Y. Jia, G. H. Lin, J. J. Wang, J. Fang and Y. M. Zhang (2016) Light Condition Estimation Based on Video Fire Detection in Spacious Buildings. Journal/Arabian Journal for Science And Engineering 41 1031-1041. [In English]
Web link: http://dx.doi.org/10.1007/s13369-015-1923-3
Keywords: ,Flame color, Adaptive segmentation model, Motion features, Fire, recognition, SUPPORT VECTOR MACHINES, FLAME DETECTION, PROBABILISTIC APPROACH

Abstract: In industrial applications in spacious buildings, video-based fire detection system needs to endure the excessive incoming light, which makes the video overexposed. So an adaptive flame segmentation and recognition algorithm is proposed to promote the adaptability and detection rate of the video-based fire detection system for a spacious building. First, moving foreground in a video is found and luminance of the moving region is calculated to estimate the light condition. For different light conditions, different flame-color segmentation models are selected adaptively. After a series of post-processes of segmentation, the suspect flame regions are extracted for feature analysis. Then, a trained support vector machine is implemented to distinguish flame and non-flame regions. The performance of the proposed algorithm is verified on a set of videos containing flames and interference. The adaptive flame segmentation model promotes the flame segmentation resulting in different light conditions. The results are compared with those of three other methods used in the literature, revealing the proposed method to have both a better segmentation result and better precision. In flame classification, the performance of five other methods has been compared with that of the proposed SVM method, and the result shows that the SVM classifier has the best stability and the accuracy is higher than most of the other tests. The proposed method achieves average detection rate of 95.0 %. The result shows that both the accuracy and robustness of segmentation have been improved and it is appropriate for industrial fire detection in spacious buildings.