Research On Some Key Issues Of Sketch Recognition-Based Chemical Piping Tubulation Modeling

In resent years, sketch recognition-based technology has become a hot issue, and a large mount of applications based on sketch recognition have been developed gradually,especially the sketch recognition-based technologies of character recognition which is widely used in tablets and phones is already quite mature. But there are a few of researches on engineering drawings,the most of existing softwares for2D-drawing or3D-modeling are the type of traditional system. The irregularity and randomness of sketch restrict its application on engineering. A prototype system with function of3D-modeling,2D-drawing and material lists outputting is developed in this paper.The source of sketch data can be mouses or android devices, it can be also read from computer disk. In the stage of preprocessing, sampling points should be simplified, smoothed and normalized. The stage of recognizing is divided into three parts:piping recognizing, fitting recognizing and command recognizing.In the stage of piping recognizing, stroke segmentation algorithm realizes recognizing of single stroke piping with muti-fold points in the axonometric direction. The concept of auxiliary point contributes to solve recognizing of piping in the non-axonometric direction. The system will add default measurements according to the piping’s size and the scale which is set by user. Some fittings will be auto-recognized by the directions of nodes’pipings.The algorithm of fitting and command recognizing is random forest.67feature values consist of dft values,length-width ratios, distances from head to tail and angle function values. The system can sort results by its probability via improving random forest. The command recognizing can realize continuous sketching measurements based on time and space interval.Sketch recognition-based chemical piping and tubulation modeling prototype system proves the following results. With the67feature values, random forest has strong classification ability. By improving random forest, the algorithm can output a set of sorting results, which can be helpful for inputting efficiency. The improved algorithm and some key issues of sketch recognition also have reference value for other area.