Table 6 Comparison with existing approaches on KTH dataset
From: A novel human activity recognition architecture: using residual inception ConvLSTM layer
Reference | Method | Publication year | Accuracy(%) |
|---|---|---|---|
Haddad et al. [42] | GF-OF and GMM | 2021 | 73.1% |
PCANet | 2020-2021 | 85.5%-93.3% | |
Ramya and Rajeswar [30] | Distance Transform + Entropy Features + ANN | 2021 | 91.4% |
Nadeem et al. [20] | SVM + ANN | 2020 | 87.57% |
Aly and Sayed [29] | Zernike Moment + BOF + SVM | 2019 | 81.03% |
Han [34] | Two-stream CNN | 2018 | 93.1% |
Nazir et al. [19] | 3DHarris + 3DSIFT + BOF + SVM | 2018 | 91.82% |
Zhang et al. [35] | Dual-channel Deep Network | 2018 | 92.8% |
Rodriguez et al. [41] | Fast-SHMM | 2017 | 74% |
Abdekkaoui and Douik [32] | DBN | 2020 | 94.83% |
Arunnehru et al. [31] | 3D CNN + 3D motion cuboid | 2018 | 94.9% |
Proposed approach | ResIncConvLSTM | 2021 | 94.08% |