camera to photograph the inkjet coding area to obtain an RGB color image with a resolution of not less than 4K. At the same time, use a hyperspectral camera to image the inkjet coding area to obtain a hyperspectral data cube in the wavelength range of 400 - 1000nm. In addition, project a grid-like stripe pattern onto the inkjet coding area through a structured light projector, and use a high-speed camera to capture the structured light image.

[0175] Next, input the collected spectral image and structured light image into a synchronous trigger circuit for timing alignment. Specifically, use a high-precision clock signal to control the exposure moments of the hyperspectral camera and the structured light camera to ensure that the acquisition time error of the two images is less than 1ms. Then, through an image registration algorithm, such as the phase correlation method, align the spatial position relationship of the two images to finally obtain the standard dual-modal image data synchronized in time and space.

[0176] Input the standard bimodal image data into a pre-trained spectral-spatial joint feature extraction network. The spectral attention module of this network first processes the spectral image: extracting the spectral features of each pixel point through one-dimensional convolution, and then using the channel attention mechanism to weight the features of different bands to highlight the key band information. At the same time, the spatial pyramid pooling layer performs multi-scale feature extraction on the structured light image: performing pooling operations on the image through pooling kernels of different sizes to obtain spatial feature maps of different scales. Finally, the cross-modal feature fusion module uses bilinear pooling to fuse the spectral features and multi-scale spatial features to obtain a feature representation of the inkjet printing area that combines spectral and spatial information.

[0177] Perform edge detection on the fused features to extract the contour information of the inkjet printing pattern. Specifically, first use the Canny operator to perform edge detection on the fused feature map to obtain a preliminary edge map. Then, use morphological operations to refine and connect the edge map to obtain a continuous edge curve. Next, use the Douglas-Peucker algorithm to approximate the edge curve into a polygon, extract the key edge points, and form an edge point set. Input the edge point set into the dynamic graph convolutional network. The adaptive graph convolutional layer of this network dynamically constructs a graph structure according to the spatial relationship between points and extracts local structural features through graph convolutional operations. Then, use the graph attention mechanism to weight the features of different nodes to highlight the important structural information. Finally, obtain a feature representation that accurately describes the contour of the inkjet printing pattern.

[0178] Match the extracted contour features with a preset inkjet printing template. First, use the graph matching network for rough registration: input the contour features and template features into the Siamese network, calculate the feature similarity matrix, and then use the Hungarian algorithm to solve the optimal match to obtain a preliminary pose deviation estimate. Then, based on the preliminary estimation result, use the iterative closest point (ICP) algorithm for fine registration: iteratively calculate the corresponding point pairs and minimize the distance between the point pairs to continuously optimize the pose estimation, and finally obtain the actual pose deviation with sub-pixel accuracy.

[0179] Input the actual pose deviation into the hierarchical reinforcement learning network to optimize the motion parameters of the metrology robot. The high-level policy network of this network first generates a parameter optimization strategy according to the pose deviation, such as adjusting the joint angles, end position, etc. Then, the low-level execution network specifically adjusts the kinematic parameters of the robot according to the optimization strategy, such as joint angles, linear velocity, angular velocity, etc. The optimization process uses a model-based reinforcement learning algorithm, such as DDPG, to continuously improve the policy network by interacting with the simulation environment, and finally obtain a robust parameter optimization scheme.

[0180] The preliminarily optimized motion parameters are input into a hybrid parameter optimizer for further optimization. First, the Bayesian optimization algorithm is used to determine the parameter search space: a Gaussian process regression model is constructed based on historical optimization data to predict the performance of different parameter combinations, and the most promising region is selected as the search space. Then, an evolutionary algorithm is adopted within this space for parameter optimization: through operations such as selection, crossover, and mutation, new parameter combinations are generated and their performance is evaluated, and finally, the globally optimal motion parameters are obtained after multiple generations of evolution.

[0181] Finally, an uncertainty assessment is performed on the optimized motion parameters. First, the uncertainty introduced by the measurement process is evaluated based on the Monte Carlo method: by repeating measurements multiple times, the distribution characteristics of the measurement results are analyzed. Then, the uncertainty of model prediction is evaluated using the ensemble learning method: multiple models are trained and the dispersion degree of their prediction results is analyzed. Finally, the sensitivity analysis method is used to evaluate the uncertainty introduced by parameter optimization: by perturbing the optimized parameters and observing the changes in the system response. The three types of uncertainties are input into an evidence fusion model based on the Dempster-Shafer theory, and the credibility of the parameter optimization results is comprehensively calculated to provide a reliable basis for subsequent decision-making.

[0182] In this embodiment, by combining hyperspectral and structured light imaging technologies, rich spectral information is acquired, and the geometric structure of the inkjet pattern is accurately captured, laying a solid foundation for subsequent feature extraction and matching. At the same time, advanced algorithms such as deep learning and graph networks are adopted to achieve precise modeling and feature expression of the inkjet pattern, greatly improving the accuracy and robustness of detection; through hierarchical reinforcement learning and hybrid optimization strategies, the motion parameters of the robot can be dynamically adjusted according to the actual detection results, effectively compensating for system errors and environmental disturbances. This closed-loop optimization mechanism significantly improves the adaptability and stability of the system, ensuring consistent performance during long-term operation; by comprehensively considering the uncertainties introduced by measurement, model, and parameter optimization, and using the evidence theory for fusion, a credibility assessment of the optimization results is given. This provides a reliable basis for subsequent decision-making, helps to timely discover potential risks and take corresponding measures, and improves the reliability and safety of the entire system.

[0183] Figure 2 FIG. is a schematic structural diagram of a real-time inkjet and position remeasurement system for a theodolite robot during signal construction in an embodiment of the present invention, as Figure 2 shown, the system includes:

[0184] The first unit is used to scan the construction area through a surveying robot, obtain the three-dimensional point cloud data of the construction area, rasterize the construction area based on the three-dimensional point cloud data to generate a raster map, collect the image information of the construction area by using the surveying robot, identify the positioning marker points in the construction area through an image recognition algorithm, map the spatial coordinates of the positioning marker points to the raster map, determine the distribution of the positioning marker points in the raster map, and adopt a time-series enhanced iterative neural network algorithm to plan the optimal motion trajectory of the surveying robot;

[0185] The second unit is used to control the movement of the surveying robot along the optimal motion trajectory, collect the surveying attitude data in real time through the inertial measurement unit of the surveying robot, control the inkjet device of the surveying robot to maintain a horizontal state based on the surveying attitude data, measure and automatically adjust the vertical distance between the nozzle of the inkjet device and the ground when the surveying robot moves to a preset inkjet position, collect the image information of the preset inkjet position through a vision positioning system, and control the inkjet device to perform inkjet operations in combination with a preset inkjet template;

[0186] The third unit is used to collect the high-definition image of the inkjet area after the inkjet operation is completed, extract the contour features of the inkjet pattern by using an image processing algorithm, compare the contour features with the preset inkjet template, calculate the actual deviation of the inkjet position, when the actual deviation exceeds a preset threshold, adaptively optimize the motion parameters of the surveying robot based on the actual deviation by using a deep learning model, and upload the actual deviation to the construction management system in real time to generate a construction quality report.

[0187] In the third aspect of the embodiments of the present invention,

[0188] A kind of electronic device is provided, including:

[0189] A processor;

[0190] A memory for storing instructions executable by the processor;

[0191] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0192] In the fourth aspect of the embodiments of the present invention,

[0193] A computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0194] The present invention can be a method, a device, a system and / or a computer program product. The computer program product can include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0195] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time inkjet coding and position re-measurement of a surveying robot during signal construction, characterized in that, Including: Scanning the construction area by a surveying robot to obtain the three-dimensional point cloud data of the construction area, rasterizing the construction area based on the three-dimensional point cloud data to generate a raster map, collecting the image information of the construction area by the surveying robot, identifying the positioning marker points in the construction area through an image recognition algorithm, mapping the spatial coordinates of the positioning marker points to the raster map, determining the distribution of the positioning marker points in the raster map, and using a time-series enhanced iterative neural network algorithm to plan the optimal movement trajectory of the surveying robot; Controlling the movement of the surveying robot along the optimal movement trajectory, collecting the surveying attitude data in real time through the inertial measurement unit of the surveying robot, controlling the inkjet device of the surveying robot to maintain a horizontal state based on the surveying attitude data, measuring and automatically adjusting the vertical distance between the nozzle of the inkjet device and the ground when the surveying robot moves to a preset inkjet position, collecting the image information of the preset inkjet position through a vision positioning system, and combining with a preset inkjet template to control the inkjet device to perform inkjet operations; After the inkjet operation is completed, collecting the high-definition image of the inkjet area, using an image processing algorithm to extract the contour features of the inkjet pattern, comparing the contour features with the preset inkjet template, calculating the actual deviation of the inkjet position, when the actual deviation exceeds the preset threshold, adaptively optimizing the movement parameters of the surveying robot based on the actual deviation using a deep learning model, uploading the actual deviation to the construction management system in real time, and generating a construction quality report.

2. The method according to claim 1, characterized in that, Scanning the construction area by a surveying robot to obtain the three-dimensional point cloud data of the construction area, and rasterizing the construction area based on the three-dimensional point cloud data to generate a raster map includes: Scanning the construction area by a surveying robot to obtain the three-dimensional point cloud data of the construction area, calculating the statistical mean and standard deviation of the three-dimensional point cloud data, determining the effective data range, removing the abnormal data outside the effective data range to obtain filtered three-dimensional point cloud data, dividing the filtered three-dimensional point cloud data into voxels according to a preset voxel size, calculating the centroid coordinates of the three-dimensional points contained in each voxel, and using the centroid coordinates as the downsampled points of the corresponding voxels to obtain downsampled three-dimensional point cloud data; Selecting the neighboring three-dimensional points within the corresponding radius range of each three-dimensional point in the downsampled three-dimensional point cloud data according to a preset radius, constructing a covariance matrix based on the neighboring three-dimensional points, calculating the normal vector of each three-dimensional point through the covariance matrix, determining the vector difference between the normal vectors, using the median of the vector difference as the Gaussian kernel parameter, constructing a Gaussian kernel function based on the Gaussian kernel parameter, and calculating the significance score of each three-dimensional point; Sorting the significance scores in descending order, selecting the top 30% of the three-dimensional points with the sorted significance scores as candidate feature points, judging the distance between the candidate feature points, removing the candidate feature points with a distance less than the preset distance threshold to obtain a feature point set; calculating the point feature histogram descriptor of each feature point in the feature point set and performing feature matching to obtain an initial registration result; Calculate the distance values and normal angle values between corresponding feature points in the initial registration result, take the median of the distance values as the distance parameter, and the median of the normal angle values as the angle parameter; construct a distance weight term based on the distance parameter, construct an angle weight term based on the angle parameter, multiply the distance weight term by the angle weight term to obtain an adaptive weight, and use the adaptive weight for iterative optimization to obtain an accurate registration result; Divide the construction area into grids of a preset size, calculate the distance between each grid and the three-dimensional point cloud data in the accurate registration result, calculate the observation likelihood of the Gaussian distribution, multiply the observation likelihood by the prior probability of the corresponding grid to obtain a probability product, calculate the reciprocal of the cumulative sum of the probability products as the normalization factor, and use the normalization factor to normalize the probability product to obtain the state probability of each grid. Based on the state probability, mark each grid as an occupied state, a free state, or an unknown state, and generate a grid map of the construction area.

3. The method according to claim 2, wherein Identify the positioning marker points in the construction area through an image recognition algorithm, map the spatial coordinates of the positioning marker points to the grid map, and determine the distribution of the positioning marker points in the grid map, including: Input the image collected by the surveying robot into the backbone feature network. The backbone feature network extracts image features through residual modules. In the residual modules, use the spatial attention mechanism and the channel attention mechanism to perform weighted reconstruction on the spatial dimension and the channel dimension of the feature map respectively to obtain the first-layer feature map; input the first-layer feature map into the feature pyramid structure, and fuse features of different scales through upsampling and downsampling operations to obtain a multi-level feature map; Input the multi-level feature map into the detection head network. The detection head network outputs the class scores and position coordinates of the positioning marker points. Based on the class scores and the position coordinates, determine the two-dimensional pixel coordinates of the positioning marker points in the image; perform distortion correction on the image based on the internal parameter matrix of the binocular camera, calculate the disparity information of the left and right images, and combine the external parameter matrix of the binocular camera to convert the two-dimensional pixel coordinates into depth information through the principle of triangulation. Combine the two-dimensional pixel coordinates with the depth information to obtain the three-dimensional spatial coordinates of the positioning marker points; Extract the corresponding feature points in the coordinate system corresponding to the camera coordinate system and the grid map, construct feature point matching pairs, calculate the rotation matrix and the translation vector, and use a non-linear optimization algorithm to optimize the rotation matrix and the translation vector to obtain a coordinate transformation matrix. Map the three-dimensional spatial coordinates to the coordinate system of the grid map through the coordinate transformation matrix to obtain the spatial distribution of the positioning marker points.

4. The method according to claim 3, characterized in that, Adopt the time-series reinforcement iterative neural network algorithm to plan the optimal motion trajectory of the surveying robot, including: Construct a time-series reinforcement iterative neural network, form a state vector by combining the real-time position information of the surveying robot, the spatial distribution of the positioning marker points, and the local information of the grid map, and form an action vector by combining the discretized linear velocity and angular velocity. The time-series reinforcement iterative neural network includes a main action network and a target value network; The active action network selects an action vector based on the current state vector, executes the action vector to obtain the next state vector, and the target value network estimates the value based on the next state vector; calculates the path length value, motion smoothness value, and marker point coverage value of the action vector, and combines them to form a reward value; Constructs state transition data with the current state vector, action vector, reward value, and next state vector, calculates the temporal difference error of the state transition data, assigns priorities to the state transition data based on the temporal difference error, stores state transition data with different priorities in the experience replay pool; samples state transition data from the experience replay pool according to the priorities to train the active action network, updates the parameters of the active action network to the target value network according to a preset number of steps, repeats the iteration until convergence, and outputs the optimal motion trajectory for the survey robot to access the positioning marker points.

5. The method according to claim 1, characterized in that Controls the movement of the survey robot along the optimal motion trajectory, real-time collects survey attitude data through the inertial measurement unit of the survey robot, based on the survey attitude data, controls the inkjet device of the survey robot to maintain a horizontal state, when the survey robot moves to the preset inkjet position, measures and automatically adjusts the vertical distance between the nozzle of the inkjet device and the ground, collects image information of the preset inkjet position through the vision positioning system, and combines with the preset inkjet template to control the inkjet device to perform inkjet operations including: Collects the acceleration component and angular velocity component through the inertial measurement unit of the survey robot, estimates the attitude quaternion through the extended Kalman filter algorithm, and calculates the roll angle, pitch angle, and yaw angle according to the attitude quaternion; Based on the attitude quaternion, establishes the kinematic model of the inkjet device, calculates the target angles of the first and second axes of the two-degree-of-freedom pan-tilt, controls the stepping motors of the first and second axes to rotate respectively through the PID controller, and sets the inkjet device to a horizontal state; When reaching the preset inkjet position, measures the vertical distance between the nozzle and the ground through the laser range sensor, compares the vertical distance with the preset vertical distance boundary to obtain the distance deviation value and the distance deviation change rate, establishes a fuzzy control rule, inputs the fuzzy control rule into the fuzzy PID controller, adaptively adjusts the PID parameters, and controls the rotation speed of the DC servo motor of the lifting mechanism to adjust the nozzle height; Collects the image of the preset inkjet position, performs grayscale and edge detection processing on the image to obtain contour features, matches the preset inkjet template with the contour features, determines the inkjet position coordinates and spraying direction angle based on the matching result, generates a spraying instruction sequence, the spraying instruction sequence includes spraying instructions containing nozzle position information, spraying pressure information, and spraying time information, and sends the spraying instruction sequence to the inkjet controller through the fieldbus; During the inkjet coding process, spray images are collected in real time, and detection parameters of spray lines in the spray images are analyzed. The detection parameters include width value, continuity value, and clarity value, and an analysis result is generated. When any one of the detection parameters is lower than the corresponding preset threshold, the spray pressure information and the spray time information are adaptively adjusted based on the analysis result until each detection parameter reaches the corresponding preset threshold, and the inkjet coding operation is completed.

6. The method according to claim 5, characterized in that, Acceleration components and angular velocity components are collected through the inertial measurement unit of the positioning and measuring robot, and attitude quaternions are estimated through the extended Kalman filter algorithm. The roll angle, pitch angle, and yaw angle calculated according to the attitude quaternions include: Acceleration components of gravitational acceleration in three orthogonal axes are collected through the three-axis acceleration sensor of the inertial measurement unit, and angular velocity components of the robot rotating around three orthogonal axes are collected through the three-axis gyroscope. The acceleration components and the angular velocity components are sampled and filtered through the signal acquisition module. Based on the angular velocity components, a system state equation is established, the attitude quaternion is set as the state vector, and the angular velocity is set as the system input to construct a nonlinear state transition model; based on the acceleration components, an observation equation is established, the projection of the gravity vector in the body coordinate system is set as the observed quantity, and a nonlinear observation model is constructed according to the relationship between the attitude quaternion and the gravity vector. The nonlinear state transition model is expanded by the first-order Taylor expansion to obtain the state Jacobian matrix, and the nonlinear observation model is expanded by the first-order Taylor expansion to obtain the observation Jacobian matrix. The extended Kalman filter algorithm is used to estimate the attitude, including: using the state Jacobian matrix and the nonlinear state transition model for state prediction to obtain the predicted attitude quaternion and the predicted state covariance; using the observation Jacobian matrix, the nonlinear observation model, and the predicted state covariance to calculate the Kalman gain, multiplying the Kalman gain by the observation residual to obtain the state correction amount, adding the state correction amount to the predicted attitude quaternion to obtain the estimated attitude quaternion, and updating the state covariance at the same time. The estimated quaternion is normalized, and each component is divided by the corresponding norm length to obtain the unit attitude quaternion; according to the four components of the unit attitude quaternion, through the conversion formula from the unit attitude quaternion to the Euler angle, three rotation angles describing the robot's attitude are calculated, including the roll angle around the X axis, the pitch angle around the Y axis, and the yaw angle around the Z axis.

7. The method according to claim 1, wherein After the inkjet coding operation is completed, a high-definition image of the inkjet coding area is collected, and the contour features of the inkjet coding pattern are extracted using image processing algorithms. The contour features are compared with the preset inkjet coding template, and the actual deviation of the inkjet coding position is calculated. When the actual deviation exceeds the preset threshold, the motion parameters of the positioning and measuring robot are adaptively optimized based on the actual deviation using a deep learning model, and the actual deviation is uploaded to the construction management system in real time to generate a construction quality report including: Collect the spectral image of the inkjet printing area through a hyperspectral camera, project a lattice pattern through a structured light projector and collect the structured light image, input the spectral image and the structured light image into a synchronous trigger circuit for timing alignment to obtain standard bimodal image data; Input the standard bimodal image data into a spectral-spatial joint feature extraction network. The spectral attention module of the spectral-spatial joint feature extraction network processes the spectral image to obtain spectral features, and the spatial pyramid pooling layer processes the structured light image to obtain spatial features. The spectral features and the spatial features are fused through a cross-modal feature fusion module to obtain the fusion features of the inkjet printing area; Perform edge detection on the fusion features to obtain the edge point set of the inkjet pattern. Input the edge point set into a dynamic graph convolutional network. Through the adaptive graph convolutional layer, extract local structure information, and use the graph attention mechanism to enhance the weight of the local structure information to obtain the contour features of the inkjet pattern; Match the contour features with a preset inkjet template, perform first-stage registration through a graph matching network to obtain the initial pose deviation, and based on the initial pose deviation, perform second-stage registration through the iterative closest point algorithm to obtain the actual pose deviation of the inkjet pattern; Input the actual pose deviation into a hierarchical reinforcement learning network. The high-level policy network of the hierarchical reinforcement learning network generates a parameter optimization strategy based on the actual pose deviation, and the low-level execution network adjusts the motion parameters of the surveying robot based on the parameter optimization strategy; Input the motion parameters into a hybrid parameter optimizer. The hybrid parameter optimizer uses Bayesian optimization to determine the parameter search space, and iteratively optimizes the motion parameters within the parameter search space through an evolutionary algorithm to obtain the optimized motion parameters; Based on a pre-constructed uncertainty evaluation model, input the optimized motion parameters into the uncertainty evaluation model, calculate the measurement uncertainty introduced by the measurement process, the model uncertainty introduced by the model prediction, and the parameter uncertainty introduced by the parameter optimization respectively. Input the measurement uncertainty, the model uncertainty, and the parameter uncertainty into the evidence theory model to obtain the credibility evaluation value of the parameter optimization result.

8. A real-time inkjet coding and position remeasurement system for a surveying robot in signal construction, which is used to implement the method described in any one of the foregoing claims 1-7, and is characterized in that, Including: The first unit is used to scan the construction area through a surveying robot, obtain the three-dimensional point cloud data of the construction area, rasterize the construction area based on the three-dimensional point cloud data to generate a raster map, collect the image information of the construction area by the surveying robot, identify the positioning marker points in the construction area through an image recognition algorithm, map the spatial coordinates of the positioning marker points to the raster map, determine the distribution of the positioning marker points in the raster map, and use the time-series reinforcement iterative neural network algorithm to plan the optimal motion trajectory of the surveying robot; A second unit, configured to control the movement of the fixed measurement robot along the optimal movement trajectory, collect real-time fixed measurement attitude data through the inertial measurement unit of the fixed measurement robot, and based on the fixed measurement attitude data, control the inkjet printing device of the fixed measurement robot to maintain a horizontal state. When the fixed measurement robot moves to a preset inkjet printing position, measure and automatically adjust the vertical distance between the nozzle of the inkjet printing device and the ground, collect image information of the preset inkjet printing position through a vision positioning system, and in combination with a preset inkjet printing template, control the inkjet printing device to perform inkjet printing operations; A third unit, configured to, after the inkjet printing operation is completed, collect a high-definition image of the inkjet printing area, extract the contour features of the inkjet printing pattern using an image processing algorithm, compare the contour features with the preset inkjet printing template, calculate the actual deviation of the inkjet printing position, and when the actual deviation exceeds a preset threshold, adaptively optimize the movement parameters of the fixed measurement robot based on the actual deviation, upload the actual deviation to the construction management system in real time, and generate a construction quality report.

9. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, The computer program instructions, when executed by the processor, implement the method according to any one of claims 1 to 7.

Images