|RESULTS: Direct Registery Structure From Motion|
The proposed SFM method computes directly the position of the camera without the necessity to
recover the inter-frame motion. The structure of the scene is formed by sets of 3D vertices characterized by
local image descriptors. In this way, by visually associating image patches extracted from camera views with
the 3D vertices, we can recover the camera position with respect to the scene model. Subsequently, the obtained
camera positions are used to update the scene model as new features are tracked. Both camera position estimation
and scene model update use robust estimation methods thus reducing the impact of poor camera position/vertex
The proposed SFM algorithm works in two stages. First, it uses a standard motion estimation technique in order to initialize a model of the scene structure. Once an initial model has been generated, all the subsequent camera positions are computed by registering 2D point projections with 3D scene structure.