Vigil: A RISC-V SoC Architecture for 2-fold Hybrid CNN-kNN based Fall Detector Implementation on FPGA

Passive, round-the-clock Health Monitoring at scale is necessary in modern society with the potential to detect adversarial health events in real-time environments. Accidental Falls constitute one of the leading causes of injuries in the elderly. This study focuses on low latency, power-efficient implementation of an Accidental Fall Detection System in dynamic environments from Accelerometer and PPG sensor data, using a hybrid CNN-kNN estimator on a RISC-V-based SoC platform. RISC-V is an open, base-plus-extension ISA suited ideally for custom SoC designs that can be configured to include the absolute necessity for an application, minimizing costs and memory usage. The hybrid model comprises a probabilistic Convolution Neural Network in conjunction with the nearest neighbor classifier for the final layer classification. Our SoC design thus focuses on an efficient Hardware (FPGA) Implementation of the custom probabilistic CNN (pCNN) architecture for motion and vital signal analysis, along with a kNN model for the final fall estimation.