Vision-Based Smoke Detection Algorithm for Early Fire Recognition in Digital Video Recording System

The digital video recording (DVR) closed circuit television (CCTV) system is quickly becoming one of the most accepted security, surveillance documentation, and monitoring in service today. Particularly a lot of catastrophes such as tsunami, volcano, and terrorism cause huge casualty and property loss. In this paper, we concentrate on the smoke detection system in video for early fire alarming. The proposed block-based smoke detection algorithm consists of three basic steps, which has simple operation and provides good performance. In the first step, we discover motion vector utilizing several motion estimation schemes and discern the suspect as smoke region. This process does not require huge computational cost because of using the H.264/AVC during the recoding in DVR system. In the second step, we conduct block based chromatic detection. In the third step, we employ motion information for detecting the correct smoke block by using the characteristics that smoke goes almost upward. Experimental results provide that the proposed method gives good performance for real smoke detection.