engleski

Comparative analysis of semantic segmentation and object detection methods for early fire detection

Fire is an event that can be extremely dangerous since it can result in human casualties, as well as catastrophic property damage. To reduce the amount of damage that fires can cause, early fire detection and prompt coordination with the appropriate authorities are necessary. This issue has inspired a number of works on fire detection systems, which can be categorized into two groups: sensor-based technologies and image-based techniques. Image-based systems employing machine learning algorithms have been demonstrated to be superior to sensor-based systems due to faster fire detection and additional fire-related information they provide. The information includes the fire’s location, intensity, and progression. It is challenging for image-based systems to reliably detect the shape and limits of flames due to several factors, including a background, varying sizes of the fires, and interference from objects that resemble flames. In this study, our objective is to evaluate the performance of semantic segmentation and object detection models using a custom-made image dataset which contains fire in both indoor and outdoor environments.

early fire detection ; machine learning ; object detection ; semantic segmentation

nije evidentirano