The integration of advanced control systems in prostheses necessitates the accurate identification of human locomotion activities, a task that can significantly benefit from EEG-based measurements combined with machine learning techniques. The main contribution of this study is the development of a novel framework for the recognition and classification of locomotion activities using electroencephalography (EEG) data by comparing the performance of different machine learning algorithms. Data of the lower limb movements during level ground walking as well as going up stairs, down stairs, up ramps, and down ramps were collected from 10 healthy volunteers. Time- and frequency-domain features were extracted by applying independent component analysis (ICA). Successively, they were used to train and test random forest and k-nearest neighbors (kNN) algorithms. For the classification, random forest revealed itself as the best-performing one, achieving an overall accuracy up to 92%. The findings of this study contribute to the field of assistive robotics by confirming that EEG-based measurements, when combined with appropriate machine learning models, can serve as robust inputs for prosthesis control systems.

The control of active prostheses and orthoses requires the precise classification of instantaneous human activity and the detection of specific events within each activity. Furthermore, such classification helps physiotherapists, orthopedists, and neurologists in kinetic/kinematic analyses of patients’ gaits. To address this need, we propose an innovative deep neural network (DNN)-based approach with a two-step hyperparameter optimization scheme for classifying human activity and gait events, specific for different motor activities, by using the ENABL3S dataset. The proposed architecture sets the baseline accuracy to 93% with a single hidden layer and offers further improvement by adding more layers; however, the corresponding number of input neurons remains a crucial hyperparameter. Our two-step hyperparameter-tuning strategy is employed which first searches for an appropriate number of hidden layers and then carefully modulates the number of neurons within these layers using 10-fold cross-validation. This multi-class classifier significantly outperforms prior machine learning algorithms for both activity and gait event recognition. Notably, our proposed scheme achieves impressive accuracy rates of 98.1% and 99.96% for human activity and gait events per activity, respectively, potentially leading to significant advancements in prosthetic/orthotic controls, patient care, and rehabilitation programs’ definition.

Accurate recording of hand joint kinematics is crucial for fields such as clinical diagnostics, ergonomics, and sign language recognition. It relies on the precise placement of markers and the number of markers used to capture joint movements effectively. This study presents a simplified marker-based biomechanical model of the hand designed to compute 22 joint angles using only 24 passive retroreflective markers, following a one-marker-per-segment approach. Kinematic data was collected from ten participants performing predefined physiological movements. The proposed biomechanical model was compared with three established biomechanical models employing one, two, and three-markers-per-segment on the fingers. To the best of the authors’ knowledge, this study is the first to directly compare these marker configurations simultaneously under identical experimental conditions. Range of motion (ROM) and joint variability were analyzed using repeated-measures ANOVA, while temporal trajectory differences were assessed via one-dimensional Statistical Parametric Mapping (SPM). Comparative analysis revealed no significant differences in ROM, variability, and joint angle trajectories across models, offering a robust and efficient approach for hand motion analysis suitable for functional, ergonomic, and AI-driven applications.