Research On Flame Recognition Algorithm For Video Image

Fires pose a serious threat to people's property safety and life safety,so it is crucial for fire flame recognition.Traditional fire detectors have temperature sensing,smoke sensing,and light sensing.They can only judge a specific information feature,and the recognition effect is not good due to environmental and spatial conditions.As computer vision and image processing technologies matured,fire detection systems for video images were produced.The recognition of flames is conducive to the discovery of early fires,and flame recognition of video images has become one of the hot research directions.In this paper,two important steps of suspected flame region detection and flame image features description in flame recognition are studied in depth.A new flame recognition algorithm for video images is presented.The main work of the thesis is as follows:A flame recognition algorithm for video image based on spatial features is studied.Firstly,the suspected flame region is determined according to the background method and color space.Secondly,the geometric features and texture features of the suspected flame region are extracted and fused.Finally,the flame recognition of video images based on SVM.A large number of experiments show that accurate combination of spatial features has a higher flame recognition rate.A complex frequency domain feature extraction method based on M-DTCWT(multidirectional dual-tree complex wavelet transform)is proposed,combined with multi-feature fusion to achieve flame recognition in multiple scenes.Firstly,the foreground region is obtained by the improved Gaussian model.Secondly,the suspected flame region is detected for the foreground region by the RGB-HSI mixed color space.Then,combining the filter bank in the M-DTCWT with the hourglass filter bank to construct more M-DTCWT in the diagonal direction,M-DTCWT decomposition on the suspected flame region image,and extracting the improved LBP(Local Binary Patterns)texture feature and circularity feature in the low frequency coefficients and extracting the edge feature in the high frequency coefficients.Finally,through feature fusion,the SVM(Support Vector Machine)using the cross grid search method recognizes the flame.A large number of experimental results verify that the algorithm improves the accuracy of flame recognition compared with the spatial features recognition algorithm.A flame recognition algorithm for video flames combining superpixel and MHSW(Maximal HSV and SILBP of Windows)feature is proposed.First,In order to improve the integrity of the suspected region of the flame,the SLIC superpixel segmentation method is used to detect the suspected flame region and the suspected flame region is segmented based on the superpixel block in YCbCr color.Secondly,MHSW feature and dynamic features were extracted from the suspected flame region images and fused.The MHSW feature is a combination of two SILBP(Scale Invariant Local Ternary Pattern)statistical histograms and HSV color histogram corresponding mode maximum values of local windows in the same level.The flame dynamic features include the area change rate of the flame and the flame flicker frequency.Finally,Flame recognition is realized by using SVM.The experimental results show that the algorithm has higher recognition rate and lower false reduction rate,and improves the flame recognition rate and the robustness to illumination changes.