A Method for Detecting the Contour Lines of Two-Dimensional Grating Mechanical Drawing Parts

In computer-aided design systems, the most critical step when analyzing the geometric structure of a part from a two-dimensional raster drawing is to extract the contour lines of the part. Traditional line detection algorithms cannot effectively distinguish between the contour lines and auxiliary lines in the drawing, and contour lines contain more than just line segments. Circles and arcs are also commonly seen in the drawing. Therefore, this article proposes a two-stage algorithm for detecting part contour lines. Firstly, preprocess the drawing and lock the part view area in the drawing. Secondly, the LCNN model that integrates EMA attention mechanism is adopted to detect line segments. Compared with existing wireframe detection algorithms, this method improves the accuracy index of vectorized nodes on the dataset established in this paper. Subsequently, a process of classification detection, segmentation, and fitting is adopted to obtain vector information of circles and arcs, and finally coordinate transformation is calculated to obtain the position information of these three types of graphic elements in the entire drawing. This technology route is relatively novel and can effectively complete the automatic extraction of part contour lines.