Intrinsic Property-Based Soft Biometric Classification Using Wavelet CNNs and LRP

In this presentation, we explore a novel approach to improving soft-biometric recognition and identification by integrating wavelet transform with wavelet deep learning techniques from the gait attributes. Our method leverages the multi-resolution analysis capability of wavelet transform to enhance the sequential context in sensor data, addressing challenges such as sensor biases and position alignment. We validate our approach on publicly available OMoCaP and IMU-based human activity datasets, demonstrating superior performance and robustness. Join us to discover how this innovative method can advance the field of human activity recognition and soft-biometric identification.