Object Extraction using Segmentation
This project explores a method for extracting unknown objects from piles using computer vision techniques. By utilizing Graph Segmentation combined with the Lucas-Kanade algorithm, the goal is to identify and grasp an unknown target item in a cluttered environment. The project acknowledges challenges such as slow processing speeds of graph segmentation and limitations of 2D image analysis. Future work aims to explore faster segmentation algorithms and to address the complexities of grasping 3D objects to improve extraction efficiency.
Object Extraction using Segmentation
E N D
Presentation Transcript
Object Extraction using Segmentation ECE 847 Final Project Bryan Willimon
Overview • Background of Project • Main Idea and Focus • Drawbacks and Obstacles • Results • Conclusion/Future Work
Background • Current Research Project • Using computer vision to grab an unknown object from within a pile given a target item • Unknown = no previous knowledge of object • Working with PUMA arm in EIB • Previous projects have achieved similar results but with known surroundings and known objects (STAIR)
Main Idea • Extract an unknown object using various types of segmentation working together • Graph Segmentation and Lucas-Kanade • Graph Segmentation breaks image into many regions • Largest region (not touching a border) is determined to be object on top
Main Idea (cont.) • Determine centroid of the object to be the point of grasping (for 2D image) • Use Lucas-Kanade to give feature points in the whole image • Only track the feature points on current object • Continue tracking until all points are gone and repeat process until target is found
Drawbacks and Obstacles • Drawbacks • Graph Segmentation is slow • Only using 2D image for grasping • Obstacles • Find and use a faster algorithm • Grasping a 3D object requires 3D modeling and/or motion • Also check if any other objects will be damaged in any way once current item is being moved
Conclusion/Future Work • Results provide a way to extract an unknown object from a pile • Explore other segmentation algorithms and find something faster • Using color and/or clustering • Work on grasping 3D objects