Abstract-This paper presents a feature-based 3D recovery system. First the scene model is recovered by the factorization algorithms from the digital images sequence. This approach also provides the result error estimation. Next the preliminary mesh is refined as follows. We choose an image from the sequence and search for the additional features using the Harris corner detector. Then we choose the second image from the sequence and estimate the correspondences positions for the additional features using the small set of reliable features. Finally estimations are defined more accurately by the Kanade-Lucas tracker.