Research On Video-based Flame And Smoke Detection Algorithms

Due to fire occurrence poses a serious threat to the human life and property safety, and traditional fire detection method based on sensor also has many disadvantages such as alarm time delay,non-applicability at a wide-range condition. Therefore, video-based fire detection method proposed is very significance. This method has many advantages such as rapid response speed, largely monitoring range and provide more information, even can adopte latest pattern recognition and machine learning for analysis.As a result of flame occurrence and smoke came into being,this paper will research video-based flame and smoke detection respectively.The main research work is as follows:(1) Researching the framework of video-based flame and smoke detection,and comparative analysising detection methods include frame difference,optical flow and background subtraction in the framework procedure of foreground detection.Finally,in order to eliminate the influence of shadow,this paper proposes a method combines Gaus-sian Mixture Model and Statistics non-parametric approach.(2) Putting forward video flame detection algorithm, firstly detect the motion re-gion and get it, then segment it using color feature in the detected region and get the flame candidate region.In order to further improve the accuracy of color segmentation, this paper combines the decision rules in RGB and YCbCr color space and then judges dynamic characteristics of flame including strobe and color variation in the candidate region. The experimental results show that, the video flame detection proposed in this paper can not only obtain very high accuracy rate but also meet the real-time require-ments.(3) Putting forward video smoke detection algorithm based on color moments, this algorithm firstly uses Gaussian Mixture Model and dynamic three frame differ-ence method for foreground segmentation. Also statistics non-parametric approach is introduced for eliminating the affect of the shadow, then using the rules of color judg-ment for foreground detection. For furhter verification of the smoke region, we use the smoke of the histogram and translucent dynamic characteristics, after judging the static and dynamic characteristics of smoke, we extract color moment characteristics in the smoke candidate region and classified with SVM. The experimental results show that, the smoke detection algorithm based on color moment can not only obtain very high accuracy rate but also meet the real-time requirements.(4) In order to further improve the smoke detection accuracy, from another point of view,this paper putting forward video smoke detection algorithm based on image sepa-ration. This algorithm firstly use the method combining Gaussian Mixture Model, Statistics non-parametric approach and dark channel for detecting smoke area in the video image frames, then separating component of smoke in the candidate smoke area. In order to accelerate the speed of separation, this paper puts forward the concept of image bright-ness channel and sets it as the initial image separation value, getting component of smoke and extracting LBP features. Eventually classifying with SVM. The experimen-tal results show that, the video smoke detection algorithm based on image separation can greatly improve the detection accuracy rate.