42 results about "Chamfer distance" patented technology

The invention provides a diffusion-driven channel adaptive point cloud semantic communication method, which belongs to the technical field of wireless communication, and comprises the following steps: obtaining a training point cloud data set, constructing a point cloud feature extraction network based on a hypergraph convolutional neural network as a semantic encoder, and constructing a channel adaptive enhancement module as a channel decoder, a channel adaptive recovery module is constructed at a receiving end as a channel decoder, a diffusion reconstruction network is constructed as a semantic decoder, end-to-end joint training is performed on the point cloud feature extraction network, the channel adaptive enhancement module, the channel adaptive recovery module and the diffusion reconstruction network, the diffusion reconstruction network predicts original point cloud distribution, and the point cloud feature extraction network and the channel adaptive enhancement module are subjected to end-to-end joint training. And a loss function is constructed based on the chamfering distance between the predicted point cloud distribution and the original point cloud. In the communication process, the reconstructed semantic features are input into the diffusion reconstruction network to reconstruct an original point cloud structure. The high-order semantic features of the point cloud can be effectively extracted so as to improve the adaptive capacity of the method to the incomplete point cloud.

The invention discloses a laser radar point cloud anomaly detection method and system, and the method comprises the steps: constructing a multi-task self-supervised learning framework comprising point cloud reconstruction, rotation prediction and contrast learning tasks, optimizing feature learning under unlabeled data through employing Chamfer Distance and InfoNCE loss functions, and reducing the dependence on manual labeling; a Transform Encoder is combined with a cross attention mechanism, a global relationship between objects in the point cloud is captured, and multi-frame spatio-temporal context information is fused, so that the recognition capability of structural anomalies is enhanced; a sliding window mechanism and an online incremental learning method are introduced, statistical distribution of normal features is dynamically updated, scene changes are adapted in real time, and the detection precision of unknown anomalies is improved; through a knowledge distillation technology and TensorRT-based GPU sparse convolution optimization, the model calculation complexity is significantly reduced, and lightweight deployment and real-time reasoning are realized.

The invention provides a multi-scale leaf point cloud completion method based on multi-attention mechanism collaboration, and belongs to the technical field of deep learning, and the method comprises the steps: segmenting independent leaf three-dimensional point cloud data according to obtained plant three-dimensional point cloud data, and carrying out the diversity and preprocessing of the leaf three-dimensional point cloud data, taking the preprocessed three-dimensional point cloud data of the blade as the input quantity of a point cloud complementation network, and then extracting the local structure and global semantic features of the incomplete blade point cloud through a multi-resolution encoder integrating a triple attention mechanism and a coordinate attention mechanism; afterwards, a pyramid decoder integrated with a multi-scale attention mechanism realizes progressive complementation of a blade point cloud incomplete part based on multiple resolutions, a discriminator receives point clouds with different resolutions, generation quality is optimized through adversarial training, chamfering distance loss is adopted as a multi-stage reconstruction loss constraint complementation whole process, and the quality of the blade point clouds is improved. The defect that an existing method can only guarantee similar geometrical shapes and cannot achieve detail completion is overcome.

The invention discloses construction and completion of a three-dimensional point cloud completion model based on multi-modal hierarchical features, and the method comprises the steps: carrying out the projection of an input point cloud, and obtaining a multi-view depth map; constructing a three-dimensional point cloud completion model; the input point cloud and the multi-view depth map serve as input, the complete point cloud serves as output, Chamfer Distance serves as a loss function, the constructed three-dimensional point cloud completion model is trained, and a trained three-dimensional point cloud completion model is obtained; multi-view depth map semantic features are extracted through a ResNet18 network and are effectively fused with point cloud geometric features extracted by a DGCNN through a multi-head cross attention mechanism in an encoding stage, and meanwhile, an encoder is introduced to capture local geometric details in a shallow layer and explicitly fuse semantic information in a deep layer and keep feature consistency, so that the accuracy of the multi-view depth map is improved. And the decoder reconstructs the missing region on the basis of the fused features, so that high-quality point cloud completion is realized, and the technical problems of insufficient geometric detail recovery and weak missing region completion capability in the existing point cloud completion method are solved.

The invention relates to a spine three-dimensional reconstruction network implementation method and system based on a biplane X-ray image, and aims to accurately and efficiently reconstruct a three-dimensional spine model with a smooth surface from a low-dose X-ray image by combining a deep convolutional neural network and differentiable parameterized surface representation. The method comprises the following steps: firstly, preprocessing an input biplane X-ray image, and fusing the biplane X-ray image into a multi-channel three-dimensional volume representation; then, inputting the three-dimensional volume into a neural network based on a 3D U-Net architecture, and training by adopting a space weighted loss function to output a three-dimensional voxel probability field with an accurate boundary; then, a group of BG-Triple parameterized surface primitives are initialized based on the probability field; and finally, through a geometric loss function containing Chamfer distance and normal consistency, iterative optimization is directly carried out on the geometric control point of the BG-Triangle, so that a vectorized smooth three-dimensional spine model is obtained. According to the invention, accurate mapping from the voxel probability field to the smooth geometric grid is realized.

The invention discloses a point cloud self-supervision completion method based on image feature guidance, which comprises the following steps: step 1, carrying out encoding and decoding operation on an acquired camera image to extract features, step 2, acquiring an original sparse point cloud Point, carrying out random point sampling on the original sparse point cloud to obtain N points, constructing a feature tensor for the N points, and recording the feature tensor as Points1; step 3, carrying out feature coding on the feature tensor Points1 constructed in the step 2 to obtain features; 4, performing feature expansion on the features obtained in the step 3 to obtain point cloud features; 5, performing point cloud reconstruction on the point cloud features obtained in the step 4 by using a full connection layer to obtain complemented point cloud features, and recording the complemented point cloud features as Points2; and step 6, during network training, carrying out farthest point sampling on the original sparse point cloud Points to obtain N points, constructing a feature tensor for the N points, recording the feature tensor as Points3, and calculating a chamfering distance between the Points2 and the Points3 as a loss function during network training.

The invention discloses a sheep body three-dimensional point cloud registration method and system based on a local overlapping area. According to the method, uniform down-sampling is carried out on filtered point cloud data, a proportional relation between a target detection body overlapping area covered by a vertical camera and a horizontal camera and the overall point cloud is estimated by using a camera field angle, and judgment of a chamfering distance is introduced into a CPD algorithm, so that a transformation matrix of a local overlapping area is obtained; and performing corresponding rotation matrix and translation vector change on the left and right complete point clouds by applying the local change matrix, and finally, merging the changed left view point cloud, right view point cloud and top view point cloud to obtain registered three-dimensional sheep body point clouds. According to the method, the problem that registration cannot be carried out due to the fact that position sensitivity and asymmetric geometrical shapes are initialized to be caught in a local optimal solution is effectively solved, and meanwhile, due to the fact that the chamfer distance is introduced, compared with a CPD algorithm before improvement, the registration speed of the method is increased by three times.

The invention discloses a DFG-PCN model-based point cloud completion method, and relates to the field of point cloud completion, and the method comprises the following steps: carrying out the hierarchical feature extraction of an input point cloud P through a feature extraction module, and generating a shape feature vector f, a lower sampling point cloud Pp, and a point feature vector Fp; fusing the shape feature vector f, the lower sampling point cloud Pp and the point feature vector Fp through a seed generation module to generate a low-resolution seed point cloud P0; inputting the seed point cloud P0 into at least one variable point diagram module to generate a step-by-step lifting high-resolution point cloud Pi; and applying Chamfer distance to the multi-stage generated point cloud to supervise loss, and gradually optimizing a point cloud completion result. According to the method, through adaptive connectivity distribution, double-graph structure collaboration and cross-scale attention fusion, the key region reconstruction precision is improved, and meanwhile, the consistency of multi-scale feature expression and overall geometry is enhanced.

The application discloses a sheep three-dimensional point cloud registration method and system based on a local overlapping area. The application estimates the proportion relationship between the overlapping area and the whole point cloud of the target detection body covered by the vertical camera and the horizontal camera by uniformly down-sampling the filtered point cloud data, introduces the judgment of the chamfer distance in the CPD algorithm, obtains the transformation matrix of the local overlapping area, applies the local transformation matrix to the corresponding rotation matrix and translation vector changes of the left and right complete point clouds, and finally, the changed left view, right view and top view point clouds are merged to obtain the registered three-dimensional sheep point cloud. The application effectively solves the problem that the initialization position is sensitive and the asymmetric geometric shape is trapped in a local optimal solution, thereby causing the registration failure. Moreover, the registration speed is improved by 3 times compared with the CPD algorithm before the improvement because the chamfer distance is introduced.

The present invention discloses a point cloud self-supervised completion method based on image feature guidance, step 1: acquiring a camera image Image Perform encoding and decoding operations to extract features. Step 2: Obtain the original sparse point cloud Points , random point sampling is performed on the original sparse point cloud to obtain N point, and for this N Points construct feature tensors, denoted as Points 1; Step 3: For the feature tensor constructed in step 2 Points 1. Perform feature encoding to obtain features; Step 4: Expand the features obtained in step 3 to obtain point cloud features; Step 5: Point cloud The feature uses the fully connected layer to reconstruct the point cloud and obtain the completed point cloud feature, which is recorded as Points 2; Step 6: When training the network, the original sparse point cloud needs to be Points Sampling the farthest point N point, and for this N Points construct feature tensors, denoted as Points 3. Calculation Points 2 and Points The chamfer distance between 3 is used as the loss function during network training.

The application discloses a point cloud model detection method and device, electronic equipment and a readable storage medium, and belongs to the technical field of point cloud models. The point cloud model detection method comprises the following steps: acquiring a point cloud model, and pre-processing the point cloud model. Fusion parameters of the pre-processed point cloud model are acquired. Chamfer distances of the pre-processed point cloud model are acquired. Local consistency parameters of the pre-processed point cloud model are acquired. A first weight of the fusion parameters, a second weight of the chamfer distances and a third weight of the local consistency parameters are acquired. A detection result is acquired based on the fusion parameters, the chamfer distances, the local consistency parameters, the first weight, the second weight and the third weight, so that the detection of the point cloud model is completed.

The invention discloses a Kerr black hole equatorial plane particle trajectory rapid prediction method and system. The method comprises the steps of S1, data construction and screening; s2, carrying out sequence standardization and pretreatment; s3, constructing a trajectory prediction model; s4, training a mixed physical constraint loss function; and S5, performing rapid reasoning and trajectory reconstruction. According to the method, complex Kerr space-time three-dimensional motion is simplified into two-dimensional motion of the equatorial plane, end-to-end mapping from an initial condition to a complete trajectory is realized by using deep learning, a time-consuming process of solving a geodesic equation is replaced by single forward propagation of a neural network, and the calculation efficiency is far higher than that of a traditional numerical integration method. According to the method, the robustness of the geometrical morphology of the track is ensured through the chamfering distance loss, and geometrical characteristics such as recent point precession angles can be accurately predicted; the common noise jitter of the neural network is eliminated by the smoothness loss, and the physical rationality of the trajectory is ensured.

The utility model relates to a near-field antenna with strong directivity. The near-field antenna comprises a circuit board and a coaxial cable, an oscillator unit and a feed network are formed on the front surface of the circuit board; the oscillator unit is a square with two opposite corners forming chamfers, the length L of each side of the oscillator unit is 80 + / -3mm, the chamfer angle of each chamfer of the oscillator unit is 45 degrees, and the chamfer distance range of each chamfer of the oscillator unit is 13 + / -1mm; the number of the oscillator units is 16, and the 16 oscillator units are distributed to form a rectangular oscillator array with four rows and four columns. In the rectangular oscillator array, the size range of the distance D1 between every two adjacent columns of oscillator units is 58 + / -2mm, and the size range of the distance D2 between every two adjacent rows of oscillator units is 25 + / -2mm; an inner conductor of the coaxial cable feeds each oscillator unit through the feed network, and an outer conductor of the coaxial cable is connected and conducted with the ground. The antenna has the characteristics of simple structure, scientific design, low production cost, small lobe width, strong beam directivity and the like.

According to the intelligent non-convex asteroid shape inversion method based on the optical variation curve and constrained by physical information, physical prior information is fused through a deep learning network (PINAS-Net), and rapid and accurate non-convex asteroid three-dimensional shape inversion is achieved. Specifically, light variation curve data and a scattering parameter c serve as input, features of all curves are extracted through a light variation curve feature extraction module (LCFEM), interaction features among multiple curves are captured through a Transform encoder, and 1024-dimensional global features are obtained; meanwhile, expanding a scalar scattering parameter c into 16-dimensional features, fusing the 16-dimensional features with global features, and importing the 16-dimensional features into the network as physical constraints; and rough point prediction is carried out based on the fused features, and a fine point cloud shape is obtained through a Transform decoder and three-layer up-sampling. The method provided by the invention is verified on observation data and simulation data of a plurality of asteroids (433 Eros, 9 Metis and 21 Luteria), and a reconstruction result is highly consistent with a reference model. Assessment indexes show that the intersection-to-union ratio (IoU) of the method for the non-convex region can reach 0.81, the Chamfer distance is 0.046, and compared with a traditional KTM method, the inversion speed and precision are remarkably improved. The method provides a new technical means for asteroid shape inversion and physical parameter estimation, and has an important planet defense value.

The invention discloses a tactile point cloud completion method based on a dexterous hand topological structure and multi-level feature coding, and belongs to the technical field of robot control. The objective of the invention is to improve the operation capability and perception precision of a robot in a non-visual scene. The method comprises the steps that a mechanical arm carries a dexterous hand to move to a target position, the dexterous hand is gradually closed until an electronic skin sensor records a contact signal, and tactile point cloud data of a target is collected; constructing a tactile point cloud completion model based on a dexterous hand topological structure and multi-level feature coding; a hierarchical loss function of a tactile point cloud completion model based on a dexterous hand topological structure and multi-level feature coding is designed, the hierarchical loss function is composed of coarse point cloud loss and dense point cloud loss, and a chamfering distance CD is adopted as a geometric similarity index for core measurement to measure the spatial deviation between a predicted point cloud and a real point cloud. According to the method, high-quality complementation of a complex object is realized, so that the operation capability and the sensing precision of the robot in a non-visual scene are improved.

The present application relates to the field of computer vision, in particular to a point cloud completion method and device based on shape prior and deep learning. The method comprises the following steps: firstly, feature extraction is performed on the input incomplete point cloud data; then, the multi-scale features are refined through the prefix embedding tree to obtain refined features; next, the multi-scale features and the refined features are connected through the semi-connected U-Net to obtain the preliminary completed seed point cloud; subsequently, the preliminary completed seed point cloud is up-sampled to generate the point cloud reaching the target resolution; finally, the generated point cloud is constrained by using a composite loss function, and the composite loss function comprises a chamfer distance, a double orthogonal constraint loss and an elastic potential energy loss, so as to obtain the complete point cloud with balanced density. Through the method and device, the technical effect of balanced density, geometric accuracy and complete detail recovery in the point cloud completion process is realized, and the problems of local density imbalance and geometric information loss are solved.

The invention discloses an infrared image three-dimensional reconstruction method and system based on a neural radiation field, and belongs to the technical field of computer vision and three-dimensional reconstruction. According to the method, the image quality is improved through dual-stage enhancement preprocessing, accurate estimation of the camera pose is realized by combining SIFT feature extraction and a beam adjustment method, a local-global joint infrared radiation field model is constructed, and a light adaptive SDF optimization strategy is introduced to balance the radiation field and symbol distance function convergence process. According to the method, an infrared imaging process is simulated through cumulative transmissivity modulation, and total loss function optimization model parameters including control loss and structure loss are designed. Experiments show that the infrared scene three-dimensional geometric reconstruction precision and the new view synthesis quality can be remarkably improved, the average chamfering distance is reduced by 87.31%, the peak signal-to-noise ratio is improved by 55.03%, and an efficient solution is provided for infrared three-dimensional reconstruction in complex scenes such as low visibility.

The present invention discloses a method for deep point cloud compression encoding based on a full self-attention network. The method comprises: constructing a point cloud full self-attention network, the point cloud full self-attention network comprising an encoder and a decoder; acquiring training data, constructing a chamfer distance objective function to train the point cloud full self-attention network; inputting point cloud data into the trained point cloud full self-attention network, performing feature sampling processing on the point cloud data using the encoder to obtain point cloud encoding, and completing point cloud compression; and reconstructing the point cloud data using the decoder based on the point cloud encoding to complete point cloud decompression. The present invention strengthens the learning of local and global correlations between each point in the point cloud through network training based on the chamfer distance objective function, and samples the features of the point cloud through an encoder to obtain a point cloud encoding that can accurately represent the semantic information of the point cloud, while ensuring the security and stability of the storage and transportation of the point cloud information. The method can be widely applied in the field of point cloud compression encoding technology.

The application discloses a bridge damage monitoring system and method based on image processing, which extracts the structural contour of a two-dimensional image through a deep convolutional neural network, and compares it with the rendered edge generated by perspective projection of a preset three-dimensional information model, and uses the chamfer distance to construct a closed-loop feedback mechanism to iteratively optimize the camera pose, so as to eliminate the pose projection drift caused by GPS rejection or weak texture in a complex environment, thereby accurately mapping the two-dimensional microscopic damage to a three-dimensional macroscopic coordinate system. In this way, the reliability of damage physical size quantification and global positioning is significantly improved, the problems of damage monitoring positioning difficulty and low mapping accuracy in complex inspection conditions are effectively solved, and efficient seamless integration of detection data and digital modeling is realized.

The invention discloses a mechanical dog quality distribution analysis method based on machine vision. The method comprises the following steps: obtaining complete mechanical dog three-dimensional digital model data; according to the material attributes and the geometric volume parameters, generating a mechanical dog overall mass mapping matrix; an optimized depth map is generated by using the improved ZoeDepth model; back-projecting the optimized depth map into dense robotic dog three-dimensional point cloud data; forming a mechanical dog coarse registration result; an IPC algorithm and a Chamfer Distance algorithm are adopted, and a final error measurement result is obtained; obtaining an updated mechanical dog three-dimensional digital model; forming a three-dimensional mass distribution data set; according to the method, a mechanical dog mass distribution thermodynamic diagram, a mechanical dog mass center scatter distribution diagram and a mechanical dog mass center vector diagram are generated, visual display and data storage are carried out, nondestructive testing and high-precision modeling analysis of the internal mass distribution of the mechanical dog are achieved, and the method has the advantages that an entity sensor is not needed, the modeling precision is high, and the visual effect is achieved.

The invention provides a vehicle classification processing method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a candidate vehicle data set and a sample vehicle data set, the candidate vehicle data set comprising first point cloud data of at least one candidate vehicle; according to each first point cloud data in the candidate vehicle data set and each second point cloud data in the sample vehicle data set, a total chamfering distance matrix between each sample vehicle and each candidate vehicle is determined, and the total chamfering distance matrix comprises chamfering distances of z rows and y columns; determining the K neighbor distance of each candidate vehicle according to the total chamfering distance matrix; and determining whether each candidate vehicle is a real vehicle or not according to the K neighbor distance of each candidate vehicle and a preset distance threshold. According to the method, the chamfering distance suitable for the point cloud similarity is combined with the K nearest neighbor distance algorithm to determine to classify the candidate vehicles, so that the vehicle classification precision can be improved, and meanwhile, the classification speed can also be improved.

The present invention discloses a method and system for completing a ship point cloud based on multimodal data fusion. The method comprises: mapping a two-dimensional ship image into a globally complete reconstructed point cloud through an encoding-conversion-decoding network architecture; performing coarse and fine registration of the original ship point cloud and the reconstructed point cloud to obtain a target registration point cloud; sampling the target registration point cloud using Poisson disks to obtain a uniformly distributed registration point cloud; dividing the uniformly distributed registration point cloud into a high-precision point cloud and a low-precision point cloud based on the chamfer distance; extracting features from the two-dimensional ship image, the high-precision point cloud, and the low-precision point cloud and performing feature fusion to obtain fused features; optimizing and completing the high-precision point cloud and the low-precision point cloud based on the fused features, and merging them to obtain a completed ship point cloud. This method utilizes multimodal data fusion and offset prediction and movement based on the fused features to achieve ship point cloud completion, significantly improving the integrity and accuracy of the ship point cloud data.

The application discloses a kind of based on similarity gradient guide three-dimensional molecular point cloud generation method and related device, belong to drug molecule design technical field;The method extracts the simplified point cloud representing the space profile of target molecule from its three-dimensional structure as sketching condition information;The sketching condition information is input into the pre-trained three-dimensional molecular point cloud diffusion model;Three-dimensional molecular point cloud diffusion model includes forward noise adding process and reverse denoising process;In the reverse denoising process of three-dimensional molecular point cloud diffusion model, introduce point cloud similarity calculation similarity gradient, to guide sampling process to generate the molecular point cloud matched with the sketching condition information on three-dimensional space configuration.The present application extracts simplified and key sketching information as condition, and calculates the condition gradient using differentiable point cloud similarity measure (such as soft chamfer distance and Gaussian volume overlap degree), realizes the accurate guidance to diffusion process, to generate the sketching structure with similar profile to target molecule.

This invention discloses a method and system for rapid prediction of particle trajectories on the equatorial plane of a Kerr black hole, including steps S1, data construction and screening; step S2, sequence standardization and preprocessing; step S3, trajectory prediction model construction; step S4, training of a hybrid physical constraint loss function; and step S5, rapid inference and trajectory reconstruction. This invention simplifies the complex three-dimensional motion of Kerr spacetime to two-dimensional motion on the equatorial plane, utilizing deep learning to achieve an end-to-end mapping from initial conditions to the complete trajectory. It replaces the time-consuming process of solving geodesic equations with a single forward propagation of the neural network, achieving computational efficiency far exceeding traditional numerical integration methods. This invention ensures the robustness of trajectory geometry through chamfer distance loss, accurately predicting geometric features such as the perihelion precession angle; smoothness loss eliminates common noise jitter in neural networks, ensuring the physical rationality of the trajectory.

The invention relates to the technical field of computer vision and three-dimensional reconstruction, particularly discloses a neural unsigned distance field (UDF) learning method and a surface reconstruction method based on cross product geometric prior, and aims to solve the technical problems that the gradient direction is unstable and the reconstructed surface is fragmented and discontinuous due to the fact that an existing neural UDF cannot be differentiated at a zero level set. The method is characterized in that a geometric prior constraint mechanism based on an original point cloud cross product is introduced; a local surface normal vector is estimated by calculating the cross product robustness of three adjacent points in an input point cloud, and a neural UDF gradient direction learned by constructing a loss function forced constraint is strictly aligned with a normal vector direction; in the optimal scheme, near surface optimization is focused in combination with an adaptive weight strategy, and a network is jointly driven to learn a more accurate and smoother UDF in combination with unsupervised chamfering distance loss and surface distance constraint. According to the method, the fuzziness of the UDF gradient near a zero level set is effectively overcome, the continuity and integrity of a reconstructed surface are remarkably improved, the method is particularly suitable for high-quality reconstruction of an open surface and an object with a complex internal structure, and the key indexes of the method are superior to those of an existing advanced technology through verification of multiple standard data sets.

This invention discloses a training-free multi-target 6D pose estimation method, system, medium, and device, relating to the field of computer technology. The method includes: acquiring an RGB texture map of the target to be tested; determining the texture map and corresponding segmented point cloud of the target to be tested; determining a scaled texture map of the target to be tested; acquiring texture maps from multiple offline reference frames; determining an initial similarity ranking between each texture map in the offline reference frames and the scaled texture map of the target to be tested; selecting point cloud data from multiple offline reference frames with the highest similarity to the target to be tested based on the initial similarity ranking, and determining the Chamfer distance between the point cloud data of the offline reference frames and the segmented point cloud to determine the most similar reference frame; extracting feature points and descriptors from the texture map of the most similar reference frame, matching them with feature points and descriptors in the texture map of the target to be tested to obtain feature matching point pairs; and combining RANSAC and SVD to determine the pose estimation result of the target to be tested.

The invention discloses a point cloud scene reconstruction method based on a vector neural network, and belongs to the technical field of computer vision and three-dimensional reconstruction. Aiming at the defects of the existing point cloud scene reconstruction method in processing rotation change and complex scene reconstruction, the method comprises the following steps of: performing instance segmentation on point cloud scene data by utilizing PointNet, and identifying and segmenting each object in a scene; establishing a CAD model library covering common indoor scene objects, training a VN network, and performing feature extraction on each point and a neighborhood thereof by using a vector convolution layer; obtaining feature representations of different scales by stacking a plurality of vector convolution layers and combining residual connection and jump connection; then calculating the similarity between the point cloud features and the CAD model features, performing retrieval by adopting a Chamfer distance, and selecting the most similar CAD model as a candidate; and finally, aligning CAD model candidates with scene point clouds one by one, and introducing a physical violation check module to ensure the accuracy and physical compliance of the reconstructed scene.

The invention discloses an improved 3DGS scene reconstruction and rendering method based on depth and curvature supervision, and the method comprises the steps: calculating a chamfering distance through combining the real depth of a sparse point cloud generated by SFM and a predicted depth generated by rapid microrasterization, and taking the chamfering distance as a loss function; the optimization enhances the geometric constraint on the position of Gaussian and ensures the correct position of Gaussian in the space, so that the wrong shielding relation and rendering artifacts are avoided, the sense of reality of the scene is improved, and compared with the prior art, the method has the advantages that through the combined supervision of depth and curvature, the real-time performance of the scene is improved, and the real-time performance of the scene is improved. The limitation of geometric constraint processing in the existing 3DGS technology is solved, wrong shielding and excessive small Gaussian in a traditional method are avoided, and the quality and efficiency of scene rendering are remarkably optimized.

The application discloses a kind of autonomous aerial refueling cone visual feature detection and pose measurement method, comprising: defining and calibrating the relative relationship between required coordinate systems, coordinate system includes world coordinate system, camera coordinate system, image coordinate system and pixel coordinate system;Collect the motion video of cone, label cone picture key point and construct cone dataset;Build model infrastructure, train to form the key point detection model of cone for the subpixel level key point pixel detection result of key point feature on non-cooperative target cone;Based on the subpixel level key point detection coordinate result fitting cone inner and outer circle ellipse equation;Based on the relative relationship between coordinate system and inner and outer circle ellipse equation of cone, the initial value of cone pose is solved by double space conical pose measurement method;Based on the double space circle pose measurement optimization method of chamfer distance loss, the initial value of cone pose is optimized by key point feature pose re-projection. Corresponding system, electronic equipment and computer readable storage medium are also disclosed.

This invention proposes a remote sensing rotating target detection algorithm based on dynamic orientation-aware attention. First, addressing the problem of insufficient multi-scale feature fusion, a cross-scale dense fusion feature pyramid network is proposed. Through full-scale dense connections and feature distillation mechanisms, the deep fusion of semantic and detailed information is enhanced. Second, addressing the problem of insufficient target orientation awareness, a dynamic orientation-aware attention module is proposed. This module dynamically generates rotating convolutional kernels guided by angle information, achieving alignment between the receptive field and the target orientation. Finally, to address the training instability caused by angle periodicity, a Gaussian weighted chamfered distance loss is constructed, transforming angle error into a geometric distance penalty. An angle-aware Gaussian weight is introduced to alleviate angle discontinuities. This invention effectively solves the problems of insufficient multi-scale feature fusion, insufficient orientation awareness, and training instability caused by angle periodicity in remote sensing rotating target detection.