Research On Video Smoke Detection Method Based On Combined Multiple Features

Because of the detection range of traditional smoke detectors based on sen sor technology has limitations,so the detection effect of traditional smoke detectors is not good in an open environment.In order to improve the detection effect,researchers devoted themselves to the study of smoke intelligent detection technology based on video image processing and pattern recognition.The intelligent detection technology is to implement smoke detection by three steps involved in motion detection,feature extraction and classification.Firstly,motion area is detected by existing Gaussian Mix Model.On the basis of motion detection,feature extraction and classifier selection are need to be analyzed.A smoke detection method based on combining the wavelet feature,texture feature and luminance feature and an integrated BP neural network classifier based on Adaboost algorithm is proposed in this paper.The key works are involved in the following aspects.On the basis of the motion detection,smoke feature extraction method based on combining wavelet feature,texture feature and luminance feature is proposed.Smoke once appears in the video,the area covered by smoke becomes blurred.This fuzziness is described by wavelet energy decrease.The high frequency coefficient features of smoke images and non-smoke images are quite different.Smoke image texture is more uniform than non-smoke image.Luminance component is characterization of the white smoke.The result is a new fashioned peculiar 22 dimensions smoke feature vector.After feature extraction,the next is selection of classifier.A smoke detection method by an integrated BP neural network classifier based on Adaboost algorithm is employed.Test results of single BP neural network could happen by chance.An integrated BP neural network is formed by 10 weighted BP neural networks,which is applied to smoke detection.This method is more effective than the single BP neural network classifier detection method.