Early Detection of Wildfires using Object Recognition from National Park Surveillance Camera Footages

Abstract

Annually occurring worldwide, wildfires cause substantial damage to animals and plants in nature, as well as humans living in nearby affected areas. When a wildfire grows or spreads to a certain extent, it becomes difficult to extinguish it; therefore, early detection of wildfires is important to prevent or alleviate environmental damage. However, in most national parks or mountain ranges, it is still humans who are responsible for monitoring. This is not a feasible way to ‘cover’ the vast land that may become a target for fires. To efficiently and effectively detect wildfires early, this study trains a model that detects a wildfire utilizing photo data from mountain range CCTV surveillance feed. For training, EfficientDet D0 Model from Tensorflow Object Detection API was used. The results of this study were successfully validated through 92% accuracy in the validation set.