A Research Of Defect Visual Inspection System For Specular And Irregular Surfaces

Visual inspection is one of the key technologies in industrial automation,and its market application is develop vigorous.Compared to the human eye,machine vision is more precise,more efficient,more objective and so on.But it is deficient obviously that it did not have a fixed pattern applied to all operating conditions because of the limited technology.It is different from the ordinary survey that focused on flat plane or regular curved surface with diffused reflected feature,the object studied in this paper is specular and curved,which is the difficult in the filed of visual inspection.The highlight-curved object mainly include plastic film,ceramic,fine grinding metal and painted surface.The properties will cause local exposure and projection of surrounding environment which will affect the quality of image.It will make it more complicated that the irregular surface and highlight properties coupled with each other.The defect studied in this paper is scratch on painted surface,and finally,the validity of this research is verified by the experiment with a motorcycle tank.The main research contents as following.First of all,from the background suppression and defect highlight,we studied the various factors affected the image in-depth,mainly including the interaction between the light source,the object,and the camera.Then we designed the illumination system according these factors.In addition,the full surface image of the irregular surfaces is accomplished by using the robot with 6 degree freedom.Secondly,firstly analysis the characteristic of the scratches and painted background after the pretreatment of the image captured.Then the defect is segmented integrity and continuity under the algorithm proposed namely edge and region combination.The defect feature extract is to reduced the dimension of the data.The region,gray-level,and gradient feature of the defect are calculated.Finally,proposed a new algorithm based on the bad value elimination principle to exclude the false defect caused by entity edge,and realizes the qualitative judgment to the quantitative index judgment.Furthermore,the effective matching of the defects between different image is realized according to the features.Among them,the matching based on the defect area feature is simple and efficient that it is suitable for real-time occasions.Inaddition,the real-time performance of the algorithm is improved with combining the FLANN Library based on the gradient feature.Finally,the 3D position of the defects are calculated based on the multiview geometry.Finally,the feasibility of the algorithm,including the scheme of imaging system,the defect segmentation and feature extraction,the defect matching and triangulation positioning,is verified by experiments with a motorcycle tank.The method to calculate the camera extrinsic parameter in real time is proposed and realized,which is the unite of camera calibration and robot system calibration.