Improving Dynamic Hand Gesture Recognition based IR-UWB using Offline Data Augmentation and Deep Learning
Djazila Souhila Korti, Zohra Slimane · 2022 IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET) · 2022
This paper presents a dynamic hand gesture recognition method that relies on combining a deep learning model with a traditional classifier, with the aim of eliminating the manual feature extraction phase. To this end, we proposed to use a multi-stream CNN-LSTM for automatic feature extraction in conjunction with a multi-class Support Vector Machine (SVM) for classification. The proposed model consists of an efficient and lightweight architecture based on depth-separable convolution layers, which effectively reduces the computational cost and learning time while maintaining high recognition performance. The model is trained on the public UWB Gestures dataset and can automatically learn range-time information to identify different types of hand gestures. In order to achieve optimal results and avoid overfitting, we adopted offline data augmentation strategies to expand the training sample size. The results obtained showed that our proposed model achieves high performance on several measures, including precision, recall, F1-score and accuracy.