Smoke detection in video using wavelets and support vector machines

Early warning systems are critical in providing emergency response in the event of unexpected hazards. Cheap cameras and improvements in memory and computing power have enabled the design of fire detectors using video surveillance systems. This is critical in scenarios where traditional smoke detectors cannot be installed. In such scenarios, it has been observed that the smoke is visible well before flames can be sighted. A novel method for smoke characterization using wavelets and support vector machines is proposed in this paper. Forest fire, tunnel fire and news channel videos have been used for testing the proposed method. The results are impressive with limited false alarms. The proposed algorithm is evaluated for its characterization properties using motion segmented images from a commercial surveillance system with good results.