Depth estimation and modeling of a tree from a single image

Recovering 3D shape of an object from one or a series of images is an interesting but challenging problem in computer vision and computer graphics. Single-image based shape recovery techniques, as such shape from shading, need to make a number of assumptions, most of which can hardly be satisfied in a natural scene. This paper presents a new method that recover the 3D shape of branches from a single image. By making use of more stable shading characteristics due to the special structure of trees, it removes most of the restrictions in traditional shape from shading and thus, is robust enough to deal with pictures of nature trees. Furthermore, a method of modeling a tree from its single static image is proposed. In stead of defining branches with mathematical models. it directly extract branches from the image and reconstruct them in 3D space. As a result, the approach is not only independent to the species of the tree, but also capable of producing more realistic models.