Three Dimensional Measurement Of Edge Contour For Workpiece Of Blur Image

With industrial automation technology rapidly developed,unattended vision detection system which has the merits of quick detection speed,Non-destructive Inspection,high measure precision and simple operation is widely used in more and more industrial areas.However,because the vision detection system may be negatively affected by mechanical vibration,human disturbance,Power System and other unfavorable factors,the captured images in measurement system may be the degraded images with low quality.Hence,the useful edge information of workpieces or scene in the degraded images is difficult to be detected accurately,which bring the nagetive influence for the three dimentional measurement about workpiece.In order to finish the task of efficiently detecting the 3D structure information of workpieces in degraded images,some key technologies of three dimentional measurement are researched in this paper including technology of image restoration,edge automatic detection and edge matching.Finally,the major content is as follows:Firstly,to improve the unattended and automatic ability of vision detection system,an image captured by the vision detection system is required be automatically judged whether this image is seriously blurred or not.In the paper,an approach which can rapidly judge the blurry degree of a captured image using to compute the max average width of horizontal and vertical directions edges and a threshold is proposed.When the blurry degree of a captured image is bigger than the prior threshold,this image is seriously blurred and need to use the procession of image restoration.Secondly,when the image captured by vision detection system is seriously blurred,the low-quality degraded image is needed to be automatically restored to a high-quality image.In order to fulfill this task efficiently,we research the restoration technology from the following two aspects to improve quality of the restored image.Because the restoration procession is a serious ill pose problem,the regularization technique is used to solve the problem.Hence,the prior knowledge of restored image detected from the degraded image is used to be constraint information for the objective function.With the error between the information of the degraded image and prior knowledge of the restored clear image becoming smaller,the quality of restored image becomes higher.Hence,the efficient prior information detected from the degraded image is used in this work.a)We use sum of the horizontal and vertical directions gradient magnitudes L1 norm in the degrade image as a prior information,because the gradient distribution of the original clear image and the degraded image both obey the similar heavy-tailed distribution.The global prior information is regarded as weak constraint in the restored proceesssion;b)to gain the local prior information of the restored image,a scoring metric function used to automatically indentify the useful local regions is proposed.Because the local smooth region of the degraded image is similar to that of the restored clear image,the local smooth region of the degraded image can be regarded as the efficient local priorinformation and needed to be detected;The local salient edge region of the degraded image is also similar to that of the original clear image,hence,The local salient edge region of the degraded image can be used to improve the estimation accuracy of blurry kernel and needed to be kept.An improved target function of image restoration using the above constraint information and the new noise model is proposed.Ultimately,the high-quality restored image is got by using the prior information of restored images obtained from the degraded images and an improved noise model.However,the target function has obvious non-convex feature,the alternate optimization technology is used to solve minimum problem of the non-convex target function.Finally,expriments prove that the proposed image restoration method has more excellent than other methods.Thirdly,an automatic edge detection method is needed to be used for high-quality restoration image obtained.The Canny method is more efficient than other edge detection methods,hence the Canny method is used in the paper.Howerver the Canny method has the drawback of lack of robustness.To solve this problem,the Canny algorithm is needed to be improved.Firstly,the improved Canny algorithm uses small-scale Gaussian function to smooth image to decrease some image noise and keep location accuracy of edge detection;Secondly,to gain auto-daptive double threshold of Canny operator,Otsu threshold algorithm is used to calculate high threshold adaptively,because the Otsu threshold algorithm is not sensitive to noise and the edges segmented by that method always are real edges.Moreover,the low threshold of Canny operator is adaptively obtained using a method based on ROC(receiver operation curve)theory,because the threshold method is always sensitive to image edges and considerably stable for getting low threshold in images including various geometry transformations.Hence,using the above high and low threshold methods can be used to detect image edge efficiently,rapidly and adaptively.Finally,experiment's results prove that the improved canny edge detection algorithm avoids set parameters artificially and has good anti-making performance,edge location accuracy and real-time,so the improved Canny method can meet the needs of industrial inspection.Fourth,in order to discard the meaningless image edges,the edges obtained are needed to be analyzed and segmented to several edge curves.Then some criterions which are got by using the prior knowledge about workpieces are proposed to discard fake and meaningless edges,which can increase the efficiency of edge matching.Fifth,to improve the efficiency of edge matching,an edge matching method constrained by matched feature points on the edge curves is proposed.We firstly finish the detecting and matching task of the feature points on the edge curves.In this paper,point-to-chord distance accumulation technique is used to detect the feature points of edge curves.Secondly,to matching those candidate feature points,an improved binary feature descriptor is proposed in this paper.In the first place,a new sample pattern is proposed based on anaylizing those of FREAK and BRISK methods.The new sample pattern is used to improve the speciality of feature descriptor.In the second place,to build the descriptor including more useful information,the new descriptor uses not only gray values but also the information of sorting gray value in the local region,which has less similarity.Finally,the descriptors are matched using Hamming Distance and the matching strategy.Finally,experiment proves that the proposed feature matching method has more excellent than other binary descriptor methods.Sixth,with the matched feature pairs are obtained,some the external constraints and some prior information about the edge of workpieces,the edge curves are matched.Ultimately,the edge curves matched is modeled and we can obtain three-dimensional information about workpieces.Finally,the example about a real standard workpiece with a pair of degraded images is given to validate effectiveness of the proposed algorithm.The detection result shows that the max measurement error of workpiece is 0.42 mm and the running time costs 15.44 seconds.Hence the proposed method is suitable to the vision inspection system that does not require high rel-time running,which can improve the automation level and unattended ability of the vision inspection system.