WAFER: Wearable, Ambient-Aware Adversarial Fall Event Detection System Using a RISC-V SoC Architecture

Individuals suffering from neurodegenerative diseases, neurological disorders, or going through pregnancy together make up an extensive group of individuals who are affected by adversarial fall-related accidents, many of which result in fatalities. Thus, a real-time deployable, non-invasive, low latency, ambient-aware passive event detection system is of utmost necessity. It must be capable of detecting adversarial falls using inertial and physiological information from biomarkers in dynamically shifting environmental scenarios. Our study aims at the development of an ambient-aware model-switching fall event estimator system on a RISC-V SoC platform. RISC-V is an open, base-plus extension ISA suited for lower memory consumption applications as necessary components for architectural development. An Area Under a Curve-based Logistic Regressor and an attention-gated CNN architecture are designed for different context applications of the system. The system architecture comprises efficient hardware (FPGA) implementation, hardware accelerators, and IP core design. Context awareness is provided by an ambiance detection module comprised of pinged GPS and ambient-sound estimators to leverage model-switching capabilities.