2016 13th Conference on Computer and Robot Vision (CRV) (2016)
Victoria, BC, Canada
June 1, 2016 to June 3, 2016
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/CRV.2016.46
The combination of outputs of single classifiers for human action recognition is proposed and evaluated in this paper. Combining the outputs of individual classifiers leads to having a more accurate and robust-applicable framework by reducing the risk of a weak choice of a classifier or a set of features. The weakness of single classifiers becomes more bold when the problem difficulty increases, particularly while having numerous action types or similarity of actions. In this paper, the individual support vector machines (SVMs) are trained using diverse feature sets from different perspectives. Then, the outputs of single SVMs are fused by employing the algebraic combinations and Dempster Shafer fusion methods. The experimental results show that the action recognition accuracy is improved while employing the algebraic combinations to fuse the outputs of single classifiers.
Feature extraction, Trajectory, Encoding, Visualization, Vocabulary, Optical imaging, Histograms
E. Mohammadi, Q. M. Wu and M. Saif, "Human Action Recognition by Fusing the Outputs of Individual Classifiers," 2016 13th Conference on Computer and Robot Vision (CRV), Victoria, BC, Canada, 2016, pp. 335-341.