94 results about "Viewing cone" patented technology

The invention relates to the field of scene rendering, and particularly discloses a three-dimensional scene optimization method and system for collaborative rendering of dynamic LOD and view cone rejection based on space-time prediction, and the method comprises the steps: dynamically calculating the visibility frequency of an object, and dynamically adjusting the loading and rendering modes of objects with different priorities; predicting a view cone range of multiple frames in the future by using an LSTM network architecture; dividing the scene into uniform grid blocks, and then performing grading elimination; optimizing a heterogeneous computing pipeline; rendering the execution process; the system comprises a dynamic LOD and visual cone rejection collaborative optimization module, an LSTM visual cone prediction module, a block-level mixed rejection module, a heterogeneous calculation pipeline module, an optimization CPU-GPU task allocation and data transmission module and a rendering execution control module. According to the method, the technologies of collaborative optimization of dynamic LOD and view cone removal, block-level mixed removal, heterogeneous calculation assembly line optimization and the like are adopted, so that unnecessary rendering calculation and resource loading are reduced, and the rendering efficiency is improved.

The invention relates to an OpenGL (Open Graphics Library)-based large-scale three-dimensional model object quick pickup method. The method comprises the following steps: converting an interactive operation of a user into a three-dimensional space structure; a CPU constructs octree space acceleration structures based on a bounding box of an object in a large-scale three-dimensional model, and leaf nodes, intersecting with a pickup ray or a pickup view pyramid, of an axis alignment bounding box in each octree space acceleration structure are screened out to serve as candidate leaf nodes; collecting triangles contained in all candidate leaf nodes, and constructing a candidate triangle set; dynamically allocating thread groups and threads of the GPU based on the number of candidate triangles in the candidate triangle set; and performing highlight display on the candidate triangles intersected with the pickup ray or the pickup view cone, and feeding back IDs of the corresponding candidate triangles. According to the method, the data transmission quantity and the video memory access pressure are remarkably reduced, the pickup response time is greatly shortened in a large-scale three-dimensional model scene, and the requirements of real-time interaction and visual operation can be met.

The invention provides a camera external parameter calibration method and device based on image and point cloud matching, and the technical scheme of the invention is that a point cloud rendering view corresponding to an original point cloud is rotated, so that the point cloud rendering view and a camera image are superposed visually, and initial space association is established; and directly learning and solving accurate camera external parameters from the local relevance between the original point cloud intensity information and the image RGB information in an end-to-end manner by using a deep learning model. According to the scheme, the strict camera-radar orientation consistency requirement in a traditional method is not needed through rough matching, manual participation in the whole process of traditional calibration is replaced, transition from visual alignment to geometric alignment is achieved, and the calibration efficiency and scene adaptability are remarkably improved. Meanwhile, the scheme of the invention initiates a process of visual field cone cutting-perspective projection rasterization-intensity-RGB end-to-end matching, and compared with a traditional feature point method, the calculation complexity is greatly reduced, and the memory occupation is reduced by 80%.

The invention discloses a multi-precision dynamic loading and performance optimization rendering method for a hydroelectric generating set model. The method comprises three stages of multi-precision model construction, multi-precision model dynamic loading and rejection, and adaptive rendering optimization. And a multi-precision model construction stage: performing component extraction on the original model file to generate high, medium and low resolution multi-precision models, and converting the models into a GLTF format to improve the loading efficiency. And a dynamic loading and rejecting stage: reading the multi-precision GLTF file, generating a geometric model by applying the conversion matrix, constructing an LOD model according to the sight distance, and rejecting only the visible part in the loading visual field through a view cone to reduce the rendering calculation amount. And an adaptive rendering optimization stage: monitoring the frame rate in real time, dynamically adjusting the resolution when the frame rate is relatively low, simplifying material illumination, forbidding special effects such as shadow and the like, ensuring stable rendering performance, and finally outputting an optimized three-dimensional scene. According to the method, the rendering speed and fluency of the water turbine model are remarkably improved, and the method is suitable for the real-time visualization requirement of a complex three-dimensional scene.

The invention provides a game rendering method and system capable of eliminating volume cloud and mist in real time and electronic equipment, and belongs to the technical field of game development. The method comprises the steps that a cloud and mist area is set, the set cloud and mist area is divided into a plurality of equal-size grids, and each grid stores an elimination state and an elimination progress value; dynamically updating vertex concentration textures; calculating and storing cloud and mist concentration values of vertexes around each grid based on the grid elimination progress; view cone optimization rendering: determining a visible cloud range, combining adjacent grids to be rendered into a rectangular region in a two-dimensional plane, expanding the rectangular region into a cube, and projecting the cube to a screen space to calculate a bounding box; and shader sampling: for pixels in the screen bounding box, calculating cloud transparency and color through ray stepping sampling. According to the method, through grid dynamic management, rendering range optimization and shader efficient sampling, low-consumption and real-time elimination high-quality volume cloud and mist rendering can be realized at a mobile terminal, and performance and game interaction experience are considered.

The invention discloses a digital twinning-based three-dimensional visual management system and a digital twinning-based three-dimensional visual management method for an anode assembly workshop. The system comprises a triple binding index module, a unified time axis alignment module, a view cone inverse solution positioning module, a safety linkage arrangement module and a resume playback module, and is provided with a process template library and a network isolation / main / standby module. Three-dimensional positioning and PTZ linkage of an AI / threshold event are realized by establishing mapping of a three-dimensional object, a camera and a PLC point location; the batch / bracket / station events, the PLC time sequence and the video slices are indexed in a unified mode; shadow verification, interlocking check, instruction issuing, read-back and video consistency confirmation and trace leaving are executed according to the process, and a safe closed loop from warning to disposal is formed. The method comprises the steps of access and calibration, triple binding, time axis alignment, event positioning and stream taking, linkage processing and resume filing. The positioning and handling efficiency can be improved, the misoperation risk is reduced, and cross-production-line reuse and tracing evidence obtaining are supported.

The invention discloses a three-dimensional target detection method based on view cone point cloud clustering and 2D detection, and the method comprises the steps: generating a plurality of target bounding boxes with categories according to a 2D image, converting the target bounding boxes to an image coordinate system, and carrying out the screening of 3D point clouds, so as to obtain view cone point clouds; performing a point cloud clustering operation on the view cone point cloud, and calculating a result of the point cloud clustering operation to obtain a 3D frame meeting a preset requirement; and performing IOU calculation of different visual angles on the 3D frames to merge the 3D frames of the same object under different visual angles so as to obtain a 3D bounding box. According to the method, the problem of geometric information loss can be effectively avoided, and the accuracy of the finally obtained 3D bounding box is ensured; the problem that global search is needed due to the fact that 3D point cloud is directly adopted is avoided, the complexity of the calculation process is reduced, and the overall calculation efficiency is effectively improved; the preset frame requirement can be set according to the requirement of a use scene, so that different requirements are met, and the wide use range is ensured.

The invention discloses an ocean terrain digital display method based on three-dimensional modeling. The ocean terrain digital display method comprises the following steps of 1, constructing a multi-source ocean data set; 2, performing space fusion by adopting a self-adaptive interpolation method based on position weight to generate a three-dimensional space density field; 3, performing contour surface extraction by adopting an improved Marching Cubes algorithm to generate an initial three-dimensional terrain model; 4, performing local modeling by adopting a Delaunay triangulation network reconstruction method, and constructing a complete three-dimensional terrain model; 5, obtaining a texture terrain model; step 6, constructing an LOD tree structure and realizing dynamic loading in combination with view cone cutting and a screen space error control strategy; and step 7, based on a dynamic loading result of the LOD tree structure, driving node switching through a time controller, and performing visual display. According to the method, three-dimensional modeling and an LOD tree structure are fused, and high-precision reconstruction and evolution display of the ocean topography are achieved.

The invention discloses a mechanical arm fine positioning and closed-loop calibration method and system based on heterogeneous visual fusion, and the method comprises the steps: heterogeneous data synchronous collection and cross-modal joint calibration, semantic segmentation based on prompt learning and high-precision mask generation. Point cloud refining and edge noise elimination in a view cone space; and differential compensation and closed-loop feedback control based on a golden model. A heterogeneous visual fusion framework is constructed, and 3D point cloud noise is constrained by using a sub-pixel edge of a high-resolution 2D image, so that the positioning precision breaks through a sub-millimeter level; a segmentation large model with zero sample generalization ability and a prompt learning mechanism are introduced, so that the system can adapt to various workpieces under a complex background without repeated training; refining point cloud and a standard three-dimensional gold model are registered and fed back to the mechanical arm in a closed loop mode, so that the system can compensate and calibrate drift and accumulative errors in real time, and long-term stable operation is achieved.

The invention provides a 3D point cloud segmentation method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining 3D point cloud data and image data corresponding to the same sampling moment and in the same scene; converting the image data into a set of 2D masks, each 2D mask identifying a contour region of an object contained in the image data; the pixel coordinates of each 2D mask are converted into a 3D point cloud coordinate system, so that the corresponding view cone space of each 2D mask in the 3D point cloud coordinate system is determined; and for each 2D mask, segmenting 3D point cloud data in a view cone space corresponding to the mask from the 3D point cloud data to determine a point cloud block where the object is located. Therefore, 3D point cloud data does not need to be manually labeled, point cloud segmentation can be guided only by using the mask generated by the image, and the manual labeling cost is reduced; the method does not depend on specific sensor parameters, adapts to different hardware configurations, reduces the adaptation cost of different platforms, and enhances the generalization ability.

The invention discloses a scene data scheduling and rendering method based on dual PVS, and relates to the technical field of data processing. According to the method, the target scene unit where the camera viewpoint is located is determined, the rendering scene tree corresponding to the target scene unit is obtained, the nodes of the rendering scene tree are the target visible objects corresponding to the target scene unit, and then the view cone corresponding to the camera viewpoint is determined; and selecting a target node in the rendering scene tree as a target rendering object according to the view cone, determining a hierarchical detail model corresponding to the target rendering object according to the distance between the target rendering object and the camera viewpoint, loading the hierarchical detail model to the rendering pipeline from the rendering scene tree, and rendering the target rendering object. According to the method, the target visible objects are screened through the rendering scene tree, and then the target rendering object is determined in the target visible objects through the view cone, so that the objects needing to be rendered are further reduced, and the computing resource consumption of model rendering is reduced.

The invention relates to a landslide visual monitoring method and device, and belongs to the technical field of landslide monitoring, and the method comprises the steps: obtaining the disaster scene data of a landslide monitoring area, constructing a view cone of the landslide monitoring area, dividing the disaster scene data through a plurality of LOD layers of the view cone, and obtaining the landslide scene data of the landslide monitoring area. Initial disaster scene data of each LOD level are obtained, error calculation and judgment can be carried out on the initial disaster scene data of each LOD level, and therefore multi-level visual analysis of the disaster scene data can be achieved; and the initial disaster scene data of each target LOD level can be subjected to adaptive scheduling and visual display, so that the intelligent degree and the adaptive capability of the disaster scene are improved, and the visualization efficiency is improved.

The invention provides a water conservancy twinborn model rendering optimization method for dynamically generating gray level image frames based on a view cone. The method comprises the following steps: obtaining visible area model data after the current view cone is dynamically cut; outputting the visible area model data to a multi-layer rendering target comprising a color buffer area, a vertex buffer area and a depth buffer area through off-screen rendering so as to generate intermediate frame data isolated from screen display; performing gridding processing on the intermediate frame data, and displaying the equipment resolution according to the physical range size of the water conservancy project or the drainage basin scene; the average color value of each grid unit is calculated, desaturation processing is carried out to generate a basic grey-scale map, and adaptive feature enhancement processing is carried out to generate an optimized grey-scale map or a gradient map; and a GPU rendering pipeline is input and acts on hydrological feature analysis of the post-processing stage of the current frame and a preloading data source rendered by a twinborn model of the next frame at the same time, historical frame data are reused in a static region through motion detection, and real-time rendering is triggered in a dynamic region.

The invention discloses an ultrahigh-resolution immersive scene rendering method based on hexagonal picture segmentation. The method comprises the following steps: 1, determining a hexagonal segmentation style of an immersive display surface; 2, unfolding the immersive display surface into a plane; 3, acquiring 3D point information of each hexagonal surface, respectively calculating the width W and the height H of the enveloping quadrilateral surface, and calculating a local coordinate system V of the hexagonal surface S; 4, using W, H and V to construct a view cone and a viewpoint matrix of the three-dimensional scene; 5, setting a shade M of a plane expansion view for the hexagonal surface S; 6, calling a rendering scene of the 3D scene for the mask region by using information of the view cone and the viewpoint matrix, and rendering the mask region into a picture at the position S of the hexagonal surface; and 7, repeating the steps 5-6, and sequentially carrying out the same treatment on each hexagonal surface. According to the method and the device, the ultrahigh-resolution picture can be rendered in a scene segmentation form, and the method and the device are suitable for ultrahigh-resolution picture rendering under immersive display equipment.

The invention discloses a roadside multi-mode fusion sensing method. According to the method, an initial 3D view cone tensor with depth information is generated through a depth estimation network, features of pixel points are projected to a coordinate system in combination with internal and external parameter matrixes of a camera, then a three-dimensional space point coordinate set is generated according to projected feature position information, a Gaussian function is adopted to distribute weights of neighborhood grids, and a three-dimensional space point coordinate set is obtained. And finally, inputting the image features and the features into a fusion unit, respectively carrying out uncertainty modeling on the two types of features, automatically generating a fusion weight, and finally, carrying out normalization weighted aggregation on the features in the neighborhood so as to reduce the position approximation error and obtain the calibrated image features, and finally, inputting the image features and the features into the fusion unit, and respectively carrying out uncertainty modeling on the two types of features and automatically generating a fusion weight. And dynamic optimization is carried out according to a self-supervised loss function in the fusion process, so that the stability of multi-modal feature fusion is ensured, the detection accuracy is high, the performance is relatively good, and the problem of global feature dislocation caused by external parameter deviation and depth error can be avoided.

The invention discloses an obstacle detection method and device, a fusion processing method and device, electronic equipment and a storage medium, and the method comprises the steps: responding to 3D laser point cloud data of a laser radar, and restoring a corresponding visual cone of a target in a 3D world according to a 2D detection frame of a camera and camera parameters; a pre-trained target detection model is adopted to process the point cloud data of the target in a corresponding view cone in the 3D world, and target point cloud data of the laser radar in the view cone at the same moment are obtained; and according to the target point cloud data, processing the target point cloud data through the target detection model to obtain an obstacle detection result. The foreground and background point cloud distinguishing capability of the target detection model is improved.

The present application relates to a kind of WebGL-based pipe network digital twin model light weight rendering method, method includes: resource preconfiguration: the grid data of different three-dimensional model is generated corresponding instance template;Through instance template generation instance is stored in instance pool;In scene graph, create include call data and dynamic data lightweight placeholder, call data is used to extract instance, and dynamic data is used to adjust the dynamic display of instance;Lightweight rendering: set the visibility screening strategy of placeholder, whether current placeholder is visible is judged in turn through the screening condition of level, distance and perspective, view cone intersection;Set instance management strategy, for the placeholder screened as visible, corresponding instance is called from instance pool and is bound, and the binding relationship is recorded, for the placeholder screened as invisible, unbind, and the instance after unbinding is marked as idle;The placeholder of bound instance is rendered and output.Effectively reduce the memory occupation when rendering.

The present application relates to a kind of infrared scene quantitative radiance calculation methods based on light ray tracking module, comprising: obtaining target mask information image, the radiance image of infrared scene, depth information image and position information image, sensor position information, sensor field of view and sensor observation direction;Infrared scene is divided into several view cones according to the resolution of target radiance area image, wherein several view cones and the object surface in infrared scene intersect to form several face elements;The radiance of each face element at the sensor is calculated by the theorem of solid angle projection;The radiance of several face elements at the sensor is accumulated, and the radiance of infrared scene to the sensor is obtained.The method supports full scene, single target, multiple targets as radiation energy source, and can be applied to high complex infrared scene simulation work;Without additional data support, the radiance at the sensor can be calculated without performing redundant work.

The application discloses a multi-dimensional demonstration method and device based on time domain technology and a storage medium, and relates to the technical field of data interaction. The application first acquires a time domain file containing multiple perspectives and / or multiple time points, and encapsulates multi-dimensional index information containing a space index and a time index. Then, a demonstration narrative stream including a key frame sequence of the demonstration and a camera motion path of the key frame is constructed based on the multi-dimensional index information. In the process of playing and rendering, the current camera state is acquired in real time, the prediction information of the camera motion path is combined, and the target display area in the current view cone is calculated. Finally, the corresponding target pyramid tile data is called from the multi-dimensional index information, and progressive decoding and rendering are performed in the order from low to high resolution, so as to realize the non-susceptible loading of the demonstration scene from the overall overview to the local micro details. In this way, the loading efficiency of the multi-dimensional demonstration scene is improved, and the resource consumption is reduced.

The invention relates to a volume fog rendering method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring light source range information of at least two light source identifiers in a three-dimensional virtual scene, and calculating projection of the light source range information in a camera view cone space corresponding to the three-dimensional virtual scene to obtain light source view cone spaces respectively corresponding to the at least two light source identifiers; determining the total number of threads and light source identification information; running each thread according to the total number of the threads, determining a target light source identifier corresponding to the thread and a target voxel identifier corresponding to the thread for each thread, and performing volume fog illumination calculation based on the target light source identifier and the target voxel identifier to obtain volume fog illumination information corresponding to the thread; and fusing the volume fog illumination information corresponding to each thread, and performing volume fog rendering in the three-dimensional virtual scene to obtain target volume fog in the three-dimensional virtual scene. By adopting the method, the volume fog rendering efficiency can be improved.

The invention relates to the technical field of computer graphics and geological exploration visualization, and discloses a geologic body space penetration processing system based on a domestic digital twin engine technology, and the system comprises a heterogeneous data mapping module which is used for carrying out the normalization interpolation of discrete geological data, and generating a continuous three-dimensional body texture; the view cone depth weighted statistical analysis module is used for carrying out view cone sampling on the body texture and generating a weighted histogram in combination with a depth weight; the time domain damping adaptive threshold value control module is used for solving an instantaneous threshold value based on the histogram and generating a smooth final application threshold value in combination with a historical threshold value; and the multi-dimensional gradient mapping and penetration rendering module is used for calculating comprehensive visual potential energy in combination with the attribute value and the surface orientation factor, and executing fragment elimination or output based on a final threshold value. According to the method, a time domain damping self-adaptive threshold control mechanism is constructed, the time sequence stability of a rendered picture in the dynamic roaming process is ensured, and visual jump is eliminated.

The invention relates to a robot vision automatic scanning path planning method based on a three-dimensional grid model, and belongs to the related field of machining and manufacturing. According to the method, a scanning view cone model of a scanner is obtained according to a series of constraints of a surface structured light scanner, and through Gaussian mapping and clustering analysis, through segmentation and combination of triangular patches, a three-dimensional grid model is obtained; the method comprises the steps of obtaining a plurality of areas convenient to scan of a workpiece, generating candidate viewpoints according to a scanning view cone model, introducing viewpoint measurement quality parameters, selecting optimal viewpoints, and carrying out path planning on all the obtained optimal viewpoints according to a greedy algorithm, so that automatic viewpoint arrangement and automatic path generation are realized; the problems of repeated measurement, measurement omission, low efficiency and the like of manual teaching viewpoint workpiece measurement are solved, and the method has the advantages of high precision, high efficiency, high integrity, automation and the like.

The application relates to the technical field of target detection, in particular to a space passable area detection method and device, a detection equipment and a storage medium, wherein the method comprises the following steps: obtaining three-dimensional point cloud data and an environment image of a target area; performing plane projection and view cone projection on the three-dimensional point cloud data respectively to obtain multiple surface features and view cone projection features; fusing the multiple surface features and the view cone projection features to obtain a space feature map; fusing the space feature map and the environment image under the same time and space to obtain a fused feature map; inputting the fused feature map into a trained detection model; and outputting a space passable area of the target area. Thus, the related art is generally used for detecting and identifying grounded obstacles, pedestrians, vehicles and the like, and the resolution and generalization are not strong, the spatial position of the obstacle cannot be accurately determined, and the problems of avoiding collision and causing accidents and the like are solved.

The application discloses a fuselage skin viewpoint generation and detection method and device based on an unmanned vehicle, and the method comprises the following steps: acquiring a fuselage point cloud model of the airplane; performing cell decomposition on the contour of the fuselage point cloud model according to the fuselage point cloud model, a camera coverage range and a shooting overlap rate requirement; acquiring a normal vector and a center point of an effective cell; calculating initial pose information of a camera according to orthoview and equidistance shooting constraint conditions; generating an optimized target viewpoint according to minimum safety distance constraint conditions and lifting device height constraint conditions; and detecting the viewpoint coverage rate of the target viewpoint according to a view cone elimination strategy and an occlusion elimination strategy. The fuselage surface coverage problem can be well solved by using the unmanned vehicle system to automatically collect data, perform cell decomposition and generate and optimize the viewpoint, so that the efficiency and safety are improved, the coverage degree is ensured, and the data collection quality is ensured.

The invention discloses a high-fidelity scene reconstruction method based on laser radar-vision-inertial navigation fusion. The method comprises the following steps: carrying out online scanning and segmentation on a laser radar point cloud according to a camera sampling interval; acquiring an initial camera attitude by a laser radar-inertial navigation odometer, and jointly optimizing a re-projection error and a luminosity error through error state iteration Kalman filtering to obtain a refined camera attitude; an octree is adopted to remove map points outside a view cone, survival map points, an image and a refined camera attitude are combined for Gaussian micro rendering, meanwhile, a Gaussian ellipsoid cloning process is improved, the influence of motion blur is eliminated, overlapped fog shadow and back propagation loss are avoided, Gaussian ellipsoid parameters are iteratively optimized, and high-fidelity scene reconstruction is achieved. According to the scene reconstruction method provided by the invention, the problem of insufficient robustness of the traditional SLAM technology in the outdoor environment is improved, and meanwhile, the performance of the 3DGS in the outdoor environment is optimized.

The application discloses a three-dimensional target detection method based on RGBD data and a view cone, and comprises the following steps: acquiring a feature image of a scene RGB image according to the scene RGB image; inputting the feature image of the scene RGB image into a region generation network to acquire a bounding box of a to-be-detected object and a category vector of the to-be-detected object; acquiring a 3D view cone point cloud according to the bounding box of the to-be-detected object and a depth image of the scene; inputting the position coordinates of points in the 3D view cone point cloud and the category vector of the to-be-detected object into a PointNet network with an attention mechanism to acquire the center point coordinates of the to-be-detected object; and further acquiring the parameters of a 3D bounding box of the to-be-detected object to detect the three-dimensional to-be-detected object. The application simultaneously considers the plane information and the depth information of a scene, improves the detection precision, and makes the detection result more accurate while ensuring the efficiency.

The application relates to a data fusion method and system based on visual deep learning and laser point cloud, and a medium, the method comprising F1: during vehicle driving, target point cloud data is acquired based on a vehicle-mounted laser radar and is time-stamped, and target image data is acquired based on a vehicle-mounted camera and is time-stamped; F2: the target point cloud data and the target image data are time-synchronized, and synchronized target point cloud data and synchronized target image data are output; F3: all point cloud data in a view cone corresponding to an image detection frame is extracted and clustered, and point cloud clustering results in the view cone are output. The application not only solves the correlation failure problem of current perception system visual and laser post-fusion, but also solves the frequent sudden braking problem caused by the fact that, when an automatic driving vehicle drives on uneven road, a rut mark or a road protrusion not affecting driving is clustered as an obstacle, and the driving fluency of the automatic driving vehicle is improved.

The invention relates to the field of image processing, in particular to an industrial gas leakage detection method based on infrared thermal imaging, which comprises the following steps of: acquiring video streams by using a plurality of infrared thermal imagers, and constructing a gas basic geometric voxel grid through a background difference and view cone intersection method; thirdly, the sight line penetration depth and a statistical variable coefficient are calculated, and adaptive normalized radiation feature mapping based on statistical moments is constructed so as to solve errors caused by traditional homogeneous hypothesis; then, measuring and calculating a three-dimensional form compactness factor, generating a heterogeneous space density field weight in combination with the relative depth, and fitting a real diffusion gradient; and finally, calibrating a physical mapping coefficient by using single-point sensor data, and performing integral calculation on the real-time mass flow in combination with a three-dimensional optical flow method. According to the invention, the complex density gradient in the gas can be effectively reduced, and high-precision three-dimensional visual display and quantitative flow monitoring of gas leakage are realized.

The invention relates to the technical field of image data processing, and discloses a YOLO and SfM point cloud-based 3DGS power equipment component three-dimensional modeling method, which comprises the following steps of: generating three-dimensional view cones with semantic tags through a two-dimensional detection frame in a multi-view image, executing space intersection operation on multiple view cones pointing to the same component to determine that a target has a high probability body, according to the method, the original coupling of semantic information and geometric reconstruction is realized, through an endogenous anti-interference mechanism of multi-view-cone geometric constraint, the modeling precision is improved, the calculation complexity is remarkably reduced, and the calculation efficiency is improved. And the point cloud quality fluctuation can be adaptively processed, and the geometric reliability of the output model is ensured.

The invention provides a visual three-dimensional target detection method and device based on normal distribution weighted feature mapping, and relates to the field of three-dimensional target detection of deep learning, the method designs a depth-guided three-dimensional to two-dimensional back projection module, the module accurately estimates the depth value of each feature point in a two-dimensional feature map, and the depth value of each feature point in the two-dimensional feature map is calculated. And constructing a normal distribution weighting function to carry out weighted mapping on the features. Compared with a traditional forward projection method, the method has the advantages that tedious steps such as view cone construction and voxel space aggregation are avoided, dense aerial view features are rapidly filled through feature indexes, and the efficiency and the real-time performance of the reasoning process are improved; meanwhile, in the back projection process, depth guidance based on normal distribution weighting is provided for feature projection, the expression ability of aerial view features is enhanced, and the accuracy and robustness of three-dimensional target detection are remarkably improved.