A novel multi-modal rehabilitation monitoring over human motion intention recognition
Saleha Kamal, Mohammed S. Alshehri, Yahya Alqahtani, Abdulmonem Alshahrani, Nouf Abdullah Almujally, Ahmad Jalal, Hui Liu
Abstract
Human Motion Intention Recognition (HMIR) plays a vital role in advancing medical rehabilitation and assistive technologies by enabling the early detection of pain-indicative actions such as sneezing, coughing, or back discomfort. However, existing systems struggle with recognizing such subtle movements due to complex postural variations and environmental noise. This paper presents a novel multi-modal framework that integrates RGB and depth data to extract high-resolution spatial-temporal and anatomical features for accurate HMIR. Our method combines kinetic energy, optical flow, angular geometry, and depth-based features (e.g., 2.5D point clouds and random occupancy patterns) to represent full-body dynamics robustly. Stochastic Gradient Descent (SGD) is employed to optimize the feature space, and a deep neuro-fuzzy classifier is proposed to balance interpretability and predictive accuracy. Evaluated on three benchmark datasets-NTU RGB + D 120, PKUMMD, and UWA3DII-our model achieves classification accuracies of 94.50%, 91.23%, and 88.60% respectively, significantly outperforming state-of-the-art methods. This research lays the groundwork for future real-time HMIR systems in smart rehabilitation and medical monitoring applications.