3D structure reconstruction from an ego motion sequence using statistical estimation and detection theory

The paper discusses the problem of estimating 3D structures from an extended sequence of 2D images taken by a moving camera with known motion. The work is mainly concerned with sparse line features and thus is a natural extension of the feature-based motion analysis paradigm. Usually such a paradigm involves several separate operations: feature detection, feature matching, structure/motion estimation, and higher level processing, such as feature grouping. The authors propose to integrate the different phases based on the statistical estimation and detection theory. They show how each operation can be formalized and, in particular, consider the structure parameter estimation and the feature matching together as the combined estimation-decision problem. The proposed algorithm is tested with both synthetic and real data