Design and Implementation of a Pose Estimation System Based on Visual Fiducial Features and Multiple Cameras

In this paper, a pose estimation system based on visual fiducial features with multi-cameras is proposed. The system aims to provide ground truth data for mobile robot indoor localization experiments. Most of the hardware components used in this system are available off the shelf, and the idea is to develop a cost-effective localization system for mobile robot tests. The embedded vision system computes fiducial feature detecting algorithm and the pose estimation results are sent through pre-synchronized distributed modules to the server. The server then retrieves the object pose by using the proposed observation decision algorithm. A glitch elimination method is applied to deal with the problems of none observation of all cameras or when the current pose data is far from the previous ones. Practical experiments on a mobile robot have been setup to verify the proposed localization system and the validation result of 6 cm accuracy is achieved.