Set-up for assessment of localisation accuracy ofStructure from Motion algorithms

This paper discusses the need for a reliable experimental set-up providing a consistent benchmark to facilitate the assessment of the accuracy of localisation in varying implementations of Structure from Motion algorithms. The physical set-up involves a KUKA LWR4+ arm robot and a USB camera, both connected to a personal computer. A key step in determining the trajectory of the camera is finding its pose relative the robot arm's end effector. Two methods are introduced in this paper to address this problem. These methods are subjected to simulated and real-world tests. One method clearly outperforms the other when dealing with noisy, real-world extrinsic camera calibration at the cost of vastly increased computational complexity.