LAUSR

As an important signature associated with human movement, human micro-Doppler (m-D) signature can provide the basis for activity classification. In particular, the m-D signal of limbs can provide a highly distinctive feature for the activity with reduced limbs movement, which can be used to detect people who are carrying weapon or injured. Fully exploiting the elaborate m-D features that correspond to the motion of limbs can improve the classification accuracy of such activities. Therefore, it is significant to separate the limb-swing micro-Doppler signature from the torso signature and process them separately. In this paper, a novel separation method is proposed, which uses the short-time fractional Fourier transform (STFrFT) with different orders and window lengths to sparsely characterize the echoes from limbs and torso, respectively. Then STFrFT based sparse representation is combined with the morphological component analysis (MCA) to realize the m-D signal separation. Simulation and experimental results verify the effectiveness of the proposed algorithm, where the real data of different activities are utilized to demonstrate its adaptability.