An Efficient Method for Vision-Based Fire Detection Using SVM Classification

In this paper, we present a new vision-based algorithm for fire detection problem. The algorithm consists of three main tasks: pixel-based processing to identify potential fire blobs, blob-based statistical feature extraction, and a support vector machine classifier. In pixel-based processing phase, five feature vectors based on RGB color space are used to classify a pixel by using a Bayes classifier to build a potential fire mask (PFM) of image. Next step, a potential fire blob mask (PFBM) is computed by using the difference between two consecutive PFM and a recover technique. In blob-based phase, for each potential blob in a potential fire blobs image (PFBI) an 7-feature vector are evaluated; this vector includes three statistical features of colour, four texture parameters and one shape roundness parameter. Finally, a SVM classifier is designed and trained for distinguish a potential fire blob are fire or fire-like object. Experimental results demonstrate the effectiveness and robustness of the proposed method.