Indoor Human Activity Recognition using Millimeter-Wave Radio Signals

Human activity recognition is crucial for civilian and security applications. Compared with the traditional wearable and optical methods, millimeter-wave sensing has advantages of wide detection range, strong environmental adaptability and no privacy issues. However, the current millimeter-wave sensing approaches are mainly based on micro-Doppler feature identification or machine learning with lots of label data, resulting in poor robustness or highly dependent on big data samples. In this article, a novel feature-driven recognition method was proposed, in which five feature metrics with physical meaning are constructed. The detailed procedures for performing the proposed method were illustrated, including pre-processing, feature extraction and classification. Experimental results show that our method can reliably recognize not only the grossly different activities, but also the similar activities such as sit and fall-down.