Surface Feature Identification Using a Vision-Based Manifold for Robotic Demanufacturing

Abstract

Robotic demanufacturing requires efficient operation on complex product surfaces from a diverse product stream. A new vision-based approach for surface identification is proposed. Vision-based information defines a surface manifold utilizing structured lighting. The surface manifold model is integrated with a surface estimation model to relate information back to the robot work space. A technique for developing surface patches related to surface objects detected in machine vision images is shown. Case study results from a scrap keyboard are presented and discussed. It is shown that the method iteratively generates a complete surface patch for any viewable surface object.