Feature Extraction And Smoke Detection In Video Images

With the rapid development of the social and economy,the people's living standards are greatly improved,and the fire safety of various places such as shopping malls,stadiums,office buildings,warehouses,hospitals,etc.is becoming more and more important.There have been many fire incidents in China almost every year.On normal circumstances,smoke will first appear in the early stage of the fire.If the smoke can be detected early,the fire warning can be successfully carried out to protect people's lives and property.Traditional smoke detection methods based on sensors.Once the fire occurs,the sensor can be used to alarm,because these sensors are sensitive to smoke and temperature.Since these sensors are sensitive to smoke and temperature,they will alarm once the fire occurs.Usually,the concentration of smoke particles and the temperature of the indoor environment need to reach a certain level before an alarm is issued.So it is likely that the fire warning cannot be performed in time,thereby missing the best rescue opportunity.In contrast,the image-level smoke detection methods are more effective by simply detecting and verifying the smoke in the video.Therefore,these methods gain considerable attention.In recent years,more and more researchers have begun to conduct smoke detection research on video images,because the smoke detection of video images can be detected by simply seeing the smoke in the video,without the need to wait for the smoke particle concentration reaching a certain level like the traditional sensor.In order to overcome the shortcomings of traditional fire smoke detection technology and improve the performance,this thesis proposes a complete smoke detection algorithm ranging from foreground extraction to smoke detection by studying the characteristics of smoke.Firstly,in view of the fact that some existing foreground extraction algorithms can not accurately extract the smoke area,this thesis proposes a novel spatial-temporal background model to extract the complete smoke foreground area from the video.Then three kinds of texture features with high discriminative capability are proposed.Finally,the related features of smoke are extracted from the foreground region,and a support vector machine is further exploited to determine whether the foreground region contains smoke.We test our method on both image and video data to verify the effectiveness of the algorithm.Under the test of the smoke image database,the average false positive rate,detection rate and error rate of the proposed algorithm are 0.145,0.978 and 0.162,respectively.The test results based on the video database show that our algorithm has strong adaptability to complex scenes and high detection accuracy.Compared with the existing smoke detection algorithms,the detection rate is increased by 2%~4%.