Next:ORB: Object Recognition for Up:Mobile Robotics Previous:Defining Islands of Reliability

Geon Recognition using a Mobile Robot Vision System

R.K.H. Ng, M.D. Levine

Geons are simple 3-D geometric primitives proposed by Biederman for object representation. The idea is to segment objects into an arrangement of simple geometric shapes. Objects represented by geons can be identified more easily since the many-to-one mapping of object recognition is reduced to the recognition of a small number of geons. In this project, a vision system is implemented for the search and recognition of geons. The vision system is composed of a color camera and a laser range finder, and it is mounted on a pan-tilt unit. The entire setup is mounted on a mobile robot. The task of the system can be divided into two phases: the first phase is focus of attention. Our focus of attention algorithm is based on color, and the color camera searches for objects with that color. Upon detection of a geon, the mobile robot moves towards it in preparation for the next phase. The second phase is object recognition. With the robot in the proximity of the geon, the laser range finder is used to obtain range data of the geon for recognition. Since only a small number of geons (N=7) is considered in this project, very sparse range data is sufficient for successful recognition.

Annual Report

Fri Nov 26 23:00:32 GMT 1999