VIMANCO: Vision Manipulation of Non-Cooperative Objects

摘要: This paper presents the work performed in context of VIMANCO on-going project. It has objective improving autonomy, safety and robustness robotics system using vision. Vision is certainly most adequate exteroceptive sensor to deal with complex varying environments for manipulation tasks non cooperative objects. The approach we propose based on an up-to-date recognition 3D tracking method that features many advantages respect other approaches. First all, it allows determine if a known object visible only one image. also compute its pose track real time along image sequence acquired by camera, even presence lighting conditions, partial occlusions, aspects changes. proposed been achieved combining efficient low level processing step, statistical techniques take into account potential outliers, formulation registration step as closed loop minimization scheme. valid camera observes object, but can be applied multi-cameras system. Finally, this provides all necessary data objects general formalism visual servoing, which control scheme expressed either image, or 3D, both spaces simultaneously. whatever vision configuration (one several cameras) robot arms (eye-in-hand eye-to-hand systems).