Table 1.

Summary of methods using hand-crafted motion features.

	YEAR	SUMMARY	DATASET
Bobick et al. [47]	2001	Use of motion-energy image (MEI) and motion-history image (MHI).	-
Schuldt et al. [49]	2004	Use of local space-time features to recognize complex motion patterns.	KTH Action [49]
Niebles et al. [55]	2007	Use of a hybrid hierarchical model, combining static and dynamic features.	Weizmann [64]
Laptev et al. [42]	2008	Use of spatio-temporal features and extend spatial pyramids to spatio-temporal pyramids.	KTH Action [49] Hollywood [42]
Chen et al. [63]	2009	Use of HOG for human pose representations and HOOF to characterize human motion.	Weizmann [64] Soccer [83] Tower [63]
Chaudhry et al. [46]	2009	Use of HOOF features by computing optical flow at every frame and binning them according to primary angles.	Weizmann [64]
Lertniphonphan et al. [45]	2011	Use of a motion descriptor based on direction of optical flow.	Weizmann [64]
Wang et al. [76]	2013	Use of camera motion to correct dense trajectories.	HMDB51 [88] UCF101 [89] Hollywood2 [90] Olympic Sports [91]
Akpinar et al. [81]	2014	Use of a generic temporal video segment representation, introducing a new velocity concept: Weighted Frame Velocity.	Weizmann [64] Hollywood [42]
Kumar et al. [15]	2016	Use of a local descriptor built by optical flow vectors along the edges of the action performers.	Weizmann [64] KTH Action [49]
Sehgal, S. [87]	2018	Use of background subtraction, HOG features and BPNN classifier.	Weizmann [64]