Video smoke detection based on deep saliency network. (April 2019)

• Record Type: Journal Article
• Authors: Xu, Gao; Zhang, Yongming; Zhang, Qixing; Lin, Gaohua; Wang, Zhong; Jia, Yang; Wang, Jinjun
• Is Part Of: Fire safety journal. Volume 105(2019)
• Page Start: 277
• Page End: 285
• Publication Date: 2019-04
• Subjects: Video smoke detection -- Deep saliency network -- Salient map -- Existence prediction; Fire prevention -- Periodicals; Incendies -- Prévention -- Recherche -- Périodiques; Fire prevention -- Research; Periodicals; 628.92205
• DOI: 10.1016/j.firesaf.2019.03.004
• Languages: English
• ISSNs: 0379-7112

Abstract

Abstract: Video smoke detection is a promising fire detection method, especially in open or large spaces and outdoor environments. Traditional video smoke detection methods usually consist of candidate region extraction and classification but lack powerful characterization for smoke. In this paper, we propose a novel video smoke detection method based on deep saliency network. Visual saliency detection aims to highlight the most important object regions in an image. The pixel-level and object-level salient convolutional neural networks are combined to extract the informative smoke saliency map. An end-to-end framework for salient smoke detection and the existence prediction of smoke is proposed for application in video smoke detection. A deep feature map is combined with a saliency map to predict the existence of smoke in an image. Initial and augmented datasets are built to measure the performance of frameworks with different design strategies. Qualitative and quantitative analyses at the frame-level and pixel-level demonstrate the excellent performance of the ultimate framework. Highlights: We investigate the performance differences between state-of-art saliency detection methods for smoke detection. We propose an end-to-end framework for salient smoke detection and prediction for smoke existence. The integration of multiple-level saliency cues is proposed for smoke detection. Evaluations at the frame-level and pixel-level demonstrate the excellent performance of the proposed method.