Early Fire Detection using SVM based Machine Learning Algorithm

---

Abstract

Every year, numerous wildfires occur worldwide, leading to detrimental impacts on forests, economies, and societies. To mitigate these losses, aside from preventive measures, early warnings and swift responses are crucial. To address this, a novel computer vision strategy for detecting fire flames, intended for early-warning fire monitoring systems is presented. An initial background elimination and color evaluation-based non-parametric technique is used to identify potential fire zones inside a frame. Subsequently, the behavior of the fire is characterized by integrating diverse spatiotemporal attributes like color prospect, flickering, and spatial and spatiotemporal energy. Concurrently, dynamic texture analysis is employed within each potential area, making use of linear dynamical systems and a variety of system techniques. Enhancing the robustness of the algorithm, the constancy of space and time for every probable fire zone is estimated. This is achieved by leveraging information about the context from neighboring blocks across both present and preceding video frames. The SVM classifier is used to find out the potential areas where fire is detected. The identified fire is processed using machine learning algorithm. It can be noted from the results obtained that the machine learning algorithm used for detection provides a better situation for the presence of fire and smoke as compared with other method and the approach is reliable under such conditions.

---

Keywords

Machine Learning (ML); Fire Detection; Support Vector Machine (SVM); human vision; camera; optimize