The project uses AlexNet as a feature extractor to learn features, which is used to recognise hand gestures.
The Hand Gesture Recognition Database can be found at Kaggle. It contains ~20,000 images in various hand poses.
It uses AlexNet, a CNN which was pretrained on the ImageNet database. The last layer was removed to extract the features, which can be further used for training.
The features can be trained on any classic machine learning model, like KNN, Naive Bayes, etc.
The result are as follows:
| Model | Accuracy |
|---|---|
| LinearSVC | 95.67% |
| KNN with 10 neighbors | 95.34% |
| GaussianNB | 80.67% |