it seams that objects in /recognized_object_array have the same frame_id as an input RGB image even if we do not use RGB image. SO we can use only registered images (aligned depth and rgb), otherwise the transformation between RGB and depth topics is not taken into account