Review Article
3D Gestural Interaction: The State of the Field
Table 1
A table summarizing different 3D gesture recognition approaches, the size of the gesture set, and the stated recognition accuracy.
| Author | Recognition approach | Number of gestures | Accuracy |
| Pang and Ding [58] | HMMs with kinematic features | 12 | 91.2% | Wan et al. [60] | HMMs with sparse coding | 4 | 94.2% | Lee and Cho [61] | Hierarchical HMMs | 3 | Approx. 80.0% |
Whitehead and Fox [68] | Standard HMMs | 7 | 91.4% | Nguyen et al. [66] | Two-stage HMMs | 10 | 95.3% | Chen et al. [63] | HMMs with Fourier descriptors | 20 | 93.5% | Pylvänäinen [67] | HMMs without rotation data | 10 | 99.76% | Chung and Yang [71] | Threshold CRF | 12 | 91.9% | Yang et al. [72] | Two-layer CRF | 48 | 93.5% | Yang and Lee [73] | HCRF with BoostMap embedding | 24 | 87.3% | Song et al. [78] | HCRF with temporal smoothing | 10 | 93.7% | Liu and Jia [80] | HCRF with manifold learning | 10 | 97.8% |
Elmezain and Al-Hamadi [83] | LDCRF with depth camera | 36 | 96.1% | Song et al. [84] | LDCRF with filtering framework | 24 | 75.4% | Zhang et al. [85] | Fuzzy LDCRF | 5 | 91.8% | Huang et al. [88] | SVM with Gabor filters | 11 | 95.2% | Hsieh et al. [89] | SVM with Fourier descriptors | 5 | 93.4% |
Hsieh and Liou [90] | SVM with Haar features | 4 | 95.6% |
Dardas and Georganas [92] | SVM with bag of words | 10 | 96.2% | Chen and Tseng [93] | Fusing multiple SVMs | 3 | 93.3% | Rashid et al. [94] | Combining SVM with HMM | 18 | 98.0% | Yun and Peng [101] | Hu moments with SVM | 3 | 96.2% | Ren and Zhang [99] | SVM with min enclosing ball | 10 | 92.9% | Wu et al. [100] | Frame-based descriptor with SVM | 12 | 95.2% | He et al. [96] | SVM with Wavelet and FFT | 17 | 87.4% | Nisar et al. [104] | Decision trees | 26 | 95.0% | Jeon et al. [105] | Multivariate fuzzy decision trees | 10 | 90.6% | Zhang et al. [106] | Decision trees fused with HMMs | 72 | 96.3% | Fang et al. [107] | Hierarchical Decision trees | 14 | 91.6% | Miranda et al. [109] | Decision forest with key pose learning | 10 | 91.5% | Keskin et al. [41] | Decision forest with SVM | 10 | 99.9% | Keskin et al. [110] | Shape classification forest | 24 | 97.8% | Negin et al. [111] | Feature selection with decision forest | 10 | 98.0% | Ellis et al. [74] | Logistic regression | 16 | 95.9% | Hoffman et al. [112] | Linear classifier | 25 | 99.0% | Kratz and Rohs [113] | $3 gesture recognizer | 10 | 80.0% | Kratz and Rohs [114] | Protractor 3D (rotation invariance) | 11 | Approx. 91.0% |
|
|