Sign Language Recognition with CW Radar and Machine Learning

Sign language is the primary communication medium for the deaf-mute community. However, the literacy of understanding and using sign language is hard to gain without professional training. In this paper, we explore the use of low power frequency modulated continuous wave radar for automatic sign language recognition. The proposed system is composed of a radar, a sound cluster and a computer for transforming signals to spectrograms. Furthermore, as the time-frequency spectrograms are high-dimensional data with redundant information, we then perform dimensionality reduction by extracting the histogram of oriented gradients features from these spectrograms. The features are finally classified by the k-Nearest Neighbour algorithm and a classification result of 95.8% is achieved on the five testing signs. The impact of the k value in the k-Nearest Neighbour is also investigated.