Abstract

Effective fire detection is crucial for safeguarding lives and property. Fires pose not only direct risks but also economic damage, especially in the case of forest fires. Traditional fire detection systems rely on electronic sensors that detect heat, smoke, or other fire-related characteristics, but their accuracy can vary based on proximity. To overcome this, video-based fire detection methods have been carried out in this paper. One approach uses the HSV color model, applying a color mask and intensity threshold to identify fire pixels in video frames. The other approach employs deep learning with the Inception V3 CNN model, training it on fire and non-fire images to identify the video frames. While both methods provide real-time detection, the deep learning method, which captures complex patterns for greater accuracy, is more sophisticated than the HSV-based method in terms of simplicity. Choosing between the two depends on requirements and available resources. Video-based fire detection shows promise and warrants further research for system enhancement in different scenarios.