Table 1.
A summary of the literature and related works about gestures and activities recognition of normal and autistic people.
| Ref. No | Sensors | Activities | Features | Algorithms and Accuracy |
|---|---|---|---|---|
| [2] | Moto 360 smartwatch | Flapping, painting, and sibbing | Discrete cosine transform, FFT, variance, bi-spectrum, z transform, entropy | Simple tree, complex tree, linear and gaussian SVM, boosted and bagged ensemble trees Accuracy: 96.7% |
| [34] | ECG, accelerometer, gyroscope, magnetometer | Walking, climbing stairs, frontal elevation of arms, knees bending, cycling, jogging, running, jump front and back, sitting, relaxing | Mean, standard deviation, and correlation | Mean prediction rate 99.69%, HMM 89.98%, DBN 92.01%, RNN 99.69% |
| [35] | Not mentioned | 9 uniform hand gestures | Not mentioned, total 576 features extracted | SVM 98.72% |
| [36] | Gyroscope, accelerometer | Hand movements, body movements | Publicly available dataset features | Convolutional neural network 87.1%, KNN 66.1%, SVM 77.1%, fully CN 88% |
| [37] | Not mentioned | Static and dynamic unistroke hand gestures | Not mentioned | SVM 97.95% |
| [38] | Accelerometer, magnetometer, gyroscope | Jogging, walking, cycling jumping, running, jump-rope | Mean, standard dev, kurtosis, skewness, range, correlation, spectral energy, spectral entropy, peak frequencies, and cross-spectral densities | SVM 26%, DT 93.24%, KNN 96.07%, RF 97.12%, Naïve Bayes 76.47% |
| [39] | Accelerometer, strain sensor | Walking, eating | Mean value, standard dev, percentiles, and correlation frequency domain (energy, entropy) | DT 93.15% |
| [40] | Camera | Gestures of alphabets | Not mentioned | KNN 94.49% |
| [41] | Flex sensor, accelerometer, camera, | Malaysian sign language gestures | Not mentioned | General algorithm for the data-glove detection system 78.33–5% |
| [42] | Camera | 24 Fingerspelling static gestures | Not mentioned | KNN classifier 87.38%, Logistic regression 84.32%, naïve Bayes classifier 84.62%, support vector machine (SVM) 91.35% |
| [43] | Leap Motion Sensor | Gestures for greetings, possessive adjectives, colors, numbers, names, etc. | Not mentioned | Hidden Markov models (HMM) 87.4%, KNN+DTW 88.4% |
| [44] | Accelerometer | Cycling, sedentary, ambulation | Mean, standard deviation, acceleration range | SVM from 88.5% to 91.6% |
| [45] | Not mentioned | ASL alphabets and basic hand shapes |
The number of fingers, the width and height of the gesture, the distance between the hand fingers, etc. | Type-2 Fuzzy HMM (T2FHMM) 100% accuracy for uniform hand images and 95.5% for cluttered hand images |
| [24] | Flex sensor | Patterns representing:
|
Not mentioned | K-nearest neighbor decision tree dynamic time warpinga verage accuracy = 90% |
| [46] | QA screening method using mobile app | Not mentioned | Age, sex, ethnicity, country of residence, etc. | RIPPER 80.95%, C4.5 82.54% |
| [47] | Not mentioned Dataset taken from UCL Machine Learning repository | Common attributes like age, nationality, sex, etc. | Not mentioned | SVM 98.30%, KNN 88.13%, CNN 98.30% ANN 98.30%, naïve Bayes 94.91%, LR 98.30% |