518 results about "Point pair" patented technology

The invention relates to an image processing technology, and discloses an efficient panoramic image splicing method and system based on multi-view fusion, and the method comprises the steps: collecting a plurality of images from different views; performing multi-scale feature extraction on each image, generating a feature descriptor for each extracted feature point, and matching a corresponding feature point pair; performing multi-view geometric constraint screening on the feature point pairs; according to the screened feature point pairs, estimating a homography matrix between adjacent images, and carrying out global optimization on the homography matrix; aligning all the images into the same coordinate system; determining an overlapping region between adjacent images; and according to the pixel information in the overlapping areas, fusing the overlapping areas by adopting a self-adaptive weighted fusion algorithm so as to splice the plurality of images into a panoramic image. The invention further discloses a control device and a computer readable storage medium. The invention aims to improve the efficiency and accuracy of generating the multi-view fused panoramic image.

The invention discloses a cross-source data three-dimensional reconstruction method and system based on improved Gaussian sputtering, and the method comprises the steps: collecting an unmanned plane inclined image and a ground panoramic image of a target region, and constructing a time-space correlation data set; based on multi-view geometric constraints, space-time coding matching point pairs are established through an adaptive feature pyramid, intelligent incremental cross-source data sparse reconstruction is carried out, and point cloud and camera parameters are output; adopting improved Gaussian sputtering, compressing a three-dimensional Gaussian kernel into a two-dimensional Gaussian primitive through double tangent vector constraint, and fitting surface geometry to realize multi-scale reconstruction; and optimizing primitive parameters by using a differentiatable renderer, completing multi-scale fine reconstruction through gradient back propagation, and generating a high-precision three-dimensional model. According to the method, multi-scale accurate geometric prior input and accurate camera poses are provided for three-dimensional reconstruction, the dependence on professional manual operation in a traditional three-dimensional reconstruction method is greatly reduced, and meanwhile, the geometric accuracy and visual fidelity of a reconstruction result are remarkably improved.

The invention relates to the technical field of image processing, and provides an OCT (Optical Coherence Tomography) image splicing method, which comprises the following steps: acquiring image pairs and matching point pairs corresponding to the image pairs, the image pairs being adjacent OCT images; calculating a homography matrix based on the matching point pairs so as to align the image pairs, and outputting preliminary aligned image pairs; constructing a fusion weight map, wherein the fusion weight map comprises a weight value of a pixel position and a region type; and according to the weight value and the region type, carrying out fusion splicing on the pixel positions so as to output a panoramic OCT image. According to the method, partition fusion is achieved by fusing the weight map, the problems of tissue dislocation and artifacts in OCT splicing are solved, and then the problem that the splicing effect of OCT image splicing is poor is solved.

The invention provides a coal mine underground point cloud accurate registration method based on dynamic weight combination, and relates to the technical field of point cloud registration, and the method comprises the steps: constructing an initial point pair set; calculating a point pair distance weight by adopting a Gaussian kernel function; calculating the shape feature weight of the point pair based on the BSC descriptor; based on the point pair distance weight and the point pair shape feature weight, constructing a point pair reliability weight dynamic combination model, and screening high-reliability point pairs from the initial point pair set; and on the basis of the screened high-reliability point pairs, optimal rigid body transformation is calculated, and point cloud accurate registration is realized. According to the invention, based on the double-weight dynamic combination model of the Gaussian kernel function and the BSC descriptor, the reliability of the point pair is accurately evaluated, and the influence of noise and local feature degradation is effectively inhibited. The adaptive weight adjustment strategy dynamically adjusts the distance weight and the shape feature weight in different registration stages, and optimizes the registration process. And an efficient point pair screening and optimization mechanism is combined with KD-tree accelerated search and Gaussian Newton method optimization, so that the registration speed and precision are improved.

The invention discloses a pose estimation method and device based on point pair feature matching, a medium and a product. The method comprises the following steps: performing coarse registration on a fusion scene point cloud and each template point cloud, and calculating point pair features; according to the corresponding hash key values, retrieving from each hash table to establish registration point pairs; calculating a translation vector and a rotation quaternion matched with each registration point pair according to the reference point coordinate system and the template coordinate system, and performing clustering processing on each registration point pair to obtain each initial corresponding set; screening out a target pose candidate transformation matrix and a matched target template point cloud from the pose candidate transformation matrixes matched with the initial corresponding sets respectively; a target optimization function between the fusion scene point cloud and the target template point cloud is constructed, iterative optimization is performed on parameters in the target pose candidate transformation matrix, and the target pose of the fusion scene point cloud is obtained at the end of iteration, so that the precision, real-time performance and reliability of six-dimensional pose estimation are improved.

The invention discloses a registration method and system for an optical image and an SAR image of the moon, and relates to the technical field of image registration, and the method comprises the steps: firstly determining the same typical moon appearance object in an image to be registered, and enabling the corresponding optical and SAR region images to form a region image pair; performing feature extraction and matching on each regional image pair by using a feature-based registration algorithm to obtain a preliminary matching point pair, and screening out a high-confidence matching point pair; and calculating a region affine transformation matrix of each region image pair based on the high-confidence matching point pairs, constructing a global optimization objective function according to all region affine transformation matrixes, solving an optimal affine transformation matrix, and finally realizing accurate registration of the to-be-registered image.

The invention relates to a multi-modal image matching method and system based on learning features and epipolar geometric constraints. The method comprises the following steps: carrying out edge enhancement processing on an input image through wavelet transform; extracting a multi-scale dense feature map based on the transformed convolutional neural network, and generating a feature descriptor with rotation and scale invariance in combination with principal direction normalization; adopting an FLANN algorithm and dynamic distance constraint to realize preliminary feature matching; and introducing a basic matrix construction and epipolar geometric consistency verification mechanism, and eliminating mismatching point pairs in combination with an RANSAC affine constraint model. According to the method, image enhancement, deep learning and geometric verification strategies are fused, the problems of radiation nonlinearity and geometric distortion caused by imaging mechanism differences among multi-modal images are effectively solved, the matching precision and robustness are improved, and the method is suitable for remote sensing application scenes such as optical-SAR registration, multi-source image fusion and earth surface change detection.

The invention discloses a satellite angular rate identification method and related equipment, and relates to the technical field of rocket launching, and the method comprises the steps: obtaining a rocket telemetry video frame sequence after satellite-rocket separation; for each frame of image of the rocket telemetry video frame sequence, determining a dynamic image area of the target satellite based on the static area mask; based on a feature detection algorithm, binary descriptors are extracted from dynamic image areas of two adjacent frames of images; based on the binary descriptor, matching corresponding feature point pairs between two adjacent frames of images; estimating a rotation matrix between two adjacent frames of images through an iterative optimization algorithm based on the corresponding feature point pairs; and determining the angular rate of the target satellite according to the rotation matrix and the time interval between the two adjacent frames of images. According to the method, rapid and high-precision identification of the satellite angular rate is realized, the problem that the traditional technology depends on indirect data and is high in delay is solved, task success or failure can be judged in real time, and the launching task evaluation efficiency is improved.

The invention provides a panoramic image intelligent splicing method based on context semantics. The method comprises the following steps: firstly, obtaining and preprocessing a sequence image; extracting a depth feature map of the image by using a pre-trained depth network; performing semantic-guided feature matching and image alignment based on the feature map; an optimal suture line is generated according to the semantic region boundary, and fusion is carried out by adopting a semantic weighted multi-band fusion algorithm; and finally, detecting and repairing the semantic inconsistent region to ensure the semantic coherence of the panoramic image. According to the method, matching is carried out by using the high-dimensional feature tensor rich in semantic information extracted by the deep neural network, the problem of feature matching ambiguity caused by repeated textures and weak texture regions such as sky or white walls is solved, the semantic information is used as a strong constraint, regions with similar appearances but different semantics can be effectively distinguished, and the accuracy of feature matching is improved. Therefore, correct matching point pairs with consistent semantics are screened out from massive candidates, and the matching accuracy is greatly improved.

The invention relates to a pathological section splicing method, system and device and a storage medium, and the method comprises the steps: collecting a pathological section image, converting the pathological section image into a plurality of modes, and carrying out the histogram equalization and denoising processing; calculating an entropy value of each pixel point region in the image, and extracting pixel points with high entropy values as feature points; carrying out preliminary matching on the image under different scales, and carrying out bidirectional matching operation in a matching region to obtain a feature point pair; constructing a deformation model according to the feature point pairs, calculating parameters of the deformation model, and enabling the to-be-registered image to deform based on the deformation model through minimizing deformation energy; gradient information of an image overlapping region is calculated, a weight is allocated to each pixel point in the overlapping region according to a weight allocation rule, and the edge of the image overlapping region is processed to obtain a spliced image; and carrying out quantitative evaluation on the spliced image. According to the method, the problems of low registration precision, large calculation amount and poor adaptability to complex backgrounds in the prior art can be solved.

The invention relates to the technical field of multi-view three-dimensional reconstruction, and discloses a multi-view three-dimensional reconstruction method, device and equipment and a storage medium, and the method comprises the steps: obtaining a multi-view image, generating a diffusion feature map for the multi-view image through a diffusion model, carrying out the feature matching according to the diffusion feature map, and generating a matching point pair; calculating the pose of a camera through a Colmap algorithm according to the matching point pairs, extracting the deep features of the multi-view image by using a convolutional neural network to obtain a deep feature map, projecting the deep feature map to a three-dimensional space according to the pose of the camera to generate a preliminary three-dimensional feature map, and extracting the preliminary three-dimensional feature map according to the preliminary three-dimensional feature map. Fusing the preliminary three-dimensional feature map by adopting a U-Net network model to obtain a fused three-dimensional feature map, generating a three-dimensional voxel grid according to the fused three-dimensional feature map, generating a three-dimensional model according to the three-dimensional voxel grid, and finally outputting a globally optimal solution to improve generalization ability and precision.

The invention discloses an underwater target identification and positioning method based on fusion of a physical model and deep learning, and belongs to the technical field of computer vision, and the method comprises the steps: carrying out the transmissivity estimation and physical restoration of a collected underwater monocular image according to an underwater light propagation model, and carrying out the image correction; key feature points are extracted, and high-quality matching point pairs are screened in combination with the joint similarity; further deriving a basic matrix through the high-quality matching point pairs meeting the epipolar geometric constraint relation, solving an essential matrix, and obtaining relative attitude parameters between the cameras in combination with weighted re-projection-LM optimization; and finally, refraction correction triangulation is carried out according to the Snell's law, pixel-level fusion is carried out after scale normalization and space alignment are carried out on the refraction correction triangulation and dense depth output by the MiDaS, three-dimensional space coordinates of the target are inverted, and high-precision recognition and positioning of the underwater target are achieved. The system is light in structure, efficient in calculation, suitable for being integrated on various underwater autonomous or remote control robot platforms and used for tasks such as target recognition, tracking and positioning.

The embodiment of the invention provides a three-dimensional space change detection method and device, equipment and a storage medium, and relates to the field of digital image processing. The method comprises the following steps: acquiring image sequences of the same area shot by the unmanned aerial vehicle in two periods; for each image sequence, performing point cloud reconstruction on the region based on the image sequence to obtain the group of point cloud data; point cloud registration is carried out according to the similarity between the feature descriptors of all the points in the two sets of point cloud data, a matching point pair set comprising a plurality of matching point pairs is obtained, and two points in each matching point pair belong to two pieces of point cloud data and are closest to each other; and for each matching point pair, geometric information of two points in the matching point pair in respective point cloud data is acquired, and if it is determined that the difference between the two pieces of geometric information meets a preset difference condition, it is determined that the corresponding position of the matching point pair in the region is changed. According to the embodiment of the invention, the problem of insufficient spatial change detection precision in a complex scene is solved.

The application is suitable for the field of visual measurement, and provides a target, an information detection method and device, a terminal and a storage medium, wherein the method comprises: controlling a camera to take a picture of the target to obtain a first imaging image; based on the first imaging image, extracting a first pixel point pair with the farthest relative distance on the inner circumference of a circular ring, a second pixel point pair with the farthest relative distance on the outer circumference of the circular ring, a center pixel point of three circles and a vertex pixel point of a first triangle; based on the first pixel point pair and the second pixel point pair, determining a first straight line where the long diameter of the circular ring in the first imaging image is located; based on the center pixel point and the vertex pixel point, determining a second straight line where the bottom side of the first triangle in the first imaging image is located; and determining the intersection of the first straight line and the second straight line as the imaging center point of the target in the first imaging image. The scheme can improve the image data processing efficiency and the practicability of the target.

The embodiment of the invention provides an underwater concrete scene three-dimensional reconstruction method, system and device and a storage medium, and relates to the technical field of underwater three-dimensional reconstruction, and the method comprises the steps: extracting feature points in each key frame image from a plurality of key frame images of video data, and obtaining feature matching point pairs; calculating relative poses of adjacent key frame images and three-dimensional coordinates of the matched feature points according to the pixel coordinates of the same matched feature point; calculating a pixel depth value of each pixel point according to the relative pose; and performing multi-view fusion according to the relative pose, the three-dimensional coordinates of the matched feature points and the pixel depth value of each pixel point, and generating a three-dimensional grid model of the underwater concrete scene. According to the method, the feature points in the enhanced video image are obtained, the moving distance and the translation vector of the shooting camera are calculated, and three-dimensional reconstruction is carried out in combination with the pixel depth value, so that the image acquisition quality is improved, point cloud data are saved, and the precision and efficiency of three-dimensional reconstruction of a large-range underwater scene are improved.

The disclosure provides a method of relocalization, including: in response to a determination that a current image frame satisfies a relocalization condition, acquiring feature points of the current image frame and descriptors of the feature points; performing, based on the feature points of the current image frame and the descriptor of each feature point, feature matching on the current image frame and each stored key frame respectively to obtain feature point pairs after matching the current image frame with each key frame respectively; determining a matching degree of the current image frame and each key frame respectively based on the feature point pairs; determining a key frame with the highest matching degree with the current image frame as a target key frame; and replacing a camera pose corresponding to the current image frame with a camera pose corresponding to the target key frame.

The invention relates to the technical field of internet big data, in particular to an image registration method for extracting image deep feature information based on deep learning, which comprises the following steps: S1, acquiring a first source image and a second source image to be registered; s2, inputting the first source image and the second source image into a trained deep feature extraction model, and outputting a feature description graph; s3, performing feature point pair matching on the feature description graphs of the first source image and the second source image through similarity calculation; s4, generating a homography transformation matrix as a spatial mapping relation between the first source image and the second source image based on the successfully matched feature point pairs; and S5, registering the first source image and the second source image based on the spatial mapping relation to obtain a registered image. According to the method, the deep features of the image are extracted and optimized through the deep learning model, and meanwhile, the homography transformation matrix is generated based on the successfully matched feature point pairs to complete image registration.

The invention relates to the technical field of three-dimensional map construction, in particular to a three-dimensional map reconstruction method and device. The method comprises the following steps: acquiring a plurality of image data and performing feature extraction to generate initial sub-maps, performing local sparse reconstruction on the plurality of initial sub-maps with overlapped visual angles to generate a plurality of sub-maps, performing global feature extraction on the sub-maps to generate a global feature database, and determining similar image pairs according to the similarity of the sub-maps. According to the method, local feature matching is carried out on similar image pairs, corresponding feature points are determined, and for the 3D-3D corresponding relation, namely when corresponding 3D feature points exist in the two sub-maps, the system adopts a rigid body transformation solving algorithm, and the relative pose between the two sub-maps can be accurately calculated by minimizing the distance between the 3D feature point pairs.

The invention provides a non-contact measurement-oriented three-dimensional point AI registration and tolerance analysis method and system, and relates to the field of three-dimensional point AI registration and tolerance analysis, and the method comprises the steps: synchronously collecting point cloud data through multiple sensors, and carrying out the adaptive preprocessing, and obtaining a preprocessed point cloud pair; constructing a multi-scale geometric descriptor by using a local curvature and a normal vector, and performing multi-scale hierarchical feature extraction to obtain an enhanced point feature set; calculating a bidirectional matching probability matrix by using an optimal transmission theory, screening high-confidence point pairs through spatial compatibility constraint to obtain a high-confidence matching point pair set and a matching score, performing transformation matrix estimation and quantification uncertainty, and outputting optimal rigid body transformation and a covariance matrix thereof; self-adaptive optimization is executed in a layered mode, the search range is dynamically adjusted, transformation parameters are verified in a cross-scale mode, and a convergent accurate registration result is output. The method is used for overcoming the defect that in the prior art, low-overlapping-rate registration precision is low.

The invention relates to the technical field of defect detection, in particular to a crankshaft quality inspection method and system based on image data, and the method comprises the following steps: extracting crankshaft coordinates to generate a serial number list, binding an image center to generate a list, constructing a path table by an attribution image, extracting a gray difference to generate an array, recognizing the position of an abnormal mapping structure, and outputting crack positioning information. According to the method, the accurate positioning basis is obtained by establishing the corresponding relation between the structure number and the space coordinate and image acquisition, the mapping between the image content and the structure area is realized by combining the imaging center point extraction and the image attribution binding, the traceability of data archiving is enhanced, the gray value point pairs are extracted by using the symmetric center line, and the gray difference matrix is constructed. According to the overall scheme, image attribution and defect identification under structure dominance are achieved, the detection accuracy and the positioning precision are improved, the false detection rate is reduced, and the stability and the engineering adaptation capacity are enhanced.

The invention discloses a visual feature detection and matching method fusing ORB and SIFT. The method comprises the steps that system parameters are initialized; performing DoG extreme point detection in a scale space by using the response significance of a FAST operator; fusing the mixed descriptors of ORB and SIFT to generate a matching strategy; and finally, outputting a matching point pair containing key points in the left and right images and corresponding three-dimensional point information, and caching the key points of the current frame and ORB / SIFT descriptors thereof as initial values of matching of the next frame. According to the method, the response significance of a FAST operator is combined, DoG extreme point detection in a scale space is guided, a fusion mechanism is utilized to enable descriptors to have higher scale adaptability and matching stability, a weight function with smooth change is utilized to dynamically adjust the proportion of SIFT and ORB descriptors, the robustness of SIFT is reserved in a large-scale area, and the robustness of the large-scale area is improved. And the high efficiency of ORB is exerted in a small-scale region.

The invention discloses a real-time image feature matching method based on spatial distribution consistency. The method comprises the following steps: acquiring a high-resolution to-be-matched image with an overlapping region; performing SIFT feature point detection on the image, and extracting descriptor vectors corresponding to SIFT feature points; calculating the Euclidean distance between the descriptors corresponding to the SIFT feature points; matching the descriptors corresponding to the SIFT feature points to obtain initial matching point pairs; two-dimensional sample points are constructed for clustering, correct matching point pairs and mismatching point pairs are separated, and coarse screening of the matching point pairs is achieved; constructing a four-dimensional sample vector based on the residual matching point pairs after coarse screening; and clustering the constructed four-dimensional sample vectors, separating correct matching point pairs from mismatching point pairs, and realizing fine screening of the matching point pairs. The method can adapt to complex application scenes such as large view angle change, non-rigid deformation and non-parametric geometric constraint, and the matching accuracy, stability and matching efficiency and precision are remarkably improved.

The invention provides a defect detection method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring a template image and a product image of a to-be-detected product; performing feature matching on the template image and the product image, and performing non-rigid alignment on the product image based on feature point pairs obtained by feature matching to obtain an alignment image of which the geometric view angle is consistent with that of the template image; and defect detection is carried out based on the aligned image and the template image to obtain a detection result, so that the problem of image deformation caused by the fact that the surface of the to-be-detected product is a curved surface and the shooting angle is not correct in a traditional scheme is effectively solved, and local and global deformation of the product image can be corrected, thereby greatly improving the alignment precision of defect detection, and improving the detection accuracy of the to-be-detected product. Therefore, the real difference of the product surface can be more focused during defect detection, non-defect geometric deformation interference is effectively shielded, and the accuracy and robustness of defect detection of complex surfaces of household electrical appliances and the like are greatly improved.

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 discloses a target frame automatic tracking drift judgment method and device, equipment and a storage medium, and relates to the field of image recognition. Extracting a target frame region of two adjacent frames of images, and converting the target frame region into a grey-scale map; angular point detection is carried out on the previous frame grey-scale map, and a first angular point group is extracted; performing optical flow calculation on the current frame grey-scale map based on the previous frame grey-scale map and the first angular point group, and matching a second angular point group; based on the distance and angle information of the successfully matched feature angular point pairs in the angular point group, judging the drift state of the tracking target corresponding to the current frame image; and when it is judged that the tracking target drifts, predicting the target position in the current frame image based on Kalman filtering and re-identifying the tracking target, and determining the real target position in the current frame image by combining a re-identification result and a response diagram obtained after correlation calculation of the tracking template. According to the scheme, feature level drift judgment is realized through corner detection and optical flow matching, effective re-identification is carried out, and the problem of target frame drift is solved.

The invention relates to the technical field of SLAM, in particular to a dynamic environment robust positioning method based on improved ORB-SLAM2, and the method comprises the steps: obtaining RGB-D scene data containing an indoor moving object, and extracting ORB feature points described by FAST angular points and BRIEF descriptors; extracting feature points with unreliable heights according to geometric features of ORB feature points of adjacent frames; pixel analysis is carried out on the RGB-D scene data based on a Fast-SAM instance segmentation model, and a mask matrix containing environmental object space boundary information is generated; performing point-prompt extraction on the height unreliable feature points and the mask matrix to obtain a dynamic object mask; removing ORB feature points existing in the dynamic object mask to obtain a reliable feature point set; and calculating matching point pairs in the reliable feature point set to obtain an essential matrix and a corresponding rotation matrix and translation vector, and refining the pose based on a nonlinear optimization method.

The invention discloses an image staged matching and positioning method, and relates to the technical field of image analysis. Comprising the following specific steps: constructing a model architecture, and inputting an unmanned aerial vehicle image u; candidate images {u, Si} are screened through global semantic feature matching; the method comprises the following steps: extracting image key points and description words, establishing an initial matching point pair set, introducing a mismatching screening mechanism, executing homography transformation on geotagging information in a satellite image, and determining ground target positioning under the view angle of an unmanned aerial vehicle. According to the method, through a mode of combining global feature optimization and local mismatching screening, accurate positioning of a ground target under the view angle of the unmanned aerial vehicle is realized, and an optimal matching candidate is rapidly screened out; in the local matching stage, mismatching point pairs are eliminated, and the geotagging information in the satellite image is accurately mapped to the unmanned aerial vehicle image coordinate system through homography transformation, so that the image matching problem under the large view angle change is effectively solved, and the accuracy and efficiency of image matching and positioning are improved.

The invention belongs to the technical field of computer vision and three-dimensional image processing, and particularly discloses a mismatching mark point correction method and system based on an improved RANSAC algorithm. According to the method, congruent triangle criteria are introduced into the RANSAC algorithm, and potential matching point pairs conforming to geometric consistency are screened. Only when the potential matching point pairs meet the geometric consistency, the transformation matrixes R and T are solved, that is, mismatching points existing in the potential matching point pairs are effectively eliminated, so that on the premise that the point cloud splicing precision and the success rate are not changed, wrong calculation is reduced, the point cloud registration efficiency and robustness are improved, and the point cloud registration accuracy is improved. The method is suitable for various application scenes such as industrial part measurement, medical modeling, cultural relic digital protection and virtual reality.

The invention discloses an iterative registration optimization algorithm based on angle clustering, and the algorithm comprises the following steps: inputting an initial scene point cloud and an initial model point cloud, simplifying the initial scene point cloud and the initial model point cloud, and obtaining an initial corresponding point pair set; constructing a compatibility constraint, calculating a compatibility score of each pair of corresponding point pairs, and sequencing to construct a compatibility matrix; performing outer layer circulation: sequentially selecting the foremost point pair from the compatibility matrix as a first corresponding point pair; inner layer circulation: selecting a second corresponding point pair; clustering the corresponding point pairs based on the six rotational degrees of freedom, and selecting all significant clusters to generate a conversion hypothesis; and verifying and selecting an optimal conversion hypothesis as an output conversion matrix. The compatibility among all the corresponding point pairs is evaluated by constructing a compatibility matrix; a simple and effective clustering strategy is adopted, and all significant clusters are considered to generate a conversion hypothesis; the simplified point clouds and the key points are effectively combined through a hypothesis verification strategy, and the accuracy of alignment of the low-overlapping point clouds is improved.

The invention provides a bullet train bottom part loss detection method and a bullet train bottom part loss detection device based on deep learning, which can solve the problems of low registration precision and false report and missing report caused by easy interference under complex working conditions in the prior art, and the method comprises the following steps: obtaining a to-be-detected image at the bottom of a bullet train and a preset template image, the template image comprises at least one template frame for identifying the position of the standard part; extracting depth features of the to-be-detected image and the template image, and performing feature matching to obtain an initial key point pair set; performing abnormal point elimination on the initial key point pair set to obtain an effective key point pair set; according to the effective key point pair set, mapping a template frame in the template image into the to-be-detected image to obtain a corrected template frame; based on a preset comprehensive cost function, matching a target detection frame in the to-be-detected image with the corrected template frame to obtain an optimal matching pair; and according to the result of the optimal matching pair, identifying whether there is part loss in the to-be-detected image.