3D Skeletal Human Action Recognition Using a CNN Fusion Model

Table 2

Comparisons with other skeletal action recognition methods on the Toyota Smarthome dataset.

Approach	CS	CV1	CV2
HMM [24] (2016)	25.7	10.2	13.4
DT [25] (2011)	41.9	20.9	23.7
LSTM [26] (2016)	42.5	13.4	17.2
I3D [27] (2017)	53.4	34.9	45.1
I3D + NL [28] (2018)	53.6	34.3	43.9
Optical spectra [14] (2018)	53.7	31.2	43.5
JDM [15] (2017)	54.0	34.5	44.6
Separable STA [18] (2019)	54.2	35.2	50.3
DPI [29] (2019)	54.5	36.0	52.8
Proposed (spatial)	54.4	35.7	52.5
Proposed (temporal)	51.1	30.2	47.9
Proposed (fusion)	55.0	36.3	53.2