Deep neural network-based system for detection and classification of construction elements in construction engineering drawings

Abstract

Method of recognizing elements and symbols in construction engineering drawings detailing heating, ventilation, and air conditioning (HVAC) and / or plumbing elements, comprising: receiving a construction blueprint 301; recognising ducts, pipe fittings and duct fittings using a first neural network 310; performing text detection using a second neural network 320 and text recognition using a third neural network 340; filtering the recognized text for known fitting type and size 345; performing pixel tracing to detect pipes using the bounding boxes of text and symbols as a starting point; generating a list of construction elements 370; and providing an interactive visualization of an annotated construction engineering schematic 380. The first neural network may be a convolutional network (CNN), the second neural network a Character-Region Awareness for Text detection (CRAFT) CNN, and the third neural network a convolutional recurrent neural network (CRNN). The first neural network may recognise element symbols, the symbols being associated with the filtered text 370 by matching the bounding box coordinates of the text with the bounding box coordinates of the symbol. The user may alter construction elements in the interactive visualization. A costing list of individual construction elements, via quantity takeoff, may be provided to the user 390.

Images