Multi-camera Configuration And Image Mosaic Of Hot Strip Surface Quality Inspection

As an efficient and reliable non-destruction technology,surface quality inspection technology of rolled strips based on machine vision has been widely concerned by iron and steel enterprises.In order to obtain high quality strip image,the surface defect detection system usually needs to configure multiple cameras,and it needs to carry on the image mosaic to eliminate the redundant information and to obtain the complete defect image,which contributes to the following defect feature extraction and classification.Therefore,this thesis studies the following aspects of the vision sensor system and the hot strip image mosaic:Firstly,the distributed-processing overall hardware structure of the strip surface detection system based on machine vision is designed according to the technical parameters of the surface quality detection of rolled hot strips,and the problem of multi-camera configuration is solved.In detail,the mathematical model of the visual sensor system is established,and the parameters selection and installation position of the lights,cameras and lens are given,then the practical solution to the problem is put forward.Secondly,the thesis describes the better feature-based registration in the current image mosaic art in detail.By comparing the performance of the feature extraction algorithm and combining the characteristic of hot strips defect images,the ORB algorithm,known to be fast and accurate,is introduced and improved to apply to the fast mosaic of defect images.Strip defect image registration algorithm based on improved ORB feature matching limits the range to detects keypoints and computes features of the keypoints,then searches and removes several-to-one keypoints to improve BF algorithm,finally uses RANSAC to calculate the affine transformation model,and realizes the fast registration and mosaic of strip defect image.Experiments show that the improved ORB registration algorithm is more accurate than the original algorithm,which can effectively eliminate the error matching points.And the results of the improved ORB algorithm are the same as the classic SURF_FLANN algorithm,but its speed is nearly 10 times faster than SURF_FLANN.Thirdly,For the feature-poor of strip images,we make the best of sensor parameters,including the camera's internal parameters and multi-cameras external parameters,and get the transform model between the images in the multi-cameras based on the imaging principle of cameras and its spatial geometrical relations.Then we achieve an efficient image mosaic that is not based on image information.Finally,Because of the gap which may appear in the direct mosaic results,we use the image fusion technology to reconstruct image.And we put forward a kind of fusion method which is based on the improved weighted average algorithm,the distance weighted average algorithm,and the linear interpolation method that takes into account effect and speed,removes traces of the direct stitching.