Research On Radiation Consistency Of Virtual-real Fusion In Infrared Road Scene

Infrared imaging simulation technology can effectively overcome the limitations of time,environment,and region to obtain infrared images under different environmental conditions.Traditional simulation methods based on scene modeling have disadvantages such as high modeling complexity and long modeling time.With the development of augmented reality technology,adding virtual simulation objects to the real scene can avoid large-scale scene modeling work and generate infrared simulation images conveniently and quickly.In order to ensure that the visual effect of the simulated image is more realistic,it is necessary to focus on the radiation consistency problem of the virtual and real fusion,that is,the virtual scenery and the real background should have consistent infrared radiation.In the real environment,the radiation value received by the detector is the value after the atmospheric environment disturbed.Therefore,when adding virtual objects to the real scene,it is necessary to analyze the environmental radiation characteristics of the real scene from the image or video,calculate the influence of environmental radiation and render it to the virtual object,so as to make the environmental radiation of the virtual target consistent with the real scene,and then complete the seamless fusion of the virtual target and the real scene.At present,most of the common research results of infrared virtual-real fusion calculate the environmental radiation impact based on real measurement data,which is largely limited by equipment and cannot meet the virtual-real fusion task in a specific environment.The existing image-based environmental radiation parameter estimation methods are often used in the field of image restoration,and the environmental radiation parameters are estimated based on the grayscale value statistical information of the image.For single-color channel infrared images,the lack of color information often makes it difficult to obtain reasonable and accurate parameter estimation results,and because the groundtruth of environmental radiation parameters are difficult to obtain,it also has a certain degree of unfavorable influence on solving the optimal solution of parameters.In order to solve the above problems,on the basis of fully studying the imaging principle of infrared image and the transmission characteristics of infrared radiation,this thesis proposes a virtual-real fusion radiation consistency method of infrared road scene,which combines the scene structure and grayscale value information of infrared road image,estimates the environmental radiation parameters from the image,and applies it to augmented reality.(1)In order to provide sufficient prior reference information for the estimation of environmental radiation parameters,this thesis segmented the road area of the road image,and proposed a road segmentation algorithm based on vanishing point and random forest.The algorithm first divides the road area through the detection of vanishing points and road boundary lines,and then uses the random forest algorithm to further optimize the initial segmentation results to obtain more accurate segmentation results.In the random forest optimization stage,this thesis uses super-pixel-level feature expression for model training,which improves the training efficiency and reduces the computational cost while ensuring the segmentation accuracy.Aiming at the vanishing point detection of infrared images,this thesis proposes a vanishing point detection algorithm based on the combination of texture orientation and improved Hough Transform,which uses the texture feature of the image and the straight line feature of the voting accumulation space to detect the vanishing point.(2)A method for estimating environmental radiation parameters in infrared images is proposed.This thesis uses the dark pixel method to estimate the overall atmospheric radiation value of the image,and innovatively proposes an atmospheric transmittance estimation algorithm based on the grayscale value attenuation prior.According to the results of road segmentation,this thesis uses different algorithms for different segmented regions to estimate atmospheric transmittance.For the road area,the proposed grayscale value attenuation prior algorithm is used to estimate the atmospheric transmission value according to the grayscale value attenuation degree of the pixels at different positions in the road scene.For non-road areas,this paper uses a method based on minimizing the loss function to estimate the atmospheric transmission value of each pixel.At the same time,the upper and lower thresholds are set according to the atmospheric transmittance estimation result of the road area and the inclusion relationship between the pixel and the image block to prevent the atmospheric transmission value too high or too low.Finally,the WLS filtering algorithm is used to regularize the initial atmospheric transmission image to obtain the final transmission image.In order to verify the validity of the experimental results,this thesis conducted experiments on the image restoration application.First,estimate the environmental radiation parameters from the image,and then remove the above environmental radiation influence from the infrared image to obtain the restored image.The experimental results prove that the algorithm in this paper can achieve better experimental results for image restoration applications.At the same time,this thesis also evaluates the intermediate experimental results of the proposed algorithm.In terms of vanishing point detection,the improved vanishing point detection algorithm in this thesis can detect the vanishing point of infrared road images more accurately and quickly;in terms of image segmentation,the segmentation method based on vanishing point and random forest algorithm also has higher segmentation accuracy.rate.Finally,this thesis applies environmental radiation estimation results to augmented reality.The experimental results show that the environmental radiation parameters estimated by the algorithm in this thesis can ensure the consistency of the radiation fusion of virtual objects and real scenes.