Linear And Circular Arc Detection Algorithm For Engineering Drawing Vectorization

Because the linear and circular contour feature is important in engineering drawings,so how to extract the straight line and arc plays an important role in the intelligent recognition of graphic symbols.Hough transform has been used as an effective pattern recognition algorithm,and got extensive attention and application.But the Hough transform can easily affected by noise,digital image discretization error and so on.There are a lot of circular arc detection methods.The effective voting method algorithm is widely used now.The algorithm uses two pixels which have been selected to determine the third pixel to form a circular model,rather than select three pixels randomly,to avoid the blindness of randomly selected.However,the algorithm will traverse to all features pixels in the image,this will increase the arc detection time.Random sample consensus algorithm can handle noise amount accounted for the entire data set the number ratio of more than 50%,but the algorithm depends on the parameters very much.We can find that it will make the arc detection time longer,and the algorithm is very dependent on the parameters.As for low accuracy,poor robustness and slow detection speed and other aspects of Hough transform and effective voting method algorithm and random sample consensus algorithm,this paper puts forward the line vectorization method based on dynamic step length seed segments,and a fast algorithm for circular arc identification of random sample consensus based on the ring region.The main research contents are as follows:As for the existing line detection problem,this paper puts forward line vectorization method based on dynamic step length seed segments.The algorithm directly uses the original image information of the image to detect the lines.Firstly,the algorithm determines the direction of scanning the image,and then along the scanning direction to build the seed segment,and get the information of the seed segments' starting point,end point and the slope.Then tracking seed segment,and identifying the line.The algorithm directly uses the original image,without any pretreatments,so as to improve the speed of the algorithm,and the accuracy of line detection.Based on line detection using the above algorithm,this paper proposed a fast algorithm for circular arc identification of random sample consensus based on the ring region.By analyzing the characteristic of the algorithm from the graphics pixels spatial relations,this paper firstly selects three random pixels to correspondingly detect the minimum arc width of the three points in the horizontal,vertical direction respectively,and make the midpoints of the detected line width as hypothetical points for getting a hypothetical circle.Secondly,the mean of the three minimum widths are calculated as the line width,which is used to expand the circle into the ring domain for detecting circle and arc based on the estimated threshold.In addition,the two rectangular region are constructed inside and outside of circular ring to judge remote image pixels from ring domain for avoiding invalid calculation and reducing the calculation amount of arc detection.Finally,the experimental results illustrate that the proposed method can not only improve accuracy and robustness of arc detection but also shorten the arc detection time,compared with the effective voting method algorithm that widely used.The proposed straight line and the circular arc detection algorithm are verified by experiments.And through the experiments,it proves that the algorithms that proposed have advantages in the accuracy,the robustness and the speed of detections.What's more,for the complicated background,straight lines and arcs exist at the same time,the algorithms can also detect the lines and arcs accurately and it proves that the validity and practicability of the algorithm.