Classification algorithm for wheelchair operation in real conditions using POF-based pressure sensors

A real-time implementation of a classification system is proposed here to operate an electric-powered wheelchair. The proposed framework innovatively integrates Polymeric Optical Fiber (POF)-based pressure sensors, positioned within a neck pillow, to effectively capture critical user-specific pressure data emanating from the user's neck region. The system considered the mechanical variations presented in the dataset construction owing to the movement generated by the wheelchair. To mitigate this inherent noise, an indispensable preprocessing stage is introduced, and a feature extraction stage is performed considering the variance, Root Mean Square (RMS), Crest Factor Margin (CMF), and Margin Factor (MF) in the time domain. This augmentation enhances the discriminative capacity of the feature representation and, consequently, enhances the classification performance. A fine k-Nearest Neighbors (kNN) was able to distinguish four neck movements: forward, right, left, and back, with performance metrics of: 80.39% (Accuracy), 87.09% (AUC), and 19.40% (False Positive Rate). This work contributes to the development of new strategies for the implementation of more robust power wheelchair controllers by integrating machine learning and optical fiber sensors, in order to assist people with disabilities.