Deep Learning-Driven Monitoring of Physiotherapy Steps Using UWB Radar Technology

The growing demand for efficient and accessible physiotherapy solutions has highlighted the need for innovative monitoring systems that ensure accuracy, privacy, and ease of use. Traditional human activity monitoring systems are predominantly based on camera-based setups, which pose several limitations, including poor performance in low light conditions, privacy concerns, and the high cost of implementation in daily life. This study presents a robust, contactless, and privacy-preserving recognition system to classify basic physiotherapy exercise steps, utilizing UWB radar and advanced deep learning (DL) techniques. Five activities were performed: Hamstring-Stretch, Hands-Stretch, Neck-Exercise, Legs-Exercise, and Basic Stepper Exercise. The radar collected data was represented as spectrograms, enabling a detailed analysis. To classify the activities, deep learning models such as ResNet50, VGG16, VGG19 and MobileNet were employed. The simulation results demonstrated excellent performance, with all models achieving a 100% classification accuracy on the provided dataset. Such systems may enable physiotherapists to remotely monitor and assess patients' progress, providing scalable and inclusive healthcare solutions.