We discuss the modeling of environments having a planar ground through a set of sparse views or wide baseline images. The camera parameters are assumed to be known approximately within some range according to the error margins of the sensors used such as inertial devices. We propose a junction detection operator that is based on determining circumferential anchors. Point matching among sparse views is achieved through reconstructing the junction planes in 3D space. Hence the so-called homographic correlation can be applied. 3D point sets can be estimated and clustered. An object can then be represented through different ways; as a bounding box, as a planar patch or 3D modeled through a voxelization scheme.