Development of vision–based SLAM: from traditional methods to multimodal fusion