Table 2 Convolutional neural network based action recognition.
Method | Data type | Dataset | Performance | References |
|---|---|---|---|---|
PoseConv3D | RGB + Depth | NTU-RGBD | Accuracy: 69.4, 94.2 | |
Temporal difference networks | RGB | Something-SomethingV1, Kinetics | Accuracy: 68.2, 79.4 | |
CNN | RGB | UCF101, HMDB51, FCVID, Activity Net | Accuracy: 98.6, 84.3, 82.1, 84.4 | |
2-Stream convolution network | RGB | UCF101, HMDB51 | Accuracy: 91.5, 65.9 | |
3-Stream CNN | RGB | KTH, UCF101, HMDB51 | Accuracy: 96.8, 92.2, 65.2 | |
Multi-stream CNN | Skeleton | NTU-RGBD (CS), NTU-RGBD (CV), MSRC-12 (CS), Northwestern-UCLA | Accuracy: 80.03, 87.21, 96.62, 92.61 | |
3D CNN | RGB | UCF101, HMDB51 | Accuracy: 90.2 | |
Actional-graph-based CNN | Skeleton | UCF50, UCF101, YouTube action, HMDB51 | Accuracy: 86.8, 94.2, Top-5 acc: 56.5, Top-1 acc: 34.8 | |
CNN | RGB | UCF50 | Accuracy: 92.5, 65.2 | |
CNN | RGB | UTD-MHAD, NTU-RGBD (CV), NTU-RGBD (CS) | Accuracy: 96.4, 94.33, 96.21, 70.33 | |
CNN-genetic algorithm | RGB | UCF50 | Accuracy: 99.98 | |
CNN | Skeleton | UTD-MHAD, NTU-RGBD (CV), NTU-RGBD (CS) | Accuracy: 88.10, 82.3, 76.2 |
