Vision-Based Robot Bin Picking: Recognition and Localization Of Multiple Objects

Abstract

Bin Picking is one of the most difficult tasks for a robot to perfonn I specially for unorganized parts. Machine vision is a vital tool that enables I robots to perfonn this important task. This thesis introduces two vision based systems which give a robot the ability to recognize and localize automatically unorganized parts from a pile. The object recognition I technique in the first system is based on shape contour and region I features. These derived features are invariant with respect to scaling, I rotation, translation, and affine transfom1ation of the objects. The extracted features are used for training both Self Organizing Map (SOM) I and Multi Layer Perceptron (MLP) neural networks for classification I purpose. The object recognition technique in the second system is based

I on the important object features. Important object features are obtained

I in two steps: firstly; by segmenting the object boundary at multiple scales through the use of its Iterative curvature scale space (ICSS) and secondly; I by concentrating on each scale separately in order to search for groups of I segments which distinguish an object from others. These groups of I segments are; then, used to build a model database through the use of artificial neural networks (ANNs).The developed vision systems are

implemented using Matlab software. The efficiency of the proposed