3D Recovery with Simultaneous Accuracy Estimation and the Following Mesh Refinement

This paper presents a feature-based 3D recovery system. We start from the small set of reliable features those were detected and tracked on the sequence of digital images. Using the factorization algorithms we recover a model of the scene and estimate its accuracy. Next the preliminary mesh is refined as follows. We choose an image from the initial sequence and search for new features using Harris corner detector. Using the initial set of features we build hypothesizes about the positions of corresponding points on another image of the sequence. These estimations are defined more accurately by Kanade-Lucas tracker.