Dumbbell Calibration for a Multi-Camera Tracking System

Camera calibration is a preliminary and key work for building up a multi-camera visual tracking system. The purpose of calibration is to obtain transformations between several camera frames, so that each camera knows "where" the other cameras are with respect to itself. In this paper, we describe and implement a technique to determine the transformation of a two-camera tracking system by using a dumbbell as the calibration object. The dumbbell prop is novel and easy to be handled. By putting it within the common view field of two cameras, we firstly extract conies of dumbbell balls from raw images by color separation and ellipse fitting. The intrinsic and extrinsic parameters of each camera are recovered from the conies using linear approaches. Then, we present a method to calculate camera coordinates transformation based on the extrinsic parameters and the multiple view geometries. Experimental results are carried to further demonstrate the practicality of the proposed methods.