501 results about "Characteristic point" patented technology

The invention relates to the technical field of digital twinning control, and discloses a composite structure prefabricated segment digital twinning-adaptive assembly correction method and system, and the method comprises the steps: reading BIM geometric model data, and constructing an assembly reference coordinate system and a digital twinning geometry of a prefabricated segment; composite data are collected in real time, and end tooth groove boundary feature point cloud is extracted; matching the actually measured point cloud with the digital twinborn geometry through a dynamic point cloud registration technology, calculating a six-degree-of-freedom pose error vector, and generating a predicted total pose error vector in combination with a pre-constructed pose drift prediction model; based on the error vector, a mechanical fine adjustment jack is driven to execute position and posture adjustment until the position and posture are converged, and then tooth groove precise meshing and mechanical locking are completed; according to the method, through the synergistic effect of the dynamic mapping of the digital twinborn model and the self-adaptive correction algorithm, high-precision dynamic correction of the multi-combination structure segment assembly process is achieved on the premise that an original mechanical connection structure is not changed.

The invention discloses a visual inertial positioning method based on dynamic target detection and semantic information constraint, and belongs to the field of motion estimation and dynamic environment processing. According to the method, a dynamic target detection mechanism is introduced, the dynamic target is effectively detected based on the target detection network, inertial navigation information and geometric constraints, the dynamic target is effectively recognized in the image processing process, the corresponding dynamic feature points are screened out, the mismatching rate of the dynamic features is remarkably reduced, and high-quality observation input is provided for back-end optimization. In the back-end sliding window optimization stage, a semantic information consistency constraint method is constructed, and the estimation stability of the system in a weak texture area or a repeated texture area is enhanced by utilizing the consistency of feature points in the same semantic area on a geometric structure. Visual inertia pose estimation is realized based on dynamic target detection and semantic information constraint, and high-robustness and high-precision pose estimation can still be realized in a complex environment with dynamic interference of pedestrians, vehicles and the like and severe scene change.

The invention relates to a defect positioning method based on fusion of weld defect features and trajectory tracking data, and belongs to the technical field of weld defect detection and positioning. The method comprises the following steps: capturing welding seam track dynamic data and defect feature data, constructing a dynamic coordinate system based on a welding seam initial feature point, and establishing double-data-set reference mapping; performing multi-physics field interference decoupling correction on the trajectory data, and performing cross-modal feature purification and core feature consistency verification on the defect data; converting the preprocessed data into a feature form adaptive to fusion, and constructing a welding process-defect formation mechanism association network to regulate and control fusion weight; and finally, reconstructing a three-dimensional dynamic contour of the welding seam, calling dynamic positioning logic to position the defect, and outputting a result carrying the process-defect causal confidence coefficient. The positioning precision is improved through multi-dimensional data fusion and mechanism association, and technical support is provided for welding quality management and control.

The invention relates to a point cloud registration method based on improved ISS-TOLDI features in combination with ICP, and belongs to the technical field of laser point cloud application. The method comprises the steps of obtaining a source point cloud and a target point cloud, and performing preprocessing; extracting feature points from the point cloud data by using an internal shape descriptor ISS algorithm; performing feature description on the extracted feature point set by using an improved TOLDI algorithm; performing coarse registration on the point cloud by using a sampling consistency initial registration algorithm; and performing fine registration on the coarsely registered point cloud data by using an iterative closest point ICP algorithm to complete point cloud registration. According to the method, the TOLDI algorithm is improved, so that the calculation complexity is reduced, the feature integrity is ensured, and the point cloud registration precision is higher.

The invention discloses a dynamic scene robust vision SLAM (Simultaneous Localization and Mapping) method based on multi-feature collaborative optimization, which comprises the following steps of: acquiring an image sequence, carrying out dynamic target detection and segmentation through an instance segmentation network, generating a segmentation mask containing a dynamic region mark, and identifying and separating a dynamic object and a static background; removing feature points corresponding to the dynamic object based on the segmentation mask to obtain static feature points; carrying out pose estimation based on the static feature points, and for the key frame, carrying out feature matching with the previous key frame by minimizing a re-projection error, and solving to obtain the camera pose of each key frame; for non-key frames, performing camera pose tracking and data association on the previous frame by adopting an optical flow algorithm, and accumulating solving results to obtain pose tracks of all the non-key frames; the key frames and the non-key frames are subjected to differential processing by fusing feature matching and an optical flow algorithm, so that the calculation efficiency is remarkably improved while the positioning precision is ensured, and the real-time performance is improved.

In various examples, systems and methods are disclosed relating to generating each initial feature map of a plurality of initial feature maps based on a respective input image of an input dataset, each initial feature map, incorporating depth data of the respective input image, corresponds to a plurality of pixels of the respective input image, generating a sparse feature point cloud including a plurality of features determined using the plurality of initial feature maps, transforming the sparse feature point cloud into multi-resolution sparse grids, each of the multi-resolution sparse grids comprising a plurality of voxels, modeling, using a plurality of neural networks according to a hierarchal architecture, the multi-resolution sparse grids to construct a hierarchical volume representation, and providing constructed content based on the hierarchical volume representation.

In a dynamic environment, a visual SLAM (Simultaneous Localization and Mapping) system often causes the problems of large positioning error and inaccurate map construction due to dynamic target interference. In order to improve the robustness and precision of the system, the invention provides a dynamic scene SLAM optimization method based on improved YOLOv11 and geometric consistency constraint. Firstly, ORB features in a scene are extracted, and meanwhile a prior dynamic object and feature points on the prior dynamic object are removed through a YOLOv11 semantic segmentation model; secondly, eliminating feature points on the potential dynamic object by utilizing geometric consistency constraint, and recovering a background shielded by the dynamic object through a semantic perception Gaussian filter; and finally, selecting a high-quality key frame and applying the key frame to loopback detection and global optimization, constructing a basic Gaussian graph through a group of determined poses and point clouds, and finally fusing repair frame information to realize new view rendering and three-dimensional scene optimization.

The invention discloses a 3D GS cultural relic digital reconstruction method and system based on a block chain, and the method comprises the steps: collecting the RGB image data and depth perception data of a cultural relic, eliminating the influence of different shooting conditions through an illumination separation processing technology, and building a standardized image data set; recognizing a surface area suitable for reconstruction based on image analysis, determining feature point distribution by using kernel density estimation, and generating initial three-dimensional representation through Gaussian ellipsoid fitting; performing gradient calculation and feature extraction on the depth data, and combining with Gaussian representation to form a geometric constraint mechanism; self-adaptive encryption based on visual importance is executed for a sparse region, and a layered rendering effect is achieved through opacity parameter adjustment; the rendering characteristics and the conversion relation of different view angles are analyzed, key observation points are determined through stability analysis, and a smooth multi-view-angle display sequence is constructed; and integrating multi-view rendering information to generate volumetric representation, and completing right confirmation of the high-quality three-dimensional digital model through digital signature.

The invention relates to the technical field of ship detection, and discloses a ship lockage safety detection method and system, and the method comprises the steps: obtaining and analyzing ship image data in real time, and separating a ship target; key feature points representing the core form and structure of the ship target are identified, and a spatial distribution diagram containing the relative position relation of the key feature points is constructed; matching the spatial distribution map with a pre-stored standard ship three-dimensional model library, and acquiring passing information of the ship target according to a matched standard three-dimensional model; according to the method, lockage safety evaluation is carried out based on passing information, deep analysis is carried out on ship image data, key feature points are identified, a spatial distribution diagram is constructed, and then matching with a standard three-dimensional model is carried out, so that accurate ship passing information is obtained; the problem of size and position information deviation caused by inaccurate hull contour extraction in the prior art is effectively solved, and the reliability of lockage safety assessment is remarkably improved.

The invention discloses a physical object three-dimensional positioning method and system based on video data, and relates to the technical field of three-dimensional positioning. The method is used for solving the problems of low object space positioning precision, unstable pose estimation and poor time sequence continuity in a video scene. Firstly, an input video stream is analyzed, an object segmentation mask is generated, feature points are extracted, camera motion parameters are calculated through inter-frame matching, and a scene sparse three-dimensional point cloud is reconstructed; a candidate three-dimensional bounding box is generated according to the segmentation mask and the point cloud, an optimal bounding box is selected by combining geometric matching degree and feature similarity evaluation, and a preliminary three-dimensional positioning result of the object is obtained; a positioning result and historical frame motion data are fused, a space-time constraint optimization model is constructed, and the six-degree-of-freedom pose of the object is solved; and finally, neural radiation field representation is established based on the pose, the pose and neural parameters are optimized through combination of micro rendering and back propagation, continuous and accurate updating of the three-dimensional position of the object is realized, and the positioning stability and robustness in a dynamic scene are improved.

The invention discloses a flow field key structure detection and feature point extraction method based on an optimized YOLO model, and the method comprises the steps: S1, obtaining flow field time sequence schlieren images of an air-breathing aircraft under different working conditions through a high-speed schlieren collection system in a wind tunnel test; s2, preprocessing the flow field time sequence schlieren image to obtain a data set corresponding to the strong shock wave of the isolation section of the air-breathing aircraft, and dividing the data set into a training set, a verification set and a test set in proportion; s3, constructing an optimized YOLO target detection model, and obtaining an optimal weight model after iterative training and verification; and S4, adopting the optimal weight model to complete target detection and feature point extraction of the test set or the schlieren image to be detected. According to the method, the target area is focused through area cutting, meanwhile, a targeted data enhancement strategy of Gaussian noise, brightness adjustment, contrast ratio adjustment and saturation is designed, a high-quality annotation data set is constructed, and the model detection precision is effectively improved.

The invention relates to a bridge structure linear change detection method based on a three-dimensional laser point cloud. The method comprises the following steps: 1) cutting a component section point cloud; 2) generating a section template library; 3) matching a section template; 4) fitting accurate key points; and 5) detecting the variable quantity of the characteristic line. According to the parameterized section template library generation method based on the large language model, dependence of design prior information is reduced, the posture size of a component is assisted in positioning in advance, meanwhile, template geometric information is considered, accurate feature points are positioned and screened out, point cloud quality is dynamically adapted, and the requirement that in the non-ideal environment of point cloud scanning in a project, the precision of the component is greatly improved can be met. The method solves the problem of local position missing of the point cloud, provides a general solution thought for automatic detection of the bridge line shape, and can be applied to bridge line shape detection in other bridge scenes.

The invention provides an ore recognition method and system based on a large model algorithm, and is applied to the technical field of computer vision and data processing, and the method comprises the steps: obtaining an ore microscopic image, and generating a feature point location candidate set through a convolutional neural network model; calculating a spectral value and a morphological parameter of each feature point location to determine a multi-dimensional data point; clustering the multi-dimensional data points to obtain a feature point cluster; and when the spatial distance between the adjacent feature point position clusters is smaller than a first preset threshold value and the spectral similarity is higher than a second preset threshold value, combining the adjacent feature point position clusters into the same mineral region. The segmentation precision and the type identification accuracy of the complex symbiotic mineral area can be obviously improved.

The invention discloses a method for identifying interference signals of weld defects under different lifts-off conditions, which comprises the following steps of: S1, de-trending processing: carrying out de-trending processing on original detection signals; s2, Gaussian wavelet transform: carrying out Gaussian wavelet transform on the detrended signal; s3, optimal wavelet basis selection: by calculating correlation coefficients or energy ratios of different wavelet basis and defect signals, selecting the wavelet basis with the highest matching degree for reconstruction; s4, envelope processing: extracting a signal envelope based on Hilbert transform; s5, mean filtering: applying sliding window mean filtering to the envelope signal; and S6, threshold processing: setting a self-adaptive threshold screening signal, and retaining the feature points of which the amplitudes exceed the threshold. According to the method, de-trending and Gaussian wavelet transform are used for de-noising enhancement. According to the method, wavelet functions of different orders are constructed and matched with defects, secondary signal enhancement is carried out by selecting a filtering method, finally, threshold stripping interference is calculated, reliable defect identification is carried out, and technical support is provided for uneven welding seam quality monitoring.

The invention discloses a feature point detection method based on edge saliency and scale sensitivity. The method comprises the following steps of performing Gaussian smooth denoising on an input image; calculating the edge saliency of the pixel points through a Laplace operator; calculating a texture change degree based on the local gradient magnitude; generating a scale sensitivity weighted value through multi-scale analysis; candidate points are screened in combination with edge saliency, texture weighting and scale weighted values, and mismatching points are removed through non-maximum suppression; and finally, key points are enhanced and marked. According to the method, the edge structure, the texture information and the multi-scale features are fused, the robustness and accuracy of feature point detection are improved, and the method is particularly suitable for image matching and target recognition tasks in complex scenes.

For a cross-modal target prompt matching task, the invention provides a training-free multi-stage matching method based on local feature points and global spatial transformation so as to realize robust positioning of the same target in different modal images. According to the method, firstly, feature points and descriptors of a target prompt image and a to-be-matched image are extracted by utilizing a pre-trained SuperPoint, then, MINIMAGlue finely adjusted on cross-modal data is preferentially adopted to perform feature point matching, and when matching fails, LightGlue is switched to serve as remediation; in the matching process, an optimal angle is selected through a rotation enumeration strategy to enhance matching robustness; in the coarse positioning stage, calculating a minimum enclosing rectangle according to matching points in the ROI, and if the matching points are insufficient, estimating a homography matrix by adopting RANSAC and carrying out geometric correction on a rectangular frame; in the fine positioning stage, the central point of the coarse positioning result is used as a prompt to input MobileSAM for target segmentation, and a final rectangular frame is corrected through area consistency constraint. Through integration and fusion of the multi-stage process, accurate matching and stable positioning of cross-modal target prompt are realized under a training-free condition, and the generalization ability and practicability of the method are improved.

The invention discloses a calibration method and system based on multi-scale feature fusion and an adaptive optimization algorithm, and belongs to the technical field of computer vision. The method comprises the following steps: collecting a plurality of images containing calibration modes; detecting and positioning feature points by adopting a multi-scale feature fusion technology, and distributing a confidence score for each feature point; carrying out feature point matching and removing abnormal values by using an improved RANSAC algorithm based on the prior geometrical relationship and confidence score; preliminarily calculating camera parameters; and a nonlinear optimization model fusing confidence weighting and adaptive regularization is constructed, joint optimization is performed on camera parameters, and a high-precision calibration result is output. According to the method, through an optimization mechanism of multi-scale feature fusion and confidence guidance, the problems of unstable feature point detection, abnormal value sensitivity and insufficient precision of a traditional method in a complex environment are effectively solved, and the accuracy, robustness and automation degree of image calibration are remarkably improved.

The invention discloses a pipeline defect positioning method, device and equipment, a medium and a product. An SPPF-LSKA optimization instance segmentation model is combined with an ORB-SLAN3 framework, and dynamic feature points are eliminated by using information of an RGB-D depth map and an instance segmentation mask; secondly, a sliding window confidence evaluation mechanism is introduced, the credibility of feature points can be dynamically evaluated, wrong feature points generated due to instantaneous false detection are removed in time, and high-precision dynamic point removal is achieved; and finally, in a back-end optimization stage, a double-layer map strategy is adopted, a static mapping layer participates in BA optimization to reduce a positioning error, and a dynamic observation layer independently records an optical flow speed and a segmentation result of a dynamic region for dynamic interference analysis and does not participate in global optimization, so that interference of dynamic factors on positioning precision is avoided.

A model and its establishment method and system, a compensation method, equipment and a storage medium, the establishment method comprises the following steps: providing a test layout, including a test pattern, including a plurality of test points; obtaining the initial pattern density of each test point according to the test pattern; taking any test point as a feature point, obtaining the effective pattern density of the feature point according to the initial pattern density of each feature point and the influence of the initial pattern density of the plurality of test points located around the feature point on the feature point, wherein the influence degree of the initial pattern density of the plurality of test points around the feature point on the feature point is set anisotropically; obtaining the actual size deviation corresponding to the feature point; obtaining the estimated size deviation of the feature point; judging whether the estimated size deviation meets the preset condition according to the estimated size deviation and the actual size deviation; when the estimated size deviation meets the preset condition, the establishment of the etching deviation compensation model is completed. The etching deviation compensation model is more accurately established.

The invention provides a real-time feature point extraction and matching method based on star operation and GateMLP, and the method comprises the following steps: S1, carrying out the preprocessing of an input image, and sending the image into a lightweight convolutional network to extract an initial feature; s2, on the basis of the output feature map, deep feature modeling is carried out by fusing the structure of a convolution branch and a star operation module, and a dense descriptor and a feature point thermodynamic diagram are output; s3, non-maximum suppression is performed on the thermodynamic diagram, and key point coordinates are screened in combination with confidence score; s4, performing rough matching on the key points of the left view and the right view by adopting cosine similarity and a mutual nearest neighbor strategy to obtain an initial matching pair; and S5, inputting the paired descriptors obtained by rough matching into the gated multilayer perceptron for fine-grained matching, and outputting accurate matching point pairs. The method provided by the invention realizes end-to-end efficient reasoning while ensuring high extraction and matching precision, has good real-time performance and robustness, and is suitable for unmanned aerial vehicle navigation, SLAM (Simultaneous Localization and Mapping) and other actual scenes needing quick response.

The invention provides a multi-modal large-model long-sequence information compression retrieval method, and relates to the technical field of data processing, and the method comprises the following steps: receiving multi-modal long-sequence original data of a target patient; all modal data in the multi-modal long sequence original data are uniformly mapped to the same hidden space through a pre-trained multi-modal large model, and a unified semantic representation sequence of semantic association between modals is obtained; performing time sequence feature analysis on the unified semantic representation sequence, regarding feature points in the semantic representation as a spatial point set, determining boundary distribution by constructing convex hulls of feature clusters, identifying key feature clusters in the semantic representation, and determining a key time sequence interval based on a convex hull distribution mode of the key feature clusters. According to the method, the problems of weak cross-modal semantic association, easy loss of key time sequence information and low compression and retrieval efficiency in the existing multi-modal long sequence data retrieval are solved.

The invention discloses a step line detection method and device for a strip mine and a storage medium. The step line detection method comprises the steps of obtaining point cloud data of the strip mine; extracting target point cloud data in the point cloud data, wherein the target point cloud data comprises ground points in the point cloud data; step line feature points meeting the constraint condition of each dimension in the ground points are selected based on multiple constraint conditions of different dimensions; determining a feature point type of each step line feature point based on the satisfied constraint condition, wherein the feature point type comprises a strong feature point and a weak feature point; and taking the strong feature points as control points, taking the weak feature points as reference points, and performing curve fitting on the strong feature points and the weak feature points to obtain step line information of the strip mine. According to the scheme, the screening accuracy of the step line feature points can be improved, and accurate detection of the step line of the strip mine can be realized through a method of performing curve fitting by combining the strong feature points and the weak feature points.

The invention provides a face recognition access control management method and system for a smart community. The method comprises the following steps: constructing a face illumination quality evaluation model to recognize an abnormal area, performing repair by adopting a multi-stage compensation mechanism, and constructing a face feature density field to recognize a sensitive area; dynamically calculating the grid size of each area of the face; extracting texture, shape and semantic features of various scales in the re-divided grids, and calculating a feature contribution value of each grid unit; and constructing a multi-modal input tensor, inputting the multi-modal input tensor to the multi-branch convolutional neural network for training, establishing a closed-loop updating mechanism based on data generated during operation of the access control system, and continuously optimizing the convolutional neural network and parameters of the previous steps. According to the method, the output quality is ensured through multi-dimensional verification, including feature point spatial distribution rationality inspection, density field continuity verification and sensitive area and facial anatomical structure consistency inspection, and the accuracy and reliability of an identification result are effectively guaranteed.

The invention relates to the technical field of electromagnetic field numerical values, and discloses a grid generation method, an electromagnetic simulation method, a device, equipment, a medium and a product, and the method comprises the steps: extracting a plurality of feature points from a metal geometric model, the plurality of feature points being used for representing key points enabling an electromagnetic field to generate singular behaviors in the metal geometric model; determining the maximum grid step length according to the shortest wavelength and the longest side of a bounding box of the metal geometric model; the minimum grid step length, the grid encryption area, the encryption step length and the grid smoothing factor are determined, the encryption step length is the grid step length of the grid encryption area, and the grid smoothing factor is used for representing the change rate between every two adjacent grid step lengths; and according to the plurality of feature points, the maximum grid step length, the minimum grid step length, the grid encryption area, the encryption step length and the grid smoothing factor, generating a plurality of grid points so as to divide the metal geometric model into a plurality of grids. According to the method, the calculation efficiency and the result precision of the FIT algorithm can be effectively improved.

The invention discloses an AI driving-based precision cutter appearance dimension data detection method and system, and the method comprises the steps: obtaining image data of the surface of a cutter, and carrying out the preprocessing, and obtaining optimized image data; key points are detected through an S-IFT algorithm, and a key feature point set is generated; an RANSAC algorithm is used to eliminate mismatching points and generate a matching result set, effective point pairs with the distance between the matching point pairs smaller than a preset distance threshold value are reserved, corresponding positioning data are searched to lock a key area, and a boundary area is obtained through pixel gradient change detection; extracting a coordinate extreme value of the edge point coordinate of which the discontinuous proportion is smaller than a preset proportion threshold value to obtain a size parameter; meanwhile, surface texture information is extracted, an initial texture feature set is generated through a gray-level co-occurrence matrix algorithm, and a texture description set is obtained through analysis; and integrating the size parameters and the texture description set, and generating a comprehensive appearance detection result through feature splicing and fusion. According to the method, high-precision cutter appearance detection can be realized.

The invention discloses a self-adaptive compression and lossless fast transmission method and system for high-frequency test data. The method comprises the following steps: extracting multi-dimensional dynamic feature vectors from high-frequency test data in real time and marking feature points; based on the features, the network bandwidth and the target compression ratio, dynamically selecting a compression algorithm combination and parameters through a pre-training decision model; performing data framing and typed preprocessing according to the algorithm combination; compression is executed through a multi-stage assembly line, and a correction link based on statistical characteristics is inserted between stages; constructing a three-dimensional transmission state matrix fusing priorities, links and quality scores, dynamically allocating links and adaptively adding forward error correction codes; and after error correction and decompression of a receiving end, realizing lossless verification through multi-dimensional comparison of feature point sequences of the reconstructed data and the original data. According to the method, the whole-course self-adaption of the compression strategy and the transmission scheduling is realized, and the lossless and reliable recovery of the data is ensured while the compression and transmission efficiency is improved.

The invention belongs to the technical field of three-dimensional geometric measurement, and particularly relates to a three-coordinate detection data processing method and system for automobile parts, and the method comprises the steps: obtaining an actual point cloud and a standard point cloud; by calculating the maximum angle gap of the data points, screening to obtain actual and standard feature point sets; anti-noise geometric descriptors based on tangential weights are constructed for the feature points in the two sets of feature point sets respectively; matching the two groups of feature points based on the descriptors and generating an initial matching set; solving global optimal transformation from the initial matching set by adopting a random sampling consistency algorithm; and the transformation is applied to the actual point cloud, fine registration is carried out by using an iterative nearest point algorithm, and finally deviation analysis is completed. The technical problem that traditional registration depends on initial alignment and is prone to falling into local optimum is solved, and the robustness and accuracy of registration are improved.

The invention discloses a point cloud registration method based on an improved SHOT descriptor, and aims to solve the defects of a traditional point cloud SHOT descriptor in the aspects of sparse point cloud or high-density point cloud feature extraction, descriptor and matching optimization. The method comprises the following steps: preprocessing a source point cloud set and a target point cloud set, calculating an average point spacing and extracting feature points; the radius of a descriptor support region is adjusted by introducing interpolation, so that the robustness of key point features is improved; then, a weighted covariance matrix is calculated, eigenvalue decomposition is carried out, and an accurate local coordinate system and normal information are obtained; sHOT descriptor matching point pairs based on the radius of the dynamic support area introduces spectrum matching to be combined with ICP, and error point pairs are eliminated. And the registration precision is further improved. Compared with the prior art, the method has the advantages that the descriptive property of the descriptor structure and the optimization of the matching process are enhanced through the dynamic adjustment of the support area, the precision, robustness and real-time property of point cloud registration are improved, and the method is suitable for the fields of computer vision, three-dimensional point cloud registration and the like.

The application relates to a feature point mismatch processing method and device, electronic equipment and a readable medium. The method comprises the following steps: matching initial feature points of two pictures to obtain a matching result; the matching result comprises a plurality of matched first feature point pairs; the moving vectors of the first feature point pairs in the two pictures are calculated; the target feature point pair is determined according to the moving vectors of the first feature point pairs; and the target feature point pair is removed from the matching result. The scheme provided by the application can reduce the probability of feature point mismatch and improve the matching accuracy.

The application discloses an end-to-end cooperative target feature extraction and matching method based on a reference frame. In view of the problems that target feature extraction is easy to be disturbed and cross-view matching is poor in robustness in a complex industrial environment, a unified end-to-end neural network is constructed: background disturbance is inhibited through a semantic shunt backbone network, sub-pixel level ellipse center positioning and morphological parameter regression are realized by using a dense offset field and a continuous parameter field; multi-modal features are extracted based on a dynamic region of interest, local visual information and geometric parameters are fused; and a multi-modal Transformer matching module with a fault-tolerant mechanism is introduced. The application realizes 100% feature detection rate and matching rate under a test set of complex light, large viewing angle, local occlusion and other working conditions, and can be widely applied to feature point extraction and matching in high-precision real-time posture measurement of industrial robots such as aviation hole making.