An Efficient Method for Detecting Electrical Spark and Fire Flame from Real Time Video

Abstract

Prompt fire detection and localization is an essential requirement for saving lives and reducing damages caused by fire accidents. The main source of 39–45% fire accidents is electrical origin. As electric fire accidents are increasing, detecting electrical spark and fire flame from the common origination point-electrical socket is imperative and vital. In this paper an efficient method for detecting electrical socket spark and fire flame from real time video processing is proposed. At first, extracted image frames from video are converted from RGB to YCbCr. After that, any change in average Luminance Y, average Blue component Cb, average Red component Cr triggers moving foreground detection step. From the detected moving foreground by frame differencing, regions having highest Luminance are considered as suspects. Finally, the actual spark or fire flame regions are detected by taking only those suspects whose area changes in consecutive frames. For evaluating the proposed method, two types of dataset (electric spark and fire flame) are used. Experimental result shows 80% and 100% accuracy in spark and flame detection respectively. The proposed algorithm worked properly in the five tests for flame and in the four tests for spark. In addition, we have compared the performance of proposed method with a state of art model and experimental results show that the proposed model outperforms existing system with 60% more accuracy. The proposed system can assist in fire accident investigations and in proper prevention action decision making by early fire and fire source detection.