Key points are not available for this paper at this time.
The use of radar technology in the field of human activity recognition (HAR) has garnered considerable interest due to its notable benefits in terms of accuracy, resilience, and safeguarding of privacy. As the back-scattered radar returns are composed of multiple components & nonstationary in behavior, the time-frequency analysis like Wigner-Ville distribution (WVD) can play a significant role, but the presence of cross-terms limits the suitability of WVD. This study introduces a time segmentation and frequency windowing-based novel strategy for cross-terms free WVD of multicomponent nonstationary signals. The effectiveness of the proposed method is validated on synthetic signals which outperforms the existing related methods for different performancemeasures. The proposed time-frequency representation (TFR) is also applied for HAR from the received signals of frequency-modulated continuous-wave (FMCW) radar. A novel analysis of extracting slow-time signatures from the received signal captured by FMCW radar is developed and the proposed technique is utilized to generate cross-term free TFRs from these signatures for each activity. Threshold-based time windowing followed by 2-D segmented features extraction is performed on the TFRs for activity recognition system development. The optimization of ensemble classifier yields an impressive classification accuracy rate of 99.51%, surpassing the performances achieved by previous methodologies on the same data set.
Faisal et al. (Tue,) studied this question.