Single Image based Fog Information Estimation for Virtual Objects in A Foggy Scene

Abstract

Synthesizing virtual objects in a real scene is an important technique in augmented reality. Immersing these virtual objects into the scene can be challenging, for instance in environments with the presence of fog. This paper investigates strategies for augmenting virtual objects on foggy scenes. Firstly, we synthesize the foggy image including original virtual objects, then depth map of single image is obtained by adapting Conditional Generative Adversarial Networks, which takes the original foggy image containing virtual objects as input. The estimated depth cues can be utilized for generating the foggy-effect on the object according to the atmospheric light model, making the virtual object exhibit a similar degree of foggy effect on real objects at a similar distance range. Experimental results demonstrate that the proposed technique presents improved visual effect for synthesizing virtual objects on foggy scenes.