Representation for Action Recognition Using Trajectory-Based Low-Level Local Feature and Mid-Level Motion Feature

Table 1

The accuracy comparisons of representation for action recognition on J-HMDB dataset and YouTube dataset.

Datasets	Features	Low-level encoding	Two-layer model
JHMDB	HOF	52.2%	56.1%
	-HOF	53.9%	57.3%
	MBH	58.4%	62.9%
	HOF + -HOF + MBH	62.9%	67.4%
YouTube	HOF	77.6%	77.9%
	-HOF	75.2%	77.0%
	MBH	84.2%	85.0%
	HOF + -HOF + MBH	85.3%	87.6%