313 results about "Point projection" patented technology

The invention relates to the technical field of computer vision, in particular to a point projection type three-dimensional reconstruction and segmentation method and system based on semi-Gaussian pruning. The method comprises the following steps: respectively obtaining an SFM point cloud and a consistency label mask of a cross-view label based on an obtained multi-view image; initializing the obtained SFM point cloud into an identity semi-Gaussian point cloud, and performing rendering optimization by using a differentiable renderer; densifying the initial sparse point cloud by using a localized semi-Gaussian point management method, and identifying a local error region for resetting and repairing; using the obtained consistency label mask to supervise Gaussian identity feature learning by using cross entropy loss, and using unsupervised 3D regularization loss to force spatially adjacent gauss to maintain identity consistency; according to the method, the identity coding semi-Gaussian kernel method is adopted, the inherent representation fuzziness of a single opacity formula is eliminated, and the positive influence on the identity coding precision is generated.

The invention belongs to the field of three-dimensional surveying and mapping, and particularly relates to a multi-precision three-dimensional surveying and mapping data fusion method based on dynamic modeling, which comprises the following steps of: assigning low-level semantic tags to LiDAR point cloud geometric features, and assigning high-level semantic tags to optical image texture features; defining a semantic tree structure, and establishing a cross-scale semantic association initial anchor point; constructing a cross-modal graph structure, projecting LiDAR point cloud nodes to optical image neighborhood superpixel nodes, and connecting and aggregating multi-scale semantic features; inputting geometric and image residual texture features by using a U-Net generator and outputting virtual textures; geometry and texture feature fusion and semantic and geometry collaborative optimization are realized through gating weighted feature fusion. According to the method, the problem of inconsistent multi-precision data semantic expression is systematically solved, semantic consistency is improved, texture deficiency is filled, and dynamic balance between geometric fidelity and semantic enrichment is realized.

The invention discloses a product model detection method and system based on welding spots. The method comprises the steps that a welding spot identification mark is obtained, a welding spot file matched with the welding spot identification mark is identified from a product model, and the welding spot file is a data file comprising a welding spot model; identifying a part model having an interference relationship with the welding spot model based on the position relationship of the welding spot file in the product model; taking an intersection surface obtained by intersecting the part models as a welding surface of the part models, projecting the welding spot model onto the welding surface to obtain a welding spot projection position, and detecting a shortest spot edge distance from the welding spot projection position to a welding surface boundary; based on the shortest point edge distance, the width of the welding face is determined through an auxiliary face intersection method; and detecting whether the product model is qualified or not based on the shortest point edge distance and / or the width of the welding surface. The technical problems that in the prior art, welding spot detection mainly depends on manual estimation, so that subjectivity is high, efficiency is low, and missing detection is prone to occurring are solved.

The invention relates to the technical field of multi-sensor fusion, in particular to an automatic calibration method and calibration device for laser radar-camera external parameters, and the method comprises the steps: synchronously collecting the point cloud data and image data of a road scene based on a vehicle-mounted laser radar and a camera; performing semantic segmentation on the image data to generate a semantic mask set of each image; based on the point cloud data, geometric attributes of a road scene are extracted from the vectorized map, a laser radar coordinate system of the vehicle-mounted laser radar and points in the vectorized map are projected into the image in combination with the semantic mask set, and a target consistency function is constructed; and adjusting the external parameter matrix based on the target consistency function until the target consistency function reaches a maximum value, and completing automatic calibration. Therefore, the problems that the calibration precision is insufficient and the high-precision perception requirement of automatic driving is difficult to meet due to the fact that related technologies depend on a large amount of labeled data or a specific environment, the generalization ability is poor, and a high-precision map and road prior information cannot be combined are solved.

The invention discloses a live pig body size and weight measuring method based on RGB-D. The method comprises the following steps: S1, obtaining a left vision rotation translation matrix and a right vision rotation translation matrix; s2, acquiring a three-view RGB image and a depth image of a single pig; s3, acquiring a body scale measurement key point; s4, acquiring a left viewpoint cloud, a right viewpoint cloud and an overlook point cloud, and performing point cloud registration to obtain a complete pig point cloud C1; projecting the body size measurement key points to the complete pig point cloud C1; s5, obtaining a final complete pig point cloud C2; s6, calculating the body length, the body width, the body height and the abdominal girth of the pig by using the final complete pig point cloud C2, and forming body size data; and S7, substituting the body size data of the pig into the multiple linear regression prediction model to predict the weight. According to the method for measuring the body size and the weight of the live pig, the accuracy and the stability of extracting the position of the body size measuring point are improved, and the accuracy of the body size and the weight measuring result of the pig is improved.

An apparatus for generation of a 3D representation of a scene includes a memory for storing a plurality of images depicting a scene; and processing circuitry in communication with the memory. The processing circuitry is configured to generate a depth map for the plurality of input images depicting a scene and unproject the depth map to a three-dimensional (3D) point cloud. The processing circuitry is also configured to project a plurality of 3D points from the 3D point cloud to a plurality of two dimensional (2D) planes and extract a plurality of triplane features from the plurality of 2D planes. The processing circuitry is further configured to generate a 3D representation of the scene based on combining the plurality of triplane features and a plurality of image features extracted from the plurality of input images.

The invention discloses a weld ROI segmentation method based on active and passive vision fusion. The method comprises the following steps: synchronously acquiring a two-dimensional image and a three-dimensional point cloud of the surface of a welded workpiece; inputting the trained deep learning network, and outputting a weld joint candidate region and a pixel-level mask graph thereof; establishing a cross-modal mapping relation through geometric calibration, and projecting the three-dimensional point cloud to a two-dimensional image to complete registration and fusion; cutting from the three-dimensional point cloud by using a cross-modal mapping relation to obtain a candidate point cloud region; generating a gray matrix for ray odd-even discrimination according to the pixel-level mask pattern; and each point of the candidate point cloud is projected to a pixel coordinate point by point, ray odd-even discrimination is carried out on the projection points according to the gray matrix, the points conforming to an odd-even rule are determined as welding line points and are aggregated into a welding line ROI point cloud, and the welding line ROI point cloud is output as a segmentation result. According to the method, the adaptive capacity to the unstructured welding seam is effectively improved, the point cloud processing complexity is reduced, and good real-time performance and robustness are achieved.

The invention discloses a pixel-level accurate texture generation method based on a multi-view artistic work. The method comprises the following steps: inputting an image file; performing multi-view projection on the 3d model for a plurality of views corresponding to the plane image to construct a texture space; aiming at coexisting points on the surface of the 3d model in the multiple visual angles, calculating a dot product of each visual angle direction and a normal of the surface of the 3d model so as to construct an initial texture map corresponding to the 3d model; projecting all contour points of the 3d model into a texture space to obtain corresponding texture information, and comparing the texture information with contour points contained in the initial texture map to construct a mask area; generating a shielding texture map corresponding to the mask area through a diffusion model; and carrying out rendering processing on the 3d model based on the initial texture map and the sheltered texture map so as to output a high-quality model with texture information. The invention further provides a pixel-level accurate texture generation device. According to the method provided by the invention, the high-quality texture can be generated for the grid, and meanwhile, the pixel-level consistency with the reference art is ensured.

An electronic device for performing calibration is provided. The electronic device includes a communication interface, memory storing one or more computer programs, and one or more processors communicatively coupled to the communication interface and the memory, wherein the one or more computer programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the electronic device to obtain, via the communication interface, a first image of a user captured by a first camera, and a second image of the user captured by a second camera, extract first joint feature points, which are two-dimensional (2D) position coordinate values of joints of the user, from the first image, and second joint feature points, which are 2D position coordinate values of the joints, from the second image, obtain three-dimensional (3D) joint feature points of the joints by lifting the extracted first joint feature points to 3D position coordinate values, obtain a projection relationship for projecting the 3D joint feature points onto the 2D position coordinate values of the second joint feature points, and perform camera calibration by predicting a positional relationship between the first camera and the second camera based on the obtained projection relationship.

The invention discloses a dual-path non-cooperative spacecraft pose estimation method based on uncertainty perception, and belongs to the technical field of spacecraft pose estimation. According to the method, a dual-path prediction framework based on a shared backbone network is constructed and comprises a geometric reasoning path and a global context sensing path. In the training stage, prediction results of the two paths are aligned through a geometric consistency loss function, and joint optimization is carried out by combining multi-task losses such as key point regression, classification, rotation and translation regression and the like. In the inference stage, the pose estimation result is adaptively fused according to the predicted variance output by each path, when the variance is low, the output of the global context sensing path is directly adopted, otherwise, the key point projections of the two paths are combined, and the final pose is recovered. The method has higher estimation precision, robustness and interpretability in complex space environments such as shielding, illumination variation and large-scale variation, and provides reliable visual navigation support for on-orbit autonomous tasks.

The invention discloses a building detail extraction method and device based on oblique photography and composite modal features, and the method comprises the steps: calculating a tilt parameter of a midpoint of a point cloud model based on the point cloud model of oblique photography data of a target building, and converting the tilt parameter into a color parameter; projecting points on the point cloud model to a target building view area of a two-dimensional plane to obtain a plurality of superpixel blocks, inputting features of the superpixel blocks into a variational auto-encoder, and performing optimization to obtain encoder optimization weight data; and inputting the features of the super-pixel blocks into an encoder to obtain corresponding optimized latent variables, clustering the super-pixel blocks based on the optimized latent variables, and mapping a clustering result to the three-dimensional model of the target building to obtain details of the target building. By adopting the technical scheme, the three-dimensional features and the color features are compounded, and the structure details and the distribution characteristics of the complex building are captured by utilizing the relevance between the features; an encoder of a pre-trained variational auto-encoder is used for aggregation, so that the accuracy of an aggregation result is improved.

The invention relates to the technical field of robots, and discloses an obstacle sensing method and a storage medium, and the method comprises the steps: obtaining image data of an environment in front of a mobile robot and three-dimensional point cloud data generated through laser radar scanning; pixel-level semantic segmentation is carried out on the image data, and a semantic segmentation map with each pixel having a semantic tag is generated; projecting each three-dimensional point in the three-dimensional point cloud data to a semantic segmentation map based on a calibration relationship between a camera and a laser radar, and generating a semantic point cloud with a semantic tag; for a detection blind area of the laser radar in front of the robot, generating a virtual three-dimensional point cloud in the detection blind area by using obstacle semantic information corresponding to an image area of the detection blind area in the semantic segmentation map; and determining the obstacle point cloud based on the three-dimensional point cloud of which the semantic label is the obstacle category in the semantic point cloud and the virtual three-dimensional point cloud. According to the invention, the obstacle detection precision in a complex environment can be improved.

The invention provides a power equipment three-dimensional modeling and route planning method based on laser point cloud and image fusion. The method comprises the steps that an original point cloud sequence and an original image sequence collected by an unmanned aerial vehicle and pose information of the unmanned aerial vehicle are acquired; preprocessing the original point cloud sequence to generate a clean point cloud sequence; preprocessing the original image sequence to generate an enhanced image sequence; performing rough matching on the clean point cloud sequence according to the pose information to generate a panoramic scene point cloud; projecting points of the panoramic scene point cloud to the enhanced image sequence; mapping the pixel color to the panoramic scene point cloud by applying an L-M algorithm to generate a color point cloud; inputting the enhanced image sequence into a pre-trained SegFormer semantic segmentation model to obtain semantic probability distribution of each image; mapping the color point cloud to a corresponding image according to the semantic probability distribution to generate a semantic point cloud, the semantic point cloud including color information and semantic information; and generating a three-dimensional scene model according to the semantic point cloud to output an inspection route.

The present invention discloses a multi-point projection device for improving the accuracy of laser ranging and its measurement method, belonging to the technical field of laser ranging; it first emits at least three laser beams in sequence through a laser emission unit, and the laser irradiates the surface of the object to be measured at different angles in sequence through an optical scanning unit, and after being reflected by the surface of the object to be measured, it is received by a laser receiving unit through an optical receiving unit; the data processing unit obtains the ranging value of each laser beam according to the received data, and calculates the spatial coordinate data of each measurement point by combining the emission angle of each beam; constructs a Hermite matrix from the obtained spatial coordinate data to obtain the minimum eigenvalue and eigenvector, and obtains the coefficients of the plane equation according to the minimum eigenvalue and eigenvector; substitutes the obtained coefficients of the plane equation into the projection distance formula from a point to a plane to obtain the actual distance between the laser ranging device and the object to be measured; maximally eliminates external errors and improves the ranging accuracy.

The invention discloses a lunar surface target three-dimensional point cloud detection and positioning method and system based on semantic information, relates to the technical field of three-dimensional object detection, and receives three-dimensional point cloud data and a two-dimensional image detection result, the two-dimensional image detection result is obtained by detecting acquired image data based on a pre-established two-dimensional target detection model, performing ground segmentation on the three-dimensional point cloud data to obtain non-ground point cloud data, mapping the non-ground point cloud data into a two-dimensional grid with a preset size to obtain a two-dimensional clustering result, and performing clustering on the non-ground point cloud data to obtain a clustering result; mapping the two-dimensional clustering result to the three-dimensional point cloud data to obtain a plurality of point cloud clusters; the points of each point cloud cluster are projected into an image coordinate system, dynamic weighted scores are calculated based on the number and confidence of the projection points of each point cloud cluster falling into the detection frame and the geometrical characteristics of the point cloud clusters, the point cloud cluster with the highest dynamic weighted score is selected for distance calculation, and the distance of the point cloud cluster with the highest dynamic weighted score is calculated; and obtaining a lunar surface target three-dimensional point cloud detection and positioning result.

The invention provides a method for obtaining a relative position relation between linear welding seams and related equipment, and the method comprises the steps: obtaining a projection distance from an end point of a current linear welding seam to a reference linear welding seam which is any welding seam in repeated linear welding seams corresponding to the current linear welding seam; the projection distance comprises a starting point projection distance and an ending point projection distance; and determining a relative position relationship between the current linear weld and the reference linear weld based on the projection distance. The projection distances from the starting point and the end point of the current linear welding seam to the reference linear welding seam are obtained, the relative position relation between the current linear welding seam and the reference linear welding seam is determined through the obtained projection distances, and the purpose of rapidly and accurately obtaining the relative position relation between the welding seams is achieved.

The invention relates to a multi-form central point fused cloud particle detection method, which mainly comprises the following steps of: firstly, performing morphological processing of different levels on a cloud particle image to generate a plurality of morphological processing results; then, for each piece of morphologically processed image data, extracting different types of center points, including a geometric center point, a minimum circumscribed rectangle center point, a maximum connected region center point, a contour centroid and a minimum circumscribed circle center point; then abnormal center point filtering and fusion are carried out on center points in each maximum connected area, and a final target center point position is obtained; and projecting the central points onto the original image data to generate a predetermined number of anchor frames, establishing a deep learning detection model to predict the anchor frames, and screening the anchor frames to obtain a detection result. According to the method, the cloud particle region detection positioning precision is improved, and a reliable basis is provided for cloud microphysical parameter inversion.

The invention relates to a weighted adaptive anchor frame cloud particle detection method, which mainly comprises the following steps of: firstly, carrying out different morphological processing on a cloud particle image, extracting multiple types of central points and carrying out first fusion; then, according to the expansion corrosion degree of morphological processing, different weights are given to each fused center point, the weight given to the center point with the larger expansion degree is lower, and the weight given to the center point with the larger corrosion degree is higher; projecting the weighted central points back to the original image, and performing secondary central point fusion and weight updating; and finally, adjusting the size of an anchor frame according to the weight of the central point, generating the anchor frame adaptive to the size of cloud particles, and establishing a deep learning model for detection. According to the method, through adaptive weight and anchor frame size adjustment, the precision of cloud particle target detection is remarkably improved, and the use efficiency of the anchor frame in the detection process is optimized.

The invention relates to a teaching-free vertical weld joint identification method, medium and equipment, and the method comprises the steps: collecting the surface point cloud of a to-be-welded workpiece, determining a bottom plane equation according to the surface point cloud of the to-be-welded workpiece to obtain upper surface feature points, carrying out the linear fitting of the upper surface feature points to obtain a plurality of linear equations, and obtaining the inner points of each linear equation; constructing an optimized feature point cloud, carrying out vector analysis on the optimized feature point cloud to obtain a plurality of welding seam end points, obtaining intersection points of points in each linear equation, determining a welding seam type according to the welding seam end points, projecting the intersection points of the points in each linear equation to a bottom plane equation to obtain a plurality of vertical welding seams, and obtaining a plurality of vertical welding seams according to the intersection points of the points in each linear equation and the welding seam end points. And determining the orientation of each vertical welding seam to complete welding seam identification. The problems that in the prior art, computing resource consumption is high, adaptability and generalization ability in practical application are insufficient, and recognition accuracy and recognition efficiency are low are solved.

The invention discloses an unmanned aerial vehicle sensing house safety appraisal management system and method based on BIM, and relates to the technical field of data processing. The method comprises the following steps: training a U-Net + + model, and based on a house three-dimensional BIM model, generating a preset track by combining a space grid sampling algorithm with a convex hull optimization algorithm; the unmanned aerial vehicle carries the dual-mode sensing pod to fly along a preset flight path, and image virtual columns and metadata are collected; de-noising enhancement processing is carried out on the image sequence, precision alignment is realized by utilizing BIM feature point projection to guide SIFT feature bidirectional matching, after Euclidean distance threshold verification, a pre-trained U-Net + + model is input to generate a crack binary mask, and the crack binary mask is converted into a physical crack parameter; the inclination angle of the wall is calculated through the SfM point cloud and ICP registration technology; integrating crack parameters, inclination angles and BIM attributes, and automatically triggering graded early warning through a risk rule base; the full-process automation from data acquisition, crack identification to risk assessment is realized, the detection precision and efficiency are remarkably improved, and the high-risk omission ratio is effectively reduced.

The invention discloses an automobile front anti-collision beam radian detection system, and relates to the technical field of anti-collision beam detection. The arc detection device comprises a detection table, a clamping assembly, a positioning assembly and a plurality of arc detection assemblies. The front anti-collision beam is reliably fixed to the clamping table through the clamping air cylinder and the rubber head, and automatic in-place and center correction are achieved through the conductive contact and the positioning conductive inclined block. When the position deviates, judgment is carried out through the resistance value difference, and early warning is given out. The arc measuring assembly is composed of image sensors, an imaging plate, a light shield and an adjusting fluted disc mechanism, the distance between the two sensors can be automatically adjusted according to detection requirements, multi-point projection imaging is conducted on a front anti-collision beam calibration point, and an overall radian curve is calculated. Meanwhile, the system is provided with a centrifugal impeller active heat dissipation mechanism, thermal noise of the image sensor is reduced, and detection stability is guaranteed.

The application relates to the chip detection technical field, in particular to a semiconductor packaging workpiece identification detection method and device and a cutting and sorting integrated machine, wherein the method comprises the following steps: acquiring a surface image of a workpiece to be detected; the surface image comprises an identification for marking a workpiece type of the workpiece to be detected; the workpiece type is a good material type or a bad material type; image skeleton extraction is performed on the surface image to obtain a skeleton line corresponding to a pattern in the surface image; for each pixel point on the skeleton line, if the minimum distance between the pixel point and a preset identification reference line is less than a first preset distance, the pixel point is projected onto the identification reference line; the identification reference line is drawn in the surface image according to the identification shape corresponding to the bad material type; and the identification corresponding to the workpiece type on the workpiece to be detected is determined according to the proportion of the area of the identification reference line projected by the skeleton line. The application can solve the problems of low identification detection precision and weak anti-interference capability of the identification on the workpiece to be detected.

The present invention discloses a visual odometry method and system based on image depth prediction and monocular geometry. The method includes inputting two consecutive image frames, detecting and describing local features of the images using the Scale-Invariant Feature Transform (SIFT) algorithm, and then using the Fast Library for Approximate Nearest Neighbors (FLANN) algorithm to match corresponding feature point pairs between the two frames; solving for the essential matrix using epipolar geometry constraints to obtain the relative pose transformation of the camera; constructing and training a monocular depth prediction model to predict dense depth information for each input image frame; if the number of valid depth information pairs formed by two consecutive image frames is greater than a given threshold, using triangulation to estimate the scale factor to obtain the corrected relative pose transformation, otherwise using a combination of the Perspective-n-Point (PnP) projection algorithm, the Random Sample Consensus (RANSAC) algorithm, and local non-linear optimization to solve for the absolute pose transformation. The present invention effectively integrates the advantages of deep learning and traditional geometric methods, can adapt to dynamic environments, and improves the robustness and accuracy of monocular visual odometry.

This application is applicable to the field of point cloud registration technology and provides an InSAR point cloud registration method and terminal device. The method uses MT-InSAR technology to estimate the height values of InSAR monitoring points, and combines the coordinates of the InSAR monitoring points in an external DEM to obtain InSAR point clouds under different orbits. InSAR point cloud data are clustered using the DBSCAN clustering algorithm with different clustering radii to obtain multiple point cloud classes. For each point cloud class, the facade points and roof points of all buildings are extracted, and each point is projected into a north-east coordinate system to obtain a two-dimensional projection of the InSAR point cloud. Corresponding points are obtained based on the corresponding point search method and two-dimensional projection corresponding to the point cloud class. Corresponding points are matched to obtain corresponding point pairs. Based on the ICP algorithm and corresponding point pairs, InSAR point cloud data are registered. This application can improve the universality and accuracy of InSAR point cloud registration.

The invention provides a method and a device for generating a process track. The method comprises the following steps: acquiring a group of discrete three-dimensional reference track points and original point cloud data for representing the three-dimensional shape of a workpiece; generating a group of three-dimensional intermediate target points with corresponding process forms through waveform interpolation based on the group of discrete three-dimensional reference track points; for each three-dimensional middle target point, according to the original point cloud data around the three-dimensional middle target point, fitting through a moving least square method to obtain a corresponding local three-dimensional curved surface; the three-dimensional middle target point is projected to the local three-dimensional curved surface, final three-dimensional track point coordinates and normal vector postures thereof on the local three-dimensional curved surface are obtained, the process track is generated based on the multiple final three-dimensional track point coordinates and the corresponding normal vector postures thereof, and it is ensured that the generated process track is attached to the real surface of the workpiece. And meanwhile, the method has process adaptability and high fidelity.

The application discloses a projection and touch fusion voltage control sensing type operation method and system, and relates to the field of projection interaction and touch detection fusion. The method comprises the following steps: determining effective touch pressure according to the construction of a voltage control array baseline and an excitation quantity and the generation of an adaptive threshold value based on static sampling; forming a pressure feature and generating touch point coordinates and a comprehensive pressure amount according to the position and pressure of the effective touch pressure; establishing a mapping from projection coordinates to physical coordinates according to a calibration pattern and generating a touch point projection position in interaction; dividing voltage control levels according to the comprehensive pressure amount and analyzing interaction instructions in combination with the projection position; and realizing adaptive correction by recursively updating mapping parameters according to running increment residuals. Through nonlinear touch pressure recognition, integrated analysis of touch point position and force, and adaptive correction of mapping parameters, stable, accurate and long-term consistent voltage control interaction in a projection and touch fusion scene is realized.

The invention relates to the field of image processing, and provides an image generation method, electronic equipment and a vehicle, and the method comprises the steps: obtaining a fisheye image corresponding to the current position of the vehicle, projecting each pixel point in the fisheye image to a preset spherical model, and determining a first projection point corresponding to each pixel point; projecting each first projection point to a preset cylinder model to obtain a second projection point corresponding to each first projection point; and obtaining target internal and external parameter data corresponding to the fisheye image, and performing texture mapping processing on all the second projection points according to the target internal and external parameter data to obtain a planetary visual angle top view corresponding to the current position of the vehicle. The fisheye image is sequentially projected twice through the spherical model and the cylinder model, alignment of ground textures is achieved due to the fact that the radius of the bottom face of the cylinder is large, points far away from the center can be compressed through vertex transformation of the spherical model, and therefore the problem that far ground points are seriously stretched is avoided, and the accuracy of ground texture alignment is improved. And the accuracy of the planetary view angle top view is improved.

The invention belongs to the field of multi-sensor fusion, and provides a lawn trimming robot environment sensing method and device based on vision and millimeter wave radar fusion. The method comprises the following steps: constructing a lawn data set, optimizing an activation function of a YOLOv11 instance segmentation model in a training stage, simultaneously adjusting and optimizing a model hyper-parameter and an optimizer to inhibit overfitting, and deploying the model to an AI chip end after conversion and quantification, so as to realize real-time visual reasoning of camera data; a DBSCAN algorithm is introduced at a millimeter wave radar end to cluster original point clouds so as to extract stable clustering center points, and the influence of stray points on the detection precision is significantly reduced; the system completes time synchronization of cross-sensor data by using a software timestamp in combination with linear interpolation, realizes spatial alignment through coordinate transformation, then projects a radar clustering center point to a visual detection plane, and performs position matching and decision-level information fusion through a geometrical relationship between the center point and a detection frame, so as to obtain a detection result. And the operation overhead is remarkably reduced while high-precision target positioning is ensured.

The invention discloses a three-dimensional imaging eye region processing method and device, equipment and a medium, and the method comprises the steps: obtaining a color image of a human face and an initial human face point cloud, carrying out the feature extraction of the color image of the human face, and obtaining eye socket key points; projecting the orbit key points to the initial face point cloud to obtain orbit point clouds and second face point clouds not including the orbit point clouds; constructing an eyeball three-dimensional model based on a least square method, and filtering point clouds of the eyeball three-dimensional model according to the orbit key points; and smoothing the filtered eyeball three-dimensional model, and adding the eyeball three-dimensional model to the second face point cloud to obtain a target face point cloud. According to the invention, the point cloud depression or discontinuity phenomenon of the eyeball area can be improved, and the accuracy and integrity of the depth camera in capturing and reconstructing the depth information of the eyeball area can be improved.