Robust mark-detection for the realization of a robot gripping action

Autonomous gripping of (known) objects is one of the main requirements to the functionality of rehabilitation robotic systems. To simplify the necessary process of object detection and localization, marks can be attached to the objects. This paper presents the robust detection of such marks in dynamic environments with changing illumination situation. For the realization the visual feedback information from the classification level is used to adapt the binary threshold for the segmentation process. The classification can be reduced to a search in a neighborhood-graph due to the structure of the used marks. A combination of the two principles enables the robust mark detection before and during the gripping process.