A robust and accurate color-based global vision recognition of highly dynamic objects in real time

Abstract

A color-based global vision system used to recognize 23 highly dynamic objects of interest using a firewire camera mounted at about 2.7m above a 2.8m × 2.3m field is presented. The developed algorithm used to identify the eleven home robots which was first shown to work on a webcam implementation by the authors' previous work is further evaluated in this paper using more testing experiments like identifying the home robots under intentionally varied illumination, shadow influence, object rotation, fast motion, collision and prolonged system operation. It is found that despite the 2.08 pixels per centimeter ratio versus the 3.8 of the webcam implementation, the eleven home robots were still accurately identified in all of these experiments while also identifying the ball and the eleven opponent robots.