30 results about "Stereo reconstruction" patented technology

The invention relates to an image three-dimensional reconstruction method, in particular to a multi-view three-dimensional reconstruction method based on frequency perception feature enhancement and cost aggregation. The objective of the invention is to overcome the defects of lack of frequency sensing capability and limited processing capability for problems of weak texture, noise, illumination variation, color distortion and the like in an existing learning-based multi-view three-dimensional reconstruction method. According to the method, multi-view three-dimensional reconstruction is realized through the steps of acquiring a multi-view image, calculating a multi-scale frequency sensing feature, calculating an initial cost body, embedding frequency information into the initial cost body, calculating depth estimation, performing back projection and the like in sequence; when multi-scale frequency sensing features are calculated, a double-branch frequency component enhancement module is arranged to process wavelet transform layer decomposition to obtain lossless approximate low-frequency components and high-frequency components of an input image, so that global consistency and local detail expression of depth estimation are improved.

The invention relates to the technical field of three-dimensional reconstruction, and provides a local feature adaptive fusion and high-fidelity new view angle synthesis method based on three-dimensional reconstruction. According to the method, a local feature module is introduced into a point cloud network, robust matching of local features is realized, a self-adaptive matching strategy is combined, high-resolution images can be efficiently processed, and the precision and stability of three-dimensional reconstruction are remarkably improved. The method is suitable for application scenes such as three-dimensional rendering, three-dimensional reconstruction and new view angle synthesis. The specific process comprises dense reconstruction and initial rendering; and carrying out missing region repairing and iterative optimization. Generating a high-precision point cloud based on an improved stereo matching algorithm, and rendering an initial image of a target view angle in combination with camera parameters; a conditional video generation model is adopted, and holes, distortion and artifacts in initial rendering are repaired; and carrying out progressive complementation on the occlusion region by using the view angle consistency constraint, and improving the reconstruction integrity by jointly optimizing the point cloud geometry and texture. According to the method, the final picture quality is remarkably improved through feature matching optimization and efficient processing capacity. Local feature branches and global context information are coordinated, and the problem of matching ambiguity in a complex scene is solved; the adaptive fusion strategy gives consideration to high-resolution image processing efficiency and detail retention; and the detail fidelity and the overall quality of the rendering result are effectively improved.

The invention discloses a photovoltaic module construction quality identification method based on computer vision and three-dimensional modeling. The method comprises the following steps: S1, collecting an original image sequence of a photovoltaic construction site and carrying out image preprocessing; s2, outputting a semantic tag graph and a two-dimensional space boundary graph through an improved BiSeNet network; s3, using a mask constraint SIFT algorithm to extract cross-view key point features, generating a dense three-dimensional point cloud according to multi-view stereo reconstruction, and outputting a structure three-dimensional geometric model; s4, mapping the semantic tag graph and the two-dimensional space boundary to the structure three-dimensional geometric model to obtain a semantic three-dimensional model; s5, performing spatial registration on the semantic three-dimensional model and the reference model, and calculating spatial error parameters of the photovoltaic module; s6, performing compliance discrimination on the spatial error parameters, and outputting a quality discrimination result; and S7, generating a construction quality evaluation report. According to the invention, automatic identification and accurate evaluation of the photovoltaic construction quality are realized, and the detection efficiency and the discrimination accuracy are improved.

The invention provides an unsupervised multi-view three-dimensional reconstruction method based on a bidirectional attention mechanism, and mainly solves the problem of low three-dimensional reconstruction quality in a weak texture region and a complex shielding scene in the prior art. The scheme comprises the following steps: 1) inputting a multi-view image and camera parameters; 2) preliminarily extracting multi-scale features of an input image through a bidirectional feature pyramid network Bi-FPN, taking the multi-scale features as input data, performing enhancement through an attention mechanism, and then calculating a fusion feature map; 3) performing multi-stage depth estimation from coarse to fine by using the fused feature map; 4) calculating an initial depth map of a data enhancement branch; 5) fusing multi-source information to optimize the initial depth map, and generating a final refined depth map; and 6) converting the refined depth map into a point cloud model to obtain a reconstruction result. According to the method, the multi-scale features of the image can be extracted and fused more efficiently, so that the three-dimensional reconstruction quality of a weak texture region and a complex shielding scene is remarkably improved.

The invention discloses a flexible metal material three-dimensional length measurement system based on a rotary camera mechanism, which is used for solving the problems that the existing contact measurement is easy to cause material deformation, the non-contact measurement is poor in adaptability to high-reflection / low-texture materials and the like. Comprising a rotary camera module (an industrial camera collects images at multiple angles along a semicircular guide rail), a visual calibration module (an ArUco code and a photoelectric sensor realize pose calibration and triggering), a texture enhancement module (structured light / random noise projection optimization feature extraction), a three-dimensional reconstruction module (a multi-view three-dimensional reconstruction algorithm generates point cloud) and an error compensation module (an error compensation module). And a material perception weighted least square fitting and neural network dynamic correction error) and a control splicing module. The bending deformation error of the material is eliminated through non-contact three-dimensional measurement; in combination with texture enhancement and error compensation mechanisms, the measurement precision of high-reflection materials such as aluminum foils and stainless steel bands is improved; and a modular integrated design is adopted, and automatic production line deployment is adapted.

The invention provides a feature pyramid network in multi-view three-dimensional reconstruction and a multi-view three-dimensional reconstruction method. Comprising an input image module, a camera parameter module, a shallow feature extraction module, a multi-scale feature extraction module, a global representation sub-network module, a position coding module, a feature fusion module, an FPN encoder module, an FPN decoder module, a matching cost body construction module, a cost body regularization module, a depth sampling range estimation module and a depth map generation module. A depth map optimization module, a point cloud generation module and a loss function module. The invention provides a feature pyramid network in multi-view three-dimensional reconstruction, which comprises the key technologies of multi-scale feature extraction, global context information extraction, feature fusion, feature pyramid construction, matching cost body construction, cost body regularization, depth sampling range estimation, depth map generation and optimization and the like. And high-efficiency and high-quality multi-view three-dimensional reconstruction is realized.

The invention discloses a multi-view three-dimensional reconstruction method based on multi-scale structure perception fusion and confidence weighted optimization, and the method specifically comprises the following steps: 1, obtaining a multi-view data set, and dividing the data set into a training set and a test set; 2, constructing a depth estimation network, wherein the depth estimation network sequentially comprises a multi-scale feature extraction module, a structure perception and attention guidance feature fusion module, a confidence coefficient weighted feature enhancement module and a multi-stage depth estimation and refinement module; step 3, training the depth estimation network by using the training set in the step 1 to obtain a trained depth estimation model; and step 4, inputting the test set constructed in the step 1 into the model trained in the step 3 to obtain a predicted depth map of each reference view. The method solves the problems that an existing reconstruction method is insufficient in detail reduction in a weak texture region and fuzzy in boundary in a shielding and illumination variation region.

The invention discloses a multi-view reconstruction method and system based on geometric perception and attention fusion. The reconstruction method comprises the following steps: designing a multi-scale feature enhancement network and a geometric perception feature fusion network; the multi-scale feature enhancement network extracts a multi-scale feature map and inputs the multi-scale feature map to a multi-stage depth estimation module to generate a preliminary depth map; inputting the initial depth map into a geometric perception feature fusion network to generate a geometric perception enhanced feature map; inputting the geometric perception enhanced feature map into a multi-stage depth estimation module, and performing iterative optimization to output a refined depth map; and training a model constructed by the network by adopting a joint loss function of a pixel-level cross entropy loss function and a depth distribution similarity loss function. According to the method, the expression ability of image features and the utilization efficiency of geometric information are effectively improved, the problem that depth estimation is difficult in a low-texture area and a structure repeated area in an existing multi-view three-dimensional reconstruction method is mainly solved, and the accuracy and integrity of three-dimensional reconstruction are remarkably improved.

Computer-implemented procedure comprising the following steps: Performing a stereo alignment on a pair of images; Straightening (204) the image pair so that epipolar lines become either horizontal or vertical; Applying (206) a stereo alignment to the straightened image pair to establish an initial correspondence between pixels from the image pair; Generating a homography matrix transformation; Defining photo consistency between a template window in a base image of the image pair and a warp window of a translated image of the image pair as an implicit function of the homography matrix transformation; Improve the photo consistency of the warp window; Generating a translated pixel from a basic pixel, wherein the generation comprises repeatedly applying (208) the homography matrix transformation to the basic pixel until the absolute value of an increment step is less than a predetermined value; Triangulating (210) correspondence points to create a three-dimensional scene; and Providing the three-dimensional scene for display.

The invention provides a multi-view three-dimensional reconstruction system for multi-stage depth estimation and propagation. Comprising an image acquisition module, a primary depth estimation module, a depth refinement module, a cross-view propagation module, a feature matching module, a noise filtering module, an abnormal value processing module, a visual angle selection module, a three-dimensional model generation module, a data preprocessing module, a visual feature enhancement module, a computing resource management module, a user interface module and a result evaluation module. The image acquisition module outputs image data to the data preprocessing module; the data preprocessing module receives the image data of the image acquisition module and outputs the preprocessed image data to the feature matching module and the primary depth estimation module. According to the multi-stage depth estimation and propagation mechanism provided by the invention, the precision and efficiency of multi-view three-dimensional reconstruction are remarkably improved, the calculation cost is reduced, and the method has a wide application prospect.

The application discloses a high-precision three-dimensional reconstruction method for a complex scene, which comprises the following steps: obtaining the position and orientation of a camera from multi-view images through polarization camera pose estimation; obtaining a scene surface normal vector through an MLP network; correcting an ambiguous polarization normal vector calculated by using a polarization image by using the scene surface normal vector, thereby obtaining a prior normal vector; inputting the prior normal vector into a pre-trained improved NeuS network to provide additional constraints, so as to alleviate the geometric blur problem of a textureless area and a mirror reflection area; and compared with depth prior, the network based on the prior normal vector can generate smooth geometry, can avoid the depth scale blur problem between multiple views, and can obtain more accurate three-dimensional information. The application can also perform high-precision reconstruction on an object with occlusion and an object with a thin structure, and can also perform better geometric reconstruction on a scene with sudden depth changes.

The application discloses an efficient multi-view stereo reconstruction method and system for an outdoor scene, and belongs to the field of multi-view three-dimensional reconstruction. The specific steps are as follows: preparing a multi-view stereo reconstruction data set for model training; constructing a multi-view stereo reconstruction network including an adaptive feature extraction converter module based on a large kernel attention, a cost regularization module and a depth estimation module; training and fine-tuning the network model to obtain a final model; and building a multi-view stereo reconstruction system and equipment. The application can solve the problems of poor reconstruction result accuracy and low completeness of existing multi-view stereo reconstruction methods when facing outdoor large-scale scenes, reduce the training and inference time and calculation cost of multi-view stereo reconstruction, and lay an important foundation for the application of outdoor large-scale scene image data in the field of multi-view three-dimensional reconstruction and the development of three-dimensional reconstruction technology.

The application discloses an artificial intelligence generated video authentication method based on multi-view consistency. The method comprises the following steps: constructing a large-scale real-world simulation video dataset, using the previous frame as a prompt to generate a future frame video with high consistency in semantics, color and physical laws, and improving data authenticity; designing a detection model based on multi-view consistency physical prior, extracting video features through real-time multi-view matching, combining a time sequence memory module to dynamically store and update inter-frame stereo reconstruction features, and using an attention mechanism to enhance long-time consistency analysis; using a multi-modal large model to generate unified prompts and label data, and improving the model generalization ability; scoring each frame feature through a scorer and globally averaging to output a video authenticity judgment result. The application uses multi-view consistency prior and memory enhancement mechanism to effectively capture the subtle defects in AI generated videos that violate the real physical laws, and significantly improves the detection robustness and accuracy of high-quality generated videos.

The present application relates to image stereo reconstruction method, specifically to multi-view stereo reconstruction method based on frequency perception feature enhancement and cost aggregation. In order to solve the problem that the existing learning-based multi-view stereo reconstruction method lacks frequency perception ability and has limited processing ability for weak texture, noise, illumination change and color distortion, the present application realizes multi-view stereo reconstruction through the steps of acquiring multi-view images, calculating multi-scale frequency perception features, calculating initial cost volume, embedding frequency information into the initial cost volume, calculating depth estimation and back projection in turn. When calculating multi-scale frequency perception features, a double-branch frequency component enhancement module is set to process the lossless approximate low-frequency component and high-frequency component of the input image decomposed by wavelet transform layer, so as to improve the global consistency and local detail expression of depth estimation.

This invention discloses an intelligent magnetic resonance holographic imaging method and system. The system includes a data acquisition device, an image reconstruction device, and an image display device. The data acquisition device acquires undersampled magnetic resonance data in real time through undersampled scanning and sends it to the image reconstruction device. The image reconstruction device inputs the undersampled magnetic resonance data into a deep learning model and outputs a high-dimensional stereoscopic reconstructed image. The deep learning model takes the undersampled magnetic resonance data of the sample as input and the corresponding fully sampled magnetic resonance data as output, and is obtained through self-supervised training. The image display device renders the stereoscopic reconstructed image in stereo and displays it in stereoscopic form through holographic projection. This invention enables rapid magnetic resonance imaging and real-time stereoscopic magnetic resonance image display, thereby better assisting doctors in surgical interventions and treatments.

The present invention relates to a multi-view stereo reconstruction method based on adaptive learning and aggregation, comprising the following steps: S1. Feature extraction, extracting image features from an input image and removing unimportant information from a large amount of information; S2. Matching cost construction, calculating the matching cost between each pixel in a reference camera and the corresponding matching pixel in its adjacent camera under each sampling depth assumption; S3. Matching cost regularization, denoising the cost volume in the matching cost; S4. Depth map estimation, weighting the result of the regularized matching cost using function regression to obtain an initial depth map; S5. Depth map optimization, reducing the effects of oversmoothing on the edges of the initial depth map. The present invention aims to provide a multi-view stereo reconstruction method based on adaptive learning and aggregation, which has higher reconstruction accuracy and completeness than MVSNet and significantly reduces time consumption and graphics card memory consumption.

The application discloses a multi-view reconstruction method and system based on geometric perception and attention fusion, and the reconstruction method designs a multi-scale feature enhancement network and a geometric perception feature fusion network; the multi-scale feature enhancement network extracts a multi-scale feature map and inputs the multi-scale feature map into a multi-stage depth estimation module to generate an initial depth map; the initial depth map is input into the geometric perception feature fusion network to generate a geometric perception enhanced feature map; the geometric perception enhanced feature map is input into the multi-stage depth estimation module to perform iterative optimization and output a refined depth map; and a joint loss function of a pixel-level cross-entropy loss function and a depth distribution similarity loss function is used to train a model constructed by the above network. The application effectively improves the expression capability of image features and the utilization efficiency of geometric information, focuses on solving the problem that the existing multi-view stereo reconstruction method has difficulty in depth estimation in a low-texture area and a structure repeated area, and significantly improves the accuracy and integrity of three-dimensional reconstruction.

The invention provides a three-dimensional root system measurement method and system, computer equipment and a storage medium, and belongs to the field of plant growth science, and the method comprises the following steps: planting crops in different growth periods in transparent containers of different specifications, collecting images by using a root system scanning platform, generating dense point clouds through structured light projection and multi-view three-dimensional reconstruction technologies, and carrying out three-dimensional root system measurement on the dense point clouds; a three-dimensional grid model is constructed through a Poisson reconstruction algorithm, a Medial Axis Transform algorithm is adopted to extract a main root and branch paths to construct a topological structure, a time sequence corresponding relation of model topological nodes in different periods is established by means of an ICP algorithm, then configuration parameters are extracted, a trend curve is fitted, a shielding region interpolation mechanism is introduced, and a shielding region is constructed. According to the method, the early-stage model and the physiological growth function are combined for structure completion, dynamic analysis is achieved, the VTK and Matplotlib technologies are used for achieving root growth visualization, morphological structure changes of crops in different growth periods can be visually analyzed, the root growth rule can be reserved, and accurate guide parameters are provided for research.

The application discloses a kind of based on attention mechanism correction Kuroshio front stereo center line reconstruction method, including the following steps: (1) based on Soble algorithm obtains three-dimensional Kuroshio front area;(2) extract Kuroshio initial stereo center line;(3) using the characteristics of flow field and sea surface height anomaly field to remove outliers;(4) using attention mechanism to adjust correction field weight.The application first uses Soble algorithm to obtain the initial stereo center line of Kuroshio front, then uses the characteristics of flow field and sea surface height anomaly field to correct the stereo center line, and introduces attention mechanism to adjust the correction field weight, realizes the stereo reconstruction of Kuroshio front center line, and effectively describes the vertical swing characteristics of Kuroshio axis, provides an effective tool for the description of Kuroshio front three-dimensional characteristics.

The application discloses a method for stereoscopic reconstruction and visualization of nasal polyp tissue based on artificial intelligence recognition of continuous pathological sections, which comprises the following steps: strict continuous section of a reconstruction object; HE staining of the continuous sections; affine registration of the continuous pathological sections; one-by-one analysis of the registered continuous pathological sections; stacking according to the registration sequence, then processing, to obtain three-dimensional model data of the nasal polyp tissue structure; and visual rendering of the three-dimensional model data of the nasal polyp tissue structure, combined with a preliminary stereoscopic model, to obtain a final stereoscopic model of the nasal polyp tissue structure. The method can reconstruct and visualize the nasal polyp tissue, and can show the three-dimensional structure of the distribution characteristics of inflammatory cells and tissue structure of the nasal polyp in the spatial structure, so that the morphological structure content of the nasal polyp tissue is enriched, and morphological basis is provided for further exploring the occurrence and development mechanism of chronic rhinosinusitis with nasal polyps.

The application provides a multi-view stereoscopic reconstruction method based on segmentation driving and edge alignment deformation, comprising: acquiring multiple images of different views under the same scene, taking each image as a reference image, and taking the rest of the images as source images; fusing estimated depth maps of all reference images to obtain three-dimensional point cloud data of the scene; estimating each depth map, comprising: segmenting regions of each object instance in the reference image, and constructing occlusion relationships between the regions of each object instance; iteratively optimizing an initial recovered depth map of the reference image multiple times to obtain a depth map, each iteration comprising: acquiring a recovered depth map of this time; for each pixel of the reference image, detecting multiple boundary pixels corresponding to the region to which the pixel belongs with the pixel as the center, sequentially connecting the multiple boundary pixels to obtain a deformation block, estimating a depth value of the pixel based on a constraint condition preset based on the occlusion relationship, the deformation block, the reference image, the recovered depth map and the source image; and obtaining a depth map of this time based on the estimated depth values of all pixels.

The application provides a multi-view stereo reconstruction method cooperating with depth edge and visibility prior, comprising: acquiring multiple images collected from different views of the same scene, taking each image as a reference image and the rest as source images and performing multiple iterations, estimating the depth map of each reference image, each iteration comprising: updating the depth estimation value of reliable pixels in this iteration; reconstructing the depth estimation value of unreliable pixels in this iteration, comprising: dividing the reference image into multiple regions; evaluating whether all pixels in the reference image are visible in the source image; for unreliable pixels, only selecting multiple anchor points from a specified pixel set in the same region as the unreliable pixels for patch deformation to obtain an anchor point set corresponding to the source image, the specified pixel set refers to a pixel set that is visible and reliable relative to the source image; according to the anchor point set corresponding to each source image, reconstructing the depth estimation value of the unreliable pixels; and obtaining the depth map of this iteration based on the depth estimation value of all pixels.

The application provides a three-dimensional space trajectory recognition method based on multi-view stereoscopic reconstruction, which comprises the following steps: using multiple cameras to shoot the same scene from different angles, determining the picture intersection part of the multiple cameras as the research scene; performing frame extraction and pedestrian detection on the video of each camera, recognizing the pedestrians and their position information in each frame of video, and assigning a uniform identity number to the same pedestrian; registering the video frame images of different cameras to obtain the pixel position of each pedestrian in each camera; solving the three-dimensional point coordinates of each pedestrian in the camera coordinate system by a multi-view triangulation method combined with the internal and external parameters of the camera; converting the three-dimensional point coordinates into the coordinates in the world coordinate system to obtain the world coordinate system coordinates of each numbered pedestrian point position in each frame; connecting the lines of each numbered pedestrian point position in each frame to form a trajectory according to the time information, and completing the reconstruction of the three-dimensional space trajectory of the pedestrian in the world coordinate system. The application can reduce the labor and calculation cost.

The invention provides an unsupervised multi-view three-dimensional reconstruction method fusing semantic information, and mainly solves the technical problem of low three-dimensional reconstruction quality of partial regions in a complex scene. Comprising the following steps: 1) inputting a multi-view image and camera parameters; 2) extracting multi-scale features of the input image through a feature pyramid network FPN; 3) performing multi-stage depth estimation from coarse to fine by using the multi-scale features; 4) calculating an initial depth map of a data enhancement branch; 5) calculating a semantic segmentation graph of the reference graph by using the pre-training model SAM; 6) fusing multi-source information to optimize the initial depth map, and generating a final refined depth map; and 7) converting the refined depth map into a point cloud model to obtain a reconstruction result. According to the method, the semantic information of the scene is used for assisting depth estimation, continuous and consistent depth estimation can be generated in the same semantic region, meanwhile, depth mutation is kept at the semantic boundary, and the reconstruction quality of key regions such as object boundaries is effectively improved.

The invention discloses a monocular depth prior-based stereo reconstruction and obstacle detection method, which comprises the following steps of: after a binocular image is calibrated and corrected, fusing semantic features extracted by a lightweight neural network and depth prior information extracted by a monocular depth estimation network; constructing a multi-scale parallax cost body based on a deformable offset prediction module, and generating a geometric feature body and an initial parallax map of corresponding scales through three-dimensional convolution regularization; according to the semantic feature and the geometric feature, fusing a disparity map and a geometric feature corresponding to the multi-scale disparity cost body to obtain an initial disparity map and a geometric feature; and inputting the initial disparity map, the fused geometric features, the monocular depth priori and the context features into a convolution gating cycle unit, guiding the direction and the amplitude of disparity update to obtain a high-precision disparity result, converting the high-precision disparity result into a three-dimensional point cloud, and detecting an obstacle in real time by using Euclidean clustering and voxel expansion. The method has high reconstruction precision and robustness, and is suitable for stereo reconstruction and obstacle detection tasks.

A measuring system for checking a measured-image state and / or a calibration quality for a stereo measured image, comprising: a first image capturing apparatus; a second image capturing apparatus spaced apart from the first image capturing apparatus; an output apparatus; and an evaluation apparatus designed to generate a stereo measured image from the at least first measured image and the at least second measured image by way of stereo reconstruction; wherein the evaluation apparatus is designed to check a measured-image state and / or to check the calibration quality of the stereo measured image and, should the at least one capturing defect overshoot or undershoot a predetermined threshold value, to transmit a first alert to the output device, and / or, should the calibration quality overshoot or undershoot a predetermined quality limit value, to transmit a second alert to the output device.

The invention discloses a double-reflection non-vision-field scene three-dimensional reconstruction method based on a neural illumination field, and provides a neural illumination field model, three-dimensional coordinates and illumination attributes are used as input, and spatial continuous volume density and projection light intensity are parameterized through multi-layer perception, so that high-resolution three-dimensional representation is realized; the proposed intensity rendering equation further integrates dynamic lighting and static scene attributes to synthesize a shadow image, thereby eliminating the necessity of making error-prone binary segmentation. A large number of experiments show that high-fidelity three-dimensional reconstruction is realized in simulated and real scenes, and the resolution limit of 1cm is reached in a hidden space of 144 cubic meters. Even under the challenging conditions such as complex ambient light interference and low shadow contrast (the pixel difference is smaller than 10), the hidden object can still be recovered by the method, and strong environmental adaptability is shown.

The present disclosure relates to a camera calibration method, comprising: acquiring a plurality of captured images, the captured images being images of a calibration board captured by the camera from different distances and / or different angles, the calibration board containing a calibration image, the calibration image containing a plurality of discrete or continuous calibration points; acquiring calibration points corresponding to each pixel of the camera in the plurality of captured images; calculating ray information of the pixel in the physical world according to the coordinates of the calibration points in the physical world, and acquiring a set of ray information corresponding to each pixel of the camera to complete the calibration of the camera. To at least partially solve the lack of universality and accuracy of traditional calibration methods.

The application discloses an appearance defect detection system and method based on rapid three-dimensional reconstruction, and belongs to the technical field of optical three-dimensional reconstruction defect detection. The method improves the traditional photometric stereo method by flexibly constructing a multi-direction light source illumination image acquisition and processing system, accurately positioning a target area, and calibrating and correcting light intensity, rapidly solves target surface three-dimensional information images by using prior information, and improves the contrast of the metal workpiece surface defect image; the normal vector graph and the depth graph of different channels are used to design an image enhancement algorithm based on multiple channels by using the direction sensitivity advantages of the normal vector graph and the depth graph, and finally, a multi-level YOLO detection model based on multiple views is proposed, the detection results of each view are fused and decided, so that high-precision detection of the metal workpiece surface is realized.