HSV Histogram-Based Method for Fire Detection in Smart Cities

Abstract

Global warming and fires are the big problems we face today. Fires, in particular, have caused loss of life and property worldwide in recent years. Early detection of fires in cities and forest areas is very important. In this study, machine learning methods were utilized for the early detection of fires. Histogram values obtained using the HSV color space were processed for the purpose of fire detection. Separate results were obtained using the Zenmuse and Phantom cameras from the FLAME dataset. It was observed that the HSV color space provided better results than the RGB and grayscale representations. For the Zenmuse Camera, the best result was achieved with the Fine KNN model, with an accuracy rate of 99.90%. For the Phantom Camera, the best result was achieved with the Cubic SVM model, with an accuracy rate of 99.64%. © The Institution of Engineering & Technology 2024.