Fire and disaster detection with multimodal quadcopter By machine learning

Abstract

Our thesis research is consisted of developing a model that can detect early fires and, mapping the area for fire and disaster detection using a UAV (Unmanned Aerial Vehicle) or quadcopter if a fire break out. Furthermore, MLP uses fire or no fire detection, sound analysis, and input sensor to create a multimodal system architecture. First, surveillance cameras detects the early stages of a fire using luminous smoke and textured flame. However, if the fire has already started, an alarm will sound, activating the quadcopter operation. Due to the quadcopter’s camera and sound system input, it obtains an aerial perspective and maps the fireaffected region while indicating human life. Finally, disaster detection provides us with a map indicating the safe zone where the less damaged part of the building will assist the fire department in saving human lives. The unique aspect of our thesis is that it designs a complete fire detection and rescue model. It will effectively detect a fire before an incident occurs and map the fire-affected region after the incident with human life signs and the safest path to rescue. The main goal here is to prevent or mitigate damage by immediately alerting the fire department. We have collected primary dataset of Fire and Disaster. Moreover, we increased the accuracy of our fire dataset to 80.32% and increased the accuracy of our disaster dataset to 9.2%. We tried to reduce the false detection of fire. Added to that, we have integrated all the five models in graphical user interface.