Video Smoke Detection And Fire Source Estimation

For the past few years, property damage and environment problem caused by fire disaster is very serious. With the development of mankind technology and the rapid growth of economy, people pay more attention to the prevention of fire. The traditional fire detection technology applies to small space, and poor conditions will lead to sensor failures, it is bad for fire detection. But video fire detection is available for big space and will not failure under poor conditions.On the basis of existing video fire detection algorithms, this paper presents a new video smoke detection and estimating the location of fire source method. This thesis research includes the follows contents:Firstly, through analysis and comparison, background subtraction was selected from several commonly used methods in moving object detection, and moving regions were extracted by background subtraction. And then, to make a good foundation for the work in the future, removing the false target produced by interference factors, and motion area was processed by regional holes filled.Secondly, considering the color feature of smoke, dark channel prior was adopted to exclude the brightly colored non-smoke regions and remained the area which color is similar to smoke. As the texture features of moving object is invariable in sequential frames, LBP feature was used to estimate the orientation of moving object. The non-smoke regions were excluded for smoke’s motion orientation is not downward. And then, to differentiate the non-smoke object which color and motion feature is similar to smoke, support vector machine was used to classify smoke from smoke candidate regions by extracting LBP and HOG features of smoke. Our experimental results and comparative experiments show that the proposed method is effective for smoke detection, and it has lower detection error rate.Finally, the location of fire source in smoke video was estimated. Around fire source, the concentration of smoke is large, through analyzing the luminance, we could obtain the dense smoke figure. Considering the variability of smoke, timing diagram was used to compute the timing dense smoke figure. And then, segmenting the dense smoke region and estimating the motion orientation of each block, the motion orientation histogram was built to statistic main motion orientation. It is easy to estimate the location of fire source based on the timing dense smoke figure and the main orientation, experiments show that the proposed method is accurate for fire source estimation.