Efficient Detection of Humans in Flames Using HOG as a Feature Criterion in Machine Learning

Abstract

Detection of humans in flames is a challenging task. The task in this work is classified into two stages. The first is detection of fire, and the second is detection of human. The proposed method involves fire detection based on colour format YCbCr for image preprocessing. It further uses a histogram of oriented gradient (HOG) and support vector machine (SVM) to detect a human in the fire. It evaluates several motion-based feature sets for human detection in the form of videos. In this work, both modules were integrated to make them work together. For the detection of fire, four different rules involving colour thresholding were used and background differencing was used for moving object detection. The main objective of this work is to spot the humans in the flames who are trapped in it so they can be rescued quickly. This can help the firefighters in rapid planning and serious zone detection. The proposed model has 81% efficiency, which has outperformed the existing models for detection of humans in flames.