Smoke-Detection Framework for High-Definition Video Using Fused Spatial- and Frequency-Domain Features

Video-based smoke detection is an effective method for fire alarm systems. Given the widespread use of high-definition cameras, a smoke detection method for high-definition video is needed. This paper proposes a smoke-detection framework for high-definition video, in which the main idea is to use the small smoke image blocks to match the image features of the motion area in the video and to use the support vector machine classifier for smoke recognition. The ViBe algorithm and other methods are used to effectively extract the areas for classification. This detection framework consists of spatial- and frequency-domain features. In the extraction of frequency domain features, we use local phase quantization (LPQ) features. In the local texture features of the spatial domain, we add the compensation of adjacent pixels and consider the gradient of the symmetrical pixels using the center-symmetric local binary pattern feature. To improve results, we also propose the trisection feature fusion scheme for features in the spatial and frequency domains. The experiments show that using the feature extraction and fusion schemes, our smoke-detection framework achieves the good performance in the detection of smoke in the video from different datasets.