REALISTIC 3D SCENE RECONSTRUCTION FROM UNSCONSTRAINED AND UNCALIBRATED IMAGES TAKEN WITH A HANDHELD CAMERA

摘要: We address the problem of reconstructing 3D scenes from a set unconstrained images. These image sequences can be acquired by video camera or handheld digital without requiring calibration. Our approach does not require any priori information about cameras being used. The camera's motion and intrinsic parameters are all unknown. use novel combination advanced computer vision algorithms for feature detection, matching, projection matrix estimation in order to reconstruct point cloud representing location geometric features estimated input In second step full model is reconstructed using triangulation process. demonstrate with data sets different structures obtained under weather conditions that our algorithm stable enables inexperienced users easily create complex content simple consumer level camera.