ROI Detection Method Of High Temperature Slab Surface Based On 3D Morphology Reconstruction

As an important part of basic equipment manufacturing industry in China,metal slab has been widely used in national defense industry,transportation,petroleum and petrochemical industry and other fields.In the process of high temperature form of metal slab,the surface quality of metal slab is very complex,the background is messy and there is a lot of interference information due to the influence of manufacturing equipment,process,environment and other factors.The background of random distribution makes it more difficult to detect the surface of high temperature metal slabs.At the same time,the gray feature of target texture is similar to its background feature,which leads to the problem of weak contrast edge of target.Regions of Interest(ROI)detection is the core of surface quality detection technology.In this thesis,a solution combined grayscale and depth information is proposed to solve the problem of ROI detection on the complex surface of high temperature metal slab.At the same time,aiming at the problems of grating stripe sharpness,reconstruction efficiency and reconstruction accuracy in the 3D morphology reconstruction of high temperature metal slab,the stripe image imaging technology,multi-source stereo vision 3D reconstruction technology and stereo matching technology are studied and solved respectively.The main research contents are as follows:(1)According to the thermal radiation characteristics and imaging characteristics of the high temperature metal slab,it is impossible to obtain the clear grating stripe images by using the projector directly.In this thesis,the stripe image processing method combining physics and digital filtering technology is adopted.Then,the projection of blue grating fringes and the addition of physical filters are selected,and the digital filtering technology is introduced to filter the interference components which have great influence.Furthermore,the contrast between the high temperature metal slab and the structural light source is enhanced,the clear grating stripe image is obtained,and the 3D morphology reconstruction of the high temperature metal slab is realized.(2)In order to solve the problem of poor robustness to radiation difference of input image in stereo matching,the performance of various matching cost calculation methods is studied,the Census transform method is improved,and a stereo matching method based on Hg_Census transform is proposed.The proposed method takes advantage of the outward diffusion of hexagonal mesh and fully considers the information of the related neighborhood pixels of the central pixel to improve the robustness of the matching cost.(3)In order to improve reconstruction efficiency and reconstruction effect,a threedimensional reconstruction method of passive stereoscopic matching and active stripe is adopted.This method combines passive three-dimensional matching technology and active phase shift profile measurement technology.Furthermore,The method combines passive stereo matching technology with active phase shift profile measurement technology to reduce the number of projected stripe patterns,combines the initial parallax image with package phase for fine matching,and obtains high precision three-dimensional reconstruction effect.(4)From the point of view of image object detection theory,a ROI detection method based on gray and depth information is proposed to solve the problems faced by ROI object detection of metal slab complex surface(i.e.,cluttered background interference,pseudo-ROI region interference and weak contrast edge of ROI).According to the surface complexity and the number of ROI,different samples are selected and the Sunk-70 image dataset is constructed.The experimental results of 2D saliency detection method and fusion depth information method are compared and analyzed to verify the effectiveness of the proposed method.(5)The experimental system of 3D morphology reconstruction and ROI detection of high temperature metal slab is designed and constructed including hardware system and software system.The hardware system is mainly composed of acquisition system,stripe projection system and auxiliary system.The software system includes the main functions of camera calibration and object reconstruction,and implements the auxiliary functions of file operation,camera control,reconstruction result display and so on.In addition,GPU parallel computing technology is introduced to speed up the processing of 3D data.