An Approach of Line Vectorization of Engineering Drawings Based on Semantic Segmentation

In order to meet the needs of design or survey, the existing engineering drawings often need to be redrawn and reedited according to the paper documents. In order to reduce the workload of redrawing and realize the vectorization of line segment combinations such as walls in engineering drawings, we proposed a line segment vectorization method based on semantic segmentation. Aiming at the bitmap scanned from paper engineering drawings, this method first uses the feature extracted from a non-local semantic segmentation algorithm based on dilated convolution and self-attention to segment the lines and endpoints in the image. The endpoints extracted from the segmentation mask are combined to rasterize the point group. Finally, according to the rasterization result between the two points and the line segmentation result, the IOU is compared to determine whether there is a line segment connection between the two points, and finally the line segment such as the wall in the drawing is vectorized. The results show that the algorithm we proposed can effectively vectorize the line segment objects in drawings, and is robust to common problems such as line breakage and blurring in drawings. The line segment geometry and topology information extracted based on the algorithm in this paper can lay the foundation for the sub-sequent drawing vectorization, and the exploratory research has certain inspiration for the sub-sequent engineering drawing vectorization research.