Research works of 2D CAD conversion of engineering drawings are presented. We developed a sparse pixel vectorization algorithm for crude vectorization, which is fast and accurate for preprocessing of graphics recognition. We also proposed a generic, flexible and adaptive graphics recognition framework and applied it to the detection of lines of various shapes (straight, circular, and polyline) and various styles (solid, dashed, dash-dotted, dash-dot-dotted), text, arrowheads, leaders (straight and arc), dimension sets, and hatched areas with groups of hatching lines. Evaluation of these algorithms proved their high performances. The performance evaluation protocol we developed has also been used in many pattern recognition competitions. The vehicle for these researches has been the Machine Drawing Understanding System (MDUS) we developed for recognition of line drawings in general and mechanical engineering drawings in particular, which won first place in the first GREC workshop in 1995 and was evaluated as the best system for mechanical engineering drawings recognition in the second GERC workshop in 1997. It also won a Third Prize in the first ACM Quest for Java contest in 1997.