Exploiting Contextual Objects and Relations for 3D Visual Grounding