Research On Multi-scale Spatio-temporal Visual Saliency Fire Detection And Firefighting Decision Approach

Fire is a fast-spreading and destructive multiple disaster.The implementation of firefighting operations at the fire site greatly threatens the safety of firefighters.Firefighting robots,as a new firefighting equipment,can replace firefighters to detect and extinguish fires in the harsh environment of high temperature and dense smoke,and have been applied in the field of fire protection.At present,firefighting robots need manual remote control to implement firefighting operations.They cannot autonomously detect fire and make firefighting decisions,so the firefighting efficiency is not high.Therefore,it is of great significance both in theory and practice to research fire detection and fire firefighting decision approach in a complex fire environment for improving the efficiency of firefighting robots and promoting the intelligentization of firefighting robots.Based on the analysis of existing fire detection methods,this thesis proposes a multi-scale spatio-temporal visual saliency fire detection method.In this method,multi-scale simple linear iterative clustering is used for to improve the region division method of Region-based Contrast saliency detection,and the color difference is used as the weight to fuse saliency maps at different scales,avoiding over-segmentation problems lie in saliency maps,improving the robustness of visual saliency detection.The prominent motion and color features of fire are used to complete the description of the the spatio-temporal features of fire.Using multi-scale spatio-temporal visual saliency to identify fire in salient areas,improving the accuracy of fire detection;On this basis,aiming at the problem that the calculation result of the visual features of fire in a complex environment is wrong,a fuzzy structure-variable dynamic Bayesian network firefighting decision method based on feature selection is proposed.The method uses Local Outlier Factor(LOF)data anomaly detection for feature selection to eliminates abnormal data features,and realizes the discretization of continuous observation variables based on Fuzzy Sets,reducing the influence of subjective factors on the division of variable states.The degree of danger of the fire is assessed according to the fuzzy variable structure dynamic Bayesian network and the discretization state of continuous observation variables in order to decide the firefighting sequence of multiple fire,improving the firefighting efficiency of the firefighting robot;A fire detection and firefighting decision system has been developed,which realizes the real-time detection of multiple fire and the decision of the sequence of fire firefighting with multiple fire.Using the fire data set of Bilkent University to experimentally verify the proposed multi-scale spatio-temporal visual saliency fire detection method,the experimental results show that the method can effectively detect fire with fluctuating and complex backgrounds.Compared with fire detection method based on multi-feature fusion,the accuracy of the method proposed is increased by 6.84%and 6.36%,and the false detection rate is reduced by 6.65%and 0.90%,respectively;Experiments on the proposed fuzzy structure-variable dynamic Bayesian network firefighting decision method based on feature selection when the feature data of multiple fire is stable and abnormal,the experimental results show that the methods proposed can effectively complete the decision of the sequence of fire firefighting when the feature data of multiple fire is stable and abnormal;The developed fire detection and fire firefighting decision system can detect simulated fire sources in real time,and accurately give the degree of danger of multiple fire sources.It can be effectively used for fire detection and fire firefighting decision in a complex fire environment to improves the fire firefighting efficiency of firefighting robots.