Usage of Computer Vision algorithms to automatically extract information from Piping and Instrumentation Diagrams

Abstract

Piping and Instrumentation Diagrams (P&IDs) are widely-used schematics describing the core information of piping networks in plants or buildings. Extracting information from P&IDs is in demand but is an expensive and time-consuming task. Therefore, using Computer Vision algorithms to automatize information extraction is beneficial.

This thesis report proposes methods that recognize instrument and detect (equipment) symbols in P&IDs. The instrument recognition is implemented with a proposed algorithm. Hough Circle Transform is utilized to detect the instrument symbols because they are circular. An Optical Character Recognition engine named Tesseract is then applied to the detected instruments to recognize the texts inside them. For symbol detection, two Computer Vision models, YOLO and Faster R-CNN, are applied in this thesis project. The detection results are evaluated and compared by the mean Average Precision (mAP) of each model.

The circular instrument detection shows good results with an accuracy of 97.72%. But only 47.20% of the detected instruments are correctly recognized. The better model for symbol detection is Faster R-CNN whose mAP achieves 78.97%.