43 results about "Local feature descriptor" patented technology

The invention provides a bridge structure crack position identification method based on deep learning and computer vision, and the method comprises the steps: obtaining a multi-view visible light image and a depth image, fusing the visible light image and the depth image according to registration parameters, and generating a fusion image containing the texture and depth information of a bridge; for the fused image, extracting a crack region by adopting an adaptive threshold segmentation algorithm, and dynamically adjusting the threshold to obtain a crack binary image; according to the crack binary image, calculating three-dimensional coordinates of crack pixel points through a three-dimensional reconstruction algorithm, and performing reconstruction to obtain three-dimensional point cloud data containing crack space position and form information; on the basis of the three-dimensional point cloud data, three-dimensional local feature descriptors are adopted to describe geometric attributes of the surface of the crack, crack fragment parameters are calculated, and a three-dimensional crack model is constructed; based on a deep learning model, a transfer learning strategy is adopted, a self-adaptive transferable bridge crack three-dimensional detection method is formed, and model parameters are updated so that the model parameters can adapt to the characteristics of a target bridge.

The invention discloses an obstacle sensing method and system based on laser radar data analysis, and relates to the technical field of radar data obstacle analysis, and the method comprises the steps: scanning a surrounding environment through a multi-mode sensor, obtaining sensor data, synchronously collecting the data of the multi-mode sensor, and carrying out the preprocessing, an obstacle target is obtained from a camera image, sensor data are integrated by using a data fusion algorithm, and obstacle features are extracted by using a segmentation threshold function based on density clustering in combination with an obstacle local feature descriptor and global feature description to form complete obstacle feature description. The laser radar can provide high-resolution distance data, so that the system can accurately identify and position obstacles in the surrounding environment, algorithm processing is combined, the method can complete scanning and analysis of the environment in a short time, rapid response to the obstacles in the dynamic environment is achieved, and the system can be widely applied to the field of dynamic environment monitoring. And the accuracy and reliability of obstacle detection are greatly improved.

The invention discloses a point cloud data splicing method, which comprises the following steps of: aiming at point cloud data splicing of an initial pose, performing pre-splicing by adopting a nearest point iterative cutting method combining local features and a nearest point distance, and then performing further accurate splicing on a pre-splicing result by adopting a point-to-surface nearest point iterative algorithm. And thus, an accurate and complete point cloud data splicing model is obtained. The method comprises the following specific steps: scanning a three-dimensional entity to obtain original point cloud data; performing noise reduction processing on the original point cloud data to obtain a to-be-spliced point cloud; calculating a local feature descriptor of each position point in the to-be-spliced point clouds, and obtaining a local feature of each position point; pre-splicing the to-be-spliced point clouds based on the local features of the position points to obtain a pre-splicing result; and the pre-splicing result is precisely spliced, and a point cloud data splicing model is obtained.

The invention discloses a non-directional dynamic patrol street lamp visual coding method and system. The method comprises the following steps: acquiring continuously acquired road images and geographic position and attitude data thereof, and grouping after primary processing and spatial index construction; screening candidate image pairs by calculating image similarity, extracting multi-dimensional features of street lamps and environments thereof, constructing local feature descriptors including street lamp bodies and environment context features, and forming small-range environment fingerprints to determine successfully matched street lamp entities; hybrid codes are generated for the entities, and the codes are combined with geographic coordinates, the fused local feature descriptors, the fused fixed object list and the fused identifier of the nearest fixed object distance, and a final code value is obtained through Hash calculation. By implementing the method provided by the invention, visual features, small-range environment fingerprints and spatial priori knowledge can be effectively fused, so that the problems are solved, and intelligent management of urban lighting facilities is promoted.

The present invention discloses a three-dimensional registration and reconstruction method based on multi-domain multi-dimensional feature maps. First, a specified number of key points are randomly selected from the scene point cloud and the model point cloud. Then, each key point is traversed, and the corresponding multi-domain multi-dimensional feature maps are calculated as local feature descriptors. Based on the multi-domain multi-dimensional feature maps, the correspondence between key points is obtained, and the key point pairs are sorted according to the matching degree. In each iteration, a compatible triple is selected in order to generate a hypothesis, and it is checked whether the hypothesis can correctly register the model into the scene. The present invention has a short calculation time and a low time cost.

This invention discloses a point cloud registration method and system based on statistical local feature description and matching, relating to the field of point cloud registration. The method includes: determining the local feature descriptor of each query point and the local feature descriptors of its corresponding neighboring points in 3D point cloud data; statistically weighting the local feature descriptor of each query point and the local feature descriptors of its corresponding neighboring points to obtain the statistical local feature descriptor of the corresponding query point; acquiring the statistical local feature descriptors of the source point cloud and the target point cloud; and improving the ICP algorithm based on the feature differences between point pairs in the source point cloud and the target point cloud and the average matching distance to perform point cloud registration. This invention can improve the accuracy and robustness of point cloud registration.

The invention relates to the technical field of robot autonomous navigation and visual identification, and discloses a visual scene identification method for domain invariant information mining, which is realized by forming a DIMNet model by using two parts of a backbone network VGG-DIMM and an aggregation head NetVLAD, and comprises the following steps of: for a given input image, preliminarily extracting a local feature descriptor of the given input image by using the VGG-DIMM; taking the local feature descriptors extracted by the VGG-DIMM as input, and aggregating the local feature descriptors into a one-dimensional global feature descriptor F (I) by using the original NetVLAD; comparing the global feature descriptors F (I) of the query image with the global feature descriptors F (I) of all database images one by one through a similarity algorithm, so as to find out all database images belonging to the same scene as the query image; according to the method, an adopted domain invariant information mining module (DIMM) is included in a cascaded convolution unit group, and the purpose of enhancing the attention of a DIMNet model to a domain invariant static object is achieved.

The invention relates to the technical field of question test paper composition, and discloses a medical theory examination intelligent test paper composition method based on an artificial intelligence technology, and the method comprises the steps: question information preprocessing, question similarity calculation, and intelligent test paper composition. In the topic processing link, the topics are subjected to disassembling and structured storage according to multiple modes of text, image and case analysis, and features are extracted in combination with technologies such as keyword extraction and local feature descriptor methods; in question similarity calculation, cosine similarity and a weighted summation mode are utilized to accurately quantify the similarity of questions of different question types, and question repetition in test paper is effectively avoided; during test paper composition, screening is performed according to parameters such as preset question type structures, knowledge point coverage and difficulty levels, reasonable difficulty distribution and comprehensive knowledge point coverage are ensured in combination with a similarity threshold judgment and replacement mechanism, a replacement object is determined according to historical use times of questions, test paper composition scientificity and question bank resource utilization rate are improved, and test paper composition efficiency is improved. And the medical theory examination paper composition flow is optimized.

This invention provides a method, apparatus, and electronic device for vehicle exterior image registration. The method includes: determining a target region from a 3D point cloud corresponding to a template image; determining a region to be matched from a 3D point cloud corresponding to an image to be detected; obtaining a first local feature descriptor corresponding to points in the target region and a second local feature descriptor corresponding to points in the region to be matched; obtaining a first transformation matrix based on the first and second local feature descriptors; obtaining a third transformation matrix based on the first transformation matrix; performing a two-dimensional perspective transformation on the third transformation matrix to obtain a pseudo-3D transformation matrix corresponding to the 2D image to be detected; and transforming the 2D image to be detected using the pseudo-3D transformation matrix to obtain the registered image to be detected. This invention can accurately register 2D images to be detected, improving the accuracy of vehicle exterior image detection.

The invention relates to the technical field of machine vision detection, in particular to an inner package detection and recognition method based on vision detection, which comprises the following steps: synchronously acquiring a multi-view image sequence of a to-be-detected piece through an annularly arranged acquisition device; independently analyzing each image, and generating an initial defect area mask and a local feature descriptor; a three-dimensional defect space distribution model is constructed by fusing multi-view masks, and geometric topology and depth information of the three-dimensional defect space distribution model are extracted to form a comprehensive feature vector. And performing cross-modal correlation analysis on the local features and the three-dimensional comprehensive features to generate an enhanced defect feature spectrum. And calling a pre-training classification model to analyze the atlas, outputting an inner package type, a defect type and a severity level, and generating a quality evaluation report and a processing instruction according to the inner package type, the defect type and the severity level. According to the method, three-dimensional space accurate positioning and multi-feature deep fusion of the defects are realized, and the recognition accuracy and evaluation comprehensiveness of the complex defects are improved.

The invention belongs to the technical field of image processing, and relates to a protection method and system for the design copyright of an automobile cushion. The method comprises the following steps of: extracting a digital image contour of an automobile cushion design drawing, calculating a normalized Fourier descriptor of the contour as a shape feature, dividing a drawing area into grid units according to radial sectors and concentric rings, converting a digital image into a CIELAB color space in each grid unit, and for each grid unit, carrying out CIELAB color space conversion on the CIELAB color space; extracting a gradient direction histogram of a brightness L * value taking gradient intensity as a weight and a quantization histogram of statistical a * b * two-dimensional color distribution of pixels in the grid unit, combining the two histograms into a local feature descriptor of the grid unit, serializing and splicing the local feature descriptor from inside to outside, and combining shape features with a local feature sequence to form copyright feature data, and performing digital signature by using an asymmetric encryption private key, and storing the signature and a timestamp in a distributed account book together to realize copyright protection.

This invention discloses a global localization method based on multi-layer 3D point cloud bird's-eye view, belonging to the field of mobile robot localization and navigation technology. The method first divides the LiDAR point cloud into multiple overlapping slices along the height direction and projects them to generate a multi-layer bird's-eye view. Then, the multi-layer bird's-eye view is input into a global localization network containing a shared feature extraction module, a vertical slice attention module, a slice feature fusion module, and a NetVLAD aggregation module to extract global and local feature descriptors. Finally, a reference frame is obtained through global feature retrieval, and the pose of the query frame is calculated by combining keypoint matching and random sampling consensus algorithms. This method can preserve the vertical structure information of the point cloud, reduce localization ambiguity in complex scenes, and balance localization accuracy and computational efficiency.

A method and system for selecting descriptor invariance based on an attention mechanism, comprising the following steps: preparing a data sequence, inputting an image into a multi-scale feature extraction module, generating four local feature descriptors with different invariances; sending the local feature descriptors into a parallel self-attention module to extract four meta-descriptors with different invariances; calculating the similarity of the four different invariant meta-descriptors of a pair of key points, and using the similarity as the weight coefficient of each local descriptor through a softmax function; and finally calculating the final local descriptor distance of the key point by weighted calculation to complete the process of selecting the local descriptor invariance.

The application discloses a loop closure detection method based on block feature uniform weighting and distance sorting in outdoor complex environment, and belongs to the technical field of visual camera simultaneous localization and mapping and computer vision. The technical scheme of the application comprises the following steps: performing gray processing and Gaussian filtering pretreatment on the collected image pair, uniformly dividing the image into a plurality of sub-image blocks, extracting local feature descriptors from each sub-image block, calculating the distance between the feature vectors of the corresponding sub-image blocks based on cosine similarity, uniformly weighting the feature similarity of all sub-image blocks, sorting the calculated feature distances of all sub-image blocks from small to large, screening out the key matching block with the smallest distance by setting a threshold to eliminate the false matching block caused by illumination change noise, and comprehensively determining the global similarity of the feature distance of the key matching block to confirm the success of loop closure detection and output loop closure information to optimize the global pose of the visual camera.

The invention provides a visual position identification method of a pre-training model based on feature aggregation and selection, and belongs to the technical field of visual position identification. According to the method, global feature descriptors and local feature descriptors are extracted, an adaptive feature aggregation technology is adopted, and the extraction of landmark information with discrimination is emphasized, so that the difference between a pre-training model and an actual visual position identification application is effectively made up. And for the global feature descriptors, integrating image global information into descriptors with rich semantics through a feature aggregation method by adopting a DINOv2 + ViT model structure. For extraction of local feature descriptors, a patch-level descriptor extraction module is designed, and dense and distinctive local features can be output. The features play an important role in two-stage tasks, firstly, the features are used for local matching of candidate images, and then the matching precision is further improved through a reordering strategy.

The application discloses a kind of multi-modal remote sensing image registration method, system, terminal equipment and storage medium, based on the structure similarity between multi-modal remote sensing image pair, local feature descriptor (edge feature) is extracted using pre-trained edge detection network, and feature matching is carried out using traditional template matching method, accurate and efficient registration under multiple modal images is realized.The application obtains higher matching accuracy and better robustness in multi-source image matching.

The present invention relates to the fields of computer vision and robotics, and in particular to a point cloud registration method based on graph-structured local feature descriptors. The method comprises the following steps: S1: performing patch extraction and normalization processing on training sample data and data to be registered; S2: building a deep neural network model based on a graph structure, so that the model is suitable for processing point cloud data; S3: using the preprocessed training sample data in step S1 to train the deep neural network built in step S2; S4: inputting the preprocessed data to be registered into the deep neural network trained in step S3, where the data to be registered is a pair of point clouds to be registered, and obtaining feature descriptor vectors of corresponding points respectively; S5: using the feature descriptor vectors of corresponding points generated in step S4, calculating the rigid transformation matrix [R, t], and using the final rigid transformation matrix [R, t] to achieve point cloud registration, where R represents a rotation matrix and t represents a translation vector.

The present application proposes a method, device, equipment and storage medium for retrieval of duplicate images, the method comprising: obtaining a feature vector and a first local feature descriptor of an image to be retrieved; obtaining image identifiers corresponding to a preset number of nearest neighbor images of the image to be retrieved according to a preset feature index and a feature vector; obtaining a second local feature descriptor of each nearest neighbor image according to each image identifier; and determining images that are duplicates of the image to be retrieved from each nearest neighbor image according to the first local feature descriptor and each second local feature descriptor. The present application roughly screens out a plurality of nearest neighbor images from a large number of images, and determines duplicate images therefrom by matching local feature descriptors, thereby narrowing the retrieval scope and improving the retrieval efficiency. And based on the rough screening of a deep learning feature embedding model trained with artificially modified similar images, it can accurately recall images that have been artificially processed, thereby improving the accuracy of duplicate image retrieval.

The application discloses a non-directional dynamic patrol street lamp visual coding method and system. The method comprises the following steps: acquiring continuous road image and its geographical position and attitude data, grouping after preliminary processing and spatial index construction; screening candidate image pairs by calculating image similarity, and extracting multi-dimensional features of street lamps and their environment, constructing local feature descriptor including street lamp body and environmental context features, forming small-range environmental fingerprint to determine the matching successful street lamp entity; generating hybrid coding for the entity, which combines geographical coordinates, fused local feature descriptor, fused fixed object list and fused identifier of the distance from the nearest fixed object, and obtaining the final coding value through hash calculation. Through the implementation of the method of the application, visual features, small-range environmental fingerprint and spatial priori knowledge can be effectively fused to overcome the above problems and promote the intelligent management of urban lighting facilities.

The present invention relates to the technical field of question composition, and discloses an intelligent composition method for medical theory examinations based on artificial intelligence technology, comprising the steps of: question information preprocessing, question similarity calculation, and intelligent composition. In the question processing link, questions are decomposed and structured according to multiple modalities such as text, image, and case analysis, and features are extracted by combining techniques such as keyword extraction and local feature descriptor methods. In question similarity calculation, cosine similarity and weighted summation are used to accurately quantify the similarity of questions of different question types, effectively avoiding question duplication in the examination paper. When composing the examination paper, questions are screened according to preset question type structure, knowledge point coverage, difficulty level and other parameters, and similarity threshold judgment and replacement mechanism are combined to ensure a reasonable difficulty distribution and comprehensive coverage of knowledge points. The replacement object is determined according to the number of historical usage of the question, thereby improving the scientific nature of the composition and the utilization rate of the question bank resources, and optimizing the composition process of the medical theory examination.

The application relates to a pavement crack identification method based on point cloud-RGB heterogenous image multi-stage registration mapping. The method comprises the following steps: collecting pavement point cloud data and pavement image data for time and space synchronization, creating a projection image pair for feature extraction, obtaining a local feature descriptor, matching feature points in the local feature descriptor, obtaining an actual matching point pair, solving the parameters of a direct linear transformation equation according to the actual matching point pair, obtaining the mapping relationship between the RGB image pixel coordinates and the three-dimensional point cloud coordinates, traversing the coordinates of each point in the pavement image data, assigning the mapping relationship depth information to the pavement image data, obtaining a point cloud projection image, generating a depth image from the point cloud projection image, labeling cracks and backgrounds, constructing a data set, training a crack identification model using the data set, obtaining a trained crack identification model, and identifying pavement cracks, thereby improving the efficiency and accuracy of automatic crack identification.

This invention belongs to the field of computer vision technology and discloses a local feature extraction method. A novel local feature learning network framework is established, using a fully convolutional network L2-Net as the backbone. Features from the spatial structure enhancement module are extracted and fused into the intermediate and deep layers of the backbone network. A dataset with depth information is used as the training set. The correspondence between two images is obtained through the inherent depth information of the dataset, serving as the ground truth for local feature discriminative and robust training. The novel local feature learning network framework outputs feature heatmaps for discriminative training and feature detection heatmaps and local feature descriptors for robust training. This invention balances robustness and discriminability; the network still contains rich structural information in its deep layers, thereby improving discriminability. A joint reinforcement learning method is used to ultimately obtain better local feature descriptors. Finally, this method was evaluated on a feature matching task and achieved state-of-the-art results.

The invention relates to a quick identity authentication method and device based on physical unclonable features, equipment and a medium, and the method comprises the steps: taking a query feature descriptor as a retrieval object, and carrying out approximate nearest neighbor matching retrieval in a local feature descriptor set of a global feature library, after screening, determining one or more candidate physical unclonable tag identities through voting; for candidate physical unclonable tag identities, corresponding registered key point space coordinates are called from a geometric feature library, and a re-projection error threshold is combined through a random sampling consistency algorithm; checking whether the space coordinate of the query key point and the space coordinate of the registered key point of the candidate physical unclonable tag meet a preset geometric transformation model or not, and calculating a geometric consistency score according to the preset geometric transformation model; and if it is detected that the geometric consistency score is greater than or equal to a preset score threshold, outputting an identity authentication success result. According to the invention, the identity authentication efficiency is greatly improved, and the false rejection rate of identity authentication is remarkably reduced.

This invention presents an image retrieval method based on the fusion of attention enhancement and autoencoding. It uses an improved ResNet50 network to extract global and local feature maps, obtains global and local feature descriptors based on these maps, calculates image similarity, and derives the target image through similarity comparison. This image retrieval method improves upon traditional residual blocks by employing automaton encoding, effectively unifying local and global features into a single network. It uses an attention mechanism to extract regions of greater interest, avoiding algorithmic overhead and resulting in faster retrieval speed and higher accuracy.

The invention belongs to the technical field of image processing; the image local feature description method constructed based on gradient information has poor robustness for some scenes such as complex illumination changes in the field of image matching, and the matching effect is not ideal; the image matching method constructed based on pixel grayscale information is better adapted to illumination change scenes, but cannot cope well with nonlinear complex illumination change scenes; the invention provides an image matching method based on a spatial ordered gradient histogram, constructs a gradient vector of each sampling point in a feature point support area, calculates the amplitude angle between the gradient vectors in turn in a counterclockwise direction, establishes an ordered gradient amplitude angle histogram in a calculation order, generates an image local feature descriptor, completes image matching, and obtains a mapping matrix between images; the invention can be widely applied to the field of image matching, and effectively improves the robustness in coping with complex illumination changes.

The invention discloses a multi-mobile robot large-area environment map establishing method based on implicit neural points, which is characterized in that 3D point clouds acquired by mobile robots are preprocessed, and neural point sets of local maps of the mobile robots and corresponding neural point features are constructed; secondly, a signed distance field of the local map is predicted based on the neural point features corresponding to the neural points, sub-maps of geometric representation displayed by the mobile robots are constructed through the generated signed distance field, then the sub-maps of the mobile robots are uploaded to a central processing unit to be associated, overlapping areas are screened, and the local map is obtained. The method comprises the following steps of: overlapping a plurality of sub-maps, constructing a local feature descriptor of each neural point in an overlapping region, matching and clustering the neural points of each sub-map by utilizing the constructed local feature descriptor to obtain a clustering point cluster matched and corresponding to each sub-map, and finally fusing each local map according to the clustering point cluster matched and corresponding to each sub-map to obtain a fused local map. And a complete large-area environment map is established.

The application belongs to the technical field of three-dimensional point cloud processing, and particularly relates to a VGICP point cloud registration algorithm based on an FPFH feature descriptor. The application comprises the following steps: 1) obtaining point cloud data and performing initialization processing on the point cloud data; 2) calculating normal vectors of each point in the point cloud data, thereby constructing a local feature descriptor of a histogram; 3) performing key point detection on all the point cloud data, thereby obtaining adjacent points; 4) extracting features represented by each point and adjacent points from the point cloud data; 5) performing coarse registration on source point cloud according to the extracted features; 6) performing voxelization on target point cloud; and 7) performing fine registration on the source point cloud. The application is aimed at selection and improvement of a point cloud registration algorithm, so that the speed of point cloud registration is accelerated and the accuracy is increased.

The invention provides an ROS trolley 3D reconstruction method and system based on lightweight deep learning. The ROS trolley 3D reconstruction method comprises the steps of S1, data preprocessing, wherein original data of a sensor are received and preprocessed; the sensor original data comprises laser radar point cloud data and RGB-D camera data; s2, lightweight feature extraction: performing multi-view projection processing and back projection mapping on the preprocessed point cloud data to obtain a local feature descriptor of each point cloud; s3, point cloud registration and key frame management: searching historical key frame features nearest to the current frame features and obtaining feature point matching pairs for coarse registration and fine registration; judging whether the current frame feature is a key frame or not; s4, incremental map construction: updating a local map based on the obtained new key frame, managing the key frame by adopting a sliding window strategy, and executing closed-loop detection and global optimization; and S5, visualization and interaction. The method effectively balances the contradiction between precision and efficiency, and is strong in resource adaptability and high in engineering practicability.

The invention discloses a cross-modal image feature matching model, a cross-modal image feature matching method and electronic equipment. Wherein the cross-modal image feature matching model fuses a local feature descriptor generated by a neural network with a high-dimensional semantic feature map which comes from a visual basic model and encodes high-level semantic context information through a feature fusion module, and through the fusion strategy, a novel mixed feature descriptor is generated; the mixed feature descriptors have the sub-pixel-level positioning precision of local features and the strong robustness of a semantic feature map to modal changes at the same time, so that a high-quality initial corresponding relation can be more effectively established between images of different modals, and the huge visual domain difference caused by different imaging technologies can be effectively overcome; the initial matching quality is obviously improved; in combination with a learning type geometric confidence module, wrong matching caused by complex non-rigid deformation can be effectively filtered out, and finally the matching precision far higher than that of an existing method is achieved.

The present invention relates to the field of virtual simulation technology, and in particular to a virtual simulation anesthesiologist motion capture method, comprising: obtaining a frame image of each action video of the anesthesiologist, obtaining a local feature descriptor of the key point according to the grayscale value of the pixel point within the preset window range of the key point in the frame image, judging whether different key points in the frame image are in the same topological structure, and then obtaining a number of topological structures of the frame image, obtaining a topological structure set according to the grayscale value of the key point in the topological structure of different frame images and the local feature descriptor of the key point, obtaining the missing important key points in the topological structure according to the centroid coordinates and the number of the key points of the topological structure in the topological structure set, and then completing the establishment of the virtual simulation model of the anesthesiologist. The present invention effectively modifies the SIFT algorithm according to the importance of different key points, so that the missing key points can be effectively detected, and the capture of the anesthesiologist's actions is completed.