Generation of geometric Digital Twins of existing buildings relies on point cloud datasets and is still a manual-intensive and time-consuming process. This paper identifies the most frequent object types in buildings, analyses how current commercial software and state-of-the-art research methods to generate these objects. We summarise the main advantages of these methods and highlight limitations that limit these methods from broader adoption by the industry. Later, we identify the open challenges and discuss future directions to enable automating geometric Digital Twin generation.