Smart Insole with Deep Learning for Real-Time Classification of Gait Abnormalities

Abstract

This study introduces a novel, real-time gait abnormality detection system that leverages deep learning and an intelligent wearable insole to overcome limitations associated with conventional gait assessment methods. Gait abnormalities are prevalent among older adults and individuals with neurological or musculoskeletal conditions, yet existing diagnostic tools are often costly, non-portable, and confined to clinical settings. To address these challenges, we developed a compact, wireless insole system integrating force-sensitive resistors (FSRs), an inertial measurement unit (IMU), and an ESP32 microcontroller for continuous acquisition of plantar pressure and motion data. The acquired sensor data is wirelessly transmitted via Wi-Fi and analyzed by a ResNet-18 deep learning model to distinguish between normal and abnormal gait patterns with high accuracy. Real-time visualization and interactive feedback are provided through a Streamlit-based dashboard, while remote monitoring capabilities are enabled via the Blynk mobile application. Experimental results demonstrate a classification accuracy of 97.56%, with high precision and recall, validating the system’s robustness and effectiveness. The proposed solution offers a cost-effective, scalable, and clinically relevant platform for early detection of gait impairments and continuous home-based rehabilitation monitoring.