Research And Implementation On Video-based Outdoor Smoke Detection

The early smoke detection in outdoor space concerns to safety issues andhas been realizedin video surveillance systems of many sites, e.g.: nearbuildings, on bridges, ships, into tunnels and etc. Smoke of an uncontrolledfire can be easily observed by a camera even if the flames are not visiblethatmake possibledetection of fire in the very beginning, before conflagrationspreads around. On the initial stagesignition source may be hidden by naturalrelief: trees, bushes, stones, tallgrass and etc.Thereby for the purpose of earlydetection of wildfires and especially forest fires only smoke detectiontechnique can be applied.We have presented a novel approach of video smoke disclosurebasedontextural features of smoke corresponded to its physiochemical nature–CEA (Color and Edge Analysis) Based Approach. Refining it bymathematical dynamic model of smoke propagation we proposed STC/DPTBased Approach that combines mathematical simulation and dynamicdetection of smoke regions.This approach isbased on spatial-temporalclustering (STC) of moving regions and analysis by dynamic patterntemplates (DPTs) for automatic outdoor fire surveillance on long-distancevideo. The spatial-temporal clustering consists of firstly spatial separation ofregions of interests (ROI) for every analyzed video frame andsubsequentproducing of temporal correspondence between ROI of adjacentframes. DPTs are estimated by a proposed mathematical model of smokepropagation in open space and can accurately predict characteristic of smokeregions.Visual smoke detection system developed on basis on the proposedapproach (STC/DPT), which has been proved in experiments with betterperformance:(just1.098times slower than the simple color-based method),accuracy of area marking (on the average it is about80.9%of the systemmarked area matches human marked area) and stable low false alarm ratio(about4.62%of false alarms) in a whole number of video sequences.