68 results about "Normal estimation" patented technology

The invention discloses a historical block scene three-dimensional reconstruction method and system based on Gaussian sputtering, and the method comprises the steps: collecting a historical block video sequence through a lightweight panorama camera, extracting a multi-frame panorama, and generating an image data set through a projection converter; monocular depth and normal estimation is carried out through a pre-training visual model, and a prior depth and normal graph data set of a historical block scene is constructed; sparse reconstruction is carried out on the multi-view image data set based on the SfM technology, initial point cloud and camera pose information are acquired, and a Gaussian ellipsoid is optimized in combination with prior depth and normal information; dynamically calculating the geometric width estimation value of the street, and guiding and adjusting the adaptive density of the Gaussian ellipsoids of the vertical surfaces on the two sides; rendering the optimized Gaussian ellipsoid through an improved rasterization renderer; and designing a multi-modal loss function and a regularization mechanism to optimize reconstruction and rendering results. According to the method, high-fidelity three-dimensional reconstruction of the historical block scene is realized, and the adaptability of the Gaussian sputtering method to the complex block scene is enhanced.

The invention provides a scene reconstruction method based on delayed rendering and three-dimensional Gaussian. The method comprises the following steps: S1, generating initial three-dimensional point cloud data based on a multi-view image; s2, constructing a trainable structural body for three-dimensional Gaussian modeling; s3, normal initialization and residual optimization are carried out on the Gaussian ellipsoid primitives, depth consistency constraint is combined, and a differentiable and learnable normal reconstruction mechanism is realized, so that the geometric expression ability of illumination modeling is enhanced; s4, introducing a reflection training mechanism based on ambient light and a reflection direction, and generating a Gaussian attribute based on a visual angle; and S5, a final image is generated through a differentiable Gaussian sputtering rendering algorithm, and optimization is carried out through pixel loss of the final image and a real image. According to the method, the reality sense and geometric consistency of the Gaussian sputtering model under the complex illumination condition are remarkably improved, and the technical problems of unreal rendering effect, inaccurate surface normal estimation, weak propagation capability and the like of the existing three-dimensional Gaussian sputtering model under the complex illumination condition are solved.

The invention discloses an unmanned aerial vehicle dynamic environment 3D target detection method and device based on multi-modal fusion, and the method comprises the steps: calibrating a 4D millimeter-wave radar and a camera, and obtaining a pose relation between the millimeter-wave radar and the camera; the method comprises the following steps: preprocessing 4D millimeter wave radar point cloud data to obtain a denoised point cloud image, and performing feature extraction on the denoised point cloud image; using hidden space normal estimation to obtain a pose matrix between adjacent frames, and then through global discretization analysis, realizing multi-frame point cloud pose correction and fusion in a motion scene, and enhancing point cloud data; processing the radar point cloud and the camera image, and obtaining a BEV feature map through a feature extraction network; using a double-end channel attention mechanism and a cross attention mechanism to fuse BEV features; a detection head module based on an anchor frame is designed, and differentiation setting is carried out for different detection targets; and verifying by adopting different data sets and different test scenes to obtain a 3D target detection result. According to the invention, the hidden space normal estimation method is provided to correlate spatio-temporal information, multi-frame splicing is carried out, the density of point clouds is increased, and the performance of target features is improved.

The invention discloses an occlusion perception 3DGS and neural SDF semantic level hybrid reconstruction method, which comprises the following steps: image preprocessing: initially obtaining an image video frame through a camera, starting from image preprocessing, processing an input image, including semantic segmentation, depth estimation, normal estimation and internal and external parameter calibration, a corresponding semantic graph, a depth graph, a normal graph and camera parameters are generated; aligning the 3DGS with the neural SDF, after preprocessing is completed, performing spatial alignment on geometric information generated based on the 3DGS and a neural symbol distance function, and performing restoration in combination with regional densification and a hierarchical depth map; and finishing the final three-dimensional object modeling and rapid reconstruction by adopting a Marking Cubes algorithm. According to the occlusion perception 3DGS and neural SDF semantic level hybrid reconstruction method provided by the invention, each object in a scene can be quickly reconstructed with high quality, and applications such as downstream editing tasks, interaction and the like are supported at the same time.

The invention discloses an intelligent low-altitude unmanned aerial vehicle obstacle avoidance path planning method and system. The method comprises the steps of multi-modal data acquisition, space-time alignment fusion, three-dimensional dynamic trajectory reconstruction, obstacle avoidance path planning and system verification. The invention belongs to the technical field of path planning, and particularly relates to an intelligent low-altitude unmanned aerial vehicle obstacle avoidance path planning method and system.According to the scheme, an incremental dynamic updating mechanism is adopted, and the three-dimensional trajectory reconstruction precision is improved through a double-correction mechanism of hidden space normal estimation and global discretization constraint; bEV feature optimization fusion is adopted to generate a double-enhanced feature map; constructing a mixed bounding box for the trajectory prediction points, calculating a dynamic minimum distance, designing a dynamic safety threshold fusing the real-time speed of the unmanned aerial vehicle and the priority, constructing a space-time coupling risk index in combination with the potential collision time and the priority to complete refined risk grading, and constructing a reinforcement learning model of priority weighted rewards; the problem that the passing right distribution is not uniform in the low-altitude multi-priority unmanned aerial vehicle obstacle avoidance process is solved.

The invention discloses a point cloud normal estimation method based on an asymmetric twin network, and the method comprises the steps: constructing a multi-view normal estimation data set, and forming a noiseless point cloud data set and a noise point cloud data set; performing point cloud rotation on the data set to obtain a rotated data set; constructing an asymmetric twin network, wherein the asymmetric twin network comprises two normal estimators with the same structure; inputting the rotated noiseless point cloud data set into one normal estimator for training to obtain a noiseless normal estimation model; inputting the rotated noise point cloud data set into another normal estimator for training to obtain a noise normal estimation model; obtaining a feature matching loss function, and adjusting parameters of the noise normal estimation model to obtain an optimized noise normal estimation model; inputting the test set into the optimized noise normal estimation model to obtain a normal estimation value of the point cloud; according to the method, the relevance of different noise level feature representations in point cloud normal estimation is obtained, the performance of the model is improved, and the problem of overfitting is solved.

The invention provides an ancient building damaged component AR matching repair method and system, and the method comprises the steps: obtaining the surface point cloud data of a damaged component through an augmented reality technology, generating a high-precision three-dimensional point cloud model through a laser scanning and multi-view stereoscopic vision fusion method, and obtaining the digital expression of the complex geometric shape of the component; for the point cloud model, adopting voxel filtering and a normal estimation algorithm to carry out denoising and surface smoothing processing, and extracting component edge features and curved surface geometric features to obtain a smooth three-dimensional geometric model; and carrying out space mapping on the optimized mortise and tenon joint structure model and an actual component position through an augmented reality technology, generating a virtual joint guide line and stress distribution visual image, guiding a component installation process, and obtaining a high-precision component installation result.

The invention relates to the technical field of industrial nondestructive testing, in particular to an automatic welding seam tracking system of a nondestructive testing robot based on multi-modal sensing. Comprising a data acquisition module, a data preprocessing module, a data fusion module, a track generation module and a motion control module. Point cloud data are obtained through a laser scanner, image information is obtained through an RGB-D camera, attitude data are obtained through an inertial measurement unit, after multi-modal data are subjected to denoising, down-sampling and correction processing, data fusion is achieved through Kalman filtering, and continuous weld space trajectory points are output. And the track generation module generates a tail end pose sequence through three-dimensional B-spline fitting and surface normal estimation, and performs pose smoothing processing by adopting quaternion interpolation. The motion control module is combined with differential path keeping and mixed position-force control, it is ensured that the robot accurately tracks along the center line of the welding seam and maintains constant contact force, meanwhile, the process error correction function is achieved, and the stability and reliability of the system in the complex welding seam environment are improved.

The invention relates to the technical field of three-dimensional modeling, in particular to a method for establishing a three-dimensional live-action model with an autonomous updating function, which comprises the following steps: performing multi-view point cloud data acquisition on a to-be-modeled scene to obtain an initial three-dimensional point cloud set, and performing curvature estimation on the initial three-dimensional point cloud set according to a singular value decomposition algorithm to obtain an initial three-dimensional point cloud set; the curvature value of each data point in the initial three-dimensional point cloud set is obtained; and performing normal estimation on each data point in the initial three-dimensional point cloud set according to the curvature value of each data point in the initial three-dimensional point cloud set to obtain a normal of each data point in the initial three-dimensional point cloud set. According to the method, initial curvature estimation and normal estimation are carried out by using the singular value decomposition algorithm, details in the three-dimensional point cloud can be analyzed more accurately, manual intervention is avoided, and the automation degree of the modeling process is improved.

The invention provides a method and system for evaluating the integrated photovoltaic potential of buildings in a high-density urban area, and the method comprises the steps: obtaining the color point cloud data of a city through the photographing and surveying of an unmanned plane, introducing the physical priori knowledge through algorithm design, and assisting a neural network in extracting a feature vector for representing the characteristics of a building group from the data; comprising the steps of point cloud sampling, normal estimation, window-wall ratio estimation and geographic orientation angle calculation, so that the model can learn implicit relation between the urban form and a photovoltaic potential prediction result; aiming at a scene of comprehensively realizing photovoltaic integration popularization of medium and large-scale buildings, the power generation efficiency of building facades and photovoltaic modules mounted at different positions of windows and walls is considered, feature vectors are input into a regression prediction framework formed by a one-dimensional point cloud neural network, evaluation errors are reduced, systematic deviation of resolution is avoided, and the evaluation accuracy is improved. Data acquisition and prediction are facilitated, and the function of predicting the maximum photovoltaic power generation potential after the building is comprehensively paved with the BIPV is realized.

The invention discloses a polishing path planning method and device of a visual language, a storage medium and a terminal, and the method comprises the steps: obtaining a visual language prompt which comprises a prompt word to-be-polished object image and a camera internal reference matrix; inputting the cue word and the image of the object to be polished into a pre-trained visual language model to obtain a semantic region mask; based on a depth map corresponding to a to-be-polished object image, using the camera internal reference matrix to perform spatial back projection on the semantic region mask to obtain a three-dimensional point cloud region; geometric analysis and normal estimation are carried out on the three-dimensional point cloud area to obtain a redundant material thickness sequence, and a normal difference weight coefficient is utilized to carry out priority division on the three-dimensional point cloud area according to the redundant material thickness sequence to obtain a first priority grinding area and a second priority grinding area; and optimizing the local path direction and the residence time of the first priority polishing area to obtain a polishing path. The polishing precision and efficiency of the complex curved surface workpiece are greatly improved.

The invention provides a point cloud normal estimation method and device based on local subspace clustering, electronic equipment and a storage medium, and the method comprises the steps: introducing a feature point weight evaluation mechanism based on covariance features and geometric structure indexes into a local neighborhood, and screening out feature points with high confidence; carrying out multi-subclass division and plane fitting on the local point cloud by adopting low-rank subspace clustering with priori weight constraint in the neighborhood of the feature points, so as to re-estimate the normal direction of the feature points on the subclass level; meanwhile, noise point filtering and a normal diffusion strategy based on adjacent non-noise points are combined, so that the method can more accurately describe a local geometric structure in a complex scene with noise, multi-structure aliasing and non-uniform sampling, the precision and stability of point cloud normal estimation are remarkably improved, and the method is suitable for popularization and application. And sharp features and edge details of the model can be better maintained.

The present disclosure provides a point cloud smoothing method and device based on local feature estimation, electronic equipment and storage medium. By introducing the noise point elimination of local density features, the normal estimation based on weighted local statistical analysis, and the neighbor update weight construction mechanism of fusing spatial distance, normal difference and normal distance, adaptive and structure-preserving smoothing processing of point cloud is realized. The method can effectively eliminate outliers and noise points, reduce the thickness of the point cloud, improve the stability of the normal estimation, and at the same time, preserve the object edges and geometric details in the smoothing process, significantly improving the overall quality and applicability of the point cloud.

The invention relates to the technical field of natural language processing and medical artificial intelligence, and discloses a large model optimization method and system fusing a medical knowledge graph, and the method comprises the steps: constructing a time fuzziness-confidence joint modeling network model, time information corresponding to a current chief complaint symptom, context chief complaint content and patient individual information are input, a normal estimation time prediction result is output, joint calibration is conducted on the normal estimation time prediction result based on causal path information in the medical knowledge graph, and time prior distribution on the graph side is obtained; and constructing joint distribution based on the normal estimation time prediction result and the time prior distribution at the map side, taking the expectation of the joint distribution as a time point prediction result, inputting the time point prediction result into a large model, outputting a structured suggestion, and feeding back the structured suggestion to the medical knowledge map for updating. Therefore, the problem that inference is wrong due to the fact that semantic dependence cannot be recognized or patient background factors are ignored is solved.

The invention provides a rock structure-oriented point cloud normal vector estimation method based on voxel geodesic distance, and the method comprises the steps: obtaining original point cloud data of a three-dimensional rock structure, carrying out the voxelization processing of the original point cloud data, and dividing the point cloud into a plurality of three-dimensional voxel units; a sparse voxel graph is constructed based on voxel space adjacency relations, each voxel is a node of the graph, and adjacent voxels are connected through edges. Furthermore, the voxel geodesic distance between any two points is calculated by using a graph search algorithm, namely the shortest distance along a voxel communication path. On the basis, a target point is taken as a center, a neighborhood point set which is communicated with the target point is screened based on geodesic distance, the distance between the target point and the neighborhood point set is smaller than a threshold value, and a principal component analysis (PCA) method taking voxel geodesic distance as a weighting factor is adopted to estimate a normal vector of the target point. According to the method, pseudo neighborhood point interference caused by cracks, holes and the like can be effectively reduced, and the robustness and precision of normal estimation are improved.

A system for enhancing an image using hand surface normal estimation is provided. In a model training phase, 3D data of the hand at various positions are used to generate a 3D model of the hand. In one embodiment, target normal training data is generated, the target normal training data including a normal of a surface of a 3D model, and synthetic 2D image training data corresponding to the 3D model and the normal is generated. The target normal training data and the composite image training data are used for training a normal estimation model. The interactive application uses normal estimation to generate enhancements applied to the hand image data.

The invention relates to the technical field of computer vision, in particular to a surface normal estimation method and device of a target object, equipment and a medium. The method comprises the following steps: acquiring a first image of a target object under a flash lamp condition and a second image of the target object under a non-flash lamp condition; the first image and the second image are preprocessed; inputting the preprocessed image into a trained preset network for normal estimation to obtain a normal graph of the target object; wherein the preset network comprises an encoder module, a diffusion denoising module and a decoder module. According to the method, the surface normal of the target object can be accurately estimated under the complex illumination condition by combining the flash lamp image and the non-flash lamp image and utilizing the diffusion prior denoising method.

The application discloses a kind of weakly supervised image semantic understanding methods based on multi-task learning, comprising the following steps: obtaining task missing image, constructing multi-level task sharing encoder, extracting high-level semantic information layer by layer, input corresponding decoder branch;Construct public space-task space feature mapping module, through the unaligned task fusion module and task interaction mapping module, update each subtask feature by mapping;Task adaptive feature update module is constructed, and multi-level iterative update unaligned task feature;Task adaptive weakly supervised image semantic understanding framework is constructed, model loss function is established, image data with task missing is input into model, and obtains multi-task prediction result such as semantic segmentation, depth estimation, surface normal estimation.The application is according to the data information of task label unaligned, through the mapping interaction of public space and task space, fully fuses unaligned task feature, iteratively generates high-quality multi-task prediction result, can effectively handle weakly supervised problem with task missing, and simultaneously improves each task prediction accuracy.

The application discloses a kind of based on plane Gauss's color model construction method, system, device and medium, method includes the position parameter of initialization Gauss ball, depth order, shape attribute, color attribute and opacity;According to shape attribute, the shortest axis of Gauss ball is determined, and then flat loss is calculated;According to the shortest axis and observation direction, the local normal of Gauss ball is obtained and then normal alignment loss is calculated;According to opacity, sparsity loss is calculated;Gauss ball is colored and color loss is calculated, according to flat loss, normal alignment loss, sparsity loss and color loss, construct final loss function, and shape attribute, local normal, opacity and color attribute are updated by back propagation algorithm.The application improves normal estimation accuracy by flattening Gauss ball, and then improves the color quality, by combining color model, high-quality rendering of a variety of materials is realized, and can be widely applied in computer vision technical field.

The invention discloses a coloring model construction method, system and device based on plane Gaussian, and a medium. The method comprises the following steps: initializing a position parameter, a depth sequence, a shape attribute, a coloring attribute and opacity of a Gaussian ball; determining the shortest axis of the Gaussian ball according to the shape attribute, and further calculating flattening loss; obtaining a local normal of the Gaussian sphere according to the shortest axis and the observation direction, and further calculating normal alignment loss; calculating sparsity loss according to the opacity; and coloring the Gaussian ball and calculating color loss, constructing a final loss function according to flattening loss, normal alignment loss, sparsity loss and color loss, and updating the shape attribute, local normal, opacity and coloring attribute through a back propagation algorithm. According to the method, the normal estimation precision is improved through the flattened Gaussian ball, so that the coloring quality is improved, high-quality rendering of various materials is realized through the combined coloring model, and the method can be widely applied to the technical field of computer vision.

This invention discloses a single-view 3D photometric reconstruction method, system, device, and medium based on machine learning, relating to the field of computer vision technology. The method includes acquiring single-view images of a target object under different light source conditions and light source direction information; preprocessing the images and light source direction information to generate standardized multi-dimensional input data; performing feature extraction and fusion to obtain a fused feature representation; performing preliminary 3D surface normal estimation; outputting preliminary normal vectors; iteratively optimizing the preliminary normal vectors; combining light source reliability assessment and material prior knowledge constraints to obtain accurate normal vectors; and constructing a 3D model of the target object using a 3D geometric reconstruction algorithm. This invention significantly improves the accuracy, robustness, and direct usability of 3D reconstruction in complex lighting and heterogeneous material scenarios by combining multi-scale feature learning with channel attention and incorporating physical optimization based on light source reliability assessment and material prior knowledge.

A real-time photometric stereo vision method and device based on an event camera, belonging to the field of computer vision, includes: lighting pattern design; simultaneously acquiring lighting signals and synchronization signals generated by lighting changes in the test object using an event camera; using the synchronization signals to calibrate the lighting direction when other events occur, achieving synchronous acquisition; constructing null vectors directly related to the surface normals of the test object from each pair of consecutively occurring events; and fusing the null vectors to solve for the surface normals. This invention reconstructs the surface normals of the test object from the basic model triggered by the event camera, reducing the cost of multiple cameras and synchronization; it utilizes the high event resolution of the event camera to acquire data under continuously changing light sources, increasing data sampling density and improving the quality of normal estimation; and it leverages the high dynamic range of the event camera to improve data acquisition speed and quality through native high dynamic range and compressed event representation.

The invention discloses a noise robustness and detail fidelity-oriented point cloud lightening method, which relates to the technical field of three-dimensional point cloud processing, and comprises the following steps: S1, carrying out multi-scale noise suppression on original point cloud data by adopting a self-adaptive sliding window polynomial fitting filtering algorithm to obtain smooth point cloud data; s2, constructing a spatial index structure for the smooth point cloud data by adopting a multi-level axial bounding box adaptive recursive subdivision method, and realizing sparse region memory optimization and high-density region feature retention; s3, extracting a skeleton structure from the point cloud data processed in the step S2 based on an improved central axis transformation algorithm; s4, performing normal estimation and correction on the point cloud data processed in the step S3 so as to retain and enhance sharp features, by means of the adaptive sliding window polynomial fitting filtering method, key details and feature distribution of a local geometric structure of the point cloud are reserved while noise interference is effectively suppressed, and the method has high robustness. And high-fidelity point cloud data is provided for subsequent normal estimation and three-dimensional reconstruction.

The present disclosure provides a point cloud normal estimation method and device based on local subspace clustering, electronic equipment and storage medium. By introducing a feature point weight evaluation mechanism based on covariance characteristics and geometric structure indicators in the local neighborhood, high-confidence feature points are screened out, and a low-rank subspace clustering with prior weight constraint is used in the feature point neighborhood to divide and fit the local point cloud, so as to re-estimate the normal of the feature point at the subclass level. At the same time, combined with the noise point filtering and the normal diffusion strategy based on the adjacent non-noise points, the method can more accurately depict the local geometric structure in the complex scene with noise, multi-structure aliasing and non-uniform sampling, significantly improve the accuracy and stability of the point cloud normal estimation, and better maintain the sharp features and edge details of the model.

The invention discloses a redefined 3D point cloud normal estimation method based on sample selection, and the method comprises the steps: collecting three-dimensional point cloud data, and carrying out the evaluation and screening of the three-dimensional point cloud data through a confidence estimation strategy; performing random sampling on the screened three-dimensional point cloud data through a k-nearest neighbor algorithm and a probability-based sampling strategy to obtain local point cloud blocks and global point cloud blocks; a PCA algorithm is introduced to carry out alignment processing on the point cloud blocks to obtain a point cloud data set; constructing a multi-constraint joint optimization solving network; inputting a point cloud data set for training, optimizing the network through a loss function to obtain an optimized multi-constraint joint optimization solving network, and outputting a predicted point cloud normal; a neural gradient field module is introduced to carry out directional correction on the predicted point cloud normal to obtain a final point cloud normal; according to the method, a network with higher robustness is constructed through multi-constraint integration, data of various noise levels can be processed, and normal estimation results with consistent directions and higher precision are obtained.

The present invention relates to the field of three-dimensional modeling technology, and specifically to a method for establishing a three-dimensional real-scene model with an autonomous updating function, comprising: collecting multi-viewpoint point cloud data of a scene to be modeled to obtain an initial three-dimensional point cloud set; estimating the curvature of the initial three-dimensional point cloud set according to a singular value decomposition algorithm to obtain a curvature value for each data point in the initial three-dimensional point cloud set; and estimating the normal of each data point in the initial three-dimensional point cloud set according to the curvature value of each data point in the initial three-dimensional point cloud set to obtain a normal for each data point in the initial three-dimensional point cloud set. By using a singular value decomposition algorithm for preliminary curvature estimation and normal estimation, the present invention can more accurately analyze details in the three-dimensional point cloud, avoid manual intervention, and improve the degree of automation in the modeling process.

An embodiment of the present application discloses a planar normal estimation method, device, and storage medium, including: obtaining an RGB image; inputting the RGB image into a planar normal estimation model to obtain normal information corresponding to the RGB image; the planar normal estimation model is a model trained according to a first RGB sample image with a first sample normal label and a second RGB sample image without a sample normal label.

The present application relates to the technical field of industrial nondestructive testing, in particular to a nondestructive testing robot welding seam automatic tracking system based on multi-modal sensing. It comprises a data acquisition module, a data preprocessing module, a data fusion module, a trajectory generation module and a motion control module. Point cloud data is obtained by a laser scanner, image information is obtained by an RGB-D camera, and attitude data is obtained by an inertial measurement unit. After denoising, downsampling and correction processing of multi-modal data, data fusion is realized by Kalman filtering, and continuous welding seam spatial trajectory points are output. The trajectory generation module generates end pose sequences through three-dimensional B-spline fitting and surface normal estimation, and adopts quaternion interpolation for attitude smoothing processing. The motion control module combines differential path keeping and hybrid position-force control to ensure that the robot accurately tracks along the welding seam centerline and maintains constant contact force, while having process correction function to improve the stability and reliability of the system in complex welding seam environment.

The application discloses a kind of multi-storey building point cloud automation segmentation method, system, medium and device based on geometric feature related to building digitization technical field, to realize the high-precision, automation segmentation of multi-storey building point cloud from whole to local.The application first reduces the calculation complexity by voxelization preprocessing, combined with normal estimation screening horizontal voxel, then utilizes Z-axis occupancy rate peak detection to realize accurate multi-storey layering;Second, extract the top point cloud to construct two-dimensional occupancy map, input the improved region growing strategy into two-dimensional occupancy map, realize room segmentation;Finally, based on random sample consensus algorithm, by introducing geometric constraint, and combined with plane merging strategy, realize wall surface segmentation.Consequently, floor-room-wall surface three-level hierarchical segmentation method is constructed, to provide high-precision structured data support for subsequent point cloud data application.

The application discloses a kind of robot multilayer multi-pass welding path adaptive generation method and system, method includes using time flight sensor to the three-dimensional point cloud of weld is collected, and the weld point cloud collected is preprocessed;The preprocessed weld point cloud is segmented to extract the feature of weld by random sample consistent weld segmentation algorithm based on density clustering, and using cubic spline interpolation method is interpolated fitting to weld feature point;According to the geometric information of welding workpiece and the weld feature point after interpolation fitting, multilayer multi-pass welding path is generated;Based on multilayer multi-pass welding path, the normal estimation method of weighted principal component analysis is used to calculate the posture of welding torch end.The method of the application can segment the three-dimensional point cloud of welding workpiece containing complex surface to extract weld feature point, and then generate multilayer multi-pass welding path through weld feature point and the geometric information of welding workpiece to solve the problem of low welding precision, slow speed and narrow application range of medium plate.