Head Gesture Recognition for Tetraplegia Patients Based on Head Orientation Feature with Polling Algorithm

Tetraplegia is a condition where all four limbs are paralyzed. It therefore makes it difficult for individuals to perform daily activities. This research presents a head gesture recognition system for tetraplegia patients based on head orientation features with a polling algorithm to enable Tetraplegia to perform normal activities through handsfree wheelchair control. The proposed model addresses head orientation in continuous gesture recognition by implementing a polling algorithm that eliminates inconsistent gesture sequences and strengthens predictions. Three deep learning models are evaluated: Proposed One-Dimensional Convolutional Neural Network (CNN1D), Long Short-Term Memory (LSTM), and Tab Transformer using head orientation features extracted from pose estimation in the authors’ own dataset. The experimental results show that the Proposed CNN1D achieves superior performance with an accuracy of 0.9844, precision of 0.9844, recall of 0.9854, and F1-score of 0.9844. The system achieves perfect classification accuracy for important navigation commands (Left, Right, Forward, and Stop) based on head orientation patterns. This head gesture recognition approach with polling algorithm offers a reliable solution to improve the independence and quality of life of tetraplegia patients through intuitive head-based wheelchair control.