Research On Matching And Fusion Technology Of Long Wave Infrared And Visible Heterogeneous Images

Night vision system is an important equipment of on-board auxiliary driving system,which can acquire environmental information under low illumination and ensure the safe driving of vehicles under bad weather conditions.Infrared and visible light sensors are widely used in night vision systems.Among them,infrared sensors can receive infrared rays emitted or reflected by objects,and can acquire environmental information under poor illumination conditions.They are especially suitable for observing pedestrians and vehicles.Visible light sensor has good performance on the details of objects,which is more in line with the naked eye observation habits.After registration and fusion,the advantages of the two images can be combined.However,the feature consistency between infrared image and visible image is weak,and the general image registration algorithm is not suitable for this kind of heterogenous image matching.At the same time,the gray level difference between the two images is great,and the traditional fusion method is not conducive to the observation of the image.To solve these two problems,based on the existing registration and fusion algorithms,this thesis proposes a registration and fusion algorithm for heterogeneous images.The final generated image can retain more regions of interest of human eyes compared with the traditional equalization fusion image,and improve the problem of weak consistency of features of heterogeneous images and unfavorable observation of fused images.Aiming at the problem of weak consistency of some features in heterologous image extraction and matching,a method of heterologous image matching based on gradient coincidence degree is proposed in this thesis.Based on the idea of integrated innovation,this method constructs a norm-based objective function to measure the gradient coincidence degree,and then completes the registration of infrared and visible images through the optimization of the objective function.At the same time,based on the camera imaging principle,according to the camera parameters and spatial position,the initial value of the matrix in this matching algorithm is derived,eliminate the image error caused by the camera structure and improve the accuracy of the algorithm.In addition,aiming at the problem that the gray level difference between infrared and visible images is large,which is not conducive to eye observation,this thesis designs an image fusion algorithm based on Laplacian pyramid and image saliency,which decomposes the image into Laplacian decomposition and calculates it on each layer.Visual saliency map in image is used to guide the weight of fused image,and finally the fusion is carried out by inverse Laplace transform.This thesis also designs a method to verify the adaptability and matching accuracy of the objective function matching algorithm based on gradient edge coincidence degree.By randomly intercepting the rectangle in the image and performing small translation and rotation operations,the transformed image is compared with the untransformed image,and the image registration error is preliminarily evaluated by four corner coordinate differences.It is found that the registration algorithm can correct the misalignment of the error within five pixels.At the same time,compared with traditional average fusion method,this fusion method can perform better in retaining the important information in infrared and visible images.