A novel smoke detection algorithm based on Fast Self-tuning background subtraction

The smoke detection in remote video surveillance is of great importance in forest fire prevention since smoke usually appears before fire. In order to solve this problem, we proposed a novel smoke detection algorithm based on fast self-tuning background subtraction segmentation and judgments of smoke analysis. First, a self-tuning background algorithm is utilized on the source image to segment the candidate smoke regions, which overcomes the drawbacks in color-based and motion-based segmentation methods. Then the static and dynamic judgments of smoke are applied on the candidate smoke regions to identify the smoke region. Experiment results demonstrate that our algorithm performs well on detecting smoke in remote video surveillance while achieving robustness to the change of environment.