Cognitive Template-Clustering Improved LineMod for Efficient Multi-object Pose Estimation