Research And Application Of 3D Reconstruction And Segmentation Of Aircraft Parts Based On Multi-Vision

Three-dimensional technology is the core component of the field of computer vision.Compared with two-dimensional vision,it has better recognizability and operability.Its application in the field of industrial production will effectively improve production efficiency and industrial intelligence.In this paper,three-dimensional reconstruction is applied to the spraying industry of civil aircraft parts,and the threedimensional model of airplane parts on assembly line is obtained by three-dimensional reconstruction.Because the spraying scene is a work for a whole tray parts,there are one or more parts on a tray.The spraying process uses the spraying robot for precise fixed-point operation,which requires that the three-dimensional model of each individual part should be acquired after each three-dimensional reconstruction.Based on this,this paper proposes the research idea of reconstruction and segmentation.Firstly,the scene is reconstructed in three-dimensional space.At the same time,the twodimensional image of the part scene is processed to obtain the two-dimensional segmentation plane.Finally,the three-dimensional scene model is segmented based on the two-dimensional segmentation plane to obtain the three-dimensional model of each part.The production efficiency is improved by one-time reconstruction and one-time collaborative segmentation.Based on this,the multi-part 3D scene reconstruction based on multi-view stereo vision and the model segmentation based on two-dimensional image to three-dimensional model are proposed and designed.Finally,the single-part 3D model acquisition is realized.Aiming at the problem of 3D scene reconstruction of parts,because of the large area of pallets,various kinds of parts and different sizes,and the way of reconstructing a whole pallet at a time,a multi-camera shooting environment is designed to collect images from multiple perspectives,obtain broader perspectives by multi-eyes,and then complete the construction of the whole three-dimensional model through acquisition of three-dimensional point clouds,geometric structure restoration and scene rendering.Jian.In the actual algorithm design,the three-dimensional scene space is restored by camera calibration,feature extraction,stereo matching,triangulation measurement,incremental reconstruction,dense reconstruction,meshing reconstruction and texture rendering,and each link is optimized to meet the actual use needs.Aiming at the problem of two-dimensional segmentation plane,based on the tray background with complex structure,the same color and parts in the scene,this paper proposes a two-dimensional image segmentation and multi-part contour extraction algorithm based on complex background.The camera is placed above the tray to collect two-dimensional images.The K-means algorithm is used to distinguish the almost identical background effectively based on RGB channel,and then automatically set the limit.Fixed area is based on Grab Cut iteration to remove background interference,get the segmented parts area,and evaluate and optimize the results.Then,the graphics algorithm is used to extract the outermost contour and two-dimensional data of each part according to the segmentation area.Aiming at the segmentation and acquisition of part three-dimensional model,a model segmentation algorithm based on two-dimensional image mapping threedimensional model is proposed based on the above three-dimensional scene model reconstruction and two-dimensional segmentation plane acquisition.Firstly,the threedimensional model is normalized,and then a PCA-based bounding box design method based on the actual scene is proposed.Then,the two-dimensional segmented image and the three-dimensional model coordinate system are used to complete the mapping from two-dimensional to three-dimensional based on the unified coordinate system and the specific bounding box,so that the contours of each part of the two-dimensional image are taken as the segmentation plane of the three-dimensional model,and the scene model is segmented.The spatial model and location of each part are obtained.