Table 4 Comparison of sensor-based hand gesture recognition works.
Author and year | Number and type of gestures | Sensor | Classifier | Number of participants | Accuracy |
|---|---|---|---|---|---|
Faisal et al.44 | Interaction-oriented 14 static and 3 dynamic gestures | IMU and flex sensors | KNN | 30 (Static) and 5 (Dynamic) | 99.53% (Static), 98.64% (Dynamic) |
Wen et al.49 | 50 ASL words and 20 sentences | Triboelectric nanogenerator (TENG) sensors | 1D CNN | Not mentioned | 86.67% |
Wang et al.41 | 51 CSL words and 60 sentences | Myo Armband (IMU, sEMG) | Multichannel CNN and attention-based encoder-decoder | 34 | 89.2% (sentence recognition) |
Saquib et al.70 | 64 ASL and BdSL alphabets | IMU, flex, and contact sensors | ANN | 5 | 96% |
Chong et al.47 | 28 ASL signs | IMU | LSTM | 12 | 99.89% |
Zhang et al.71 | 42 ASL and traffic signs | Strain sensor | DTW | 3 | 94.58% |
Yu et al.50 | 150 CSL subwords | Myo Armband (IMU, sEMG) | 2D CNN and LSTM | 8 | 95.1% (user-dependent) and 88.2% (user-dependent) |
Lee et al.45 | 26 ASL alphabets | IMU, flex, and pressure sensors | Support vector machine | Not mentioned | 65.7% (without pressure sensor) and 98.2% (with pressure sensor) |
Jani et al.46 | 26 ASL alphabets | IMU and flex sensor | DTW and nearest mapping algorithm | 8 | 96.50% |
Abhishek et al.72 | 26 ASL alphabets and 10 letter | Capacitive touch sensor | Decision Tree | Not mentioned | 92% |
Wu et al.40 | 80 ASL signs | IMU and sEMG sensors | Support vector machine | 4 | 85.24% (intra-subject) and 96.16% (combined) |
Gałka et al.48 | 40 regularly used signs | Accelerometer sensor | Parallel hidden Markov models | 5 | 99.75% |
Su et al.43 | 121 CSL subwords | Accelerometer and sEMG sensors | Random Forest | 5 | 98.25% |
Savur et al.39 | 26 ASL alphabets | Myo Armband (IMU, sEMG) | Bagged Tree classifier | 10 | 79.35% |
Proposed method | 26 ASL alphabets and 14 words | IMU and flex sensors | Parallel-path CNN | 25 | 84.42% (static) and 97.35% (dynamic) |
