Autonomous Shape and Best Grip Determination for Upper Limb Prosthetics

Abstract

The aim of our research is to develop a prosthetic hand that is as natural and easy to use as a person’s organic extremity, without the need for invasive surgical procedures, an autonomous hand capable of determining the most suitable grasping pattern for griping an object and executing that best grasping pattern with minimum human effort. By using machine vision techniques a system is developed for fully automatic object recognition and reconstruction of 3D objects from multiple images taken from single camera embedded in the palm of the hand. We assume that the objects or scenes are rigid. A camera Matrix is associated for each image, which is parameterized by rotation, translation and focal length. For Feature matching between all images we use Speeded-Up Robust Features (SURF) and by using the RANSAC algorithm noisy matches are eliminated and find those matches that are consistent with the fundamental matrix. Objects are recognized as subsets of matching images. From 3D reconstruction we can estimate the perimeters i.e. length, width and Height of detected object and by using thresh holding function that is defined by different experimentation on human hand we execute the best grasping pattern for the griping the detected object.