Vision-Based Mobile Robot Pose Tracking in RoboCup Environment

Self-localization is a fundamental and essential task for middle-sized mobile robots competition in RoboCup environment. A vision-based pose tracking approach for robot soccer is proposed in this paper firstly. In order to evaluate fitness between the prior and the estimated pose, an objective function is also formulated in the self-localization method so that the incoming odometry data and monocular vision feature can fuse effectively. According to a set of points given in the prior global map, mobile robot uses the position and orientation estimated from odometry to generate corresponding map points in robot's view, which are projected to the image plane through camera model to form the predicted scene for feature matching. After the points' searching perpendicular to the tangent of the projected curve, the correspondence between estimated and measured vision features is established and further Newton iterative algorithm is applied to estimate posterior robot pose. Experiment result implemented in real mobile robot platform and error analysis show the method's validity and practicability