Affine-Invariant Feature Extraction for Activity Recognition

Table 3

Comparison with the state of the art on the KTH and Weizmann datasets.

Method	KTH	Weizmann
Our method	93.5%	98.0%
Liu and Shah [15]	92.8%	—
Wang and Mori [16]	92.5%	—
Jhuang et al. [17]	91.7%	—
Rodriguez et al. [9]	88.6%	—
Rapantzikos et al. [18]	88.3%	—
Dollár et al. [19]	81.2%	—
Ke et al. [20]	63.0%	—
Fathi and Mori [21]	—	100%
Bregonzio et al. [22]	—	96.6%
Zhang et al. [23]	—	92.8%
Niebles et al. [24]	—	90.0%
Dollár et al. [19]	—	85.2%
Kläser et al. [25]	—	84.3%