Forest Fire Smoke Detection Based On Video Analysis

As one of the most serious natural disasters,forest fire brings serious economic losses to human beings.The traditional detection technology of forest fire is based on sensors,and the performance is not satisfied,there is a large number of false and missed detection.Therefore,researchers have carried out studies on forest fire detection technologies based on computer vision,in which fire detection and alarm based on the forest fire smoke have become the focus of attention.This thesis focuses on the video analysis based forest fire smoke detection technology,and the main contributions are as follows.Firstly,the method of moving target detection based on background modeling is studied and improved.In addition to the traditional Gaussian mixture model,the LBP texture model is applied to forest background modeling.The moving target detection results of Gaussian mixture model and LBP texture model are fused by using D-S evidence theory.The accuracy and robustness of moving target detection are improved markedly.In addition,in order to improve the running speed of the algorithm,background modeling is conducted based on image block.Compared with traditional background modeling which is conducted based on image pixel,the speed of our proposed algorithm is significantly improved.Secondly,the characteristics of smoke in the image are analyzed and extracted.The static and dynamic characteristics of smoke are considered.The static characteristics include color feature,texture feature and high-frequency energy,and the dynamic characteristics include area change,direction of motion and motion accumulation.The characteristics of smoke are analyzed by experiment,and the results show that it is difficult to accurately distinguish the smoke and non-smoke just by using a single feature.It is necessary to combine the multiple features to effectively describe the smoke.Thirdly,based on support vector machine,the smoke classifier is designed.The factors that affect the accuracy of classifier,including normalization of eigenvalues,different kernel functions,different parameters,and different combinations of features,are analyzed.On the basis of experimental results,the classifier is optimized,and the classifier with the best recognition effect is obtained.Finally,the forest fire smoke detection system is designed by combining the moving target detection algorithm,the characteristics of smoke image and the classifier based on support vector machine.The integrated system's performance is tested,and the results show that the system can timely and accurately identify the smoke.