Multifeature Fusion Human Motion Behavior Recognition Algorithm Using Deep Reinforcement Learning

Table 3

Specific test results of initial sampling affecting performance.

Randomly select the initial observation window	Data set	Recognition rate (%)
	KTH	92.65
	Weizmann	96.32
	IXMAS	92.38
	Hollywood	95.65
	HMDB51	96.57
	UCF101	94.21
After determining the person area, take the person area as the initial observation window	Data set	Recognition rate (%)
	KTH	93.63
	Weizmann	97.20
	IXMAS	93.69
	Hollywood	96.52
	HMDB51	97.32
	UCF101	95.21