Feature and viewpoint selection for industrial car assembly

Abstract. Quality assurance programs of today’s car manufacturers show increasing demand for automated visual inspection tasks. A typical example is just-in-time checking of assemblies along production lines. Since high throughput must be achieved, object recognition and pose estimation heavily rely on offline preprocessing stages of available CAD data. In this paper, we propose a complete, universal framework for CAD model feature extraction and entropy index based viewpoint selection that is developed in cooperation with a major german car manufacturer.