A Powered Prosthetic Hand With Vision System for Enhancing the Anthropopathic Grasp

The anthropomorphic grasping capability of prosthetic hands is critical for enhancing user experience and functional efficiency. Existing prosthetic hands relying on brain-computer interfaces (BCI) and electromyography (EMG) face limitations in achieving natural grasping due to insufficient gesture adaptability and intent recognition. While vision systems enhance object perception, they lack dynamic human-like gesture control during grasping. To address these challenges, we propose a vision-powered prosthetic hand system that integrates two innovations. Spatial Geometry-based Gesture Mapping (SG-GM) dynamically models finger joint angles as polynomial functions of hand-object distance, derived from geometric features of human grasping sequences. These functions enable continuous anthropomorphic gesture transitions, mimicking natural hand movements. Motion Trajectory Regression-based Grasping Intent Estimation (MTR-GIE) predicts user intent in multi-object environments by regressing wrist trajectories and spatially segmenting candidate objects. Experiments with eight daily objects demonstrated high anthropomorphism (similarity coefficient ${R}^{{2}}=0.911$ , root mean squared error $\textit {RMSE}=2.47 {^{\circ}}$ ), rapid execution ( $3.07\pm 0.41$ s), and robust success rates (95.43% single-object; 88.75% multi-object). The MTR-GIE achieved 94.35% intent estimation accuracy under varying object spacing. This work pioneers vision-driven dynamic gesture synthesis for prosthetics, eliminating dependency on invasive sensors and advancing real-world usability.