An Automatic System for Detection of Fires in Outdoor Areas

The traditional detection of fires is based on smoke sensors. However, this method is unsuitable for large and open buildings, and outdoor areas. Therefore, this study proposes an efficient approach to detecting fires in open areas based on computer vision systems. The input video is framed and decomposed by lifting wavelet transform to reduce the data size without missing important features. Then parallel color detection in three color-spaces (HSV, YCbCr, and binary) are applied. Otsu's algorithm is used to extract automated intensity pixels in binary space. The experimental results show that the approach accuracy exceeds 99% for the offline videos, and exceeds 89% for online (real-time) videos with low complexity.