Extraction of 3D structure from video sequences

Abstract

Inferring 3D information from video sequences for building scene models is a costly and time consuming task. However, newly developed technologies in video analysis and camera calibration allow us to acquire all the information required to infer the 3D structure of a scene from the recording of a video sequence of it using a domestic video camera held by a moving operator. In this paper we present a method to recover the 3D rigid structure from a video sequence. We base the method in a given set of key 2D features tracked by the Kanade-Lucas-Tomasi algorithm, and validating them by checking that they can correspond to points of a rigid scene. © Springer-Verlag Berlin Heidelberg 2003.