The Illumination Analysis Of Outdoor Scene

Augmented Reality (AR) attracts more and more interests in recently years, which integrates the virtual objects into the real scene and makes the whole scene seem to be seamless. Because of the characteristics of enhancing the display, AR is used in many areas such as urban planning, military training, medical operation and film and television industry. But in order to achieve the goal of seamless integration, there are many problems. The illumination consistence is an important part for the rendering of the virtual objects.The illumination study of the outdoor scene is a research project of computer vision. In AR technique, the virtual objects and the real scene share a common lighting environment. The outlook of the virtual objects should be consistent with the real scene. When the illumination of the real scene changes, for example, a cloud in the sky over the sun, the light becomes weak, then the illumination of the virtual objects dynamic changes, otherwise the actual fusion of the virtual and real will be affected.In order to achieve the illumination consistence, we propose a novel method based on the image decomposition to recover the illumination parameters of direct sunlight and ambient skylight for a static outdoor scene, also we can get the irradiance of sunlight and skylight for the virtual rendering of AR system. The main contributions are:1. In the offline stage, our method obtains the shadow effects previously through using k-means to analyse the images of a sunny day captured at a fixed viewpoint, and we can identify some pixels in the shadow area for the next step.2. We assume the skylight is uniform, then we can decompose an image of outdoor scene into the linear combination of the skylight basis image and the sunlight basis image, and the combination coefficients are the intensity of the skylight and the sunlight. 3. Based on the known shadow function, we raise two ways to obtain the skylight basis image-one is to solve pixel by pixel, the other is the minimum energy function. Then we can solve the sunlight basis image and the intensity successively, and scale the intensity of skylight and sunlight to providing a unified metrics for the rendering of virtual objects.4. Our method needn’t to assume the surface to be diffuse, and our method can deal with the specular reflection, which is the advantage of our work,5. In addition, our algorithm doesn’t need any geometry、material、texture information or user interactions and doesn’t need know the skylight basis previously. Experimental results successfully demonstrate the effectiveness and flexibility of our proposed method.