The Omdena team developed a computer vision model for identifying a building’s different types of elements/ symbols from 2D images on floor plans. The model aims to facilitate the creation of energy model reports for real-estate developers with the ultimate goal to reduce building emissions.

The project partner Sysconverge Inc. is a Canada-based consulting company that offers an efficient way to acquire, disseminate and analyze data to allow building designers, owners and managers to make better-informed decision.

The problem

Completing a city permit-compliant energy model report is a high cost and long duration; it requires at least three months of manual labor for a team of 4-5 people. The current process deters real estate developers from focusing on emission reduction, and the high cost of the model discourages smaller real estate developers from entering the market. This project aims to improve the efficiency of this manual effort by identifying different types of building elements (walls, railings, etc.) in the 2D image.

The project outcomes

Collecting additional datasets as well as using Sysconverge´s data, the team annotated images with 150+ objects to prepare for object detection and semantic segmentation. Next,  the team created a model to perform elements detection and dimension estimation accurately.

Project screenshot: Floor plan labeling.