Excellent Feature Points for Three-Dimensional Reconstruction of Scene from Video

This paper presents an outlier removing method for improving the performance of automatic 3D reconstruction of scene using video sequences acquired with an uncalibrated camera. Firstly, the feature points in the region of interest are tracked in video sequences via KLT tracker. Secondly, we applied a vector direction analysis (VDA) strategy in our system for outlier's removal, which can efficiently decrease the number of outliers. Some of the experimental results are provided for showing the performance of our strategy. Finally, this paper uses the singular value decomposition (SVD) method to solve the problem of 3D reconstruction of scene from video. Experiments demonstrate the reasonable results by using VDA strategy on the tracked features.