Feature Extraction for Dynamic Hand Gesture Recognition Using Block Sparsity Model

In this work, we propose a block sparse based time-frequency (TF) feature extraction method for dynamic hand gesture recognition (HGR) using millimeter-wave radar sensors. Previous studies suggested that improved recognition performance can be achieved by extracting several most representative TF points under the sparse hypothesis. In fact, the micro-Doppler features of hand gestures tend to be clustered rather than merely independent scattered points. In this paper, we investigate such a characteristic to improve the classification accuracy for HGR task. Firstly, the block sparse model is applied to model the TF distribution of hand gestures. Secondly, the TF features are extracted using the block orthogonal matching pursuit (BOMP) algorithm. Then, the extracted block features are fed into a kNN classifier for classification. At last, the effectiveness of the proposed method is validated using real data measured by a K-band radar. The results demonstrated that the block sparse model is beneficial to improve the accuracy of HGR, and the average classification accuracy for four types of hand gestures reaches 89.8%.