Vision Saliency Based Smoke Detection Method In Open Areas

Fire is one of natural disasters which have serious threat to social security. In recent years,many places such as large shopping malls, community and so on have been the frequently firesoccurred places, and have bought much loss to the social security. Since the smoke is a presentphenomenon before burning flame ,so we can send the fire alarm by detecting smoke in time toreduce the loss, In the past 30 years, along with neuropsychology and behavior science research forthe brain vision cognitive mechanism, artificial intelligence and machine vision areas appearedgreat mass fervor on biological visual calculation model research. And the research on VisualAttention Model provides a new development direction for the smoke detection. This paper aims toexplore a new smoke detection methods based on visual significant and wavelet analysis under theopen environment.The focus of this research is study on the smoke detection algorithm based on visual attentionmodel. Including improved traditional bottom-up visual attention model, bringing in top-downmission driving mechanism in smoke detection, extracting suspicious regional dynamic features ofsmoke in order to realize smoke on-line identification. This research can summarize in thefollowing respects:(1) Proposed a bottom-up visual attention module used for smoke detection based on theanalysis and summary of the Itti visual attention model. In the stage of feature extraction weintroduce the movement feature of images; also we introduce an effective integration strategy forthe integration of significant maps, making Made the model more fit with the human visualattention mechanism.(2) Establishing a top-down visual attention model to control the significant of bottom-upsignificant map. In order to simulate the task-based significant control mechanism of human eyes’,this paper combined the irregularity of smoke’s edge, using the regional image similarity togenerate the top-down significant map. This map used to control the bottom-up significant areas.(3)Building a Bayesian classifier achieved the smoke identification. The region of interestdetected based on the visual attention module is the preliminary detection of smoke. In order toidentify whether the region is smoke region or not, this paper extracted the region’s irregularity,wavelet high frequency energy, gray degree and other dynamic feature. Training a Bayesianclassifier and achieving the smoke identification accurately.(4) In the "Intelligent forest fires detection system in open areas”, I completed the developmentof remote monitoring center software. It has different functions such as real-time video playback, recording, PTZ control, alarm response and so on. Also in order to verification the effectiveness ofthe smoke detection algorithm proposed in this paper, we did the image’s bottom-up visualsignificance analysis, top-down control of regional significance and so on. At last we raised theoutlook of fire detection.