Upgrading eigenspace-based prediction using null space and its application to path prediction
This paper proposes a method for an Eigenspace-based prediction of a vector with missing components by modifying a projection of conventional Eigenspace method, and demonstrates the application to the prediction of the path of a walking person. This modification is based on domain-specific knowledge of data, and a linear combination of vectors in the null space of Eigenspace is added so that a cost function of smoothness of path is minimized. Some experimental results on actual paths are shown to demonstrate how the proposed method works.
Proceedings of Subspace 2007
|date of issued||
Asian Conference on Computer Vision
Copyright (c) 2007 by Author
Graduate School of Engineering