Emergency Floor Plan Digitization Using Machine Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Building Safety Engineering

Background:

• Special-use and high-rise buildings pose evacuation challenges during emergencies.
• Limited research exists on using autonomous vehicles for indoor emergency scenarios.
• Efficient emergency response requires accurate and accessible building information.

Purpose of the Study:

• To develop a method for classifying emergency symbols and locating them on floor plans.
• To extract geometric and semantic data from emergency floor plans for digitization.
• To enhance Building Information Modeling (BIM) based evacuation tools.

Main Methods:

• Utilized color filtering, clustering, and object detection to extract building walls.
• Generated clean, digitized floor plans by integrating geometric and semantic data.
• Trained two Faster Region-based Convolutional Neural Networks (Faster R-CNNs) on real and synthetic datasets.

Main Results:

• The synthetic model demonstrated superior performance in identifying rare emergency symbols compared to the standard model.
• The framework successfully digitized emergency floor plans, improving data quality.
• Faster R-CNN models achieved high accuracy in symbol classification and localization.

Conclusions:

• The developed framework effectively digitizes emergency floor plans for improved evacuation planning.
• Enhanced BIM tools can leverage this digitized data for better path planning and decision-making.
• The approach offers a valuable solution for advancing digital evacuation applications.