Improving the robustness of image-based tracking to control 3D robot motions

Abstract

This paper reports new techniques to render real-time image-based tracking methods more robust to enable the control of robots in an arbitrary 3D indoor environment. The methods enable robust feature tracking by integrating several cues within tracking windows. Integration is achieved using EPIC, a method of edge-projected integration of cues at field rate using a common Pentium PC. The thorough analysis of the control loop indicates this need to operate at field rate to obtain optimal dynamic performance. The demonstration verifies this and shows the application of following in 6D a grey part over grey metal background using common room lighting.