Open-Scenario-Oriented Human Gait Recognition Using Radar Micro-Doppler Signatures

The application of radar technology in human activity recognition and gait recognition has garnered significant attention due to the wealth of information pertaining to human movements that can be derived from Doppler shifts. The present study aims to investigate radar-based gait recognition, with a specific emphasis on the challenge of identifying disguised gaits. The proposed model offers a distinct advantage over previous methodologies as it can be effectively applied in open scenarios. These scenarios involve unknown specific styles of disguised gaits and lack any prior information pertaining to this particular condition. In addressing this issue, we propose an approach that conceptualizes this specific requirement as a problem of domain generalization. In our model, the integration of the maximum entropy principle serves as an effective measure to mitigate the potential risk of degradation that may arise from pseudo-invariant features. Furthermore, we present a methodology for quantifying the complexity of features using norms. This approach successfully addresses the geometric biases inherent in the classifier. The evaluation of the proposed model is conducted using the measurement radar dataset, which comprises eight distinct gait patterns. The model's outstanding performance serves as evidence of its appropriateness for tackling the disguised gait recognition problem in open scenarios. Extensive experimentation, involving ablation studies and additional exploratory analyses, has been conducted to further investigate the efficacy and generalizability of the approach.