Mining Key Skeleton Poses with Latent SVM for Action Recognition

Table 5

Comparison of our method with the others on MSR Action 3D.

MSR Action 3D
Histogram of 3D joints [24]	78.97%
EigenJoints [25]	82.30%
Angle similarities [27]	83.53%
Actionlet [1]	88.20%
Spatial and temporal part-sets [28]	90.22%
Covariance descriptors [29]	90.53%
Our approach	90.94%