Table 9.

Comparison of methods on self-made datasets. The results are presented in the order of experiments (i–v), and the result of using the completed architecture we proposed in Figure 2 is shown in bold text. Note: joint points (JP), module ratio (MR), vector angle (VA), keyframe sequences (KFS), keyframes (KF). For more details, see Section 4.2 and Section 4.3.

Experiment	Approach	Features	Accuracy
(i)	Module I+ IV	JP	82%
(ii)	Module I+ II+ IV	MR+ VA	90.25%
(iii)	Module I+ III+ IV	JP+ KFS	86.75%
(iv)	Module I+ II+ III+ IV	MR+ VA+ KFS	92.13%
(v)	(Module I+ II+ III)+ SVM	MR+ VA+ KF	77.02%