Robust Cad Model Based Visual Tracking For 3d Microassembly Using Image Space Potentials

A CAD model based visual tracking system for the flexible automation of 3D microassembly processes is presented. The system achieves six degrees-of-freedom tracking of MEMS components in real-time (30 Hz). The experimental results indicate a 1sigma precision of 1.0mum and a positioning accuracy of 2%. A major source of errors in model based tracking is due to ambiguous object configurations with closely spaced edges. We also present a real-time image space potential method that utilizes the readily available model information to reduce these errors and improve tracking robustness.