Revision history [back]

Tabletop object recognition could basically be divided into two kinds: textured object recognition and textureless object recognition. For textureless objects, people mainly use shape information extracted from point cloud (like that generated by Kinect). You can refer to Correspondence Grouping and Implicit Shape Model presented by PCL. There are both good docs and open source codes. For textured object recognition, people mainly use feature points like SIFT or SURF extracted from images. You can refer to MOPED which also contains paper and open source codes. In ICRA2011, there is a perception challenge origanized by WillowGarage. The champion of this challenge published a paper named "A Textured Object Recognition Pipeline for Color and Depth Image Data" in ICRA2012 which is also very enlightening.

Hope that helps.