Fire Smoke Detection Based On Video

Fire detection technology based on video has been paidmore and more widespread attention. However, due to the complexityand changing characteristics of the environment in a real fire scenario, itis more difficult to identify the fire. The fire has the patterns offormation and development in accordance with certain characteristics,in chronological order, the main stages: the generation and diffusion ofsmoke, and flame generation and expansion. Early fires, the flame isvery small, the fire detection is more difficult. Due to diffusion of thesmoke, it is relatively easy to be detected, the detection of smoke in thepre-warning of the fire is very important. Smoke alarm in time plays avery important role on early warning fire detection, in order to avoid therisk of fire or reduce fire losses. The key of fire detection technology ofvideo surveillance is the timely detection of fire smoke.The texture is one of the true、 inherent features in the real imagesequences. Generally speaking, the texture is composed of manymutually approaching and mutual component, and has a periodically.Dynamic texture has the time correlation of the texture in the imagesequence. Visual fire detection, dynamic texture features is an importantparameter of the fire smoke detection. Local Binary Pattern (LBP)method is an effective method of image texture analysis. In this paper,feature extraction method based on a local binary pattern is proposed todescribe and identify the dynamic texture of smoke. First of all,according to the dynamic characteristics of the smoke, Motion region inthe video sequence is extracted by a Gaussian mixture model ofbackground subtraction method, then extract the feature from the imagesequence, and establish VLBP model to make statistics for the dynamictexture characteristics of fire smoke, then calculate the distance betweeneach texture characteristic curve. Classify and identify the differentdynamic texture by using the knn (k-nearest neighbor) classifier and“leave one out” rules, thereby, detect the fire smoke. Experimental results show that the smoke dynamic texture recognition method basedon volume local binary pattern (VLBP), which is very effective toidentify the smoke of the fire, and to improve the robustness of the firedetection.