원문정보
초록
영어
Human sport action reconstruction is the current hot issue of image processing, against the feature of the problem, a new human action reconstruction method base on the combination of LLE and KRR is proposed. First, extract the human action framework against the color and deep information of the image and establish action training set and then transfer the training set into the motion vector library, calculate the low dimensional manifold using LLE algorithm and conduct correlation analysis to the manifold data, and finally make the predicted low dimensional manifold convergence action point mapped irreversibly to the low dimensional Euclidean space. This is the first application of KRR algorithm to solve traditional dimensionality reduction mapping irreversibility, so as to make the action reconstruction realized. The results show that: the construction action obtained by the convection action point is ideal, as the action process of the different degrees of freedom; the proposed human sport action construction has very good results.
목차
1. Introduction
2. LLE Algorithm
3. KRR Algorithm
4. Realization of Human Sport Action Reconstruction
4.1. LLE Mapping of Human Action
4.2. Reconstruction of Human Body Action
5. Experimental Analysis
5.1. Human Skeleton Extraction
5.2. Human Body Action Two-Dimensional Manifold Generation Test
5.3. Reconstruction Test of Human Body action
6. Conclusion
Acknowledgment
References