Localization correction using predicted objects and road geometry for collective perception