Fast segment iterative algorithm for 3D reconstruction

3D model reconstruction has many application possibilities, for example: person detection and authentication, model scanning for computer simulation, monitoring, object recognition, navigation, etc. The biggest problem of this approach is its computation complexity. More precisely the problem lies in process of searching for differences in multiple input images (e.g. stereovision). Most of existing algorithms searches for the shift in each image point to obtain most detailed disparity map. But it is possible to speed up this process by reducing the number of points that must be processed. This paper is describing a new method for a fast key-point extraction using sparse disparity. The effectiveness of the proposed algorithm comes from its ability to divide input images into segments in two steps: First initial division identifies key-points and is based on local extremes in Difference of Gaussian. Second division is used to obtain results with better detail from initial division. Therefore, it is possible control level of detail for the output 3D model so it is possible to control the computational demands.