25 results about "Essential matrix" patented technology

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 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 relates to the technical field of agent deep learning, and particularly discloses a single-agent three-dimensional visual positioning method and system based on deep learning, and the method comprises the steps: generating an optical flow field through a deep learning optical flow prediction model, judging the prediction accuracy in combination with analysis process parameters, guaranteeing the input quality, and providing a reliable basis for subsequent three-dimensional visual positioning; an essential matrix is constructed and decomposed, an initial three-dimensional visual positioning result is obtained in combination with constraints, positioning precision is improved, when the initial three-dimensional visual positioning result is unreliable, inertial measurement unit data is fused for space-time alignment, fusion reliability is verified through process parameters, and positioning robustness in a complex scene is enhanced. And the accuracy and the stability are considered, and the accuracy and the reliability of single-agent three-dimensional visual positioning are effectively improved.

Embodiments are directed to the association of concurrent tracks across multiple views for sensing objects. A sensing system that employs signal beams to scan paths across an object may be provided such that two or more sensors separately detect signals reflected by the object. Events may be determined based on the detected signals such that the events include pairs of events that correspond to pairs of sensors. Essential matrices may be generated based on the positions of each pair of sensors. The pairs of events associated with the pairs of sensors may be compared based on the essential matrices of the sensors. Scores for the pairs of events may be provided based on the comparison. If a score for a pair of events may be less than a threshold value, each event in the pair of events may be associated with a same location on the object.

This application relates to a method, apparatus, device, and storage medium for detecting camera angle shift, and pertains to the field of computer technology. The method includes: acquiring a standard image captured by the camera at a first time point and an image to be detected captured by the camera at a second time point; matching first image feature points of the standard image with second image feature points of the image to be detected; if the number of successful matches between the first and second image feature points is greater than or equal to a preset number, then for any set of successfully matched first and second image feature points, determining an essential matrix based on the first image coordinates of the first image feature points and the second image coordinates of the second image feature points; and based on the essential matrix, determining whether an angle shift has occurred in the camera between the first and second time points. This method can improve the accuracy of camera angle shift detection.

The invention relates to a method for fusing visual odometry with inertial measurement unit (IMU) data of an ego-vehicle (1) with the following steps: - Capturing an environment of the ego-vehicle (1) by means of at least one optical sensor (2) and generating sensor data, wherein the optical sensor (2) is a camera and the sensor data are consecutively captured camera images; - Applying semantic segmentation on the camera images; - Computing a visual odometry by means of optical flow; - Observing a shape and position of a road (F) and / or road elements (4) and deriving a tilt angle between an optical axis of the optical sensor (z) and the road (F); - Estimating a three axis rotation between two consecutive camera images by means of an essential matrix, which was computed through single value decomposition; - Fusing a delta rotation between the two consecutive camera images with accelerometers of the inertial measurement unit, wherein the fusion is conducted using a filter; - Adjusting rotation parameters of the essential matrix and transforming the detected objects (3) based on the fusion of the delta rotation.

The application discloses a monocular image sequence satellite attitude estimation method, and belongs to the technical field of space target perception, and comprises the following steps: acquiring satellite sequence images shot by a space-based monocular camera as input images; completing segmentation of a solar panel and a load main body; obtaining a single-frame satellite attitude based on a segmentation result; screening feature points in the satellite sequence images based on a feature extraction deep learning network and a preliminary semantic segmentation result; guiding feature point matching based on semantic information of an extracted mask; iteratively optimizing the feature point matching by using a polar line constraint; obtaining an essential matrix between adjacent images; calculating satellite attitude changes in the satellite sequence images; and combining the single-frame satellite attitude to obtain a series of attitudes of the satellite in the satellite sequence images. According to the technical scheme, the attitude estimation task can be better completed without labeled data, and the generalization performance for a non-cooperative target satellite is improved.

The application relates to a monocular vision inertial odometer method based on feature point depth, which comprises the following steps: acquiring input images of two adjacent frames and extracting feature points in the images respectively; matching the extracted feature points; predicting first camera pose changes by using acceleration measurement data and angular rate measurement data output by an inertial measurement sensor; obtaining the depth of each pair of matched feature points based on the first camera pose changes; if the number of feature points meets a preset condition, constructing a weight for each pair of feature points and obtaining a weight matrix; constructing a weighted observation equation set based on an epipolar geometry constraint equation and the weight matrix; constructing a quadratic form based on the weighted observation equation set and solving the quadratic form to obtain an essential matrix of the matched feature points in the least square sense; obtaining a second rotation matrix and a second translation vector by singular value decomposition based on the essential matrix and the second rotation matrix and the second translation vector, and obtaining the representation of actual camera pose changes.

The disclosure describes a method for calculating an essential matrix of a capsule endoscope in a tissue cavity, including: collecting a first image of a reconstruction region in the tissue cavity along a first direction in the tissue cavity having a wrinkled inner wall, and causing the capsule endoscope to collect a second image along a second direction forming an included angle with the first direction in the tissue cavity; calculating an optical flow of a first pixel region of the reconstruction region in the first image and a second pixel region corresponding to the first pixel region of the reconstruction region in the second image by a neural network based on a convolution operation, wherein the neural network based on the convolution operation extracts a first feature point set including a plurality of feature points at the first pixel region, and obtains a second feature point set corresponding to the first feature point set at the second pixel region based on the first feature point set and the optical flow; and calculating the essential matrix of the capsule endoscope based on the first feature point set and the second feature point set. Thus, the essential matrix of the capsule endoscope in the tissue cavity can be obtained.

The application discloses an image feature matching method based on gradual neighbor consistency mining, and comprises the following steps: 1, all features are constructed into an initial image feature matching set S; 2, the initial image feature matching set S is pruned to obtain a reliable candidate image feature matching set and corresponding inlier weight; 3, the corresponding essential matrix is predicted according to the candidate image feature matching set and the corresponding inlier weight; 4, the inlier probability w of all input initial image feature matching is obtained according to the essential matrix; finally, the essential matrix of two matching images and the probability of correct matching inliers are estimated. The application has obtained significant performance improvement in image matching.

The invention relates to an image matching method based on corresponding relation pruning and structured context aggregation. The method comprises the following steps: modeling an image matching task into a mixed problem combining classification and regression; feature coding is carried out on the matching point set through a vision-space fusion network, vision-space embedded features F0 are extracted and then input into a corresponding relation pruning network CPFormer, the CPFormer network constructs a feature graph firstly, then the feature graph is processed through a semantic graph network module, and local geometric features and global context features are extracted and spliced; then feature aggregation and coding are carried out, a multi-head self-attention mechanism is used for modeling a dependency relationship between features, and further processing is carried out to obtain discriminative feature representation; estimating multi-stage interior point confidence based on features output by the CPFormer network, and further calculating an essential matrix; and calculating a symmetric polar line distance of each matching point based on the essential matrix, and when the symmetric polar line distance is smaller than a preset threshold value, determining that the matching point is an inner point, thereby completing image matching. The method is beneficial to improving the image matching precision.

The invention is suitable for the technical field of computer vision, and provides a GraphMama-based dual-view image matching method, which comprises the following steps of: constructing a model which comprises an InitialProjection unit, a Preminarial Local ContextEncoding unit, a Consensus-Aware Learning Block unit, a GraphMama unit and an Inlier Predictor unit, and constructing a model which comprises an InitialProjection unit, a Preminarial Location unit, a Consensus-Aware Learning Block unit, a GraphMama unit and an Inlier Predictor unit; the method comprises the following steps of: performing high-dimensional feature mapping, local neighborhood enhancement, global consistency optimization and long-range dependence modeling by taking initial matching point pairs of a double-view image extracted by SIFT (Scale Invariant Feature Transform) as input, and performing Consensus-Aware Learning Block and GraphMama iteration for four times; and training the model by using supervised data sets such as YFCC100M, SUN3D and the like, and finally outputting correct matching point pairs and an essential matrix. According to the GraphMama-based double-view-angle image matching method, through the limitation that traditional MLP independently processes matching pairs and RANSAC robustness is insufficient, long-range dependence of long-sequence matching pairs is accurately captured, on YFCC100M and SUN3D data sets, mAP indexes of relative attitude estimation and feature matching are remarkably superior to those in the prior art, complex scenes such as view angle changes and shielding can be dealt with, and the method has the advantages of being high in adaptability, high in robustness and the like. The method can adapt to computer vision tasks such as image stitching and vision positioning.

The invention provides a double-view feature matching method based on a paired motion field, and the method comprises the following design steps: 1, for a given image feature matching image pair, employing an SIFT algorithm to extract feature points; 2, designing a pairwise motion vector field construction module; 3, designing a feature fusion module; and 4, further predicting the probability of each matching pair serving as a real matching point (namely an inner point) by using the initial matching pair processed by the constructed paired motion vector field and the feature fusion module. And 5, taking the obtained probability set and the corresponding set as input, and estimating an essential matrix by using a weighted eight-point algorithm. And F, iteratively executing the step C to the step E for five times, calculating cross entropy loss according to a predicted classification result and a real category result in each iteration, calculating regression loss in combination with a predicted essential matrix and a real essential matrix to guide network training, and finally obtaining a double-view feature matching model with optimal performance.

This invention discloses a method for 3D reconstruction of asphalt pavement texture, belonging to the fields of road detection and computer vision technology. The method first calibrates an industrial camera to obtain intrinsic parameter matrices and distortion coefficients, and then continuously acquires pavement images at short intervals of 1 to 2 centimeters on a moving vehicle. Next, the images undergo preprocessing including cropping, Gaussian kernel convolution for noise reduction, and 8 to 12 low-pass filtering convolution enhancements. Feature points are matched using the SIFT algorithm combined with a limited neighborhood range strategy of 80 to 120 pixels. Subsequently, based on the matched points, the fundamental and essential matrices are solved using the RANSAC algorithm and the 8-point method. The camera pose is optimized using bundle adjustment to obtain an initial point cloud, followed by kdtree clustering and statistical filtering for dual noise reduction. Finally, bending deformation is corrected through quadratic polynomial surface fitting, tilt is corrected using projection transformation, and the model scale is adjusted according to the scaling factor to obtain a standardized 3D model. This method effectively solves the problem of weak asphalt texture matching and achieves efficient and high-precision pavement texture reconstruction.

The application discloses a kind of methods for indoor positioning of uncalibrated camera using Aruco code.The application uses a special two-dimensional code mark-Aruco code to carry out the self-calibration of camera, and the mapping and camera positioning of indoor environment.Through the corresponding relationship of a group of Aruco code corner points under different viewing angles, the distortion coefficient of the camera is solved using epipolar geometry constraint.The camera intrinsic parameters are calculated using the special properties of essential matrix after removing the distortion of the image.In an indoor scene with a group of Aruco codes pasted on a wall and a ceiling, the positional relationship between all codes is calculated using the pose relationship between a single code and the camera, thereby establishing an indoor map and completing the indoor positioning task of the camera.

The invention discloses a point pricking control method for a PTZ camera image set, a medium and electronic equipment, and the method comprises the steps: extracting and describing feature points from the image set, and carrying out the matching of the extracted feature points; based on a matching result, estimating motion between the images by using an essential matrix, and performing rotation averaging on all the images to obtain an image relative orientation result; performing bundle adjustment to optimize a relative orientation result; querying a relative orientation result, and obtaining an internal parameter matrix and an external parameter between the images; calculating a conversion relation of corresponding points between the images by using the internal reference matrix and the external reference between the images; and according to the conversion relation, calculating and marking the predicted position of the control point on the non-pricking point image. Through the computer vision technology and the SFM technology, automatic control point pricking of the PTZ camera image set is achieved, the working efficiency and precision are improved, and the method is suitable for various monitoring and measurement application scenes.

The invention discloses a camera self-calibration method, device and equipment, a medium and a program product, and relates to the technical field of camera self-calibration, and the method comprises the steps: constructing at least one adjacent image pair according to a continuous image sequence collected by a target camera; the adjacent image pair comprises two images; the collection moments of the two images in the adjacent image pair are adjacent; for each adjacent image pair, determining a target basis matrix of the adjacent image pair; according to the target basic matrix of each adjacent image pair, determining a target function based on essential matrix manifold constraint, and optimizing the target function to obtain an optimal internal reference of the target camera; and determining the pitch angle of the target camera according to the optimal internal reference and the continuous image sequence. According to the embodiment of the invention, joint calibration of the internal reference and the pitch angle of the camera is realized, and pitch angle matching in the camera is ensured, so that the accuracy of visual navigation by using the internal reference and the pitch angle of the camera is improved.

The invention discloses an SAR three-dimensional false target suppression method and system based on an essential matrix and neighborhood consistency constraint, and relates to the field of false target suppression. The objective of the invention is to solve the problem that an existing false target elimination method is poor in false target discrimination capability. The method comprises the following steps: acquiring two-dimensional coordinates of to-be-estimated pixel points on a main image and an auxiliary image and a two-dimensional coordinate set of neighborhood pixel points, and acquiring three-dimensional coordinates of the to-be-estimated pixel points and three-dimensional coordinates of neighborhood pixel points by utilizing the two-dimensional coordinates of the to-be-estimated pixel points on the main image and the auxiliary image and the two-dimensional coordinate set of the neighborhood pixel points; constructing a measurement function by using the three-dimensional coordinates of the to-be-estimated pixel and the three-dimensional coordinates of the neighborhood pixel points so as to obtain a measurement value; and determining and storing the pixels of the real target according to the measurement function value. The method is used for suppressing false targets.

The invention relates to a method for fusing visual odometry with data from an inertial measurement unit (IMU) of a self-propelled vehicle (1) comprising the following steps: - Capturing the environment of the vehicle (1) by means of at least one optical sensor (2) and generating sensor data, wherein the optical sensor (2) is a camera and the sensor data are successively recorded camera images; - Performing a semantic segmentation of the camera images; - Calculating a visual odometry based on the optical flux; - Observing the shape and position of a roadway (F) and / or roadway elements (4) and deriving an angle of inclination between an optical axis of the optical sensor (z) and the roadway (F); - Estimating a three-axis rotation between two successive camera images using an essential matrix calculated by single-value decomposition; - Fusion of a delta rotation between the two successive camera images using accelerometers of the inertial measurement unit, wherein the fusion is performed using a filter; - Adjusting rotation parameters of the essential matrix and transforming the detected objects (3) based on the fusion of the delta rotation.

This invention discloses a non-contact strain gauge identification and precise coordinate positioning method, comprising the following steps: acquiring strain gauge image data; converting the strain gauge image data to grayscale and acquiring filtered strain gauge contour image data; determining the target strain gauge contour data; determining the essential matrix of the image acquisition device based on the image rotation matrix and translation vector; identifying the same strain gauge in the target strain gauge contour data based on the essential matrix, and performing coordinate positioning of the same strain gauge based on binocular vision, transforming it to the coordinates under the image acquisition device to obtain the coordinate information of the strain gauge under the image acquisition device; establishing the transformation relationship between the coordinate information and the target coordinate system, and transforming the coordinate information to the target coordinate system to obtain the strain gauge identification result in the target coordinate system; repeating steps S1-S5 to obtain the strain gauge identification results for all structural tests. This invention can improve the accuracy of strain gauge position coordinates.

The application relates to a robust visual odometer method, which comprises the following steps: acquiring input images of two adjacent frames based on a monocular camera, and extracting feature points in the two input images respectively; acquiring a plurality of pairs of feature points, wherein the feature points are expressed in pixel plane coordinates; constructing an epipolar geometry constraint equation based on the pairs of feature points, and solving to obtain an initial essential matrix; using the initial essential matrix, and calculating an initial rotation matrix and an initial translation vector of the monocular camera between the two adjacent input images through singular value decomposition; converting the pixel plane coordinates of the pairs of feature points into three-dimensional world coordinates based on the initial rotation matrix and the initial translation vector; calculating the difference between the feature points in the pairs of feature points based on the three-dimensional world coordinates; judging whether the feature points are static feature points based on the difference, and if yes, the feature points are reserved; re-estimating the actual essential matrix of the monocular camera based on the static feature points, and obtaining an actual rotation matrix and an actual translation vector.

The invention relates to a method for calibrating a headlight (12) of a motor vehicle (10). The headlight (12) generates a light-based calibration pattern (28) which comprises at least one pattern feature (32) and which is projected onto an object (26) in the vicinity of the motor vehicle (10). For different pattern features (32), camera coordinates are determined to map the respective pattern feature (32) onto a camera projection plane (34) in a camera coordinate system (18), and headlight coordinates are determined to map the respective pattern feature (32) onto a headlight projection plane in a headlight coordinate system (20). A translation vector (36) between an origin of the camera coordinate system (18) and an origin of the headlight coordinate system (20) is determined.From several relations forming a system of equations, a rotation matrix is ​​determined which, together with the translation vector (36), describes a coordinate transformation between the camera coordinate system (18) and the headlight coordinate system (20) by establishing, for each of the different pattern features (32), one of the relations between the camera coordinates of one of the pattern features (32) and the headlight coordinates of the same pattern feature (32) via an essential matrix.

The invention discloses a double-view feature matching method based on space and channel feature enhancement, which comprises the following steps of: 1, giving an image pair consisting of two images, and firstly extracting key points and corresponding descriptors from the two images by utilizing scale invariant feature transformation; 2, calculating an initial disordered motion vector set from the initial matching set; and step 3, designing a filtering module by using a feature enhancement technology, gradually correcting wrong motion vectors, and effectively capturing context information at the same time. And 4, recovering the ordered vector into a disordered motion vector by using a graph attention network. And 5, predicting a correct / wrong matching classification result by using a multi-layer perceptron. And step 6, estimating an essential matrix by using a weighted eight-point algorithm according to a classification result. And iteratively executing for 6 times, jointly optimizing network training through the cross entropy loss of the classification result and the real category and the regression loss of the prediction and the real essential matrix, and finally constructing a high-performance feature matching model.

The invention provides an image feature matching method and device based on selective context awareness, and a medium, and relates to the technical field of feature matching. The method comprises the following steps: acquiring a group of to-be-matched image pairs, extracting image feature points and descriptors by using a feature extraction algorithm, and generating an initial corresponding point set based on descriptor similarity; inputting the initial corresponding point set into a pre-constructed pruning network for processing; the pruning network adopts a two-stage progressive pruning strategy, global context information and local context information are captured and subjected to interactive fusion refining, so that inner point weights of corresponding points are predicted, an initial corresponding point set is pruned according to the weights, and a reliable candidate point set is obtained; calculating an essential matrix through a weighted eight-point algorithm by using the reliable candidate point set; and based on the essential matrix, utilizing a full-size verification module to calculate epipolar distances of all the corresponding points, and classifying the corresponding points of which the epipolar distances are smaller than a preset threshold value as inner points to complete image feature matching.

The invention discloses a high-quality image feature matching method based on a knowledge-guided hypergraph network, and the method comprises the steps: 1, constructing an initial matching set: extracting feature descriptors of similar images, carrying out the similarity comparison, and carrying out the normalization, thereby obtaining the initial matching set; step 2, setting a structure of a knowledge-guided hypergraph network EGH-Net, namely performing two rounds of pruning through an energy function guiding module, namely an EFG module, so as to eliminate wrong matching, namely an exterior point, calculating an essential matrix and verifying an epipolar distance; step 3, training the EGH-Net until a preset number of times is reached; and step 4, inputting a to-be-matched image pair into the trained EGH-Net to obtain a corresponding relation of feature points. According to the method, the high-order consistency between the nodes and the dependency relationship between the sub-graphs can be effectively captured, the graph node features are dynamically optimized, and mismatching points are accurately eliminated.