Light Color Flame Recognition And Localization Based On Multi-Source Vision

Fires caused by chemicals usually have great destructive power.How to achieve the detection of flame targets in the early stage of fire is of great significance to avoid the occurrence of explosions and to maintain the safety of rescue workers.There are many productions about the flame detection at the domestic and overseas.For these issues,it takes the light color flame caused by flammable chemicals as the research target,and solves the target recognition and localization under various complicated scenes such as illumination interference,similar background color,occlusion interference and mirror flame.At present,the traditional flame detection algorithm mostly performs flame target detection based on the static and dynamic characteristics of the flame.And they can't accurately recognize and locate the light color flame image with missing information.In view of the above situation,this paper combines image fusion technology,image matching technology and stereo vision technology to study a multi-source visual inspection system based on infrared and visible light images.The specific work of the paper is as follows:(1)In order to realize the light flame target recognition in complex scenes,this paper studies the flame detection algorithm based on fused image.Firstly,the fusion information is used to complement the useful information in the infrared and visible images to obtain high-quality fused images.Then the dual-threshold segmentation algorithm based on RGB-IHS color space is used to realize the recognition of light-colored flame targets.In order to effectively preserve the color information and edge detail information in the visible light image and fuse the rich gray information in the infrared image,an improved IHS-NSCT fusion algorithm is adopted.Based on the independence of each component in the IHS color space,the NSCT algorithm is used to fuse the I component of the visible image and the infrared image.In order to obtain a high-quality fused image,a fusion rule based on the local logarithmic Gabor energy model is used in the high-frequency sub-band image fusion process,and a fusion rule based on the double-channel pulse coupled neural network(D-PCNN)model is used in the low-frequency sub-band image.(2)Aiming at the low efficiency of traditional stereo matching algorithm,this paper studies the improved S-SIFT matching algorithm.In the process of generating feature points,an adaptive Gaussian filter is used to make the extracted feature points better describe the edge regions of the image.The number of extracted feature points is controlled by the adaptive threshold ? setting,and the 16-dimensional descriptor is used to replace the traditional 128-dimensional descriptor,which reduces the calculation amount and shortens the matching time.Finally,a constraint matching algorithm based on two-way matching is adopted to avoid the occurrence of mismatch.The simulation experiment shows that the flame detection algorithm based on fused images has a recognition rate of 91%for the light flame target in complex scenes,and the false recognition rate is 9%.The positioning algorithm based on S-SIFT can obtain the exact position of the light flame target,and the absolute error is less than 100mm.The experimental results show that the improved IHS-NSCT algorithm has a great improvement in image fusion effect compared with similar algorithms.The improved S-SIFT matching algorithm in this paper has further improved in the matching rate and matching accuracy.In this paper,there are 29 figures,4 tables and 40 references.