Journal of Digital Image Processing and Computer Vision

Authors- P.Tamil Mathi , Dr. L. Latha

Abstract- 

The role of fire detection is to identify fire in a timely manner. Practically automatic fire detection using image processing techniques characterizes one of the significant aspects of any fire avoidance earlier. In current fire detection systems exact location of fire, manual work such as physical inspection of fire pose a serious threat to timely detection of fire. To overcome this standard approach must be developed. Detection of fire using image and video is effective than using sensors.
A computer vision based approach for detecting the presence of fire in video is presented here. The contributions for the fire detection may be an image or a video but the input as a video is quite complex process but provides good result. The fire is detected using spatio temporal flame modeling and analysis of dynamic textures. Extraction of features such as color consistency, spatial wavelet, spatio-temporal analysis, temporal analysis is by spatio temporal flame modeling. RGB color model is used for calculating color consistency and 2DWT for calculating wavelet in space. In wavelet analysis the image is divided into sub-images (LL, LH, HL, and HH). The main feature of fire, the flickering property is identified using temporal analysis. These features are combined with features of texture analysis which includes LDS and codebook is formed which are then classified using SVM classifier. This approach performs better than other fire detection system with high true positives and less true negatives.