3D Object Localization via Stereo Vision using an Omnidirectional and a Perspective Camera

This paper describes our stereo vision method which is combining an omnidirectional and a perspective camera. It was developed for our robot soccer team 1. RFC Stuttgart, which attends RoboCup competitions every year. The common approach to stereovision leads to high deviations from the real positions when it is not possible to synchronize the cameras for tracking moving objects and when the orientations of the cameras of a soccer playing robot can be decalibrated during a game. Therefore we introduce an object localization for the RoboCup scenario combining the most accurate position information from each camera system. Our method for a reliable three-dimensional position estimation can be used to track a flying ball after being kicked by a robot.