Deep neural networks for floorplan vectorization and feature tagging for first responders

An area of exploration that can lead to a critical improvement in the way first responders can better respond to situations (e.g., fires, shootings, etc.) in indoor scenarios is in the development of intelligent indoor mapping systems that provide critical navigation details to the first responders. This enables first responders to not only plan out their strategies in handling a particular indoor incident, but also provide them with real-time navigation details to accelerate these strategies. Two important components to building such intelligent indoor mapping systems is: 1) the digitalization of floorplans and 2) the identification and location of key features (fire extinguishers, hose attachment locations, stairs, doors, etc.) based on symbols in the floorplans. Doing these two components manually is intractable given the time-consuming and laborious nature of these steps. In this project, working with Mappedln, we aim to develop deep neural networks for automating the conversion of raster floorplans into digital vector formats, and automatically interpreting symbols in raster floorplans to identify what key feature they symbolize.

Faculty Supervisor:

Alexander Wong;Mohammad Javad Shafiee

Student:

Brennan Gebotys;Saad Rasheed Abbasi

Partner:

Mappedin

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

University of Waterloo