Color-Depth-based book segmentation in library scenario for service robots

This work presents scenario oriented Color-Depth integration framework for the purpose of robust object segmentation in a real workplace scenario. The workplace is the Library of Bremen University and the objects to be grasped are books located on a shelf. The proposed framework is responsible for segmenting the book to be grasped and extract the features needed for a successful grasping operation. The proposed Color-Depth integration utilizes both disparity image and color image in the segmentation process. It segments at first the book to be grasped from the disparity image. However, due to the low-textured scene, the corresponding disparity image has a poor quality which in turn reduces the accuracy of the segmented region. The proposed integration framework increases the segmentation accuracy by utilizing the disparity segmented region in defining a region of interest (ROI) in the color image and then extracts the exact boundary of the book based on color. The developed Color-Depth integration does not only deliver high accuracy grasping features; but ensures a less straining operation of the robot by the End-user as well.