Human Activity Recognition Using Gaussian Mixture Hidden Conditional Random Fields
Table 8
Classification results of the proposed system on IXMAS action dataset (A) using ANN, (B) using SVM, (C) using HMM, and (D) using existing HCRF [30], while removing the proposed HCRF model (unit: %).
Activities
CA
SD
GU
TA
Walk
Wave
Punch
Kick
(A)
CA
65
5
7
6
5
4
3
5
SD
5
72
4
3
5
4
3
4
GU
4
3
75
5
3
2
4
4
TA
6
7
4
68
3
5
4
3
Walk
3
4
5
4
70
6
4
4
Wave
4
6
5
3
4
71
3
4
Punch
3
5
5
6
7
3
67
4
Kick
4
5
4
6
5
4
3
69
Average
69.62
(B)
CA
77
3
4
2
2
3
5
4
SD
3
79
3
2
4
5
2
2
GU
5
6
69
3
4
5
4
4
TA
2
3
2
80
4
4
3
2
Walk
3
5
4
2
71
5
6
4
Wave
2
6
3
5
4
73
4
3
Punch
1
5
3
4
1
3
81
2
Kick
3
6
7
4
3
5
4
68
Average
74.75
(C)
CA
79
3
4
1
2
4
3
4
SD
1
84
3
2
3
1
4
2
GU
0
1
88
1
2
3
2
3
TA
5
2
3
79
2
3
2
4
Walk
1
0
3
1
90
2
3
0
Wave
2
3
1
0
3
86
2
3
Punch
1
0
2
3
0
4
89
1
Kick
3
2
4
1
0
2
4
84
Average
84.77
(D)
CA
90
1
2
0
3
4
0
0
SD
3
85
2
1
3
3
2
1
GU
0
1
91
1
0
2
3
2
TA
1
3
2
87
1
2
2
2
Walk
1
0
3
1
89
2
1
3
Wave
0
2
1
0
2
90
3
1
Punch
1
4
2
1
2
3
84
3
Kick
1
2
4
1
3
2
4
83
Average
87.38
CA: cross arm, SD: sit down, GU: get up, TA: turn around.
