Research On The Key Technology Of Image-based Photorealistic Rendering For Outdoor Augmented Reality

With the development of augmented reality technology,it is gradually coming into the people's vision.Seamlessly rendering synthetic objects into a legacy photograph is one of the most popular research directions in computer graphics,and has a wide range of applications,such as military,entertainment,medical training and so on.Efficient and true interactive is the goal of all the AR applications,and with the rapid development of this technology,how to enhance the realism will be one of the most challenging difficulties.To improve the realism and meet the needs of real-time online analysis,this thesis will combine the computer vision and graphics method to analyze the key technologies and methods involved in mixed virtual reality of a single outdoor image.We improve the method based on spherical harmonic function,and propose a real-time global algorithm based on the median-cut theory.These algorithms constitute the core of illumination uniformity technology.The main contents of this thesis are as follows:(1)To ensure the geometric consistency,this thesis studies the algorithm of camera calibration based on a single image using double vanishing points,filters the correct disappearance point combination by edge detection and classification,and restores the internal and external information of the camera.A position-dependent algorithm for environment mapping is proposed,which utilizes the camera parameters obtained by calibration.In the image scene at any specified location can generate the corresponding environment mapping.(2)In terms of illumination consistency,this thesis studies the main factors of outdoor scene illumination,including the global influence of sunlight,sky light and surrounding environment on objects,and establishes global illumination model.Using the spherical harmonic function to simplify the illumination model,combines the prior knowledge constraints,and solves the illumination parameters to illuminate virtual objects.At the same time,to meet the real-time processing requirement,this thesis proposes an improved median-cut illumination estimation algorithm based on the reverse tone mapping,which transforms the LDR image into HDR images.Finally using a series of point lights to replace the use of environment maps to illuminate objects.(3)In the aspects of synthetic consistency,this thesis studies the characteristics of outdoor scene shadows,and studies the interaction between real and virtual objects.The real shadow is projected onto the surface of the virtual object through shadow mapping;the transformation between 2D and 3D is used to deal with the common virtual shadow projection to the real object surface;and the smoothing of the soft shadow is carried out at the same time.By using the differential rendering to generate the final images,the result will be obtained.(4)Implement the algorithms of the key algorithms in this thesis,achieve the goal of synthesizing augmented images base on a single outdoor image.The two different algorithms proposed in this thesis will be compared with other algorithms and the experimental results will be further analyzed for our follow-up improvement work.