Research On Smoke Detection SVM Method Based On Mutual Information Entropy Using Spatial And Temporal Analyses

Fire accident is quite harmful.However conventional point fire detectors have certain limitations in high-rise buildings and vast space.Therefore immediate valid early fire warning is a matter of great significance.In the early stages of fire,the burning objects are in smoldering state,often accompanied by the production of a lot of smoke.As artificial intelligence technology and computer vision technology developed.This research method has received more and more attention.Unlike detection of other rigid targets,smoke belongs to soft body gradient objects of non-rigid.Thus the video-based fire smoke detection still has many problems to be solved.These problems mainly focus on suspected smoke regions segmentation and feature vector extraction.Therefore,specific research work is as follows:1)A method for segmentation of smoke regions using mixed gaussian model in Lab color space is proposed.In Lab color space,the distribution of smoke pixels is concentrated,so a mixed gaussian model of L、a、b color channels is constructed.Based on the mixed Gaussian model,the 16?16 blocks in the video frames are screened for suspected smoke pixels.The 16?16 blocks with suspected smoke pixels proportion greater than or equal to 12.5% are set as suspected smoke blocks.Compared with the two common color segmentation methods,the proposed method shows better experimental results.2)The spatio-temporal features of smoke based on mutual information are proposed.The mutual information and normalized mutual information of the smoke video foreground gray 16?16 block and the background gray 16?16 block in the video frame are calculated to describe the similarity between the images and distinguish the translucency of the smoke from the interference object opacity;Additionally,analyzing one-dimensional wavelet transform of non-normalized and normalized mutual information respectively of the foreground gray 16?16 block and the background gray 16?16 block to distinguish smoke dynamic process of slowly covering the background and rapid movement of the disturbing objects.The algorithm is tested and verified in the smoke and non-smoke videos.Finally the proposed smoke description method in this paper can effectively distinguish the smoke from the disturbing objects in video frames.