Group Sparse Regression-Based Learning Model for Real-Time Depth-Based Human Action Prediction

Table 1

Comparisons with other real-time existing methods.

MSR-Daily Activity dataset
Dynamic temporal warping [5]	54.0
Actionlet ensemble [5] (skeletal feature only)	68.0
Fourier temporal pyramid [5]	78.0
Distinctive canonical poses [24]	65.7
Relative position of joints [25]	70.0
Moving pose [26]	73.8
BIPOD representation [17]	79.7
Our approach (skeletal feature only)	85.6
Our approach	88.2
UTKinect-Action dataset
HOJ3D [27]	90.9
Histogram of Direction vectors [28]	92.0
BIPOD representation [17]	92.8
Our approach (skeletal feature only)	94.3
Our approach	95.1
SYSU 3D HOI dataset
ST-LSTM(Tree)+Trust Gate [29]	76.5
Part-aware LSTM [30]	76.9
BIPOD representation [17]	77.3
Our approach (skeletal feature only)	79.1
Our approach	80.7