Abstract
The application of fire detection has gained a lot of attention due to the increasing threats posed by global warming and recent forest fires around the globe. Traditional fire detection systems that rely on sensors and changes in its physical surroundings like temperature and pressure. However, these methods trigger alerts only when the fire has already spread beyond control.
This project tries to address this challenge by leveraging modern machine learning and transfer learning methods relying on vision inputs. Furthermore, this project also incorporates an interactive dashboard that provides insights and helps the users understand data and derive patterns for decision making. It also provides active fire locations, alerts and historic fire data.
The image dataset that are used in this project to train and test the model are collected from the Roboflow. The datasets to build a dynamic dashboard is from the year 2003-2022 Residential Fires Information Data collected from US Fire Administration (USFA).
The final product is an integrated system that provides both real time detection, analysis of data and an alerting system that can be utilized by emergency teams as well as users and can benefit both to respond quickly, stay informed and be safe.