Using a Selective Ensemble Support Vector Machine to Fuse Multimodal Features for Human Action Recognition

Table 6

The comparison results between our approach and other methods.

Researchers	Descriptor	Recognition methods	Accuracy
			G3D (%)	CAD60 (%)
Dollár et al. [37]	Sparse	SVM	78	83
	Spatiotemporal features
Liu et al. [38]	PMI spatiotemporal features	SVM	82	86
Laptev et al. [39]	Spatiotemporal corner	SVM	87	84
Rapantzikos et al. [40]	Dense saliency spatiotemporal features	KNN	88	89
Rodriguez et al. [41]	Spatiotemporal template	Template matching	88	89
Our approach	Mixed features	SESVM	91.7	91.8