Research On Indoor Video Flame Recognition Based On Improved ViBe Algorithm And Deep Learning

With the improvement of fire safety requirements in the country,in the face of traditional fi.re protection technology,there are many problems such as single detection index,low real-time performance and low recognition accuracy.Video flame recognition technology has become a hot topic in the field of fire safety.Based on the detailed analysis of video flame technology and features,this paper focuses on the problem of inaccurate indoor dynamic video object extraction due to the shortcomings of traditional segmentation algorithm,such as low real-time performance and low detection accuracy,the computer vision segmentation technology and multi-feature recognition technology are combined to realize the rapid recognition of flame.Aiming at the problem that the Vi.Be algorithm has slow ghost elimination and can not adapt to sudden changes in light,this paper proposes an improved ViBe algorithm.The algorithm updates the image of the current frame segmentation by adopting the seven-frame difference subtraction method to eliminate the"ghost"caused by selecting the single-frame image as the background;At the same time,the improved algorithm establishes the judgment mechanism to replace the original fixed threshold with the adaptive threshold when judging the sudden change of light.Simulation results show that the proposed improved method not only eliminates"ghost"but also shortens the adaptation time of the algorithm to the abrupt light.Aiming at the problem of inaccurate color recognition for different saturations due to the single color model for the flame region color decision,an improved color space model is proposed.The model incorporates the HSI model into the RGB model and adds new saturation judgment rules.The simulation test results show that the model can realize the recognition of the highlight of the flame center and effectively eliminate the non-flame target.In order to solve the problem of inaccurate flame identification,the convolutional neural network is introduced into the original feature fusion algorithm.The training model is obtained by establishing the training flame image set,and the obtained model is applied to the flame identification.The simulation results show that the method can improve the accuracy of error-prone flame recognition.Comprehensive simulation results show that the fusion motion region segmentation technique and multi-feature recognition technology can achieve fast and accurate recognition of the flame.