Exploiting Unique State Transitions to Capture Micro-Doppler Signatures of Human Actions Using CW Radar

Micro Doppler phenomenon enables radars to study manifestations of micro movements on top of a moving body. Recent studies have successfully demonstrated the calibre of off body radar sensor nodes in human activity recognition, using time - frequency plots. In this paper, Recurrence Quantification Analysis(RQA) and Poincaré plot based features have been introduced for action classification using radars. Using the underlying idea that each action introduces a unique sequence of pattern in time frequency plots and subsequently Cadence Velocity Diagrams(CVD), we extract state distance matrix from CVD, considering each cadence frequency as a unique state. RQA statistics and Poincaré features try to encapsulate this uniqueness. Additionally, Recurrence plots are used to provide a novel way of representation of micro Doppler data. The signal processing pipeline is tested on experimental data collected from 9 subjects for 8 actions, providing testing accuracy on unseen data of 96.8%, thus demonstrating the efficacy of the proposed method. Average value for both precision and recall is 0.97. Standard deviation for precision and recall is 0.03 and 0.05 respectively.