A Neural Network Model for a View Independent Extraction of Reach-to-Grasp Action Features