FPGA Based Deep Learning CNN Model for Fall Detection
Shiva Siddharth Pulipaka, Keshav Sriram S, Ferents Koni Jiavana K · 2025
Falls are a major health concern, especially among the elderly, often leading to serious injuries and long-term complications. Early and accurate detection is crucial for timely intervention and improved safety. This project presents an innovative approach to fall detection by implementing a Convolutional Neural Network (CNN) directly on an FPGA platform, enabling efficient on-device training without reliance on external computing resources. By leveraging the parallel processing capabilities of an FPGA, the system processes sensor data from accelerometers and gyroscopes in real-time, distinguishing between fall and non-fall events with high accuracy. Unlike traditional machine learning models that depend on cloud-based processing, this FPGA-based approach brings inference and training directly onto the hardware, eliminating communication delays and enhancing system reliability. This localized processing ensures adaptability across different environments, from healthcare monitoring systems to assistive technologies in smart homes. Furthermore, this project explores an emerging paradigm—training deep learning models directly on FPGA hardware, an area that remains largely unexplored. By integrating machine learning and hardware in a unified framework, this work demonstrates the feasibility and effectiveness of FPGA-driven deep learning for specialized applications, paving the way for more independent and adaptive AI systems.